Datasets:

bzantium

/

LongPerplexity

like 0

in recent years , a large body of evidence suggests that the evolution and properties of active galactic nuclei ( agn ) are tightly linked not only to the properties of their hosting galaxies , but also to the environment that these host galaxies inhabit . the most clear example of this perhaps comes from radio - loud agn , which have long been known to be primarily hosted by giant , massive , elliptical galaxies , which are predominantly found in very dense environments ( @xcite ; @xcite ; @xcite ; @xcite ) . in general , x - ray agn have also been found to be strongly clustered ( @xcite ; @xcite ; @xcite ) , though x - ray agn out to @xmath12 with harder x - ray spectra , e.g. , type-2 , or obscured x - ray agn , are preferentially found in underdense regions ( @xcite ) . large redshift surveys such as the sloan digital sky survey ( sdss , @xcite ) and the 2df qso redshift survey ( @xcite ) have enabled detailed studies of optical quasars , and have shown that their clustering was larger in the past in such a way that optically selected quasars seem to be hosted by halos of roughly constant mass , a few times @xmath13 , out to @xmath14 . the advent of the _ spitzer space telescope _ opened a new , mid - infrared ( mid - ir ) window to agn populations , providing samples that are relatively insensitive to the dust extinction that affects quasar surveys in the optical , ultraviolet ( uv ) and soft x - ray ( @xmath15 kev ) bands . @xcite developed a simple selection technique based on irac colors that identifies luminous agn essentially independent of their obscuration , and thus is particularly useful for identifying the dominant population of obscured agn that were largely missed in previous surveys ( see also @xcite ; @xcite ) . however , it is the recent launch of the _ wide - field infrared survey explorer _ ( _ wise _ ; @xcite ) that has made it possible to efficiently and robustly recover agn over the entire sky , including both unobscured and obscured sources . the most widely accepted idea about the physical origin of obscuration is the presence of a thick dust torus that , when viewed sideways , blocks the central part of the agn and hides many of the quasar - like features observed in unobscured agn ( @xcite ; @xcite ) . the first indirect evidence in favor of a torus was the detection of polarized broad emission lines , characteristic of unobscured agn , in a fraction of well known obscured agn due to the scattering toward the line of sight by free electrons just above ( or below ) the torus ( see @xcite ) . as an alternative to orientation - driven or torus models of agn obscuration , it is also plausible that at least part of obscuration could be caused by the interstellar medium ( ism ) of the host galaxy or by larger , @xmath0kpc - scale clouds of cool dust ( e.g. , @xcite ) . specifically , galaxy formation simulations by @xcite predict enhanced agn activity after galaxy mergers , which is initially obscured by kpc - scale dust clouds but is later laid bare as agn feedback pushes out the obscuring material . a basic prediction of the orientation - driven agn unification models is that similarly selected agn should populate similar environments . while some differences are clearly evident based on intrinsic agn luminosity or radio - loudness ( e.g. , @xcite ; @xcite ) , the expectation is that obscured ( or type-2 ) and unobscured ( or type-1 ) agn of similar luminosity and radio power should reside in similar environments . however , relatively little is known about the clustering of obscured agn , particularly those identified at mid - ir wavelengths . @xcite studied the spatial clustering of x - ray agn at @xmath12 , finding no significant difference in clustering strength between obscured and unobscured x - ray selected agn . similarly , from a matched sample of powerful radio - loud agn at @xmath16 , @xcite found that radio - loud quasars ( e.g. , unobscured radio - loud agn ) reside in similar environments to high - redshift radio galaxies ( e.g. , obscured radio - loud agn ) . in contrast , @xcite analyzed a sample of 806 _ spitzer _ mid - ir - selected quasars at @xmath17 in the botes field . they found marginal ( @xmath18 ) evidence that obscured quasars have a larger bias and populate more massive dark matter halos . these studies , while powerful due to the availability of spectroscopic redshifts and/or a large number of photometric bands , suffer the typical limitations of deep pencil - beam surveys , providing samples of a few hundred objects at most . in this paper we adopt a complementary approach by combining the _ wise _ and sdss data sets over thousands of square degrees . we select agn based on the _ wise _ 3.4 @xmath2 m ( @xmath19 ) and 4.6 @xmath2 m ( @xmath20 ) bands using selection criteria recently developed and demonstrated by @xcite and @xcite . to quantify the clustering , we undertake a correlation analysis , which is arguably the most powerful method for studying the distribution of galaxies . the angular correlation function measures the projected clustering of galaxies by comparing the distribution of galaxy pairs relative to that of a random distribution . while a less direct probe than the spatial correlation function , @xmath21 , the angular correlation function is a powerful approach as it can be applied to wide - area surveys and large samples of galaxies , overcoming the limitations of small number statistics and cosmic variance . in this work we focus on the angular correlation of agn . adopting a preliminary estimate of the redshift distribution of _ wise_-selected agn , we derive the absolute bias and estimate the typical mass of the dark matter halos that host them . this paper is organized as follows . in 2 we describe the surveys used in this work . in 3 we describe mid - ir selection of agn using the _ wise _ survey and detail the colors , morphologies , and redshift distribution of such sources . section 4 presents the angular clustering measurements , 5 presents the results and conclusions , and 6 summarizes these results and discusses the implications of this work . throughout the paper we assume a flat concordance @xmath22cdm cosmology , with @xmath23 , @xmath24 , and @xmath25 . unless otherwise noted , all magnitudes in this paper refer to the vega system . agn_v1(0.3 ) agnnopt_v1(0.3 ) the _ wise _ satellite mapped the full sky in four bands centered at 3.4 , 4.6 , 12 and 22 @xmath2 m ( bands @xmath19 , @xmath20 , @xmath26 and @xmath27 , respectively ) , achieving 5-@xmath28 point source sensitivities better than 0.08 , 0.11 , 1 and 6 mjy , respectively . every part of the sky has been observed typically @xmath29 times , except near the ecliptic poles where the coverage is much higher . astrometric precision is better than 015 for high signal - to - noise ( snr ) sources ( @xcite ) and the angular resolution is 61 , 64 , 65 and 12@xmath30 for bands ranging from 3.4 to 22 @xmath2 m . this paper is based on data from the _ wise _ all - sky release , which comprises images and four - band photometry for over 563 million sources , and has been publicly available since march 2012 . an object is included in this catalog if it : ( 1 ) is detected with snr@xmath315 in at least one of the four bands ; ( 2 ) can be measured well in at least five frames ; and ( 3 ) is not flagged as a spurious artifact in at least one band . we refer the reader to the _ wise_all - sky release explanatory supplement for further details ( @xcite ) . the sdss ( @xcite ; @xcite ) is a five - band photometric ( @xmath32 bands ) and spectroscopic survey that has mapped a quarter of the sky , providing photometry , spectra and redshifts for about a million galaxies and quasars , and photometry for many more . the imaging reaches 50% completeness at @xmath33 ( @xcite ) . the sdss pipeline calculates several kinds of magnitudes . in this work we have adopted the model magnitudes ( _ modelmag _ ) , which perform well for both bright and faint sources and provide unbiased galaxy colors . magnitudes are corrected for galactic reddening using the dust maps of @xcite . when appropriate , sdss magnitudes ( nearly in the ab system ) are converted into the vega system using @xmath34 , where @xmath35 is estimated by projecting model stellar spectra into the sdss @xmath36-band filter ( for details , see kcorrect software ) . in addition , sdss asinh scale magnitudes are converted into pogson logarithmic scale magnitudes ( see sdss website for further details ) . mid - ir selection of agn relies on distinguishing the characteristic rising power - law agn spectrum from the black body spectrum of stellar populations , which peak at rest - frame 1.6 @xmath2 m . this means that agn tend to be redder than normal galaxies in the mid - ir . this was initially shown in _ data where simple irac - band color cuts isolate agn from other galaxy populations at @xmath37 ( e.g. , @xcite ; @xcite ) . more recently , the _ wise _ survey has proven very efficient in detecting agn using just the two shorter ( and more sensitive ) bands at 3.4 @xmath2 m and 4.6 @xmath2 m ( @xcite ; @xcite ) . using empirical agn and galaxy spectral templates , @xcite showed that even pure agn present typically red @xmath38 colors out to @xmath39 for reasonable values of dilution by the host galaxy light ( e.g. , see fig . 1 of @xcite ) . heavily extincted agn are of course even redder in @xmath38 . this contrasts with the bluer @xmath38 colors of : ( 1 ) galactic stars , as only brown dwarfs with spectral types cooler than t5 have @xmath40 ( @xcite ) ; and ( 2 ) normal galaxies out to @xmath41 . thus , the primary contaminants to the red _ wise _ color selections will be the coolest brown dwarfs , which are quite rare on the sky , and galaxies at @xmath42 , which are effectively eliminated by our brightness cut , @xmath43 . using _ wise _ data over the area covered by the cosmos survey , @xcite demonstrated that a simple mid - ir color criterion is extremely robust at selecting agn candidates . selecting sources with @xmath40 above the 10-@xmath28 flux limit of 0.16 mjy at 4.6 @xmath2 m ( @xmath43 , vega ) identifies a large population of agn that is @xmath095% reliable and nearly 80% complete with respect to the _ spitzer _ agn selection of @xcite . these criteria identify 62 agn per deg@xmath44 , as compared to the @xmath45 quasars per deg@xmath44 identified by the optical sdss quasar selection algorithm , which is sensitive to agn of similar intrinsic luminosity ( @xcite ) . we construct our agn sample by applying the same selection criteria over a much larger area covered by sdss . in our sample , we only allow sources whose @xmath19 and @xmath20 photometry is unaffected by diffraction spikes , scattered light , persistence or optical ghosts ( @xmath46 in both @xmath19 and @xmath20 ) . @xcite reports on _ wise _ selection of agn down to @xmath47 in the higher ecliptic latitude , and thus deeper botes field . we refer the reader to their work for a useful comparison of _ wise _ agn selection at various depths . we note that , ignoring @xmath38 color for the moment , typical @xmath48 galaxies can be observed by _ wise _ up to @xmath49 at a 5-@xmath28 sensitivity ( @xmath50 ; see fig . 6 of @xcite ) . with our conservative flux density cut , @xmath51 , only the brightest , several @xmath48 galaxies will be detected by _ wise _ at @xmath52 . using the selection criteria of @xmath53 and @xmath54 , we selected 176,467 _ wise _ agn candidates over an effective area of 3363 deg@xmath44 ( see [ sec : mask ] for details about the angular mask ) . the @xmath54 magnitude cut guarantees that 99.7% candidates are detected with snr@xmath55 and that 99.98% have snr@xmath56 , while the mean snr@xmath57 of the sample is @xmath45 . the 176,467 selected agn candidates are cross - matched with the sdss photometric catalog . using a matching radius of 15 , we find 152,672 ( 86.5% ) _ wise _ agn candidates with single optical matches , 6095 ( 3.5% ) sources with two or more sdss counterparts , and 17,700 ( 10.0% ) _ wise _ agn candidates without an optical source listed in the sdss database . the multiple optical matches are mostly due to spurious detections of large sources split into multiple components or , in a few cases , real interacting galaxy systems . these _ wise _ unresolved close galaxy pairs are on scales @xmath58 deg , well below the spatial scales relevant in this work . the clustering analysis of galaxies on such small angular scales is beyond of the scope of this paper , as it would require full knowledge of the deblending performance of the sdss and _ wise _ pipelines . therefore , we focus here on _ wise _ agn candidates with single or no optical counterparts . note that so far we have not applied any constraints on sdss magnitudes , so that among the 152,672 single _ wise_-sdss matches , about 5% are fainter than the @xmath59 50%-completeness limiting magnitude of sdss , but are nevertheless listed in the sdss catalog . to insure that the _ wise _ agn without sdss counterparts are all real sources and not artifacts , we have visually inspected the _ wise _ and sdss images of 1000 randomly selected objects . we did not find any artifacts from the inspection . in addition , figure [ fig : mask ] shows the equatorial coordinates of all _ wise _ agn considered in this study , as well as the _ wise _ agn lacking optical counterparts in the sdss database . in this latter case , we have closely inspected their spatial distribution projected on the sky . there are no obvious large scale patterns , suggesting that the lack of an optical identification is intrinsic to the sources , and not related to image artifacts , galactic objects , or large - scale extinction . finally , to further demonstrate that the _ wise _ agn selection is robustly identifying agn , we investigate the fraction of _ wise_-selected agn with x - ray counterparts in the 60 ks exposures of the _ xmm - newton _ wide - field ( @xmath60 deg@xmath44 ) survey of the cosmos field ( xmm - cosmos @xcite ; @xcite ) . we find that @xmath075% of _ wise_-selected agn are x - ray detected , with the remaining @xmath025% expected to be fainter and/or heavily obscured agn missed by the _ xmm - newton _ observations . indeed , deeper _ chandra _ observations of the central half of the cosmos field ( @xcite ) detect 87% of the _ wise_-selected agn . similar results were found previously in @xcite , though that work imposed an snr@xmath55 cut , as opposed to the flux density cut used here . ( @xmath61 on a side ) , sdss @xmath36 ( @xmath61 on a side ) , and _ hst f814w _ ( @xmath62 ; @xmath63 on a side ) . north is up , and east is to the left . the top row shows an example of a blue , or unobscured _ wise_-selected agn at @xmath64 . the middle row shows an example of an optically faint , red , or obscured _ wise_-selected agn at @xmath65 ; this source is x - ray detected and classified as a type-2 agn ( @xcite ) . the bottom row shows an example of the 10% of _ wise_-selected agn that are undetected by sdss . this source is detected by _ xmm - newton _ and has a photometric redshift of @xmath66 . see text for further details on the individual sources . ] as mid - ir observations are relatively insensitive to obscuration by dust and optical observations are significantly affected by dust extinction , type-2 , or obscured agn , can be isolated by comparing _ wise _ and sdss fluxes ( @xcite ; @xcite ) . @xcite applied a similar method in the botes field using irac 4.5 @xmath2 m and @xmath67-band photometry to differentiate obscured and unobscured agn . for the sake of completeness , we note , however , that there is no rigorous and unique definition to differentiate obscured and unobscured agn across all wavelengths . in this work , we divide the _ wise _ agn sample according to @xmath68 color . figure [ fig : rw2_w1w2 ] illustrates that agn show a bimodal color distribution that separates two populations of agn . those with colors redder than @xmath69 are , of course , optically faint ( or undetected in sdss ) , but nevertheless well detected at 4.6 @xmath2 m . we call these `` red agn '' , in contrast with `` blue agn '' that are relatively bright at both mid - ir and optical wavelengths . we fold agn lacking optical matches into the red agn sample . as shown in @xcite , the red population is more closely associated with type-2 agn , while the blue population is associated with type-1 agn , e.g. , agn presenting broad emission lines in optical spectroscopy . in total , about 60,000 sources are selected as red agn candidates , implying a type-1 fraction of roughly 55% , similar to the fraction found by @xcite for luminous agn with bolometric luminosities exceeding a few times 10@xmath70 erg s@xmath71 . in [ sec : zdist_sel ] we evaluate the model selection function of red and blue agn to test the reliability of the @xmath68 criteria to separate type-1 and type-2 agn . figure [ fig : w2magdist ] ( top panel ) shows the @xmath20 magnitude distribution of red and blue _ wise_-selected agn . although red agn seem slightly fainter at mid - ir wavelengths in general , both subsamples have similar distributions , suggesting there is no strong bias due to the @xmath68 color cut . the bottom panel shows the distribution of sdss @xmath36-band magnitudes ( corrected for galactic reddening , converted to vega and in the pogson scale ) . blue agn are considerable brighter , peaking at @xmath72 and falling steeply at @xmath73 . most red agn are fainter , peaking around @xmath74 and extending to fainter magnitudes , beyond the nominal sdss completeness limit . a considerable fraction ( 10% ) of _ wise _ agn candidates are simply undetected by sdss ; we indicate such sources with a single bin at @xmath75 . the botes field has considerably deeper @xmath67-band photometry available , reaching @xmath76 ( 5-@xmath28 , point source ; @xcite ) . there are 61 sdss - undetected , _ wise_-selected agn in botes . their @xmath67-band magnitude distribution peaks at @xmath77 ( gray line ) with all sources having optical counterparts . this again illustrates the optical faintness , but detectability of essentially all _ wise_-selected agn . finally , we note that the fraction of _ wise_-selected agn with x - ray counterparts in the _ xmm - newton _ wide - field at the 0.5@xmath7810 kev band is @xmath79 for blue agn ( @xmath80 ) , dropping to @xmath81 for red agn ( @xmath5 ) . these high detection rates further demonstrate the reliability of our sample . figure [ fig : agn_stamp ] shows the range of optical morphologies of _ wise_-selected agn . for three candidates in the cosmos field , we display @xmath61 on a side images in _ wise _ @xmath20 and sdss @xmath36-band , and @xmath82 on a side _ hubble space telescope _ ( _ hst _ ) images in the @xmath83 filter . the top row shows an example of a blue , or unobscured _ wise_-selected agn : wise j100025.25 + 015852.1 is an optically bright , optically unresolved sdss quasar ( @xmath84 ) at redshift @xmath64 . the middle row shows an example of an optically faint , or red , obscured _ wise_-selected agn : wise j100005.98 + 015453.1 is an optically faint source detected by sdss ( @xmath85 ) . @xcite report a redshift of @xmath65 for this x - ray detected , optically resolved source and classify it as type-2 agn based on its spectrum . the bottom row shows an example of the 10% of _ wise_-selected agn which are undetected by sdss : wise j100153.32 + 021928.3 is undetected by sdss ( @xmath86 ) , but is detected by both _ hst _ and _ xmm - newton_. the source has a photometric redshift of @xmath66 . optical morphologies offer an additional observable with which to investigate the _ wise _ agn selection . luminous , unobscured , or type-1 agn are typically unresolved at optical wavelengths , which was one of their foundational attributes that led to the name `` quasar '' , or quasi - stellar radio source . we have known for several decades now that only @xmath87 of quasars are radio - loud , with little variation in this fraction with either redshift or optical luminosity , at least at the high luminosity end ( e.g. , @xcite ) . similarly , mid - ir selection is showing that unresolved , unobscured quasars represent a minority population of luminous agn . indeed , using the sdss type flag to discriminate morphologies , we find only @xmath88 of the _ wise_-selected agn considered in this paper are classified as unresolved point sources . we use the cosmos field to characterize how morphology depends on optical - to - mid - ir color for _ wise_-selected agn . figure [ fig : morph_hist ] shows a histogram of the optical - to - mid - ir colors of the 82 _ wise_-selected agn with _ hst _ acs ( f814w ) imaging in cosmos to our @xmath51 depth , coded by optical morphology . fifteen of the sources are undetected by sdss in the @xmath36-band , and are simply plotted at @xmath89 ; all 15 of these sources are detected in the deeper _ hst _ @xmath83 imaging and are spatially resolved . indeed , of the 28 red agn candidates with @xmath5 , only 1 ( 4% ) is unresolved . this supports our expectation that red optical - to - mid - ir colors select a clean sample of obscured agn with little contamination from unobscured agn . of the 54 blue agn candidates , 35 ( 65% ) are unresolved . figure [ fig : morph_z_rw2 ] shows the @xmath68 color vs. redshift for these blue agn candidates , with symbols indicating their _ hst _ morphologies . as we will show in the next section , most of the resolved agn e.g. , likely obscured agn contaminating our blue agn selection are at lower redshift ( @xmath90 ) and , in fact , reside in the redder end of our blue agn selection . however , figure [ fig : morph_hist ] also clearly shows that it is not feasible to simply make a bluer @xmath68 cut to separate obscured ( e.g. , resolved ) and unobscured ( e.g. , unresolved ) agn . to characterize the host galaxies of _ wise_-selected candidates we performed more detailed visual classifications on the _ hst _ acs image cutouts ( independently by three of the four authors ; as we agreed for the majority of objects , we report the average here ) . for red agn , we find that 54% ( 15 ) are disk galaxies or interacting systems , 32% ( 9 ) are elliptical or point sources , and the remaining 14% ( 4 ) have uncertain morphology . this contrasts with blue agn , where we find that 20% ( 11 ) are disk galaxies and 80% ( 43 ) are either point sources ( most ) or ellipticals . these results are consistent with the work of @xcite , who studied the morphology of agn in cosmos selected at radio , x - ray and mid - ir wavelengths . that work found that the red mid - ir - selected agn consist of 63% disk galaxiess , 22% point sources / ellipticals , and 15% other morphology , while the blue agn consist of 15% disk galaxies and 85% point sources / ellipticals . the main conclusion we wish to draw here is that given its high fraction of disk galaxies , the red agn sample is not dominated by typical red sequence galaxies . in fact , the red agn have a higher fraction of disk galaxies than the blue agn . as further discussed in [ sec : hostgal ] , this suggests it is unlikely that a bias in host galaxy type ( favoring red agn in early - type hosts and blue agn in late - type hosts ) could have a large impact in the interpretation of the clustering results presented in [ sec : clust_results ] . given the difference in optical flux introduced by the @xmath68 cut , it is not unreasonable to expected differences in the redshift distribution of blue and red agn . in order to understand the redshift distribution and properties of _ wise _ agn candidates , we have matched our list to publicly available spectroscopy in the cosmos field as well as recent spectroscopic observations ( see @xcite for details about the compiled list of spectroscopic and photometric redshifts ) . figure [ fig : zhist ] shows the redshift distribution of the 112 _ wise_-selected agn available in cosmos , of which 88 have spectroscopic redshifts and 11 have photometric redshifts ( plus 13 objects with no redshift information available ) . the top panel highlights the blue agn ( @xmath80 ) , including five sources , plotted at @xmath91 , that lack both spectroscopic and photometric redshifts . the distribution peaks around @xmath92 and extends up to @xmath93 , with most of the sources at @xmath94 . there is an indication of a second smaller peak at @xmath95 , most probably ( as we will see later ) due to type-2 agn that enter into the redder part of our blue sample selection at low redshift . for reference , we also show the spectroscopic redshifts of 536 _ wise _ agn candidates within the botes field ( dashed histogram ) , obtained from the agn and galaxy evolution survey ( ages ; @xcite ) . this survey has different completeness levels for different galaxy samples ( @xmath96 for galaxies , @xmath97 for agn , but with varying priority levels based on their brightness at mid - ir , @xmath98 m , radio , and x - ray energies ) and therefore a complicated redshift selection function . however , considering the differences in target selection as compared to cosmos ( which essentially targeted every source to @xmath99 ) , the two distributions agree remarkably well . this suggests that both are not far from the true redshift distribution of _ wise_-selected agn with blue @xmath68 colors . the bottom panel in figure [ fig : zhist ] shows the corresponding distributions for _ wise_-selected agn with red @xmath68 colors , including eight sources lacking both spectroscopic and photometric redshifts plotted at @xmath100 . red agn candidates peak at lower redshift , around @xmath101 , and extend up to @xmath102 . again , the agreement with ages redshifts in the botes field is notable . to further understand the nature of the differences in redshift among the red and blue agn samples , we model their selection function by constructing mock objects using the agn and galaxy spectral energy distribution ( sed ) templates from @xcite . the parameter @xmath103 quantifies the fraction of the bolometric luminosity coming from the agn component ( see @xcite , 2013 for details ) . figure [ fig : ahat ] shows whether an object with a given host galaxy type ( e or i m ) , @xmath104 value , and reddening towards the accretion disk , parametrized by @xmath105 , would be targeted as a blue agn ( @xmath80 , blue region ) , as a red agn ( @xmath5 , red region ) , or as an inactive galaxy ( @xmath106 , white region ) . as expected , at low @xmath104 , most systems are characterized as normal galaxies . the panels at @xmath107 and @xmath108 highlight the extreme cases of a zero reddening or a heavily extincted agn ; the typical boundary between type-1 and type-2 agn corresponds to a reddening of @xmath109 ( see @xcite for details ) . the gray hatched area marks the region where _ wise _ is not sensitive due to the shallowness imposed by our @xmath43 flux density requirement , namely @xmath52 host - galaxy dominated objects . this figure shows that while it is very unlikely to misclassify a blue agn as a red one , the opposite happens for a significant fraction of parameter space , suggesting that our red agn selection constitutes a reliable yet incomplete type-2 agn sample , while our blue sample consists of a mixture of type-1 and type-2 agn . we also used the deep , multi - wavelength data available in the botes field to do detailed sed modeling of _ wise_-selected agn and explore how reddening relates to @xmath68 color for blue and red agn candidates as a function of redshift . this is shown in figure [ fig : reddz ] . the reddening parameter @xmath105 is derived by fitting the agn and galaxy sed templates of @xcite . as expected , red agn tend to show considerable reddening at all redshifts , with @xmath110 , while blue agn are mostly unreddened above @xmath95 . however , below this redshift , blue agn can sometimes show large reddening values , consistent with the idea that some of these objects might well be type-2 agn interlopers in the blue sample . as shown in the bottom panel of figure [ fig : reddz ] , the distributions of @xmath104 are strongly peaked toward high values , with most red agn above @xmath111 , and a minor fraction of blue agn with @xmath112 . this means that while the blue area in the bottom right panel of figure [ fig : ahat ] is large , only a minority of sources could be potentially biased due to the galaxy host type ( i.e. selected as blue agn due to the presence of i m galaxy host ) . a standard tool to measure galaxy clustering is the two - point angular correlation function , @xmath113 . it is defined as the probability that a given pair of galaxies separated by an angle @xmath114 on the sky are contained within a solid angle @xmath115 @xmath116 where @xmath117 is the mean number density of galaxies . in practice , @xmath113 is calculated by counting pairs of galaxies in annuli of different radii and comparing with the corresponding counts in a random sample of galaxies . to estimate @xmath113 we use the @xcite estimator , given by @xmath118 where @xmath119 , @xmath120 and @xmath121 are the normalized data - data , data - random and random - random pair counts , respectively . it is very important that the random sample has the same angular selection as the data pairs . for this purpose we constructed an angular mask using the software manglemolly / mangle/. ] that describes the survey geometry in terms of disjoint spherical polygons . this mask accounts for the holes caused by bad quality fields in the sdss survey , as well as the areas around bright stars selected from the tycho 2 catalog ( @xmath122 ) . in addition , we also remove the areas around large ( @xmath123 ) sources from the 2mass extended source catalog that in some cases appear decomposed into multiple sources in _ wise_. galactic absorption can have an impact in faint galaxy counts ( @xcite ) , so we mask out areas with @xmath124 . finally , we avoid regions contaminated by the moon and limit the sample to the rectangular area bounded by @xmath125 r.a . @xmath126 and @xmath127 dec . @xmath128 ( j2000 ) . these rather conservative limits avoid both the galactic plane , where contamination by stars could present an issue , and the ecliptic pole , where the sensitivity of _ wise _ improves substantially due to denser coverage and lower zodiacal background . our selected area has a typical _ wise _ coverage of @xmath129 frames per bandpass . at small scales , the clustering of an extragalactic source population is difficult to predict due to processes such as merging and interactions . however , at larger scales ( e.g. , @xmath31 1 - 2 @xmath130 mpc ) , galaxy interactions have little impact and the galaxy correlation function follows that of the dark matter halos . at any redshift , massive halos cluster more strongly than less massive halos . given an average redshift , this , in turn , allows one to estimate the typical mass of dark matter halos in which objects reside by estimating their absolute bias , i.e. , their observed clustering level with respect to that of the underlying dark matter . we compare our @xmath113 measurements to the predictions of the standard cold dark matter ( cdm ) model in the linear perturbation theory of structure growth along with the non - linear correction . to compute the dark matter angular two - point correlation function , @xmath131 , we use the non - linear fitting function of @xcite for the cdm power spectrum projected onto the same agn redshift distribution . the bias factor is simply defined as @xmath132 . in general , the bias is a function of scale , but under the assumption that galaxies cluster in a similar manner as dark matter , the bias factor is nearly scale - independent . this is particularly valid in the linear regime ( i.e. , large scales ; see @xcite ) . we limit the bias and the corresponding fits from @xmath133 to @xmath134 , which corresponds to scales of roughly @xmath135 kpc to @xmath136 mpc at @xmath49 . using an ellipsoidal collapse model , @xcite related the halo bias factor to its mass and calibrated a fitting relation for a large library of cosmological @xmath137-body simulations : @xmath138{+ \sqrt a b ( a\nu^2)^{1-c } } - \frac{(a\nu^2)^c}{(a\nu^2)^c + b(1-c)(1-c/2 ) } ] \\ \end{multlined}\ ] ] where @xmath139 , @xmath140 , @xmath141 and @xmath142 is the critical density ratio for collapse given by @xcite as @xmath143 , and @xmath144 . @xmath145 depends on the cosmology as @xmath146 , and @xmath147 is defined as @xmath148 , where @xmath149 is the linear growth factor , here approximated analytically using the formulae by @xcite . the rms fluctuation of the density field is given by @xmath150dk,\ ] ] where the term in brackets represents the spherical top - hat window function ( @xcite ) and the radius @xmath36 is related to the enclosed halo mass @xmath151 as @xmath152{\frac{3m_h}{4\pi\rho_0}},\ ] ] where @xmath153 is the present mean density of the universe , given by @xmath154 . the linear power spectrum of density fluctuations , @xmath155 with @xmath117=1 ( the harrison - zeldovich case ) , is constructed using the fitting formula of @xcite for the transfer function @xmath156 and normalized with the adopted value of @xmath157 for @xmath158 mpc . we begin our analysis by exploring the angular clustering for the full sample of agn selected by _ wise_. figure [ fig : acf_w1w2 ] shows that agn with @xmath40 present an angular correlation similar to that of optical quasars selected from sdss by @xcite using a photometric kernel density estimation ( kde ) technique ( @xcite ) . a power - law fit of the form @xmath159 gives a value of @xmath160 within the range @xmath161 $ ] ( @xmath162 mpc at @xmath163 ) . @xcite find @xmath164 for optically selected quasars at @xmath165 , while @xcite find a slightly shallower value for 2qz quasars , @xmath166 when averaged over scales of 1 - 100 @xmath130 mpc and after correcting for redshift distortions . these slight differences are not entirely surprising considering the very different agn selection criteria and the fact that the _ wise _ agn sample includes both obscured and unobscured agn , while the optical quasar samples are entirely comprised of broad - lined , type-1 agn . furthermore , the clustering of quasars might not be properly represented by a single power law . at scales below @xmath167 , we find that redder agn have slightly higher angular clustering . this is interesting considering that this scale ( @xmath168 mpc ) marks the transition between the 1-halo and 2-halo terms , which , in the framework of halo clustering models , arises from galaxy pairs located in either the same or in two different halos , respectively . as shown at the bottom panel of figure [ fig : acf_w1w2 ] , the absolute bias for _ wise_-selected agn with @xmath40 with respect to the underlying dark matter distribution is @xmath169 , as compared to @xmath170 for _ wise_-selected agn with redder mid - ir colors , @xmath171 . taking into account the caveat that different redshift and luminosity distributions can possibly bias the results , the simplest interpretation is that redder agn are hosted by slightly more massive dark matter halos . for type-1 agn at @xmath172 previous work has shown that the clustering depends only weakly on redshift , luminosity or color ( @xcite ; @xcite ) . however , for type-2 agn this is mostly unknown and our @xmath173 sample is expected to be a mixture of both type-1 and type-2 agn . finally , we note that our results compare well to the bias estimates obtained by @xcite for optical quasars over a similarly broad redshift range centered at @xmath174 . we explore now the angular clustering of _ wise_-selected red and blue agn . the corresponding correlation functions , shown in figure [ fig : acf_rw2 ] , display very different amplitudes . for a fixed slope @xmath175 ( that of the entire agn sample ) , blue , or unobscured agn ( e.g. , @xmath80 ) have @xmath176 , while red , or obscured agn ( e.g. , @xmath5 ) have @xmath177 , i.e. , a factor of @xmath178 larger . the bottom panel shows that the mean bias of obscured sources relative to the dark matter is @xmath7 , as compared to @xmath10 for unobscured agn . for reference , we also show in figure [ fig : acf_rw2 ] the angular clustering of type-1 quasars ( grey shaded area ) from @xcite , which is in broad agreement with our estimation for the blue agn sample . part of the difference in clustering strength could , in principle , be due to the obscured sources having a different selection , that is , since obscured sources are required to be optically faint ( or undetected ) , they could reside at slightly higher redshifts than their unobscured cousins . on the contrary , spectroscopy from both cosmos and botes demonstrates that red agn tend to be at slightly lower redshift ( figure [ fig : zhist ] ) . the caveat is that there is a @xmath179 incompleteness in the two spectroscopic samples and the sample sizes are not extremely large . while directly comparing the full and complete redshift distributions for blue and red _ wise_-selected agn would be ideal to check whether their different clustering strengths are related to different redshift distributions , we can nevertheless minimize it by selecting agn limited only to those with @xmath180 counterparts in sdss . the amplitudes of the best - fit power - law become @xmath181 for obscured agn , compared to @xmath182 for unobscured agn , for a fixed slope @xmath175 . figure [ fig : acf_rw2_r23 ] shows the corresponding angular auto - correlations , illustrating once again that , while noisier , obscured agn have a correlation amplitude a factor of @xmath60 larger than the unobscured sources . given the difference in amplitude between the correlation functions of red and blue agn , we investigate how this reflects into the masses of dark matter halos that host them . using the prescriptions described in [ sec : halomass ] , we estimate that blue agn at @xmath183 are hosted in halos of characteristic mass @xmath184 . this is in excellent agreement with the halo mass of @xmath185 reported by @xcite for sdss optical quasars at @xmath186 . @xcite finds a similar value of @xmath187 for 2qz quasars hosts . in figure [ fig : biasz ] we show the bias as function of redshift for the best - fit model ( thick blue line ) , along with models of constant halo mass ( dashed black lines ) for reference . we find that the halos of our red agn have a much larger characteristic mass of @xmath188 , i.e. over a factor of 10 larger than for blue agn . we discuss the physical implications of this result in the following section . we also note that @xcite reports a very similar mass of @xmath189 for their obscured quasar sample , though their value of @xmath190 for unobscured quasars is slightly larger than both our value and literature results for optically selected unobscured quasars . to understand the clustering result of our red and blue samples , we study the host galaxies of _ wise_-selected agn using sed fitting and the morphology classifications discussed in [ sec : morph ] . this is important because the observed difference in clustering might , in principle , be attributed to a selection effect that biases our red agn sample to being hosted by early - type galaxies and our blue agn to being hosted by late - type galaxies . such a difference might either be the result of an intrinsic difference between the populations or due to a selection function bias . in particular , figure [ fig : ahat ] suggests that our red agn sample could be biased against type-2 quasars in starburst galaxies if mid - ir selected agn had a large spread over values . first , we use the sed fitting of _ wise_-selected agn candidates in the botes field with the templates of @xcite to analyze the distribution of host light coming from each of the three galaxy templates ( e , sbc and i m ) . from figure [ fig : reddz ] , the blue agn sample contains some sources with considerable dust obscuration ( i.e. , well above the @xmath109 boundary line ) . for these misclassified type-2 agn , we find that 38% , 36% and 26% of their host galaxy emission is dominated by the e , sbc and i m templates , respectively , where we define an object to be dominated when @xmath3150% of the host luminosity is coming from a given template . these similar proportions suggest that this selection bias , if present , is not preferentially missing i m galaxies , and therefore it is unlikely to be significantly affecting our results . for completeness , we note that it is difficult to determine the dominant host for the majority of blue candidates which have @xmath191 and thus agn emission dominates the optical sed , making host sed fitting challenging . the same analysis for red agn candidates in botes gives 63% , 13% and 24% of the cases dominated by e , sbc and i m templates , respectively . we note here that an e galaxy sed template does not directly imply that the galaxy morphology is elliptical . as discussed below , sources with the e - type template also include spiral galaxies with prominent bulges . overall , there is also a large fraction of objects ( 37% if we combine sbc and i m ) dominated by late - type templates , suggesting that while early - type hosts are common , the red agn population is hosted by a mixture of galaxy types . second , we recall the morphological results from [ sec : morph ] . there we found that 54% ( 20% ) of red ( blue ) agn have disks , while 32% ( 80% ) are elliptical or point sources . this means that the red agn sample is not dominated by typical red sequence galaxies , and that blue agn have , in fact , a lower fraction of late - types than red agn . these findings strongly suggest that it is unlikely that the clustering results are driven by host galaxy differences or selection bias . instead , the observed differences in their correlation functions actually represents an intrinsic difference in the environments of type-1 and type-2 agn . furthermore , as discussed in [ sec : discussion ] , the increase in clustering while moving from blue cloud to red sequence galaxies is markedly smaller than the difference between blue and red agn . as the amplitude of @xmath113 will certainly change depending on the location and shape of the redshift distribution in any observed sample of galaxies , it is important to assess how sensitive our bias estimates are to changes in the redshift distribution . for this purpose we fit different model distributions for red and blue agn redshifts and investigate how much the inferred absolute bias would change by systematically varying the distribution parameters with respect to the best - fitting case . to describe our redshifts , we adopt the gamma statistical distribution of parameters @xmath192 ( shape ) and @xmath193 ( scale ) , although this choice is not critical ( e.g. using gaussians will lead to variations of the same order of magnitude ) . for red agn the best - fit parameters to observed cosmos redshifts are @xmath194 and @xmath195 ( @xmath196 ) , while for blue agn we obtain @xmath197 and @xmath198 ( @xmath199 ) . for our red agn sample , in figure [ fig : varbias_red ] we reproduce the different model distributions with varying @xmath192 ( top row ) and @xmath193 ( bottom row ) , and the corresponding effect on the bias , always normalized to the best - fitting case highlighted in red . changing @xmath192 or @xmath193 within the range shown means that the bias could change by @xmath020% at most . note that to estimate @xmath200 we assume a spatial correlation length @xmath201 that is constant in redshift . figure [ fig : varbias_blue ] shows the same analysis applied to our blue agn sample . the result is a similar variation of @xmath025% in bias . in addition , we also test the effect of fitting a double gamma distribution ( dashed red line ) instead of a single one . as expected , adding a second peak to the model naturally adjusts much better to the observed redshifts , yet the derived bias would decrease by only @xmath06% . to further assess these conclusions , we repeated the test by directly convolving the cosmos redshift distributions of blue and red agn with a gaussian kernel of increasing width . once again the bias changes by about 30% for any reasonable broadening . note that adopting the botes redshift distributions as reference instead of cosmos shifts these percentages by @xmath2024% , and so does not qualitatively alter our conclusions . finally , a bias uncertainty of @xmath203% in the _ least favorable _ scenario e.g. , the blue agn bias is underestimated by 20% , while the red agn bias is overestimated by 20% still translates into systematic halo mass estimates a factor of @xmath204 larger for red agn than for blue agn . in this work we have taken advantage of recently released data from _ wise _ to construct a large sample of @xmath205 mid - ir - selected agn candidates with the main purpose of analyzing their angular clustering properties . the selection is highly reliable ( @xmath3 ) , as demonstrated in @xcite and @xcite , as well as by the high rate of x - ray detections ( 3.1 ) . the median redshift of the sample is @xmath206 based on relatively complete spectroscopy in the cosmos and botes fields . by considering their optical counterparts from sdss , we distinguish those _ wise_-selected agn that are optically faint , and thus have red optical - to - mid - ir colors and are inferred to be heavily obscured agn , from those that are optically bright , and thus have blue optical - to - mid - ir colors and are inferred to be unobscured agn . we find that , as a whole , the _ wise_-selected agn population presents a similar clustering strength to optically selected quasars at comparable redshifts , with a slightly higher absolute bias with respect to the dark matter distribution for redder @xmath38 subsamples . we find that the red agn show a notably larger bias level than that of blue agn , with @xmath7 versus @xmath207 respectively . using a significantly smaller sample of few hundred sources over a much smaller area , @xcite reported a similar absolute bias of @xmath208 for obscured _ spitzer_-selected agn . our absolute bias estimates suggest that red agn ( _ i.e _ obscured sources ) are hosted by massive dark matter halos of @xmath209 , well above the halos of mass @xmath11 that harbor blue agn ( unobscured sources ) . it is possible to interpret these results in a scenario where , at least during a brief phase before the dust is removed and the agn gets `` exposed '' , the black hole mass is a factor of few below the @xmath210 relation of active galaxies . for our sample , from the sed fits of _ wise _ agn in botes we find that both red and blue agn have similar distributions of agn bolometric luminosity , with a nearly identical mean of @xmath211 . this suggests that the black hole masses of our red and blue agn do not differ much , and it is unlikely that their relative eddington ratio is much different from unity . moreover , their high luminosities are indicative of quasar - like accretion happening in both samples and we know that _ wise _ agn selection tends to pick up agn radiating at large fractions of their eddington limits ( @xcite ) . since we find that obscured sources are hosted by more massive halos , then this means that , at least during a period of time , the black hole mass growth lags behind that of the hosting halos and hence the black holes in obscured agn are temporarily `` undermassive '' until they reach their final mass . this is not entirely surprising , as , for example , @xcite find that submillimeter galaxies at @xmath212 host black holes @xmath29 times smaller than the expected for radio galaxies and quasars . the basis of such a lag argument for agn has been proposed before in the literature ( e.g. , theoretically by @xcite , and coupled to clustering by @xcite ) . @xcite suggests that the effect of rayleigh - taylor instabilities on the eddington outflows that regulate black hole growth leads to black holes masses in active galaxies a factor a few below the m-@xmath28 relation , assuming an observed agn phase represents a black hole growth phase . thus , agn should recurrently reach eddington - order luminosities in order to grow fast enough to reach the masses specified by the @xcite relation . one popular scenario for obscured quasars is that they represent an early evolutionary stage of rapid black hole growth just before the emergence of an unobscured , optical quasar . @xcite pose that in the final stages of coalescence of the galaxies , massive inflows supply large amounts of gas , increasing the gas density around nuclear regions and feeding the black hole that : ( 1 ) initially is obscured , ( 2 ) grows accordingly at high eddington rates , and ( 3 ) is small compared to the spheroid in formation . then , any possible link between ( final ) black hole mass and halo mass ( e.g. @xcite ) would predict dm halos of the same mass for obscured and unobscured sources . but , if obscured agn are an early stage where black holes are acquiring their final mass , then they would inhabit more massive halos when compared to unobscured quasars of the same black hole mass . the clustering of red and blue galaxies has been studied in detail by @xcite using deep2 survey data . they find that at @xmath213 the bias of blue cloud galaxies is in the range of @xmath214 . moving towards the red sequence , the bias increases in about 30% , so the measured bias of red galaxies is @xmath215 . our blue agn candidates have a bias that is at least comparable to luminous blue deep2 galaxies or to their less luminous red galaxies , but the bias of _ wise _ red agn is much larger than that of red galaxies , and is well more than 30% greater than that of blue agn . this suggests that our red agn candidates do not seem to cluster like typical red sequence galaxies at these redshifts and that the change in clustering is intrinsic to the two agn types . these results are in broad agreement with @xcite , who finds that mid - ir - selected agn tend to reside in galaxies slightly bluer than the green valley ; and with @xcite , who conclude that the x - ray and mid - ir agn are not dominated by early - type galaxies , but by later - type galaxies with disks . finally , our results allow to test a basic assumption of the agn unification paradigm . a fundamental prediction of orientation - driven agn unification models is that the angular clustering strength should be similar for obscured and unobscured agn . we find evidence that obscured agn are , in fact , more clustered than unobscured sources , which would appear to make simple orientation , or obscuring torus scenarios much less plausible , or , at least , not the full story for agn obscuration . alternatively , it would be interesting to compare our results against predictions of more physical agn models , where , for example , the sublimation radius changes with agn power or the covering fraction depends on other physical parameters . our data set does not allow us to test these models in detail , but larger samples with improved redshifts and spectral coverage will make it possible . our primary result is a significant difference in the clustering of optically bright ( blue ) and optically faint ( red ) mid - ir - selected agn , implying that , on average , obscured and unobscured agn reside in different halos . this surprising result has important implications for agn unification , the role of agn feedback in galaxy formation , the lifetime of quasars , and for understanding the sources responsible for the cosmic x - ray background and their cosmic evolution . we thank a. myers for useful replies to questions and extend our gratitude to the _ wise _ extragalactic science team for its continuous support and interesting discussions over the years . we also gratefully acknowlege the anonymous referee and numerous colleagues including g. hasinger and r. hickox who provided insightful comments that have improved our discussion . this publication makes use of data products from the _ wide - field infrared survey explorer _ , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration . funding for the sdss and sdss - ii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , the us department of energy , the national aeronautics and space administration , the japanese monbukagakusho , the max planck society and the higher education funding council for england . the sdss website is http://www.sdss.org/. r.j.a . was supported in part by an appointment to the nasa postdoctoral program at the jet propulsion laboratory , administered by oak ridge associated universities through a contract with nasa . was also supported in part by gemini grant number 32120009 . we also thank the nasa astrophysics data analysis program ( adap ) for its support . 99 abazajian k. n. , adelman - mccarthy j. k. , agueros m. a. , et al . , 2009 , apjs , 182 , 543 alexander d.m . et al . , 2008 , aj , 135 , 1968 antonucci r.r.j . , 1993 , ara&a , 31 , 473 assef r. j. et al . , 2010 , apj , 713 , 970 assef r. j. et al . , 2013 , apj , 772 , 26 best p.n . , kauffmann g. , heckman t.m . , brinchmann j. , charlot s. , ivezi . , white s.d.m , 2005 , mnras , 362 , 25 brusa m. , civano f. , comastri a. et al . , 2010 , apj , 716 , 348 carroll s. m. , press w. h. , turner e. l. , 1992 , ara&a , 30 , 499 coil a.l . et al . , 2008 , apj , 672 , 153 coil a.l . et al . , 2009 , apj , 701 , 1484 croom s.m . et al . , 2004 , mnras , 349 , 1397 croom s.m . et al . , 2005 , mnras , 356 , 415 cutri r. et al . , 2011 , wise explanatory supplement . donley j.l . , koekemoer a.m. , brusa m. et al . , 2012 , apj , 748 , 143 donoso e. , li cheng , kauffmann g. , best p.n . , heckman , t.m . , 2010 , mnras , 407 , 107 eisenstein , d. j. , hu , w. 1998 , apj , 496 , 605 elvis m. , civano f. , vignali c. et al . , 2009 , apjs , 184 , 158 falder j. t. , stevens j. a. , jarvis m. j. , 2010 , mnras , 405 , 347 ferrarese l. , 2002 , apj , 578 , 90 georgakakis a. et al . , 2007 , apj , 660 , 15 gilli r. et al . , 2005 , a&a , 430 , 811 gilli r. et al . , 2009 , a&a , 494 , 33 griffith r.l . , stern d. , 2010 , aj , 140 , 533 hasinger g. et al . , 2007 , apjs , 172 , 29 heisler c.a . , lumsden s.l . , bailey j.a . 1997 , nature , 385 , 700 hickox r.c . et al . , 2007 , apj , 671 , 1365 hickox r.c . et al . , 2009 , apj , 696 , 891 hickox r.c . , 2011 , aj , 731 , 117 hopkins p. et al . , 2008 , apjs , 175 , 356 jannuzi b. et al . , 2010 , bulletin of the american astronomical society meeting 215 , vol 42 . , p.513 jarrett t. et al . , 2011 , apj , 735 , 112 king a.r . , 2010 , mnras , 408 , 95 kirkpatrick j.d . et al . , 2011 , apjs , 197 , 19 kochanek c. et al . , 2012 , apjs , 200 , 8 lacy m. et al . , 2004 , apjs , 154 , 166 landy s.d . , szalay a.s . , 1993 , apj , 412 , 64 limber d.n . , 1953 , apj , 117 , 134 martnez - sansigre a. et al . , 2009 , apj , 706 , 184 matthews t.a . , morgan w.w . , schmidt m.,1964 , apj , 140 , 35 myers a.d . et al . , 2006 , apj , 638 , 622 myers a.d . , brunner r.j . , nichol r.c . , richards g.t . , schneider d.p . , bahcall n.a . , 2007 , apj , 658 , 85 navarro j. f. , frenk c. s. , white s. d. m. , 1997 , apj , 490 , 493 peacock j.a . , dodds s.j . , 1996 , mnras , 280 , 19 peebles p. j. e. , 1980 , the large - scale structure of the universe . princeton university press , princeton richards g. t. et al . , 2002 , aj , 124 , 1 richards g. t. et al . , 2004 , apjs , 155 , 257 ross n.p . et al . , 2009 , apj , 697 , 1634 schlegel d. j. , finkbeiner d. p. , davis m. , 1998 , aj , 500 , 525 shen y. et al.,2001 , aj , 697 , 1656 sheth , r. k. , mo , h. j. , tormen , g. , 2001 , mnras , 323 , 1 soltan a. , 1982 , mnras , 200 , 115 stern d. , djorgovski s.g . , perley r. , de carvalho r. , wall j. et al . , 2000 , aj , 119 , 1526 stern d. et al . , 2005 , apj , 631 , 163 stern d. et al . , 2012 , aj , 753 , 30 stoughton c. et al . , 2002 , aj , 123 , 485 strauss m. a. et al . , 2002 , aj , 124 , 1810 tasse c. , rttgering h. , best p.n . , 2011 , a&a , 525 , 127 trump j.r . et al . , 2007 , apjs , 172 , 383 urry c.m . , padovani p. , 1995 , pasp , 107 , 803 verde l. et al . , 2002 , mnras , 335 , 432 wright e. l. et al . , 2010 , aj , 140 , 1868 wylezalek d. , galametz a. , stern d. et al . , 2013 , apj , 769 , 79 yan l. et al . , 2013 , aj , 145 , 55 york d. g. et al . , 2000 , aj , 120 , 1579

this work is concerned with catalytic branching processes that model the dynamics of catalyst - reactant populations in which the activity level of the reactant depends on the amount of catalyst present . branching processes in catalytic environments have been studied extensively and are motivated , for instance , by biochemical reaction networks ; see @xcite and references therein . a typical setting consists of populations of multiple types such that the rate of growth ( depletion ) of one population type is directly affected by population sizes of other types . the simplest such model consists of a continuous time countable state branching process describing the evolution of the catalyst population and a second branching process for which the branching rate is proportional to the total mass of the catalyst population , modeling the evolution of reactant particles . such processes were introduced in @xcite in the setting of super - brownian motions ; see @xcite . for classical catalyst - reactant branching processes , the catalyst population dies out with positive probability and subsequent to the catalyst extinction , the reactant population stays unchanged , and therefore the population dynamics are modeled until the time the catalyst becomes extinct . in this work , we consider a setting where the catalyst population is maintained above a positive threshold through a specific form of controlled immigration . branching process models with immigration have also been well studied in literature ; see @xcite and references therein . however , typical mechanisms that have been considered correspond to adding an independent poisson component ; see , for example , @xcite . here , instead , we consider a model where immigration takes place only when the population drops below a certain threshold . roughly speaking , we consider a sequence @xmath0 of continuous time branching processes , where @xmath1 starts with @xmath2 particles . when the population drops below @xmath2 , it is instantaneously restored to the level @xmath2 . there are many settings where controlled immigration models of the above form arise naturally . one class of examples arises from predator - prey models in ecology , where one may be concerned with the restoration of populations that are close to extinction by reintroducing species when they fall below a certain threshold . in our work , the motivation for the study of such controlled immigration models comes from problems in chemical reaction networks where one wants to keep the levels of certain types of molecules above a threshold in order to maintain a desired level of production ( or inhibition ) of other chemical species in the network . such questions are of interest in the study of control and regulation of chemical reaction networks . a control action where one minimally adjusts the levels of one chemical type to keep it above a fixed threshold is one of the simplest regulatory mechanisms , and the goal of this research is to study system behavior under such mechanisms with the long - term objective of designing optimal control policies . the specific goal of the current work is to derive simpler approximate and reduced models , through the theory of diffusion approximations and stochastic averaging techniques , that are more tractable for simulation and mathematical treatment than the original branching process models . in order to keep the presentation simple , we consider the setting of one catalyst and one reactant . however , similar limit theorems can be obtained for a more general chemical reaction network in which the levels of some of the chemical species are regulated in a suitable manner . settings where some of the chemical species act as inhibitors rather than catalysts are also of interest and can be studied using similar techniques . these extensions will be pursued elsewhere . our main goal is to establishes diffusion approximations for such regulated catalyst - reactant systems under suitable scalings . we consider two different scaling regimes ; in the first setting the catalyst and reactant evolve on `` comparable timescales , '' while in the second setting the catalyst evolves `` much faster '' than the reactant . in the former setting , the limit model is described through a coupled system of reflected stochastic differential equations with reflection in the space @xmath3 . the precise result ( theorem [ thmcrdiffusion ] ) is stated in section [ secdiffusionlimit ] . such limit theorems are of interest for various analytic and computational reasons . it is simpler to simulate ( reflected ) diffusions than branching processes , particularly for large network settings . analytic properties such as hitting time probabilities and steady state behavior are more easily analyzed for the diffusion models than for their branching process counterparts . in general , such diffusion limits give parsimonious model representations and provide useful qualitative insight to the underlying stochastic phenomena . for the second scaling regime , where the catalyst evolution is much faster , we establish a stochastic averaging limit theorem . a key ingredient here is an ergodicity result , which says that under a suitable `` criticality from below '' assumption on the catalyst dynamics , the limiting catalyst reflected diffusion admits a unique stationary distribution , which takes an explicit form ( proposition [ propstationarydistrofx ] ) . characterization of the invariant distribution is based on a variant of echeverria s criterion for constrained markov processes @xcite . next , by constructing suitable _ uniform _ lyapunov functions , we show that the stationary distribution of the scaled catalyst branching process converges to that of the catalyst diffusion ( theorem [ thmtightnessandconvergenceofnu^n1 ] ) . these results are then used to establish a stochastic averaging principle that governs the dynamics of the reactant population in the fast catalyst limit . proofs proceed by developing suitable moment estimates that are uniform in time and the scaling parameter and by using characterization results for probability laws of reflected diffusions through certain constrained martingale problems @xcite . the limit evolution of the reactant population is given through an autonomous one - dimensional sde with coefficients that depend on the stationary distribution of a reflected diffusion in @xmath4 . such model reductions are important in that they not only help in better understanding the dynamics of the system but also help in reducing computational costs in simulations . indeed , since in the model considered here the invariant distribution is explicit , the coefficients in the one - dimensional averaged diffusion model are easily computed , and consequently this model is significantly easier to analyze and simulate than the original two - dimensional model . we refer the reader to @xcite and references therein for similar results in the setting of ( nonregulated ) chemical reaction networks . it will be of interest to see if similar model reductions can be obtained for general multi - dimensional regulated chemical - reaction networks . key mathematical challenges will be to identify suitable conditions for ergodicity of multi - dimensional reflected diffusions in polyhedral domains that arise from the regulated part of the network , and to develop uniform ( in time and the scaling parameter ) moment estimates for such multi - dimensional constrained diffusions . we consider two different formulations of models with multiple time scales . in theorem [ thmfastcatalystweaklimit ] we consider the setting where both catalyst and reactant processes are described through ( reflected ) diffusions and the time scale parameter appears in the coefficients of the catalyst evolution equation . an important step here is to argue that the generator of the two - dimensional catalyst - reactant reflected diffusion is suitably close to the generator of the one - dimensional averaged diffusion , for large values of the scaling parameter . bounds on the exponential moments of the catalyst process , obtained in lemma [ lemunifmombdx ] , play a key role in this argument . the second formulation is considered in theorem [ fastcatalystbpaveraging ] . here , both catalyst and reactant populations evolve according to near critical countable state branching processes , and the branching rate in the catalyst dynamics is of higher order than that for the reactant process . in this setting one encounters the additional difficulty of showing that the steady state distributions of the scaled catalyst branching process , for large values of the scaling parameter , are suitably close to the stationary distribution of the limiting catalyst reflected diffusion . the approach taken here is based on characterizing the limit points of a certain sequence of random measures on the path space of the catalyst process and the associated reflection process , as time and the scaling parameter together approach infinity . the model considered in this work does not incorporate any spatial dynamics of the two chemical species . as noted earlier in the , in the unregulated setting , dawson and fleischmann @xcite considered catalyst - reactant systems , with chemical species moving continuously in a spatial domain , given in terms of super - brownian motions . it will be of interest to develop analogous continuous spatial models for the regulated catalyst - reactant systems of the form considered in the current work . this question will be explored in a future work . the paper is organized as follows . we begin in section [ secdiffusionlimit ] by presenting the basic limit theorem in the setting of `` comparable time scales . '' section [ chasymptoticbehaviorofthecatalystpopulation ] studies the time asymptotic behavior of the catalyst process under a suitable _ criticality from below _ assumption . section [ secdiffusionlimitofthereactantunderfastcatalystdynamics1 ] presents our main results for the multiple time scale setting . section [ secauxresults ] collects some auxiliary estimates that are needed in our proofs . section [ sec6 ] proves theorem [ thmcrdiffusion ] , and section [ secproofofasymptoticbehaviorofthecatalystpopulation ] is devoted to the proofs of proposition [ propstationarydistrofx ] and theorem [ thmtightnessandconvergenceofnu^n1 ] . finally , in section [ secdiffusionlimitofthereactantunderfastcatalystdynamics ] we present proofs of stochastic averaging principles stated in section [ secdiffusionlimitofthereactantunderfastcatalystdynamics1 ] . the following notation will be used throughout this work . denote by @xmath5 the natural numbers , let @xmath6 , denote the set of integers by @xmath7 and let @xmath8 be the set of nonnegative real numbers . the state spaces of the scaled catalyst , reactant , and auxiliary processes , @xmath9 and @xmath10 , respectively , introduced below in ( [ eqhatw^n ] ) , are @xmath11 , @xmath12 and @xmath13 . let @xmath14 and @xmath15 . let @xmath16 denote the space of @xmath17-times continuously differentiable , real valued functions on @xmath18 , and denote by @xmath19 the space of @xmath16 functions with compact support . here , by a ( @xmath17-times ) differentiable function @xmath20 on a set @xmath21 we mean a function that can be extended to a ( @xmath17-times ) differentiable function @xmath22 on an open domain @xmath23 such that @xmath22 restricted to @xmath24 equals @xmath20 . given a metric space @xmath25 , the space of probability measures on @xmath25 will be denoted by @xmath26 , the borel @xmath27-field on @xmath25 by @xmath28 , and the space of real valued , bounded , measurable functions on @xmath25 by bm(@xmath25 ) . let @xmath29 and @xmath30 , where these @xmath24-spaces are endowed with the usual skorohod topology . let @xmath31 be the space of continuous functions from @xmath32 to @xmath32 endowed with the local uniform topology . we say a sequence @xmath33 of random variables with values in some polish space @xmath34 is tight if the corresponding probability laws are a tight sequence in @xmath35 . for a function @xmath36 , let the jump at time @xmath37 be defined as @xmath38 , @xmath39 , and @xmath40 . for a function @xmath41 and @xmath42 , let @xmath43 . for two semimartingales @xmath44 and @xmath45 , the quadratic covariation ( or bracket process ) and predictable ( or conditional ) quadratic covariation are denoted by @xmath46_t\}_{t\in{\mathbb{r}}_+}$ ] and @xmath47 , respectively ; their definition will be recalled in section [ secauxresults ] . consider a sequence of pairs of continuous time , countable state markov branching processes @xmath48 , where @xmath1 and @xmath49 represent the number of catalyst and reactant particles , respectively . the dynamics are described as follows . each of the @xmath50 particles alive at time @xmath37 has an exponentially distributed lifetime with parameter @xmath51 ( mean lifetime @xmath52 ) . when it dies , each such particle gives rise to a number of offspring , according to the offspring distribution @xmath53 . additionally , if the catalyst population drops below @xmath2 , it is instantaneously replenished back to the level @xmath2 ( _ controlled immigration _ ) . the branching rate of the reactant process @xmath49 is of the order of the current total mass of the catalyst population , that is , @xmath54 , and we denote the offspring distribution of @xmath49 by @xmath55 . a precise definition of the pair @xmath48 will be given below . we are interested in the study of asymptotic behavior of @xmath56 , under suitable scaling , as @xmath57 . to facilitate some weak convergence arguments , we will consider an auxiliary sequence of processes @xmath58 that `` shadow '' @xmath1 in the following manner . the process @xmath58 will be a @xmath7 valued pure jump process whose jump instances and sizes are the same as that of @xmath1 away from the boundary @xmath59 , whereas when @xmath1 is at the boundary , @xmath58 has a negative jump of size 1 whenever there is immigration of a catalyst particle into the system . this description is made precise through the infinitesimal generator given in ( [ eqgeneratorofwholeprocess ] ) . the process @xmath10 will not appear in the statements of the results ; nevertheless it plays an important role in our proofs . we now give a precise description of the various processes and the scaling that is considered . roughly speaking , time is accelerated by a factor of @xmath2 , and mass is scaled down by a factor of @xmath2 . define rcll processes @xmath60 and let @xmath61 , where @xmath62 is the initial number of catalyst and reactant particles and @xmath63 . then @xmath64 is characterized as the @xmath65 valued markov process with sample paths in @xmath66 , starting at @xmath67 , and having infinitesimal generator @xmath68 given as @xmath69 \mu_1^{(n)}(k ) \nonumber\\[-8pt]\\[-8pt ] & & { } + \lambda_2^{(n ) } n^2 x y \sum _ { k=0}^\infty \biggl[\phi \biggl(x , y+ \frac{k-1}{n},z \biggr)-\phi(\mathbf w ) \biggr]\mu_2^{(n)}(k),\nonumber\end{aligned}\ ] ] where @xmath70 and @xmath71 . from the definition of the generator we see that , for each @xmath72 , given @xmath73 , the process jumps to @xmath74 with rate @xmath75 and to @xmath76 with rate @xmath77 , except when @xmath78 and @xmath79 , in which case the latter jump is to @xmath80 with rate @xmath81 . this property of the generator at @xmath79 accounts for the instantaneous replenishment of the ( unscaled ) catalyst population to level @xmath2 , whenever the catalyst drops below @xmath2 . for @xmath82 , let @xmath83 we make the following basic assumption on the parameters of the branching rates and offspring distributions as well as on the initial configurations of the catalyst and reactant populations : [ conditioncr ] for @xmath82 and for @xmath84 , @xmath85 and @xmath86 . for @xmath82 , as @xmath57 , @xmath87 , @xmath88 and @xmath89 . for @xmath82 and for every @xmath90 , @xmath91 as @xmath57 , @xmath92 condition [ conditioncr ] and the form of the generator in ( [ eqgeneratorofwholeprocess ] ) ensure that the scaled catalyst and reactant processes transition on comparable time scales , namely @xmath93 . in order to state the limit theorem for @xmath94 , we need some notation and definitions associated with the one - dimensional skorohod map with reflection at 1 . let @xmath95 be defined as @xmath96 the function @xmath97 , known as skorohod map , can be characterized as follows ; see , for example , appendix b in @xcite and references therein : if @xmath98 are such that ( i ) @xmath99 , ( ii ) @xmath100 , ( iii ) @xmath101 , ( iv ) @xmath102 is nondecreasing , @xmath103 , and @xmath104 , then @xmath105 and @xmath106 . the process @xmath102 can be regarded as the reflection term that is applied to the original trajectory @xmath107 to produce a trajectory @xmath108 that is constrained to @xmath4 . from the definition of the skorohod map and using the triangle inequality , we get the following lipschitz property : for @xmath109 , @xmath110 the diffusion limit of @xmath111 will be the process @xmath112 , starting at @xmath113 , which is given through a system of stochastic integral equations as in the following proposition . [ propuniqunessofx , y ] let @xmath114 be a filtered probability space on which are given independent standard @xmath115 brownian motions @xmath116 and @xmath117 . let @xmath118 be square integrable @xmath119 measurable random variables with values in @xmath4 and @xmath32 , respectively . then the following system of stochastic integral equations has a unique strong solution : @xmath120 where @xmath97 is the skorohod map defined in ( [ eqskorohodmap ] ) . in the above proposition , by a strong solution of ( [ eqx])([eqeta ] ) , we mean an @xmath121-adapted continuous process @xmath122 with values in @xmath123 that satisfies ( [ eqx])([eqeta ] ) . the following is the main result of this section . [ thmcrdiffusion ] suppose condition [ conditioncr ] holds . the process @xmath124 converges weakly in @xmath125 to the process @xmath112 given in proposition [ propuniqunessofx , y ] with @xmath126 . proposition [ propuniqunessofx , y ] follows by standard arguments , so its proof is relegated to the . theorem [ thmcrdiffusion ] will be proved in section [ sec6 ] . stochastic averaging results in this work rely on understanding the time asymptotic behavior of the catalyst process . such behavior , of course , is also of independent interest . we begin with the following result on the stationary distribution of @xmath127 , where @xmath127 is the reflected diffusion from proposition [ propuniqunessofx , y ] , approximating the catalyst dynamics . the proof uses an extension of the echeverria criterion for stationary distributions of diffusions to the setting of constrained diffusions ; see section [ secproofstationarydistrofx ] . we will make the following additional assumption . recall the constants @xmath128 and @xmath129 introduced in condition [ conditioncr ] . [ conditionc1 ] for all @xmath84 , @xmath130 and @xmath131 . [ propstationarydistrofx ] suppose condition [ conditionc1 ] holds . the process @xmath127 defined through ( [ eqx ] ) has a unique stationary distribution , @xmath132 , which has density @xmath133 where @xmath134 . the following result shows that the time asymptotic behavior of the catalyst population is well approximated by that of its diffusion approximation given through ( [ eqx ] ) . we make the following additional assumption on the moment generating function of the offspring distribution , which will allow us to construct certain `` uniform lyapunov functions '' that play a key role in the analysis ; see theorem [ thmintegratingtillreturntime ] and the function @xmath135 defined in ( [ lyapfn ] ) . [ conditionmoment ] for some @xmath136 , @xmath137 [ thmtightnessandconvergenceofnu^n1 ] suppose conditions [ conditioncr ] , [ conditionc1 ] and [ conditionmoment ] hold . then , for each @xmath84 , the process @xmath138 has a unique stationary distribution @xmath139 , and the family @xmath140 is tight . as @xmath57 , @xmath139 converges weakly to @xmath132 . proposition [ propstationarydistrofx ] and theorem [ thmtightnessandconvergenceofnu^n1 ] will be proved in section [ secproofofasymptoticbehaviorofthecatalystpopulation ] . as noted in section [ secdiffusionlimit ] , the catalyst and reactant populations whose scaled evolution is described through ( [ eqgeneratorofwholeprocess ] ) transition on comparable time scales . in situations in which the catalyst evolves `` much faster '' than the reactant , one can hope to find a simplified model that captures the dynamics of the reactant population in a more economical fashion . one would expect that the reactant population can be approximated by a diffusion whose coefficients depend on the catalyst only through the catalyst s stationary distribution . indeed , we will show that the ( scaled ) reactant population can be approximated by the solution of @xmath141 where @xmath142 . such model reductions ( see @xcite and references therein for the setting of chemical reaction networks ) not only help in better understanding the dynamics of the system but also help in reducing computational costs in simulations . in this section we will consider such stochastic averaging results in two model settings . first , in section [ secstochaveragingindiffusionsetting ] , we consider the simpler setting where the population mass evolutions are described through ( reflected ) stochastic integral equations and a scaling parameter in the coefficients of the model distinguishes the time scales of the two processes . in section [ secstochasticaveragingforscaledbranchingprocesses ] we will consider a setting which captures the underlying physical dynamics more accurately in the sense that the mass processes are described in terms of continuous time branching processes , rather than diffusions.=1 in this section we consider the setting where the catalyst and reactant populations evolve according to ( reflected ) diffusions similar to @xmath127 and @xmath143 from proposition [ propuniqunessofx , y ] , but where the evolution of the catalyst is accelerated by a factor of @xmath144 such that @xmath145 as @xmath146 ( i.e. , drift and diffusion coefficients are scaled by @xmath144 ) . more precisely , we consider a system of catalyst and reactant populations that are given as solutions of the following system of stochastic integral equations : for @xmath42 , @xmath147 where @xmath148 , @xmath149 , @xmath150 , @xmath116 and @xmath117 are independent standard brownian motions , and @xmath97 is the skorohod map described above proposition [ propuniqunessofx , y ] . the following result says that if @xmath131 , then the reactant population process @xmath151 , which is given through a coupled two - dimensional system , can be well approximated by the one - dimensional diffusion @xmath152 in ( [ eqchecky ] ) , whose coefficients are given in terms of the stationary distribution of the catalyst process . [ thmfastcatalystweaklimit ] suppose condition [ conditionc1 ] holds . the process @xmath153 converges weakly in @xmath31 to the process @xmath152 . the proof of theorem [ thmfastcatalystweaklimit ] is given in section [ secdiffusionlimitofthereactantunderfastcatalystdynamics ] . we now consider stochastic averaging for the setting where the catalyst and reactant populations are described through branching processes . consider catalyst and reactant populations evolving according to the branching processes introduced in section [ secdiffusionlimit ] , but where the catalyst evolution is sped up by a factor of @xmath144 such that @xmath145 monotonically as @xmath146 . that is , we consider a sequence of catalyst populations @xmath154 , @xmath42 , where @xmath1 are the branching processes introduced in section [ secdiffusionlimit ] . the reactant population evolves according to a branching process , @xmath155 , whose branching rate , as before , is of the order of the current total mass of the catalyst population , @xmath156 . the infinitesimal generator @xmath157 of the scaled process @xmath158 is given as @xmath159\mu_1^{(n)}(k ) \nonumber\\[-8pt]\\[-8pt ] & & { } + \lambda_2^{(n ) } n^2 x y \sum _ { k=0}^\infty \biggl[\phi \biggl(x , y+\frac{k-1}{n } \biggr)-\phi(x , y ) \biggr]\mu_2^{(n)}(k),\nonumber\end{aligned}\ ] ] where @xmath160 and @xmath161 . we note that a key difference between the generators @xmath157 above and @xmath162 in ( [ eqgeneratorofwholeprocess ] ) is the extra factor of @xmath144 in the first term of ( [ eqcheckg^n ] ) , which says that , for large @xmath2 , the catalyst dynamics are much faster than that of the reactant . we will show in theorem [ fastcatalystbpaveraging ] that the reactant population process @xmath151 can be well approximated by the one - dimensional diffusion @xmath152 in ( [ eqchecky ] ) . once again , the result provides a model reduction that is potentially useful for simulations and also for a general qualitative understanding of reactant dynamics near criticality . [ fastcatalystbpaveraging ] suppose conditions [ conditioncr ] , [ conditionc1 ] and [ conditionmoment ] hold . then , as @xmath57 , @xmath163 converges weakly in @xmath164 to the process @xmath152 . we will prove the above theorem in section [ secdiffusionlimitofthereactantunderfastcatalystdynamics ] . in this section we collect several auxiliary results , which will be used in the proofs of our main results . recall that the quadratic covariation ( or bracket process ) of two semimartingales @xmath44 and @xmath45 is the process @xmath46_t\}_{t\in{\mathbb{r}}_+}$ ] defined by @xmath165_t:=\xi_t\zeta_t-\int _ 0^t \xi_{s- } \,d \zeta_s-\int_0^t \zeta_{s- } \,d\xi_s,\qquad t\geq0,\ ] ] where @xmath166 . the predictable quadratic covariation of @xmath44 and @xmath45 is the unique predictable process @xmath167 such that @xmath46_t-\langle\xi,\zeta\rangle_t\}_{t\in{\mathbb{r}}_+}$ ] is a local martingale . if @xmath168 , then @xmath169\equiv[\xi,\xi ] $ ] and @xmath170 are , respectively , the quadratic and predictable quadratic variation processes of @xmath44 . for @xmath171 , let @xmath172 , @xmath173 , and @xmath174 . note that for a locally bounded measurable function @xmath20 on @xmath18 @xmath175 is a local martingale with respect to the filtration @xmath176 . for the rest of the paper , we suppress the filtration , and simply refer to @xmath177 as a local martingale . let @xmath178 this process will play the role of the reflection term in the dynamics of the catalyst , arising from the controlled immigration . the following tightness result will be used in the weak convergence proofs . [ lemtighnessofcatalyst - reactant ] suppose conditions [ conditioncr ] and [ conditionc1 ] hold . then the family @xmath179 is tight in @xmath180 . if additionally condition [ conditionmoment ] holds , then the family @xmath181 is tight in @xmath182 . the proof of proposition [ lemtighnessofcatalyst - reactant ] will be based on the following results . lemma [ lemrepresentationofcrprocesses ] below gives some useful representations for the catalyst and reactant processes . lemmas [ lemevofsquaredsuprema][lemevofsquaredsupremaofshiftedprocesses ] and corollary [ corevofsquaredsupremaofshiftedhatx^n ] provide moment bounds that are useful for arguing tightness . proofs of these results are given in section [ secproofsofauxresults ] . [ lemrepresentationofcrprocesses ] suppose condition [ conditioncr ] holds . the process @xmath94 can be represented as @xmath183\\[-8pt ] & = & \gamma \biggl(\hat x_0^{(n)}+c_1^{(n ) } \lambda_1^{(n)}\int_0^\cdot \hat x_s^{(n)}\,ds+m_\cdot^{(n ) } ( \phi_1 ) \biggr ) ( t)\nonumber\end{aligned}\ ] ] and @xmath184 moreover , for @xmath185 , @xmath186\\[-8pt ] & \le&\lambda_1^{(n)}\alpha_1^{(n ) } \int_0^t\hat x_s^{(n)}\,ds\nonumber\end{aligned}\ ] ] and @xmath187 let @xmath188 then we have the following second moment estimate . [ lemevofsquaredsuprema ] suppose conditions [ conditioncr ] and hold . then there is a @xmath189 such that for all @xmath84 and @xmath190 , @xmath191 and for each @xmath192 , @xmath193 where @xmath194 . in order to study properties of invariant measures of @xmath195 , it will be convenient to allow the initial random variable @xmath196 to have an arbitrary distribution on @xmath197 . when @xmath198 has distribution @xmath199 on @xmath197 , we will denote the corresponding probability and expectation operator by @xmath200 and @xmath201 , respectively . if @xmath202 for some , we will instead write @xmath203 and @xmath204 , respectively . when considering an initial condition @xmath205 for @xmath195 , @xmath205 will always be in @xmath197 , although this will frequently be suppressed in the notation . the symbols @xmath206 and @xmath207 ( without any subscripts ) will correspond to the initial distribution as in condition [ conditioncr ] . [ lemexpmomentsonfinitetime ] suppose conditions [ conditioncr ] and , [ conditionc1 ] and [ conditionmoment ] hold . then there exist @xmath208 such that for every @xmath209 , @xmath210 [ lemsuptsupnexpmoment ] suppose conditions [ conditioncr ] and , [ conditionc1 ] and [ conditionmoment ] hold . then there exist @xmath211 such that for every @xmath212 , @xmath84 and @xmath42 , @xmath213 the following is immediate from lemmas [ lemevofsquaredsuprema ] and [ lemsuptsupnexpmoment ] . [ corevofsquaredsupremaofshiftedhatx^n ] suppose conditions [ conditioncr ] and , [ conditionc1 ] and [ conditionmoment ] hold . let @xmath214 be as in lemma [ lemsuptsupnexpmoment ] and @xmath215 . then there exists a @xmath216 such that for all @xmath212 and @xmath84 , @xmath217 the next lemma follows by combining lemma [ lemsuptsupnexpmoment ] with arguments as in the proof of lemma [ lemevofsquaredsuprema ] . the proof is omitted . [ lemevofsquaredsupremaofshiftedprocesses ] suppose conditions [ conditioncr ] and , [ conditionc1 ] and [ conditionmoment ] hold . let @xmath214 be as in lemma [ lemsuptsupnexpmoment ] . then for each @xmath190 there are @xmath218 such that for all @xmath84 and @xmath219 , @xmath220 in order to prove weak convergence results for the scaled catalyst and reactant processes , we will need to argue that the limit processes are continuous , which will be a consequence of the following bounds on the jumps . the somewhat stronger estimate on the jumps of the catalyst population in ( [ eqpsupdeltahatx^nt ] ) , below , will be used in the stochastic averaging argument in the proof of theorem [ fastcatalystbpaveraging ] . recall that for a process @xmath221 the jump at instant @xmath39 is defined as @xmath38 and @xmath40 . [ lempsupdeltahatx^nt ] suppose condition [ conditioncr ] holds . fix @xmath222 . then , as , @xmath223 if additionally conditions [ conditionc1 ] and [ conditionmoment ] hold , then , as @xmath57 , @xmath224 in this section we prove the results stated in section [ secauxresults ] . we begin with the proofs of lemmas [ lemrepresentationofcrprocesses][lemevofsquaredsupremaofshiftedprocesses ] . using these results , we will then prove proposition [ lemtighnessofcatalyst - reactant ] . the proof of lemma [ lempsupdeltahatx^nt ] is given at the end . proof of lemma [ lemrepresentationofcrprocesses ] recall that @xmath225 and that for @xmath171 , @xmath226 , @xmath173 , and @xmath174 . from ( [ locmart ] ) , @xmath227 using ( [ eqgeneratorofwholeprocess ] ) , we get @xmath228 next , since @xmath229 , we have @xmath230 and , once more using ( [ eqgeneratorofwholeprocess ] ) , @xmath231 thus with @xmath232 as in ( [ hateta^n ] ) , we get @xmath233 noting that @xmath234 and using ( [ eqhatz^n ] ) , ( [ eqhata^nphi3 ] ) and ( [ eqhatx^n - hatz^n ] ) , we have @xmath235 since @xmath232 is nondecreasing and @xmath236 , we have from the characterization given above ( [ eqskorohodlipschitz ] ) that @xmath237 next , for the reactant population , using similar calculations as for @xmath238 , we get @xmath239 finally , routine calculations then show ( see @xcite , lemma 3.1.3 ) that ( [ eqpqvm^nphi1 ] ) and ( [ eq5.6 ] ) hold . details are omitted . proof of lemma [ lemevofsquaredsuprema ] using ( [ eqpqvm^nphi1 ] ) and doob s inequality , we have @xmath240 next , from ( [ eq5.3 ] ) , @xmath241 the lipschitz continuity of the skorohod map implies @xmath242 letting @xmath243 , we now get @xmath244 combining this with ( [ eqhatn^ndef ] ) and ( [ eq5.20 ] ) , we obtain @xmath245 . \nonumber\end{aligned}\ ] ] using gronwall s inequality , we get , since @xmath246 , @xmath247 where @xmath248 . since @xmath249 and @xmath250 converge as @xmath57 , we have that for some @xmath189 and all @xmath251 @xmath252 using ( [ eq5.23 ] ) in ( [ eq5.20 ] ) , ( [ eq5.22 ] ) and ( [ eqhatx^nintemsofmartandeta ] ) , we have the estimate in ( [ eqevofsquaredsuprema ] ) by choosing @xmath253 sufficiently large . we next establish ( [ eqestimateonhaty^n ] ) . using doob s inequality once more and applying ( [ eq5.6 ] ) , we have @xmath254 thus , by ( [ eqhaty^nb^n2 ] ) , @xmath255\int_0^t e \bigl(\bigl|\hat y^{(n)}\bigr|_{*,s\wedge\sigma^{(n)}_k}^2 \bigr)\,ds \biggr).\end{aligned}\ ] ] the estimate in ( [ eqestimateonhaty^n ] ) now follows by choosing @xmath253 sufficiently large and applying gronwall s inequality . proof of lemma [ lemexpmomentsonfinitetime ] first we show , using conditions [ conditioncr](i ) and ( ii ) , [ conditionc1 ] and [ conditionmoment ] , that there are @xmath256 such that for all @xmath257 $ ] and @xmath258\mu^{(n)}_1(k ) \le-\delta d_1.\ ] ] note that @xmath259\mu^{(n)}_1(k)\\ & & \qquad = n^2 \sum_{k=0}^\infty \biggl(\sum _ { l=1}^\infty\frac { 1}{l ! } \biggl ( \frac{(k-1)\delta}{n } \biggr)^l \biggr)\mu^{(n)}_1(k ) \\ & & \qquad= n\delta \biggl(\sum_{k=0}^\infty k \mu^{(n)}_1(k)-1 \biggr ) + \frac { 1}{2 } \delta^2\sum_{k=0}^\infty(k-1)^2 \mu^{(n)}_1(k ) \\ & & \qquad\quad { } + n^2\sum_{k=0}^\infty \biggl ( \sum_{l=3}^\infty\frac { 1}{l ! } \biggl ( \frac{(k-1)\delta}{n } \biggr)^l \biggr)\mu^{(n)}_1(k).\end{aligned}\ ] ] now , as @xmath57 , @xmath260 and @xmath261 noting that @xmath130 , we can choose @xmath262 sufficiently small , and @xmath263 suitably , such that ( [ eqinstep1 ] ) holds . for @xmath264 as above and @xmath265 , let @xmath266 and @xmath267\mu^{(n)}_1(k ) \bigr)1_{\{\hat x^{(n)}_s > 1\}}\,ds.\ ] ] note that , by ( [ eqinstep1 ] ) , for any @xmath268 , @xmath269\\[-8pt ] & \le&-\delta d_1\lambda^{(n)}_1 \int _ u^t \hat x^{(n)}_s 1_{\{\hat x^{(n)}_s>1\}}\,ds.\nonumber\end{aligned}\ ] ] moreover , @xmath270 the first inequality in the last display is immediate , the second inequality can be seen as follows : @xmath271 by ( [ eqinstep1 ] ) , the first term on the right - hand side of the last display is smaller or equal to @xmath272 . thus @xmath273 we now argue that @xmath274 is a local martingale . let @xmath275 and @xmath276 here @xmath277 is the generator of @xmath195 , that is , for @xmath278 , @xmath279 \mu_1^{(n)}(k).\ ] ] note that @xmath280 \mu^{(n)}_1(k ) \bigr)1_{\{x>1\}}\ ] ] and thus @xmath281 also , @xmath282 consider the markov process @xmath283 defined by @xmath284 denote by @xmath285 the generator of @xmath283 . then the action of the generator on the function @xmath286 with @xmath275 and @xmath287 is given by @xmath288 using ( [ eq5.28 ] ) we now have that @xmath289 is a local martingale . from ( [ eq5.29 ] ) we now see that the last expression equals @xmath290 , @xmath42 , which is thus a local martingale . we next show that for every @xmath209 , @xmath265 and @xmath42 , @xmath291 note that @xmath292\\[-12pt ] & = & \biggl(m^{(n),\delta}_t+\alpha^{(n)}_\delta \int_0^t e^{-\beta ^{(n),\delta}_s}\,d\hat \eta^{(n)}_s \biggr)e^{\beta^{(n),\delta}_t}.\nonumber \ ] ] applying it s formula and using ( [ eqinstep1 ] ) , ( [ eq5.28 ] ) and ( [ eq5.31 ] ) , we see that @xmath293 the estimate in ( [ eq5.30 ] ) now follows by combining the above inequality with lemma [ lemevofsquaredsuprema ] and ( [ eqestimatealpha^bdelta ] ) . fix @xmath209 , @xmath294 and @xmath295 . then , since @xmath296 for all @xmath42 , we have for @xmath297 , @xmath298 where the last inequality follows on noting that @xmath299 is a submartingale and applying doob s inequality . now from ( [ eqestimatebeta^bdelta ] ) and ( [ eq5.30 ] ) , @xmath300 choose @xmath301 . then , by combining the above estimates , we can find a @xmath302 such that for all @xmath294 , @xmath84 and @xmath295 @xmath303^{1/2}.\end{aligned}\ ] ] dividing both sides by @xmath304^{1/2}$ ] yields @xmath305^{1/2}\le d_4(\delta,\rho , m)\ ] ] for any @xmath294 and @xmath84 . the result follows . proof of lemma [ lemsuptsupnexpmoment ] for @xmath306 , define @xmath307 and @xmath308 from ( [ eqinstep1 ] ) , we have , for some @xmath309 , @xmath310 for all @xmath265 [ with @xmath264 as above ( [ eqinstep1 ] ) ] and @xmath311 . observing that with @xmath275 , @xmath312 , where @xmath313 is the generator of @xmath195 defined in ( [ eq5.27 ] ) , we have that @xmath314 is a local martingale . fix @xmath214 and @xmath315 as in the statement of lemma [ lemexpmomentsonfinitetime ] . without loss of generality , we can assume that @xmath265 . note that on the set @xmath316 we have @xmath317&\le&\delta \bigl[\hat x^{(n)}_{j\rho}- \hat x^{(n)}_{(j-1)\rho}\bigr]-\int_{(j-1)\rho } ^{j\rho}b^{(n)}_\delta \bigl(\hat x^{(n)}_s\bigr)\,ds-\delta\kappa\rho \nonumber\\[-8pt]\\[-8pt ] & \equiv & v^{(n)}_j-\delta\kappa\rho.\nonumber\end{aligned}\ ] ] fix @xmath39 , and let @xmath318 be such that @xmath319 . then , similarly , on the set @xmath320 @xmath321\le v^{(n)}_{n}(t)$ ] , where @xmath322-\int_{(j-1)\rho}^{t}b^{(n)}_\delta \bigl(\hat x^{(n)}_s\bigr)\,ds.\ ] ] now , for a fixed @xmath323 , let @xmath324 be such that @xmath325 is the last interval in which @xmath195 visits 1 before time @xmath326 . we set @xmath327 if 1 is not visited before time @xmath326 . we distinguish between the cases @xmath328 , @xmath329 and @xmath327 , where the latter corresponds to the case where 1 is not visited before time @xmath326 . _ case _ 1 : @xmath328 . in this case @xmath330 for @xmath331 , let @xmath332 and @xmath333.\ ] ] then @xmath334 combining the above estimates , we have @xmath335 thus , in this case @xmath336 where by convention @xmath337 for @xmath338 and @xmath339 . _ case _ 2 : @xmath327 . in this case , 1 is not visited before time @xmath326 and thus @xmath340 _ case _ 3 : @xmath329 . suppose first that there is an @xmath341 $ ] such that @xmath342 . it then follows that @xmath343 now suppose that there is no such @xmath341 $ ] . define @xmath344 to be such that @xmath345 is the last interval in which @xmath195 visits 1 before @xmath346 . once again we set @xmath347 if there is no such interval . if @xmath347 , we get exactly as in case 2 that @xmath348 if @xmath349 , then @xmath350 combining the three cases , we have @xmath351 thus , for any @xmath352 , @xmath353 \biggr ) e^{-\delta\kappa\rho(n - l-1 ) } \biggr]\\ & & \qquad\quad\hspace*{227.5pt}{}+ e_x\bigl(e^{\delta\theta ^{(n)}_n } \bigr ) \biggr).\end{aligned}\ ] ] recalling @xmath354 from ( [ equ^nmart ] ) and using its martingale property , we get @xmath355\\[-8pt ] & \le & e^{-m_0}e^{(1+x)\delta } \,d(\delta,\rho,1 ) \biggl(1 + \frac { 1}{1-e^{-\delta\kappa\rho } } \biggr),\nonumber\end{aligned}\ ] ] where the last inequality follows from lemma [ lemexpmomentsonfinitetime ] and the observation that @xmath356 where @xmath357 is as in ( [ eqtheta^nj ] ) . finally , from ( [ eq5.36 ] ) , we get that for all @xmath42 and @xmath84 @xmath358 where @xmath359 . the result follows . proof of proposition [ lemtighnessofcatalyst - reactant ] we will first consider the second part of the proposition . we begin by showing that @xmath360 is tight . for that , in view of ( [ eqevofsquaredsupremaofshiftedprocesses ] ) , it suffices to show that the following condition ( aldous kurtz criterion ) holds : for each @xmath361 and @xmath362 there are @xmath262 and @xmath363 such that for all stopping times @xmath364 with @xmath365 , we have @xmath366 let @xmath367 be given . note that @xmath368 by ( [ eqestimateofshiftedhatx^noverfinitetime ] ) we have , for @xmath264 sufficiently small , @xmath369 it remains to prove that , for some @xmath262 , @xmath370 using the martingale property of @xmath371 , @xmath372 and , using ( [ eqpqvm^nphi1 ] ) , @xmath373 now , using ( [ eqestimateofshiftedhatx^noverfinitetime ] ) once more , we can choose @xmath262 such that ( [ eq5.40 ] ) holds . this proves tightness of @xmath360 and , using the continuity property of the skorohod map ( from @xmath374 to @xmath375 ) , that of @xmath376 and @xmath377 . tightness of @xmath378 now follows by using the uniform estimate in lemma [ lemsuptsupnexpmoment ] . now we consider the first part of the proposition . tightness of @xmath379 follows as before . we now consider @xmath380 . fix @xmath381 . using ( [ eqestimateonhaty^n ] ) , we get , for @xmath382 , @xmath383 & \leq&\frac{e ( \sup_{t\leq t}(\hat y^{(n)}_{\sigma ^{(n)}_k\wedge t})^2 ) } { \tilde k^2}+p \bigl(\sup_{t\leq t}\bigl(\hat x^{(n)}_t\bigr)\geq k \bigr ) \\[-2pt ] & \leq&\frac{\exp(kt^2 k^2 ) ( y^{(n)}_0)^2}{\tilde k^2}+\frac{e ( \sup_{t\leq t}(\hat x^{(n)}_t)^2 ) } { k^2}.\end{aligned}\ ] ] using ( [ eqevofsquaredsuprema ] ) , we can choose @xmath17 such that @xmath384 now choose @xmath385 such that @xmath386 the last two displays imply @xmath387 , and since @xmath381 is arbitrary , the tightness of the random variables @xmath388 , for each @xmath389 , follows . to establish the tightness of the processes @xmath390 , it now suffices to show that for each @xmath361 and @xmath391 there are @xmath262 and @xmath363 such that for all stopping times @xmath364 with @xmath365 , we have @xmath392 fix @xmath393 . then , for any @xmath394 , @xmath395 & & \qquad \leq p \bigl(\bigl|\hat y^{(n)}_{(\tau_n+\theta)\wedge\sigma ^{(n)}_k}-\hat y^{(n)}_{\tau_n\wedge\sigma^{(n)}_k}\bigr| \geq\gamma \bigr)+p \bigl(\sigma^{(n)}_k\leq m+1 \bigr).\end{aligned}\ ] ] taking @xmath396 and @xmath17 as in ( [ eqsupe|x^n|k ] ) , we have @xmath397 for all @xmath84 . for the first term on the right - hand side of the last display , we get , using ( [ eqhaty^nb^n2 ] ) and that @xmath398 , @xmath399 & & \qquad \leq p \biggl(\biggl|c_2^{(n)}\lambda^{(n)}_2 \int_{\tau _ n\wedge\sigma^{(n)}_k}^{(\tau_n+\theta)\wedge\sigma^{(n)}_k}\hat y^{(n)}_s \,ds\biggr|\geq\frac{\gamma}{2k } \biggr ) \\[-2pt ] & & \qquad\quad { } + p \biggl(\bigl|m^{(n)}_{(\tau_n+\theta)\wedge \sigma^{(n)}_k}(\phi_2)-m^{(n)}_{\tau_n\wedge\sigma^{(n)}_k } ( \phi_2)\bigr|\geq\frac{\gamma}{2 } \biggr).\nonumber\end{aligned}\ ] ] the first term on the right - hand side can be bounded as follows : @xmath400 where @xmath253 is the constant from ( [ eqestimateonhaty^n ] ) . thus , for @xmath264 sufficiently small , we get @xmath401 the second term on the right - hand side of ( [ eq5.43 ] ) can be bounded as follows : @xmath402 using ( [ eqestimateonhaty^n ] ) once more , we have that , for @xmath264 sufficiently small , the second term in ( [ eq5.43 ] ) is bounded by @xmath403 . combining the above estimates , we now see that ( [ eq5.42 ] ) holds , and thus tightness of @xmath404 follows . proof of lemma [ lempsupdeltahatx^nt ] consider ( [ eqpsupdeltahatx^nt ] ) . let @xmath405 be the number of deaths of particles of the ( unscaled ) process @xmath1 in the time interval @xmath406 $ ] . . then @xmath408 by corollary [ corevofsquaredsupremaofshiftedhatx^n ] , we can choose @xmath409 such that @xmath410 for @xmath219 and @xmath84 . next , consider @xmath411 note that on the set @xmath412 the branching rates of @xmath1 are bounded during the time interval @xmath406 $ ] , uniformly in @xmath413 and @xmath2 , and thus we can choose a @xmath414 such that for @xmath219 and @xmath84 @xmath415 finally , let , for @xmath84 , @xmath416 be i.i.d . random variables distributed as @xmath417 . then , since the variance of the offspring distribution converges , we have for @xmath363 sufficiently large and all @xmath418 , @xmath419 combining the above estimates , ( [ eqpsupdeltahatx^nt ] ) follows . the limit in ( [ eqpsupdeltahaty^nt ] ) can be established similarly , using lemma [ lemevofsquaredsuprema ] instead of corollary [ corevofsquaredsupremaofshiftedhatx^n ] ; the proof is therefore omitted . the following martingale characterization result will be useful in the proof of theorem [ thmcrdiffusion ] . the proof is standard and is omitted ; see @xcite , @xcite , @xcite and theorem 5.3 of @xcite . for @xmath420 , let @xmath421 let @xmath422 and @xmath423 be the corresponding borel @xmath27-field ( with respect to the skorohod topology ) . denote by @xmath424 the canonical filtration on @xmath425 , that is , @xmath426 , where @xmath427 for @xmath428 . finally , let @xmath429 , @xmath173 , be the coordinate processes , that is , @xmath430 . [ thmmartingalecharacterization ] let @xmath431 be a probability measure on @xmath432 under which the following hold a.s . : @xmath433 is a nondecreasing , continuous process , and @xmath434 ; @xmath435 is an @xmath436 valued continuous process ; @xmath437 ; for all @xmath420 @xmath438 is an @xmath439 martingale ; @xmath440,where @xmath441 and @xmath442 are as in proposition [ propuniqunessofx , y ] . then @xmath443 . proof of theorem [ thmcrdiffusion ] recall that for @xmath420 , we have @xmath444 where @xmath445 is a martingale , and @xmath446 is as defined in ( [ eqgeneratorofwholeprocess ] ) . also note that @xmath447 can be rewritten as @xmath448 where @xmath449\mu_1^{(n)}(k ) \\ & & { } + \lambda_2^{(n ) } n^2 x y \sum _ { k=0}^\infty \biggl[\phi \biggl(x , y+\frac{k-1}{n } \biggr)-\phi(x , y ) \biggr]\mu_2^{(n)}(k)\end{aligned}\ ] ] and @xmath450 thus , using ( [ hateta^n ] ) , ( [ eq6.1 ] ) can be rewritten as @xmath451 recall the path space @xmath452 introduced above theorem [ thmmartingalecharacterization ] . denote by @xmath453 the measure induced by @xmath454 on @xmath425 and by @xmath455 the corresponding expectation . from proposition [ lemtighnessofcatalyst - reactant ] , @xmath456 is tight . let @xmath431 be a limit point of @xmath457 along some subsequence @xmath458 . in order to complete the proof , it suffices to show that under @xmath431 properties ( i)(v ) in theorem [ thmmartingalecharacterization ] hold almost surely . property ( i ) is immediate from the fact that @xmath232 is nondecreasing and continuous with initial value 0 for each @xmath2 . also , property ( v ) is immediate from the fact that @xmath459 , a.s . , for each @xmath2 . next , consider property ( ii ) . the continuity of @xmath460 and @xmath461 follows by ( [ eqpsupdeltahaty^nt ] ) ; see @xcite , proposition vi.3.26 , page 315 . to see ( iii ) , consider , for @xmath462 , continuous bounded test functions @xmath463 such that @xmath464 note that , for each @xmath84 , @xmath465 and thus , for each @xmath462 , @xmath466 consequently , for each @xmath462 , @xmath467 , almost surely w.r.t . the property in ( iii ) now follows on sending @xmath468 . finally , we consider part ( iv ) . it suffices to show that for every @xmath469 @xmath470 where @xmath471 is an arbitrary bounded , continuous , @xmath472 measurable map . now fix such @xmath473 and @xmath107 . then by weak convergence of @xmath474 to @xmath431 and using the moment bound in lemma [ lemevofsquaredsuprema ] , @xmath475 to complete the proof , it suffices to show that the limit on the left - hand side above is 0 . in view of the martingale property in ( [ eq6.1 ] ) , to show this , it suffices to prove that for @xmath420 , @xmath476 and @xmath477 the latter is immediate upon using the smoothness of @xmath108 and the moment estimate for @xmath232 in ( [ eqevofsquaredsuprema ] ) . for ( [ eqh4forsingletypecont ] ) , we rewrite @xmath478 using a taylor expansion as follows : @xmath479\mu_1^{(n)}(k ) \\ & & \qquad\quad { } + \lambda_2^{(n ) } n^2 xy \sum _ { k=0}^\infty \biggl[\frac { k-1}{n } \,\frac{\partial}{\partial y}\phi(x , y)+\frac{1}{2 } \biggl(\frac{k-1}{n } \biggr)^2\,\frac{\partial^2}{\partial y^2}\phi(x , y ) \biggr]\mu_2^{(n)}(k ) \\ & & \qquad\quad { } + r^{(n)}(x , y ) \\ & & \qquad = c_1^{(n)}\lambda_1^{(n ) } x \,\frac{\partial}{\partial x}\phi ( x , y)+\frac{1}{2}\alpha_1^{(n ) } \lambda_1^{(n ) } x \,\frac{\partial^2}{\partial x^2}\phi(x , y ) \\ & & \qquad\quad { } + c_2^{(n ) } \lambda_2^{(n ) } x y \,\frac{\partial}{\partial y}\phi(x , y)+\frac{1}{2}\alpha_2^{(n ) } \lambda_2^{(n ) } x y \,\frac { \partial^2}{\partial y^2}\phi(x , y)+r^{(n)}(x , y),\end{aligned}\ ] ] where the term @xmath480 is a remainder term , which , using part ( iii ) of condition [ conditioncr ] , is seen to satisfy @xmath481 as @xmath57 , for any @xmath482 . furthermore , using the compact support property of @xmath483 , it follows that @xmath484 . next note that @xmath485 and therefore , in view of condition [ conditioncr ] , @xmath486 once more using the compact support property of @xmath108 , it follows that @xmath487 combining the above estimates , we have ( [ eqh4forsingletypecont ] ) , and the result follows . uniqueness of the invariant measure of @xmath127 is an immediate consequence of the nondegeneracy of the diffusion coefficient ( note that @xmath488 ) . for existence , we will apply an extension of the well - known echeverria criterion for invariant measures of markov processes @xcite and theorem 5.7 of @xcite . this criterion , in the current context , says that in order to establish that a probability measure @xmath489 is an invariant measure for @xmath127 , it suffices to verify that for some @xmath490 and all @xmath491 @xmath492 where @xmath493 we now show that ( [ eqewk - criterion ] ) holds with @xmath494 and @xmath495 . for @xmath496 and @xmath497 as in ( [ eqdensityofnu1 ] ) , @xmath498 thus ( [ eqewk - criterion ] ) follows . throughout this section we assume that conditions [ conditioncr ] , [ conditionc1 ] and [ conditionmoment ] hold . this will not be explicitly noted in the statements of the results . existence of a stationary distribution @xmath139 of the @xmath499 valued markov process @xmath195 follows from the tightness of @xmath500 , which is a consequence of lemma [ lemsuptsupnexpmoment ] . the uniqueness of the stationary distribution follows from the irreducibility of @xmath195 . in order to establish the tightness of the sequence @xmath140 , we will use the following uniform in @xmath2 moment stability estimate for @xmath195 . [ thmmomentstability ] there is a @xmath501 such that for all @xmath502 and @xmath503 , @xmath504 fix an @xmath505 , and let @xmath506 . observe that if @xmath507 for some @xmath318 , then , following arguments as in the proof of lemma [ lemsuptsupnexpmoment ] , for @xmath508 , @xmath509 thus we have that @xmath510 where @xmath511 , @xmath82 . the above estimate along with lemma [ lemsuptsupnexpmoment ] implies , for @xmath84 , @xmath512 where @xmath513 is as in lemma [ lemsuptsupnexpmoment ] and @xmath514 is some constant , independent of @xmath2 . fix @xmath503 . then , for some @xmath515 , we have @xmath516 choose @xmath517 large enough such that @xmath518 . then for @xmath519 @xmath520 the result follows . as a consequence of theorem [ thmmomentstability ] , we have the following result . for @xmath521 , define the return time to a compact set @xmath522 by @xmath523 there are @xmath524 and a compact set @xmath522 such that @xmath525 note that @xmath526 since @xmath527 . applying theorem [ thmmomentstability ] with @xmath528 , we have that there is an @xmath529 such that with @xmath530 , for all @xmath531 , @xmath532 where @xmath517 is as in theorem [ thmmomentstability ] . let @xmath533 and @xmath534 consider a sequence of stopping times defined as follows : @xmath535 let @xmath536 , and @xmath537 then @xmath538 let @xmath539 . we claim that there is a @xmath540 such that for all @xmath541 , @xmath542 @xmath543 due to the strong markov property , to prove the claim it suffices to show that for some @xmath540 and for all @xmath544 @xmath545 note that for @xmath311 , @xmath546 , where @xmath547 . using this bound along with lemma [ lemevofsquaredsuprema ] , we get , for some @xmath548 , @xmath549 the claim follows . from the estimate ( [ eqclaimaa ] ) , we now have @xmath550 note that @xmath551 is a markov chain with transition probability kernel @xmath552 where @xmath553 is the transition probability kernel for @xmath138 . using ( [ eqexpmomentoutsidecompact ] ) and lemma [ lemsuptsupnexpmoment ] , we get that for any @xmath529 there exists a @xmath554 such that for all @xmath555 @xmath556\\[-8pt ] & & { } + \sup_n \int_1^\infty y^3 p^{(n)}_{t_0 l}(x , dy ) 1_{\{x\leq l\ } } \nonumber\\ & \leq & x^3-\frac{1}{2}x^3+\tilde b 1_{[0,l]}(x ) . \nonumber\end{aligned}\ ] ] the above inequality along with theorem 14.2.2 of @xcite yields @xmath557 } \bigl(\hat x^{(n)}_{\sigma ^{(n)}_k } \bigr ) \biggr ) \biggr ) \\ & = & 2 \bigl(x^3+\tilde b 1_{[0,l]}(x ) \bigr)\le\tilde c x^3,\end{aligned}\ ] ] where the equality in the last display follows from the fact that @xmath558 for @xmath559 . the result follows now on combining the last estimate with ( [ eq7.7 ] ) . the following theorem is proved exactly as proposition 5.4 of @xcite . the proof is omitted . [ thmestimatefx^nbyv^n ] let @xmath560 be a measurable function . define for @xmath561 and a compact set @xmath522 @xmath562 if @xmath563 is everywhere finite and uniformly bounded on @xmath564 , then there exists a @xmath565 such that for all @xmath566 @xmath567 we now return to the proof of theorem [ thmtightnessandconvergenceofnu^n1 ] and establish the tightness of @xmath140 . we will apply theorem [ thmestimatefx^nbyv^n ] with @xmath568 , and @xmath569 as in theorem [ thmintegratingtillreturntime ] . since @xmath139 is an invariant measure for @xmath195 , we have for nonnegative , real valued , measurable functions @xmath570 on @xmath4 @xmath571 fix @xmath192 and let @xmath572 . let @xmath573 by ( [ eq7.9 ] ) , we have that @xmath574 . let @xmath575 by the monotone convergence theorem @xmath576 as @xmath577 . we next show that @xmath578 is bounded from below for all @xmath579 : if @xmath580 , then @xmath581 where the last inequality follows from theorem [ thmestimatefx^nbyv^n ] . if @xmath582 , @xmath583 thus @xmath584 for all @xmath311 . by fatou s lemma , we have @xmath585 by theorem [ thmestimatefx^nbyv^n ] we have @xmath586 . combining this with the last display , we have @xmath587 using the invariance property of @xmath139 once more , we see that the first term on the right - hand side above equals @xmath588 , and therefore@xmath589 . this completes the proof of tightness . the tightness of @xmath140 implies that every subsequence of @xmath590 has a convergent subsequence . call such a limit @xmath591 . theorem [ thmcrdiffusion ] and the stationarity of @xmath139 imply that @xmath591 is a stationary distribution of @xmath127 . since the stationary distribution of @xmath127 is unique , we have @xmath592 , which completes the proof.=-1 in order to prove the result , we will verify that the assumptions of theorem ii.1 ( more precisely , those in the remark following theorem ii.1 ) in @xcite , pages 78 and 79 , hold . for this , it suffices to show that for all @xmath192 , @xmath593 , @xmath594 and @xmath595 , there exists a sequence @xmath596 with @xmath597 and @xmath598}\bigl{\vert}e \bigl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \bigl(\phi \bigl(\check y^{(n)}_{t+h_n } \bigr)-\phi \bigl(\check y^{(n)}_{t } \bigr)-h_n\check l\phi \bigl(\check y^{(n)}_t \bigr ) \bigr ) \bigr]\bigr{\vert}\nonumber\\[-8pt]\\[-8pt ] & & \qquad = o(h_n),\nonumber\end{aligned}\ ] ] where @xmath599 is given as @xmath600 letting @xmath601 , @xmath42 , we see , using scaling properties of the skorohod map and straight forward martingale characterization results , that @xmath602 has the same probability law as the process @xmath127 that was introduced in proposition [ propuniqunessofx , y ] with initial value @xmath603 . the following uniform moment bound will be used in the proof of theorem [ thmfastcatalystweaklimit ] . [ lemunifmombdx ] there exists a @xmath604 , such that whenever @xmath127 is as in proposition [ propuniqunessofx , y ] with initial value @xmath605 , for some @xmath579 , we have @xmath606 we begin by establishing exponential moment estimates for the increase of @xmath607 over time intervals of length @xmath608 when the process is away from the boundary 1 , where @xmath297 and @xmath609 . fix @xmath610 . let @xmath611 , @xmath612 and @xmath613 . note that in view of condition [ conditionc1 ] , @xmath614 . define @xmath615 and @xmath616 . then @xmath617 where the inequality follows on noting that @xmath618 and @xmath619 . using the super martingale property of the stochastic exponential , we have that the second term on the right - hand side of the last display is bounded by 1 . thus , sending @xmath620 , we have , with @xmath621 , @xmath622 next , for @xmath579 , we have by application of it s formula that for @xmath623 and @xmath624 , @xmath625 where @xmath626 and the last inequality follows by an application of gronwall s lemma and the lipschitz property of the skorohod map ; see ( [ eqskorohodlipschitz ] ) . using the above estimates , we will now establish certain uniform estimates on the tail probabilities of @xmath627 , which will lead to exponential moment estimates at these time points . fix @xmath505 , and let @xmath628 @xmath629 . fix @xmath192 , and let @xmath630 if there is an @xmath631 $ ] such that @xmath632 , and set @xmath633 equal to 0 otherwise . let @xmath634 then @xmath635 . letting @xmath636 , we have , using ( [ eq8.3 ] ) , @xmath637 next note that , from ( [ eq8.4 ] ) , @xmath638 } 1_{\tau_i < i\rho } \bigr)+1=e_x \bigl ( e^{-\delta x_{\tau_i}}1_{\tau _ i < i\rho } e_{x_{\tau_i } } \bigl(e^{\delta x_{i\rho } } \bigr ) \bigr)+1 \\ & \le & e_x \bigl[e^{-\delta x_{\tau_i}}1_{\tau_i < i\rho } \bigl(e^{\delta x_{\tau_i}}+x_{\tau_i } c_1(\rho,\delta ) \bigr ) \bigr]+1\equiv c_2(\rho,\delta)+1.\end{aligned}\ ] ] hence , @xmath639 the last estimate yields , analogously to ( [ eq5.38 ] ) , @xmath640 for some @xmath641 . finally , letting @xmath642 , we have from ( [ eq8.4 ] ) for @xmath643 $ ] , @xmath644 , @xmath645 the result follows . [ remunifmombdcheckx^n ] note that lemma [ lemunifmombdx ] and the scaling property noted above that lemma say that for all @xmath579 @xmath646 we now prove theorem [ thmfastcatalystweaklimit ] by showing ( [ eqstochaveragcriterion ] ) . let , for @xmath594 , @xmath647 then @xmath648 \nonumber \\ & & \qquad = e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \int_t^{t+h_n } \mathcal{l}_{\check x^{(n)}_s}\phi\bigl(\check y^{(n)}_t\bigr)\,ds \biggr ] \\ & & \qquad\quad { } + e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \int_t^{t+h_n } \bigl(\mathcal{l}_{\check x^{(n)}_s}\phi\bigl(\check y^{(n)}_s \bigr)- \mathcal{l}_{\check x^{(n)}_s}\phi \bigl(\check y^{(n)}_t \bigr ) \bigr)\,ds \biggr].\hspace*{-15pt}\nonumber\end{aligned}\ ] ] for the second term , we have , using remark [ remunifmombdcheckx^n ] and the fact that the function @xmath570 is bounded and @xmath108 as well as its derivatives are continuous with bounded support , that @xmath649 } e \biggl|\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \int_t^{t+h_n } \bigl(\mathcal{l}_{\check x^{(n)}_s } \phi\bigl(\check y^{(n)}_s\bigr)- \mathcal{l}_{\check x^{(n)}_s } \phi\bigl(\check y^{(n)}_t\bigr ) \bigr)\,ds \biggr| \nonumber \\ & & \qquad= \sup_{t\in[t_{k+1},t ] } e \biggl|\phi \bigl(\check y^{(n)}_{t_1 } , \ldots,\check y^{(n)}_{t_k } \bigr ) \nonumber\\ & & \qquad\quad\hspace*{48pt}{}\times\int_t^{t+h_n } \biggl[c_2\lambda_2\check x^{(n)}_s \bigl(\check y^{(n)}_s\phi'\bigl ( \check y^{(n)}_s\bigr)-\check y^{(n)}_t \phi'\bigl(\check y^{(n)}_t\bigr ) \bigr ) \\ & & \hspace*{93.3pt}\qquad\quad { } + \frac{1}{2}\alpha_2\lambda_2 \check x^{(n)}_s \bigl ( \check y^{(n)}_s \phi''\bigl(\check y^{(n)}_s \bigr)-\check y^{(n)}_t\phi '' \bigl(\check y^{(n)}_t\bigr ) \bigr ) \biggr]\,ds \biggr|\nonumber\\ & & \qquad = o(h_n).\nonumber\end{aligned}\ ] ] recalling the definition of @xmath602 above lemma [ lemunifmombdx ] , the first expected value on the right - hand side in ( [ eqab ] ) equals @xmath650.\ ] ] thus @xmath651 \\ & & \qquad = e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) h_n \biggl(\frac{1}{h_n a_n } \int_{ta_n}^{ta_n+h_na_n } \mathcal{l } _ { x^\circ_s}\phi \bigl(\check y^{(n)}_t\bigr)\,ds-\check{\mathcal{l}}\phi\bigl ( \check y^{(n)}_t\bigr ) \biggr ) \biggr]\\ & & \qquad\quad{}+o(h_n ) \\ & & \qquad = e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) h_n \biggl(c_2 \lambda_2\check y^{(n)}_t\phi ' \bigl(\check y^{(n)}_t\bigr)+\frac { 1}{2 } \alpha_2\lambda_2 \check y^{(n)}_t \phi''\bigl(\check y^{(n)}_t \bigr ) \biggr ) \\ & & \hspace*{173.4pt}{}\times \biggl(\frac{1}{h_na_n}\int_{ta_n}^{ta_n+h_na_n } x^\circ_s \,ds - m_x \biggr ) \biggr]\\ & & \qquad\quad{}+o(h_n).\end{aligned}\ ] ] to complete the proof , it thus remains to show that for some sequence @xmath652 with @xmath597 @xmath653 converges to 0 uniformly in @xmath654 $ ] . from the ergodicity of @xmath127 and the moment estimate in lemma [ lemunifmombdx ] , it follows that @xmath655 the above result , along with lemma [ lemshiftintegrallimits ] , below , implies that there is a sequence @xmath652 such that @xmath656 , and the expression in ( [ eqc ] ) converges to 0 uniformly in @xmath657 $ ] . this completes the proof . the proof of the following lemma is adapted from lemma ii.9 , page 137 , in @xcite . [ lemshiftintegrallimits ] let @xmath658 and @xmath659 monotonically as @xmath57 . if for all @xmath654 $ ] @xmath660 then there is a sequence @xmath652 such that @xmath661 as @xmath662 , and @xmath663}e\biggl{\vert}\frac{1}{h_na_n}\int_{ta_n}^{ta_n+h_na_n}x_s\,ds - m_x \biggr{\vert}\to0\qquad \mbox{as } n\to \infty.\ ] ] let @xmath664 . note that @xmath665 converges monotonically to 0 as @xmath666 . for @xmath654 $ ] we have @xmath667 for all @xmath2 such that @xmath668 . note that the right - hand side of the last display is independent of @xmath654 $ ] . choosing @xmath669 , the lemma follows . as in the proof of theorem [ thmfastcatalystweaklimit ] , it suffices to show that for all @xmath192 , @xmath593 , @xmath594 and @xmath595 , there exists a sequence @xmath596 with @xmath597 and @xmath663}\bigl|e\bigl[\phi\bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k}\bigr ) \bigl(\phi\bigl(\check y^{(n)}_{t+h_n}\bigr)-\phi\bigl(\check y^{(n)}_{t } \bigr)-h_n\check l\phi\bigl(\check y^{(n)}_t \bigr)\bigr)\bigr]\bigr|=o(h_n),\ ] ] where @xmath599 is given as in ( [ eq8.2 ] ) . let for @xmath594 and @xmath670 @xmath671 \mu_2^{(n)}(k)\ ] ] and recall @xmath672 from ( [ eq8.6 ] ) . then @xmath673 \nonumber\\ & & \qquad = e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \int_t^{t+h_n } \mathcal{l}^{(n)}_{\check x^{(n)}_s}\phi\bigl(\check y^{(n)}_t \bigr)\,ds \biggr ] \\ & & \qquad\quad { } + e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \int_t^{t+h_n } \bigl [ \mathcal{l}^{(n ) } _ { \check x^{(n)}_s}\phi \bigl(\check y^{(n)}_s \bigr)-\mathcal{l}^{(n ) } _ { \check x^{(n)}_s } \phi \bigl(\check y^{(n)}_t \bigr ) \bigr]\,ds \biggr ] . \nonumber\end{aligned}\ ] ] using lemma [ lemsuptsupnexpmoment ] , we get , as in ( [ eqsecondterm ] ) , that the second term in the last display is @xmath674 uniformly in @xmath654 $ ] . thus @xmath675 \nonumber\\ & & \qquad = e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) \biggl(\int_{t}^{t+h_n } \bigl(\mathcal{l}^{(n ) } _ { \check x^{(n)}_s}\phi \bigl(\check y^{(n)}_t \bigr)-\mathcal{l}_{\check x^{(n)}_s}\phi \bigl ( \check y^{(n)}_t \bigr ) \bigr ) \,ds \biggr ) \biggr ] \nonumber\\[-8pt]\\[-8pt ] & & \qquad\quad{}+ e \biggl[\phi \bigl(\check y^{(n)}_{t_1},\ldots,\check y^{(n)}_{t_k } \bigr ) h_n \biggl(\frac{1}{h_n } \int_{t}^{t+h_n } \mathcal{l } _ { \check x^{(n)}_s}\phi \bigl(\check y^{(n)}_t\bigr)\,ds-\check{\mathcal{l}}\phi\bigl ( \check y^{(n)}_t\bigr ) \biggr ) \biggr]\nonumber\\ & & \qquad\quad{}+o(h_n ) . \nonumber\end{aligned}\ ] ] calculations similar to those in the proof of theorem [ thmcrdiffusion ] show that the first term on the right - hand side in the last display is @xmath674 uniformly in @xmath654 $ ] [ see proof of ( [ eqh4forsingletypecont ] ) ] , while the second term can be written as @xmath676.\end{aligned}\ ] ] to show that the latter term is @xmath674 uniformly in @xmath657 $ ] , it suffices to show the following result . [ th8.1 ] as @xmath57 @xmath663 } e\biggl{\vert}e_{\check{\mathcal{f}}^{(n)}_t } \biggl(\frac{1}{h_n } \int_{t}^{t+h_n}\check x^{(n)}_s \,ds - m_x \biggr)\biggr{\vert}\to0,\ ] ] where @xmath677 and @xmath678 . in order to prove this theorem , we need the following three results . let @xmath679 , @xmath26 be the space of probability measures on @xmath25 , and , given a sequence @xmath680 $ ] , @xmath681 be a sequence of @xmath26 valued random variables defined as follows . for @xmath682 , @xmath683\,ds.\ ] ] let @xmath684 . [ lemrandommeasuretight ] the family of @xmath26 valued random variables @xmath685 is tight , and any weak limit point is a @xmath686 valued random variable . let @xmath687 with @xmath460 and @xmath461 being the canonical coordinate processes on @xmath688 . [ lemrandommeasurelimit ] let @xmath199 be a weak limit point of @xmath689 given on some probability space @xmath690 . then for @xmath691 a.e . @xmath692 , @xmath693 satisfies the following : a. @xmath694 , for all @xmath42 ; @xmath695 ; b. @xmath461 is nondecreasing and @xmath696 a.s . @xmath693 ; c. @xmath697 a.s . d. under @xmath693 , for all @xmath491 @xmath698 is a @xmath699-martingale , where @xmath700 is as in ( [ eq7.2 ] ) and . we postpone the proofs of lemmas [ lemrandommeasuretight ] and [ lemrandommeasurelimit ] until after the proof of theorem [ th8.1 ] . the following is immediate from the above two lemmas , proposition [ propstationarydistrofx ] and the martingale characterization of the probability law of the process in ( [ eqx ] ) ; see theorem [ thmmartingalecharacterization ] . [ lemrandommeasurelimitcorollary ] let @xmath701 be as in proposition [ propuniqunessofx , y ] with @xmath702 and @xmath132 given as in proposition [ propstationarydistrofx ] . let @xmath703 be the probability measure on @xmath688 induced by @xmath701 . then @xmath681 converges weakly to @xmath703 . proof of theorem [ th8.1 ] it suffices to show that for an arbitrary sequence @xmath704 $ ] we have , as @xmath57 , @xmath705 since @xmath706 it suffices to show that @xmath707 for any @xmath708 , let @xmath709 be the following continuous function : @xmath710 and @xmath709 is linearly interpolated on @xmath711 $ ] . by corollary [ lemrandommeasurelimitcorollary ] @xmath681 converges weakly to @xmath703 , and therefore , for every @xmath708 , @xmath712 moreover , using the estimate in lemma [ lemsuptsupnexpmoment ] , @xmath713 and @xmath714 the last three displays imply the convergence in ( [ eq8.14 ] ) , and thus the result follows . proof of lemma [ lemrandommeasuretight ] to show the tightness of @xmath689 , it suffices to show that @xmath715 is tight , where for @xmath682 @xmath716\,ds.\ ] ] however , the tightness of @xmath717 is immediate in view of the tightness of @xmath718 which was proved in proposition [ lemtighnessofcatalyst - reactant ] . let @xmath199 be a weak limit point of @xmath681 and @xmath719 be defined by @xmath720\,du,\ ] ] where @xmath721 then @xmath722 is continuous and bounded on @xmath25 , and in order to show that @xmath199 is supported on @xmath688 , it suffices to show that @xmath723 ; see @xcite , page 147 . in turn , for the latter equality to hold , it suffices to show that for all @xmath381 , @xmath724 as @xmath57 . now @xmath725\wedge1 \bigr)\,du>\varepsilon \biggr ) \,ds.\end{aligned}\ ] ] finally , noting that @xmath726 is continuous and using lemma [ lempsupdeltahatx^nt ] , we now have that the right - hand side of the latter equation converges to 0 as @xmath57 . the result follows . proof of lemma [ lemrandommeasurelimit ] for a measure @xmath727 , let @xmath728 denote the expectation operator . for ( a ) , we show that @xmath729\\[-8pt ] & & \eqntext{\mbox{a.s . for all bounded continuous } f\mbox { on } s.}\end{aligned}\ ] ] note that @xmath730 this proves ( [ eq8.15 ] ) since we can choose @xmath596 such that @xmath731 , and thus ( a ) follows . property ( b ) is immediate from the fact that @xmath732 is nondecreasing and continuous with initial value 0 for each @xmath2 . to prove ( c ) , it suffices to show that for a.e . @xmath323 and for every @xmath733 @xmath734 where @xmath735 is defined in ( [ eqfdelta ] ) . in turn , for the above equality to hold , it suffices to show that @xmath736 the latter equality is immediate on noting that for every @xmath733 @xmath737 & & \qquad = \frac{1}{a_nh_n}\int_{a_nt_n}^{a_nt_n+a_nh_n}e \biggl(\int _ 0^t f_\delta\bigl(\hat x^{(n)}_{s+u}\bigr)\,d\bigl(\hat\eta^{(n)}_{s+\cdot}- \hat \eta^{(n)}_s\bigr ) ( u)\wedge1 \biggr ) \,ds=0\end{aligned}\ ] ] and thus @xmath738=\lim_{n\to\infty } e \biggl[\tilde e^{\mu_n } \biggl(\int_0^t f_\delta\bigl(\pi^{(1)}_s\bigr)\,d \pi^{(2)}_s\wedge1 \biggr ) \biggr]=0.\ ] ] finally , consider ( d ) . it suffices to show that for every @xmath739 , @xmath740 & & \hspace*{151.5pt}\qquad{}- \phi'(1)\bigl[\pi_t^{(2)}- \pi_r^{(2)}\bigr ] \biggr ) \biggr ) \biggr|=0,\end{aligned}\ ] ] where @xmath741 is an arbitrary bounded , continuous , @xmath742 measurable map . now fix such @xmath743 and @xmath107 . assume without loss of generality that @xmath681 converges to @xmath199 . combining this weak convergence with lemma [ lemevofsquaredsupremaofshiftedprocesses ] , we see that the left - hand side of the last display is the limit of @xmath744 & & \qquad\quad\hspace*{144.5pt}{}- \phi'(1)\bigl[\pi_t^{(2)}-\pi_r^{(2 ) } \bigr]\biggr ) \biggr ) \biggr| \\[-2pt ] & & \qquad=\frac{1}{a_nh_n}\int_{a_nt_n}^{a_nt_n+a_nh_n } e \biggl| e_{\check{\mathcal{f}}_{t_n } } \biggl(\psi\bigl(\hat x^{(n)}_{s+\cdot},\hat \eta^{(n)}_{s+\cdot}-\hat\eta^{(n)}_s\bigr ) \\[-2pt ] & & \hspace*{146.2pt}{}\times \biggl[\phi\bigl(\hat x^{(n)}_{s+t}\bigr)-\phi\bigl(\hat x^{(n)}_{s+r}\bigr ) \\[-2pt ] & & \qquad\quad\hspace*{130pt}{}-\int_r^t \mathcal{l}_1\phi\bigl(\hat x^{(n)}_{s+u } \bigr)\,du\\[-2pt ] & & \qquad\quad\hspace*{130pt}{}-\phi'(1)\bigl[\hat\eta^{(n)}_{s+t}-\hat \eta^{(n)}_{s+r}\bigr ] \biggr ] \biggr ) \biggr|\,ds.\end{aligned}\ ] ] to complete the proof , it suffices to show that the limit of the expression in the last display is 0 . note that for @xmath491 , @xmath745 is a martingale , where @xmath746 $ ] and @xmath747\mu_1^{(n)}(k).\ ] ] thus , it suffices to prove that @xmath748 and @xmath749 the proofs for the last two equalities are completed as those for ( [ eqh4forsingletypecont ] ) and ( [ eqeintd^n - phi ] ) upon using the uniform estimates in corollary [ corevofsquaredsupremaofshiftedhatx^n ] and lemma [ lemevofsquaredsupremaofshiftedprocesses ] . proof of proposition [ propuniqunessofx , y ] we will consider here only the case where @xmath750 for some @xmath751 . the general case can be treated similarly . the unique solvability of ( [ eqx ] ) is an immediate consequence of the lipschitz property of the skorohod map , lipschitz coefficients ( note that @xmath752 is a lipschitz function on @xmath4 ) and a standard picard iteration scheme ; see , for example , proposition 1 in @xcite . we next argue the unique solvability of ( [ eqy ] ) . for @xmath84 , let @xmath753 , @xmath754 and @xmath755 . consider the equation @xmath756\\[-8pt ] & & { } + \sqrt{\alpha_2 \lambda_2}\int_0^t \sqrt { \bar x^{(n)}_s f^{(n)}\bigl ( \bar y^{(n)}_s\bigr ) } \,db_s^y.\nonumber\end{aligned}\ ] ] from the lipschitz property of @xmath757 and @xmath758 it follows that , for each @xmath2 , the above equation has a unique pathwise solution . let @xmath759 and @xmath760 . note that @xmath761 solves ( [ eqy ] ) on @xmath762 $ ] . also , by unique solvability of ( [ eqcheckyinpropuniqunessofx , y ] ) , we have for all @xmath84 , @xmath763 finally , letting @xmath764 , the unique solution of ( [ eqy ] ) is given by the following : @xmath765 we thank the editor , the associate editor and the referees for their comments that led to several improvements in the manuscript .

in the last few years , a lot of progress has been made in the study of d - branes as probes of sub - stringy geometry . while the usual picture of space - time is supposed to be valid upto the string scale , it was argued in @xcite that d - branes are the natural candidates to probe the geometry of space - time beyond the string scale . as has been widely accepted by now , the standard concepts of space - time appears , from the d - brane perspective , as the vacuum moduli space of its world volume gauge theory . indeed , it has been found that the space - time appearing in d - brane probes at ultra - short distance scales has qualitatively different features from that probed by bulk strings @xcite . of particular interest has been the study of d - branes on calabi - yau manifolds . in @xcite,@xcite , the moduli space of calabi - yau manifolds was studied using fundamental strings . a rich underlying geometric structure was discovered , and the moduli space was shown to contain topologically distinct geometric ( calabi - yau ) phases , as well as non - geometric ( landau - ginzburg ) phases . the khler sector of the vacuum moduli space was shown to consist of several domains ( separated by singular calabi - yau spaces ) , whose union is isomorphic to the complex structure moduli space of the mirror manifold . it was also shown , that using mirror symmetry as in @xcite or the approach of gauged linear sigma models as in @xcite , one can interpolate smoothly between the topologically distinct points in the khler moduli space . in the fundamental string picture , the geometric and non - geometric phases of calabi - yau manifolds appear as two distinct possibilities , with the latter being thought of as an analytic continuation of the geometric phases in the presence of a non - zero theta angle in the language of @xcite . it was , however , shown in @xcite that the presence of additional open string sectors change this picture considerably , and that the d - brane linear sigma model probes only a part of the full linear sigma model vacuum moduli space , namely the geometric phase . it was explicitly demonstrated in @xcite that d - branes ( in the particular examples of the orbifold @xmath5 and @xmath6 ) ` project out ' the non - geometric phase , in agreement with the results obtained in @xcite . this was found to be generally true for calabi - yau orbifolds of the form @xmath7 @xcite . since the pioneering work of @xcite , which dealt with d - branes on abelian orbifold singularities of the form @xmath8 that was generalised to arbitrary non - abelian singularities of the form @xmath9 ( @xmath10 being a subgroup of @xmath11 ) , d - branes on abelian and non - abelian orbifold backgrounds have been extensively studied in @xcite,@xcite , @xcite,@xcite,@xcite,@xcite . d - branes at orbifolded conifold singularities have been considered in @xcite . ( see also @xcite for an analysis of a blowup of the four - dimensional @xmath12 , @xmath13 orientifold ) . of particular interest has been the applications of methods of toric geometry @xcite,@xcite to the study of d - brane gauge theories on such singularities . in the approach pioneered in @xcite , the matter content and the interactions of the d - brane gauge theory , which are specified by the d - terms and the f - terms of the gauge theory respectively , are treated on the same footing , and the gauge theory information can be encoded as algebraic equations of the toric variety . it is an interesting question to ask , if this procedure can be reversed , i.e , given a particular toric singularity , is there a way of consistently reading off the world volume gauge theory data for a d - brane that probes this singularity . a step in this direction has been recently taken in @xcite . the authors of @xcite have given an algorithm by which the geometrical data encoded in a toric diagram can be used to construct the matter content and superpotential interaction of the d - brane gauge theory that probes it , and have demonstrated this in the cases of the suspended pinch point ( spp ) singularity of the blowup of the @xmath0 orbifold and the various partial resolutions of the @xmath14 orbifold analysed in @xcite . in the present paper , we critically examine this procedure in several other examples . we first consider this algorithm , which we call the inverse toric procedure , for partial resolutions of the orbifold @xmath0 . further , we consider several blowups of the orbifolds @xmath1 and @xmath15 and present results on the extraction of the d - brane gauge theory data from the given toric geometric data . the organisation of the paper is as follows . in section 2 , we first recapitulate the essential details of the construction of the d - brane gauge theory on partial resolutions of the orbifold @xmath16 @xcite,@xcite that will also set the notation and conventions to be used in the rest of the paper and then demonstrate the inverse toric algorithm to construct the gauge theory of a d - brane probing the @xmath17 and the conifold singularities obtained by partially resolving the @xmath0 orbifold . section 3 deals with the orbifold @xmath1 and its various partial resolutions . in section 4 , we consider the example of the orbifold @xmath18 and consider some resolutions of the same . section 5 contains discussions and directions for future work . we begin this section by recapitulating the essential details of the application of toric methods to the construction of d3-branes on @xmath0 and some of its partial resolutions which will also set the notations and conventions to be followed in the rest of the paper . recall that the world volume supersymmetric gauge theory of d - branes probing a toric singularity are characterised by its matter content and interactions . while the former are given by the d - term equations , the latter are specified , via the superpotential , as the f - term equations . these two sets of equations , in conjunction , define the moduli space of the theory . in order to define the d - brane moduli space by toric methods , the standard procedure @xcite is to re - express the f and d - term constraints in terms of a set of homogeneous variables @xmath19 , and a concatenation of these two sets of equations give rise to the toric data . let us briefly review this construction for the example of d-3 branes on the orbifold @xmath20 . we start by considering the field theory of a set of d-3 branes at a @xmath20 singularity . generically , a single d - p brane of type ii string theory at a point in the orbifold @xmath3 is constructed by considering the theory of @xmath21 d - p branes in @xmath22 and then projecting to @xmath3 where the projection is defined by a combined action of @xmath10 on @xmath22 as well as the d - brane chan - paton index . the unbroken supersymmetry in @xmath23 would then be @xmath24 in the closed string sector while the open string sector will have @xmath12 supersymmetry . in our example , a generator @xmath25 of the discrete group @xmath26 acts simultaneously on the coordinates of @xmath22 and also the d-3 brane chan - paton factors . the surviving fields in the theory are the ones that are left invariant by a combination of these two actions . specifically , denoting the action of the quotienting group on @xmath22 by @xmath27 and that on the chan - paton indices by @xmath28 ( where @xmath29 denotes a regular representation of @xmath25 ) , the surviving components of the scalars @xmath30 that live on the brane world volume are those for which @xmath31 , and the components of the gauge fields @xmath32 that survive the projection are those that satisfy @xmath33 . after imposing these projections , the @xmath12 matter multiplet of the theory consists of the twelve fields @xmath34 where we have denoted the complex bosonic fields corresponding to the position coordinates tangential to @xmath22 by @xmath35 . after projection of the gauge field , the unbroken gauge symmetry in this case is the subgroup @xmath36 of @xmath37 , and , apart from an overall @xmath38 , we have a @xmath39 theory . the total charge matrix is given by @xmath40 with the last row of @xmath41 signifying the overall @xmath38 . the d - term constraints are now given by @xmath42 where @xmath43 is the charge of @xmath44 under the @xmath45th @xmath38 and @xmath46 signifies the matrix indices of the surviving components of @xmath47 . @xmath48 is the condition for the existence of supersymmetric vacua . the superpotential of the theory is given by the expression @xmath49x^3\ ] ] with the vacuum satisfying @xmath50=0 \label{superpot}\ ] ] using the expressions for the @xmath47 s from eq.([nonzerox ] ) in eq . ( [ superpot ] ) , one can derive twelve constraints on the twelve fields @xmath51 , out of which six are seen to be independent , hence the twelve initial fields can be solved in terms of six independent fields . denoting the twelve matter fields collectively by @xmath52 , and the set of six independent fields ( which we take to be @xmath53 ) by @xmath54 , the solution , which is of the form @xmath55 can be encoded in the columns of the matrix @xmath56 , given by @xmath57 the columns of the above matrix are vectors in the lattice @xmath58 , and define the edges of a cone @xmath59 . in order to make the description of the toric variety explicit , we consider the dual cone @xmath60 , which is defined in the dual lattice @xmath61 . the dual cone @xmath60 is defined to be the set of vectors in the dual lattice @xmath62 which have non - negative inner products with the vectors of @xmath59 , i.e @xmath63 the dual cone corresponds to defining a new set of fields @xmath64 , in terms of which the variables @xmath65 are solved , in a way that is consistent with the relation ( [ indep ] ) . in this example , @xmath66 and the dual cone is generated by the columns of the matrix @xmath67 the relationship between @xmath19 and @xmath65 are then of the form @xmath68 thus , there are nine variables @xmath19 in terms of which we parametrize the six independent variables @xmath65 . this parametrization , however , has redundancies , and in order to eliminate these , we introduce a new set of @xmath69 actions on the @xmath19 s , with the condition that under these actions , the fields @xmath65 are left invariant . this can be equivalently described by the introduction of a new set of @xmath38 gauge symmetries , such that the @xmath65 s are gauge invariant @xcite . in our example , since there are nine fields @xmath19 and six independent variables @xmath65 , we introduce a gauge group @xmath39 , such that the @xmath19 s are charged with respect to these , with charges @xmath70 for the @xmath71th @xmath38 . the gauge invariance condition demands that the charge matrix @xmath72 must obey @xmath73 where @xmath74 is the transpose of @xmath72 . from the matrix @xmath75 given above , the charge matrix is given by @xmath76 we now consider the constraints imposed on the original fields @xmath77 via the d - flatness conditions , and impose these conditions in terms of the new fields @xmath19 . the charges of the fields @xmath65 under the original @xmath38 charge assignments are given by the matrix @xmath78 from the relationship @xmath79 , it is clear that in order for the charge assignment @xmath80 for @xmath19 ( corresponding to the original set of @xmath81 @xmath38 symmetries ) to reproduce that for the @xmath65 s , we must have @xmath82 , and one possible choice for the matrix @xmath83 is @xmath84 with the matrix @xmath85 satisfying @xmath86 . the matrix @xmath85 is thus the inverse of the matrix @xmath87 . however , since @xmath87 is a rectangular matrix the usual definition of its inverse ( as in the case of a square matrix ) is not applicable and there are several choices for @xmath85 . in particular , for a rectangular matrix @xmath32 , using the rules of single value decomposition , we can compute the moore - penrose inverse of @xmath32 which is defined in a way that the sum of the squares of all the entries in the matrix @xmath88 is minimised , where @xmath89 is the identity matrix in appropriate dimensions . for any choice of the inverse of @xmath87 , the final toric data will of course be identical , and in this section and the next , we simply use the results of @xcite in defining the matrix @xmath85 . for this case , the matrix @xmath85 is @xmath90 now , multiplying by @xmath91 and concatenating @xmath72 and @xmath92 we obtain the total charge matrix , @xmath93 where the @xmath94 are the fayet - illiapoulos parameters for the original @xmath38 s , there being three independent @xmath94 s because of the constraint @xmath95 . the toric data can now be calculated from the cokernel of the transpose of @xmath96 , which ( after a few row operations ) become @xmath97 now from the dual cone @xmath75 , we can , via the matrix @xmath56 , find expressions for all the twelve initial fields in terms of the fields @xmath19 . these are then used to define invariant variables in the following way @xmath98 which follow the relation @xmath99 . the toric diagram of the orbifold @xmath0 is shown in figure [ 322figs](a ) . let us now consider some examples of the possible partial resolutions of this space . we will discuss three possible partial blowups as in @xcite,@xcite . the first example corresponds to blowing up the orbifold @xmath0 by a @xmath100 parametrised by @xmath101 . this gives the suspended pinch point ( spp ) singularity given by @xmath102 . the spp can be further blown up by introducing a second @xmath100 . one possibility is to parametrise this @xmath100 by @xmath103 ( or @xmath104 ) and the remaining singularity is the conifold @xmath105 ( or @xmath106 ) of figure [ 322figs](b ) . one can also introduce a @xmath100 parametrised by say @xmath107 whence the resulting blowup is a @xmath108 . two possible toric diagrams of the latter are shown in fig . [ 322figs](c ) and fig . [ 322figs](d ) . = 3.5 in let us now come to the description of the inverse toric problem . suppose we are given with a singularity depicted by a toric diagram of the form given in fig . [ 322figs](b ) , [ 322figs](c ) or [ 322figs](d ) . we wish to extract the gauge theory data of a d - brane that probes such a singularity , starting from figure [ 322figs](a ) . this issue was addressed in @xcite , and let us review the basic steps in their algorithm . recall that the toric data for a given singularity that a d - brane probes is given by the matrix @xmath109 which is the transpose of the kernel of the total charge matrix ( obtained by concatenating the f and d - term constraints expressed in terms of the homogeneous coordinates @xmath19 ) . hence , the solution to the inverse problem would imply the construction of an appropriately reduced total charge matrix @xmath110 , the cokernel of the transpose of which is the reduced matrix @xmath111 obtained by deleting the appropriate columns ( corresponding to the fields that have been resolved ) from @xmath109 . however , although @xmath111 is known , there is no unique way of specifying the reduced charge matrix @xmath110 from a knowledge of @xmath111 . further , there is no guarantee apriori that such a @xmath110 , if obtained , would continue to describe the world volume theory of a d - brane that probes the given singularity . the algorithm of @xcite gives a canonical method by which these issues can be addressed . it consists of determining the fields to be resolved by tuning the fayet - illiopoulos parameters appropriately ( so that the resulting theory at the end is still a physical d - brane gauge theory ) and then reading off the the matrix @xmath110 which can be separated into the f and d - terms and thus can be used to obtain the matter content and the superpotential determining the gauge theory . let us discuss this in some more details . the toric diagrams for the partially resolved singularities are obtained by the deletion of of a subset of nodes from the original one . conversely , a given toric diagram of dimension @xmath112 can be embedded into a singularity of the form @xmath113 where @xmath114 . in our case , the singularities mentioned above can be embedded into the toric diagram for the singularity @xmath115 . for example , the spp , and the conifold singularity can both be embedded into the @xmath0 singularity , as is obvious from their toric diagrams . we shall henceforth refer to these singularities , in which we embed the partially resolved ones as the parent singularities . a toric diagram corresponding to a partial resolution can of course be embedded in more than one parent singularity and according to the algorithm of @xcite , one can choose the minimal embedding . we will return to this issue in a while . we now determine the fields to be resolved from the parent singularity in order to reach the given toric diagram , by appropriately choosing the values of @xmath94 . the difficulty that arises here is that given the toric diagram of the singularity that we are interested in , there is no unique way of knowing exactly which fields ( in the toric data of the parent singularity ) have to be resolved in order to reach the toric singularity of interest . given a toric diagram for the parent singularity , a particular resolution , along with the elimination of one or more nodes , might also require a subset of fields from some other nodes to be resolved . hence , we have to carefully tune the fi parameters for a consistent blowup of the parent singularity to the one that we are interested in . we now illustrate this procedure by the example of the blowup of the @xmath0 singularity to the @xmath108 orbifold illustrated in fig . [ 322figs](c ) . this example is , in a sense , straightforward , because from the toric diagram of fig . [ 322figs](c ) , it is clear that this being the @xmath17 singularity , we can directly read off the matter content and the interactions from the standard results for d3-branes on the @xmath17 orbifold @xcite . however , note that this is the simplest example where the inverse procedure of calculating the d - brane gauge theory has to be valid . the @xmath17 orbifold can be thought of as partial resolutions of the orbifolds @xmath0 , or @xmath1 , and the inverse toric procedure must give familiar results for both in agreement with @xcite,@xcite . a lower dimensional toric orbifold can always be embedded in the toric diagram of one of higher dimension , and the inverse toric procedure must consistently reproduce the d - brane gauge theory on the former , starting from the latter data . as we will see , this is indeed the case . we proceed by embedding the @xmath17 singularity in @xmath0 . in order to determine the fields to be resolved , we perform gaussian row reduction on the the matrix @xmath96 to obtain @xmath116 the above set of equations imply the constraints @xmath117 where , for convenience of notation , we have labelled @xmath118 . this set of equations is now solved in terms of the fields that we know would definitely get resolved . from the toric diagram , three such fields are @xmath119 . however , from the relation between @xmath120 and @xmath121 above , it suffices to solve the above set of equations in terms of @xmath122 and @xmath121 , say . this gives the solution set , @xmath123&=&\left[x_1,(2x_9-x_7-\zeta_1-\zeta_2),(x_1 + \zeta_2+\zeta_3 ) , ( x_1+x_7-x_9+\zeta_2),\right.\nonumber\\ & ~&(2x_1+x_7 - 2x_9+\zeta_1 + 2\zeta_2+\zeta_3 ) ( x_1+x_7-x_9+\zeta_1+\zeta_2+\zeta_3),\nonumber\\ & ~&\left.x_7,(x_9- \zeta_1-\zeta_2 ) , x_9\right]\end{aligned}\ ] ] now , from the toric diagram of fig.[322figs](c ) , we choose the fields @xmath124 $ ] to have zero vev i.e these fields continue to remain dynamical . thus , in the above solution set , we impose @xmath125 to obtain @xmath126=\left[0,0,0,(x_9-\zeta_1),0,(x_7-x_9+\zeta_1),x_7 , ( x_9-\zeta_1-\zeta_2),x_9\right]\ ] ] we further set @xmath127 in order to make @xmath128 . hence , @xmath129 . also , @xmath130 . therefore , the fields to be resolved are @xmath131 $ ] , all of which have positive vevs . we now need to obtain the reduced charge matrix for this choice of variables to be resolved . the method of doing this essentially consists of performing row operations on the full charge matrix @xmath96 in eq . ( [ fullq ] ) , such that the columns corresponding to the resolved fields @xmath131 $ ] are set to zero . the reduced charge matrix @xmath110 thus obtained must be in the nullspace of the reduced toric matrix @xmath132 , that can be directly evaluated from the full toric matrix in eq . ( [ fulltoricdata ] ) by removing the columns @xmath133 . let us first discuss the general formalism for achieving this , which can then be easily applied to our present example . consider a general parent theory that has a total charge matrix @xmath96 of dimensions @xmath134 , and suppose we wish to eliminate @xmath41 fields from the parent theory in order to reach the theory of interest . this implies that we have to perform row operations on @xmath96 so that the given @xmath41 columns will have zero entries and thus be eliminated . the latter can be achieved by constructing the nullspace of the transpose of a submatrix that is formed by precisely the @xmath41 columns that need to be eliminated . further , since the toric data @xmath109 was initially in the nullspace of @xmath96 , removal of the @xmath41 columns from @xmath96 to obtain @xmath110 would mean that the reduced toric data @xmath111 with the same @xmath41 columns removed would be in the nullspace of @xmath110 . thus , the expression for @xmath110 is given by @xcite @xmath135 \label{qreduced}\ ] ] where @xmath136 is the transpose of the matrix obtained from @xmath72 containing as its columns , the deleted columns of @xmath72 and @xmath137 is the remaining matrix . in our example , we evaluate @xmath138 directly , by writing @xmath136 as the transpose of the matrix that contains columns @xmath133 of the charge matrix @xmath96 , and @xmath137 being the matrix containing the columns @xmath139 of @xmath96 . this implies that the reduced charge matrix is @xmath140 from this , the reduced charge matrices corresponding to the f and d - terms are @xmath141 denoting the kernel of the charge matrix by @xmath142 , the transpose of its dual is the reduced @xmath143 matrix , @xmath144 now , from eq . ( [ indep ] ) , the charge matrix for the initial fields @xmath77 are given , via the charge matrix for the independent fields @xmath65 by the relation @xmath145 hence , in this case , making use of the identity @xmath146 , the matter content for the d - brane gauge theory on the resolved space is obtained as @xmath147 . from which in this case we obtain the matrix @xmath148 where we have included an extra row corresponding to the redundant @xmath38 . hence , the gauge group is seen to be @xmath149 , with four bifundamentals , in agreement with the result of @xcite . also note that one adjoint field has appeared in our expression . this signals an ambiguity that exists in the toric procedure with regards to handling chargeless fields . consider , for example , the superpotential interaction , which can be determined from the matrix @xmath56 that encodes the f - flatness conditions . the number of columns in the matrix @xmath150 is the number of fields that appear in the interactions and the terms in the superpotential are read off as linear relations between the columns of @xmath151 . in this example , from the matrix @xmath56 , we have the relation @xmath152 , and from the @xmath41 matrix , we can see that each of these have charge zero . hence , we try as an ansatz for the superpotential , @xmath153 where @xmath154 is a field uncharged for both the @xmath38 s . this agrees with @xcite after the identification @xmath155 . as is clear , however , there is an ambiguity in writing the superpotential due to the presence of fields that are chargeless under both the @xmath38 s @xcite . let us also mention that we can carry out the inverse procedure just outlined for the case of the alternative blowup to @xmath156 shown in figure [ 322figs](d ) . this gives exactly the same matter content and superpotential as the case just analysed , in the region of moduli space given by @xmath157 , @xmath158 . + @xmath159 we now discuss the example of the blowup of the @xmath160 singularity to the conifold . proceeding in the same way as above , we now find that in the region of the fi parameters determined by @xmath161 , the fields to be resolved are @xmath162 $ ] , when the fields @xmath163 $ ] are chosen dynamical , i.e continue to have zero vev . therefore , from eq . ( [ qreduced ] ) , we obtain the reduced charge matrix to be @xmath164 note that in this , case , there are no f - terms and we have @xmath165 , moreover , the kernel of this matrix is the @xmath166 identity matrix , i.e @xmath167 , which implies that ( since the dual matrix of @xmath142 is also the @xmath166 identity matrix ) @xmath168 , with the charge matrix given in this case ( by adding the extra column corresponding to the overall @xmath38 ) as @xmath169 in this case , the matrix @xmath138 being zero , there are no f - terms and the entire gauge theory information is contained in the d - term . let us now consider d3-branes on the orbifold @xmath170 . in this case , the bosonic fields corresponding to the d3-brane coordinates tangential to the @xmath22 are @xmath171 matrices , and the components that survive the projection by the quotienting group constitute the @xmath12 matter multiplet , given by the set @xmath172 the f - flatness condition ( [ superpot ] ) in this case imply @xmath173 constraints out of which @xmath174 are seen to be independent . further , there are @xmath175 homogeneous coordinates @xmath176 , corresponding to the physical fields in terms of which the dual cone may be described . the total charge matrix of the homogeneous coordinates @xmath19 is defined by introducing a set of @xmath177 @xmath69 actions that remove the redundancy in expressing the original independent fields in terms of the homogeneous coordinates and a further set of @xmath178 charges expressing the d - term constraints in terms of the @xmath19 . ( the @xmath179th @xmath38 is redundant because of the relation @xmath48 ) . the total charge matrix can then be obtained by concatenating these two sets of charge matrices as in our earlier example , and is given ( with the inclusion of a column specifying the @xmath180 independent fayet - illiopoulos parameters ) by @xcite @xmath181 the toric data is the co - kernel of the transpose of this charge matrix ( when the fi parameters are set to zero ) and is given by @xmath182 the invariant variables are defined by @xmath183 from which it can be seen that @xmath184 = 4.5 in let us now discuss some examples of blowups of this space that will illustrate the procedure outlined in section 2 . we will use the notation @xmath185 in what follows . + @xmath159 our first example is illustrated in figure [ 3221 ] . from the diagram , we see that the field @xmath186 will definitely be resolved . hence , after performing gaussian row reduction on the charge matrix @xmath96 , which gives @xmath187 we use this matrix to solve for all the fields in terms of @xmath188 , the solution set being given by @xmath189&=&\left[x_1,x_2,(2x_1 - 2x_2+x_{14}-\zeta_1 -\zeta_3+\zeta_5),\right.\nonumber\\ & ~&(2x_2-x_{14}+2\zeta_1 + \zeta_2+\zeta_3+\zeta_4 ) , ( x_2+\zeta_1),\nonumber\\ & ~&(x_1-x_2+x_{14}-\zeta_1-\zeta_3),(2x_2-x_{14}+\zeta_1 + \zeta_2+\zeta_3 ) , \nonumber \\ & ~&(x_1+\zeta_1+\zeta_2+\zeta_4+\zeta_5 ) , ( x_1-x_2+x_{14}+\zeta_5),\nonumber\\ & ~&(x_1+\zeta_2+\zeta_5 ) , ( -x_1 + 4x_2 - 2x_{14 } + 3\zeta_1 + 2\zeta_2 + 2\zeta_3+\zeta_4),\nonumber\\ & ~&(x_2+\zeta_1+\zeta_2+\zeta_3 + \zeta_4+\zeta_5 ) , ( x_2+\zeta_1+\zeta_2+\zeta_3+\zeta_4),\nonumber\\ & ~&x_{14},(2x_2-x_{14}+2\zeta_1 + 2\zeta_2+\zeta_3+\zeta_4+\zeta_5 ) , \nonumber\\ & ~&\left.(x_2+\zeta_1+\zeta_2 ) , ( x_2+\zeta_1+\zeta_2+\zeta_3)\right ] \label{sol14}\end{aligned}\ ] ] now , we choose the fields @xmath190 $ ] as the ones that continue to have zero vev , and in eq . ( [ sol14 ] ) set the values of these to zero . we thus obtain conditions on the fi parameters , @xmath191 . in the new solution set obtained with these conditions , we choose @xmath192 which makes @xmath193 . this also imples that @xmath194 ( since @xmath195 ) . we also set @xmath196 to obtain @xmath197 . our choice of @xmath198 implies that @xmath199 . this gives the solution set @xmath189&=&\left[0,(-\zeta_3-\zeta_1),0,0,-\zeta_3 , ( -\zeta_1-\zeta_3),\right.\nonumber\\ & ~&\left . 0,0,0,0,0,-\zeta_1,-\zeta_1,(-\zeta_3-\zeta_1),0,-\zeta_3,0\right]\end{aligned}\ ] ] thus the fields to be resolved are @xmath200 $ ] . hence , from ( [ qreduced ] ) , we obtain the reduced charge matrix , @xmath201 in the usual way , we calculate the kernel of the charge matrix @xmath142 , and the transpose of its dual matrix is @xmath202 from the matrices @xmath203 and @xmath204 , we can read off the matter content and gauge group of the d - brane gauge theory from the matrix @xmath205 this is an @xmath36 gauge theory with @xmath174 matter fields , in agreement with @xcite,@xcite . to calculate the superpotential , we first note that the relations between the columns of the matrix @xmath143 , which can be written as @xmath206 . all these combinations are seen to be chargeless , as is the combination @xmath207 . hence , our ansatz for the superpotential in this case is @xmath208 which agrees with that calculated in @xcite after the identification with their notation , @xmath209 . in terms of the original variables of ( [ 323matter ] ) , this blowup corresponds to giving a vev to the fields @xmath210 . the invariant variables are in this case defined by @xmath211 in terms of these variables , the blown up space is given by @xmath212 . + @xmath159 we now consider the second example which is depicted in figure [ 3222](a ) . in this case , we start from the matrix ( [ grrqred ] ) , and in the solution set of eq . ( [ sol14 ] ) , keeping in mind the toric diagram of the blowup , we select the set of fields @xmath213 $ ] that remain dynamical , with zero vev . this gives the solution @xmath214 . from the resulting solution set , we now select @xmath215 , in order to obtain @xmath216 and after imposing this condition , we make the choice @xmath217 in order for @xmath218 to have zero vev . the final solution is given by @xmath189&=&\left[(\zeta_1+\zeta_4 ) , ( -\zeta_1-\zeta_2-\zeta_3),(\zeta_1+\zeta_2+\zeta_4 ) , ( \zeta_1+\zeta_4),\right .\nonumber\\ & ~&(\zeta_2-\zeta_3),(\zeta_4-\zeta_3),0,(\zeta_1+\zeta_2+\zeta_4),0 , \zeta_2,0,-\zeta_1,\zeta_4 , \nonumber\\ & ~&\left .(-\zeta_1-\zeta_2-\zeta_3),\zeta_2,-\zeta_3,0\right]\end{aligned}\ ] ] hence , the fields to be resolved are @xmath219 $ ] and the dynamical fields are @xmath220 $ ] . we can thus directly evaluate the reduced charge matrix from ( [ qreduced ] ) which is given by @xmath221 defining the invariant variables from eq . ( [ 323matter ] ) in this case as @xmath222 this blownup space can be seen to be the conifold @xmath223 by using the redefined variables in ( [ 323eqn ] ) . the gauge theory living on the d - brane is of course the same as in eq . ( [ d2 ] ) . in terms of the original variables in eq . ( [ 323matter ] ) , this blowup can be shown to correspond to giving vevs to the fields @xmath224 . since there are no f - terms in this case , all the information about the gauge theory is contained in the d - term constraint . + @xmath159 let us now consider the toric diagram given by figure [ 3222](b ) . from the toric diagram of fig . [ 3222](b ) , we choose the fields @xmath225 to have zero vev . these conditions imply that @xmath226 and also @xmath227 . substituting this in the solution set ( [ sol14 ] ) , we obtain @xmath189&=&\left[0,0,0,0,\zeta_1,\zeta_2,0,0,(\zeta_1 + \zeta_3),0,0,\zeta_3,(\zeta_2+\zeta_3),\right . \nonumber\\ & ~&\left . ( \zeta_1+\zeta_3-\zeta_5),0,(\zeta_1+\zeta_2 ) , ( \zeta_1+\zeta_2+\zeta_3)\right ] \end{aligned}\ ] ] now , we have a choice . we put @xmath228 and obtain @xmath229 which also implies that @xmath230 . hence , from the final solution set , we see that the fields that continue to have zero vev and are dynamical are , in this case , given by the set @xmath231 $ ] and the fields that pick up non - zero vev and hence are resolved are @xmath232 $ ] . hence , according to ( [ qreduced ] ) , we obtain the reduced charge matrix as @xmath233 and the dual of the kernel of the charge matrix @xmath138 is @xmath234 as before , in order to calculate the matter content , we use the matrix @xmath235 , from which the matrix @xmath236 , which specifies the matter content and gauge group of the d - brane gauge theory is calculated to be @xmath237 hence , the gauge group is @xmath36 and there are @xmath174 matter fields . using the relations @xmath238 with total charge zero and the chargeless combination @xmath239 , we write the superpotential in this case as @xmath240 where the fields @xmath241 are possible adjoints . the invariant variables ( [ invariables ] ) are defined by @xmath242 with the relations @xmath243 . the remaining singularity is seen from eq . ( [ 323eqn ] ) to be the equation @xmath244 in @xmath22 , which is recognised as corresponding to blowing up the @xmath1 orbifold to the singularity @xmath245 . let us also mention that in terms of the original variables in eq . ( [ 323matter ] ) , this blowup corresponds to giving vevs to the fields @xmath246 . + = 3.5 in @xmath159 we now come to the example shown in figure [ 3222](c ) . in this case , from the gaussian row - reduced matrix of eq . ( [ sol14 ] ) , we choose the set of fields @xmath247 $ ] to have zero vev . this implies that @xmath248 and also that @xmath214 . substituting this in the solution set implies that in addition @xmath249 , and further setting @xmath250 , so that @xmath251 is zero , we obtain the full solution set in this case as @xmath189&=&\left[(-\zeta_2-\zeta_5),(-\zeta_3+\zeta_4+\zeta_5 ) , ( -\zeta_2-\zeta_5),(-\zeta_2-\zeta_5),\right . \nonumber\\&~&(-\zeta_2-\zeta_3),(\zeta_4-\zeta_5),0,(-\zeta_2-\zeta_5 ) , -\zeta_2,0,0 , ( \zeta_4+\zeta_5),\zeta_4 , \nonumber\\ & ~&\left.(-\zeta_3+\zeta_4+\zeta_5),0,-\zeta_3,0\right]\end{aligned}\ ] ] hence , the fields that are dynamical and have zero vev is given by the set @xmath252 $ ] and the set to be resolved is given by @xmath253 $ ] . using this , the reduced charge matrix is calculated from ( [ qreduced ] ) to be @xmath254 from which the f and the d - terms as @xmath255 the dual of the kernel of @xmath138 is given by @xmath256 and the matter content and gauge group of the d - brane gauge theory is obtained as @xmath257 the superpotential is written down , after noting that the field @xmath258 is chargeless , and the relations @xmath152 from the matrix @xmath143 . it is given by @xmath259 where @xmath154 is a possible adjoint . the matter content and the superpotential are exactly the same as in ( [ compare1 ] ) and ( [ compare2 ] ) . this is of course as expected because both describe the world volume theory of a d3-brane on the @xmath260 singularity . this blowup corresponds to giving vevs to the fields @xmath261 . the invariant variables ( [ invariables ] ) are in this case @xmath262 where @xmath263 . the blown up space is , from eq . ( [ 323eqn ] ) , given by the equation @xmath264 in @xmath22 , which describes the blowup to @xmath245 . let us pause to comment here that the toric diagram of figure [ 3222](c ) can be embedded both into that of @xmath1 and @xmath0 ; in fact , this diagram is identical to that in figure [ 322figs](d ) . the fact that either embedding gives correct results for the d - brane gauge theory data shows that the inverse toric procedure is indeed consistent . + @xmath159 finally , let us discuss the toric diagram of figure [ 3222](d ) . which is the conifold obtained by partially resolving the @xmath265 singularity . following the same procedure as outlined in the previous examples , we start from the solution set of ( [ sol14 ] ) and select the fields @xmath266 $ ] which we set to zero . from this , we obtain the solution @xmath267 and @xmath268 . substituting this in the solution set and further setting @xmath192 so that @xmath269 , we finally obtain the full solution set to be @xmath189&=&\left[0,0,0,(\zeta_1+\zeta_2+\zeta_4 ) , \zeta_1,0,\zeta_2,(\zeta_1+\zeta_2+\zeta_3+\zeta_4 ) , ( \zeta_1+\zeta_3),\right . \nonumber\\ & ~&\zeta_2,(\zeta_1 + 2\zeta_2+\zeta_4)(\zeta_1+\zeta_2+\zeta_3+\zeta_4 ) , ( \zeta_1+\zeta_2+\zeta_3 + \zeta_4),\nonumber\\ & ~&\left.(\zeta_1+\zeta_3),(\zeta_1 + 2\zeta_2+\zeta_4),(\zeta_1+\zeta_2 ) , ( \zeta_1+\zeta_2+\zeta_3)\right]\end{aligned}\ ] ] hence , the fields that are dynamical are given by the set @xmath270 $ ] while the rest acquire non - zero vev and hence are resolved , and in this case the reduced charge matrix is obtained from ( [ qreduced ] ) to be @xmath271 this blowup corresponds to the conifold . to see this , we define the invariant variables ( [ invariables ] ) in this case by @xmath272 in this case , from ( [ 323eqn ] ) , we can see that the remaining singularity is given by the conifold @xmath105 . let us also mention that in terms of the original variables in eq . ( [ 323matter ] ) , this blowup corresponds to giving vevs to @xmath273 . we now consider d - branes on the orbifold @xmath18 and some of its blowups @xcite . we will consider a d1-brane at this singularity , and the world volume theory is an @xmath274 sym theory in two dimensions . the method of construction of the moduli space of the d-1 brane is similar to the method outlined in sections 2 and 3 . the unbroken gauge symmetry in this case is ( apart from a redundant @xmath38 ) , @xmath275 . the matter multiplet , after imposing the appropriate projections consist of @xmath276 surviving fields , @xmath277 the f - term constraints of eq . ( [ superpot ] ) , in this case imply that out of the @xmath276 fields , @xmath278 are independent , and the @xmath276 initial fields can be solved in terms of these @xmath278 , and the solutions can be denoted as vectors in the lattice @xmath279 . the dual cone in this case has @xmath280 homogeneous coordinates ( matter fields ) @xmath19 , and the redundancy in the definition of the initial independent coordinates in terms of these can be eliminated by introducing a set of @xmath281 @xmath69 actions . the charge matrix @xmath72 for the @xmath69 actions is given in @xcite . for the matrix denoting the original @xmath38 charges expressed in terms of the homogeneous coordinates @xmath19 , we use the moore - penrose inverse of the matrix defining the dual cone , and concatenate these two matrices to obtain the total charge matrix given in the appendix . the toric diagram is given by the columns of the matrix @xmath282 = 5.5 in in terms of these variables , the space is defined by the surface @xmath290 in @xmath291 . two of its resolutions ( corresponding to cases ( vb and vib of @xcite ) are shown in figure [ 4222 ] . + this singularity has to be analysed in the lines of @xcite in order to determine whether its partial resolutions are realised in the moduli space of the d - brane world volume gauge theory . we leave this issue for future work , and for the moment analyse the blowups of this singularity into lower dimensional orbifold singularities , using the inverse toric procedure . + @xmath159 let us consider the blowup illustrated in fig . [ 4222](b ) . performing gaussian row reduction on the total charge matrix , we find that for the range of the fi parameters @xmath292 , an appropriate initial choice determines the fields that retain zero vev as the set @xmath293 $ ] , while all others have non - zero vev and are hence resolved . the reduced charge matrix is @xmath294 from the kernel of the charge matrix @xmath138 , the transpose of its dual is obtained as @xmath295 from these matrices , we can extract the gauge theory data which is now given by the matrix @xmath296 the gauge group is as expected , @xmath36 with @xmath297 matter fields , and the superpotential is given by @xmath298 from the definition of the invariant variables in eq . ( [ inv422 ] ) , it can be seen that this space is @xmath0 defined by the equation @xmath299 in @xmath300 . in terms of the original fields , it corresponds to giving vevs to the fields @xmath301 and @xmath302 . + @xmath159 finally , we come to the example of fig . [ 4222](c ) . in this case , in the region of moduli space given by @xmath303 , we find that the dynamical fields are given by the set @xmath304 $ ] , while the rest acquire non - zero vev and are resolved . the reduced charge matrix is @xmath305 the matrix @xmath143 is in this case given by @xmath306 which specifies the matter content as a @xmath149 gauge theory with the charge matrix given , as expected , by @xmath307 with a superpotential @xmath308 with @xmath154 being a possible chargeless field . in terms of the original fields in the theory , this resolution corresponds to giving vevs to @xmath309 , and the resulting space is the @xmath260 orbifold @xmath310 that we have discussed earlier . in this paper , we have critically examined the inverse procedure of obtaining the world volume gauge theory data of d - branes probing certain toric singularities and their blowups from the geometric data of the resolution . we have shown that the algorithm of @xcite gives consistent results , for a given singularity that can be reached by partial resolutions of different parent theories . we have explicitly checked several partial resolutions of orbifolds of the form @xmath3 and @xmath4 and found the results to be in agreement with field theory calculations . however , as pointed out in @xcite , in the presence of chargeless fields in the matter multiplet , the procedure can not be used to specify the superpotential interaction uniquely , because of the inherent problem in the handling of such fields by toric methods . we have treated the simplest examples of the resolution of the + @xmath18 singularity in this paper . it would be interesting to investigate the various partial resolutions of the moduli space of d - branes on this singularity the lines of @xcite and determine the gauge theory data thereof . finally , as noted in @xcite , the ambiguities that exist in the inverse algorithm seem to imply that in some cases , different gauge theories , in the infra red limit flow to theories with identical moduli space . we have not dealt with this aspect in the present paper , and it would be an interesting direction for future work . 99 m. r. douglas , d. kabat , p. pouliot , s. shenker , `` d - branes and short distances in string theory , '' nucl . phys . * b485 * ( 1997 ) 85 , hep - th/9608024 . b. greene , y. kanter , `` small volumes in compactified string theory , '' nucl . * b497 * ( 1997 ) 127 , hep - th/9612181 . p. aspinwall , b. greene , d. morrison , `` calabi - yau moduli space , mirror manifolds and space - time topology change in string theory , '' nucl . phys . * b416 * ( 1994 ) 78 , hep - th/9309097 . e. witten , `` phases of @xmath24 theories in two dimensions , '' nucl . * b403 * ( 1993 ) 159 , hep - th/9301042 . m. douglas , b. greene , d. morrison , `` orbifold resolution by d - branes , '' nucl . b506 * ( 1997 ) 84 , hep - th/9704151 . e. witten , `` phase transitions in m theory and f theory , '' nucl . b471 * ( 1996 ) 195 , hep - th/9603150 . t. muto , `` d - branes on orbifolds and topology change , '' nucl.phys . * b521 * ( 1998 ) 183 , hep - th/9711090 . m. douglas , g. moore , `` d - branes , quivers , and ale instantons . '' hep - th/9603167 . c. johnson , r. myers , `` aspects of type iib theory on ale spaces , '' phys . rev . * d55 * ( 1997 ) 6382 , hep - th/9610140 . m. douglas , b. greene , `` metrics on d - brane orbifolds , '' adv . * 1 * ( 1998 ) 184 , hep - th/9707214 . m. douglas , a. kato , h. ooguri , adv . * 1 * ( 1998 ) 237 , hep - th/9708012 . t. muto , `` brane configurations for three - dimensional nonabelian orbifolds , '' hep - th/9905230 . b. feng , a. hanany , y. he , `` z - d - brane box models and nonchiral dihedral quivers , '' hep - th/9909125 . k. oh , r. tatar , `` branes at orbifolded conifold singularities and supersymmetric gauge field theories , '' jhep * 9910i * ( 1999 ) 0319 , hep - th/9906012 m. cvetic , l. everett , p. langacker , j. wang , `` blowing up the four - dimensional z(3 ) orientifold , '' jhep * 9904 * ( 1999 ) 020 , hep - th/9903051 w. fulton , introduction to toric varieties , princeton university press , 1993 . d. cox , alg - geom/9606016 . e. gimon , j. polchinski , `` consistency conditions for orientifolds and d manifolds , '' phys . rev . * d54 * ( 1996 ) 1667 , hep - th/9601038 . b. greene , `` d - brane topology changing transitions , '' nucl.phys . * b525 * ( 1998 ) 284 , hep - th/9711124 . s. mukhopadhay , k. ray , `` conifolds from d - branes , '' phys * b423 * ( 1998 ) 247 , hep - th/9711131 . b. feng , a. hanany , y. he , hep - th/0003085 . c. beasley , b. greene , c. lazaroiu , m. plesser , `` d3-branes on partial resolutions of abelian quotient singularities of calabi - yau threefolds , '' nucl . b553 * ( 1999 ) 711 , hep - th/9907186 . j. park , r. rabadan , a. m. uranga , hep - th/9907086 , `` orientifolding the conifold , '' nucl.phys . * b570 * ( 2000 ) 38 , hep - th/9907086 . a. uranga , `` brane configurtions for branes at conifolds , '' jhep * 9901 * ( 1999 ) 022 , hep - th/9811004 d. morrison , r. plesser , `` nonspherical horizons 1 , '' adv . * 3 * ( 1999 ) 1 , hep - th/9810201 . a. ahn , h. kim , `` branes at @xmath4 singularity from toric geometry , '' jhep * 9904 * ( 1999 ) 012 , hep - th/9903181 . k. mohri , `` d - branes and quotient singularities of calabi - yau fourfolds , '' , nucl * b521 * ( 1998 ) 161 , hep - th/9707012

the formation mechanism of compact objects , neutron stars ( nss ) and black holes ( bhs ) , is an unsolved problem in high - energy astrophysics . a model for the formation of such objects requires to perform physically - motivated simulations of the core - collapse supernova , which is computationally challenging ( see e.g. @xcite ; @xcite ; @xcite ) . another possible way to investigate the formation of nss and bhs is to study the birth and evolution of x - ray binaries ( xrbs ) hosting a bh or a ns accreting from a stellar companion . the orbital parameters , peculiar velocities and galactic position of these binaries directly follow from their evolutionary history , and are affected in particular by the conditions at the moment of compact object formation ( see e.g. @xcite ; @xcite ; ; @xcite ) . the measurement of pulsar proper motions ( see e.g. @xcite ; @xcite ; ; @xcite ) , combined with the study of ns - xrbs ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) , has exposed evidence that some nss receive a low velocity , whereas others a high velocity at formation ( so called _ natal kicks _ , nks ) . the prevailing idea is that nss are formed either in a standard core - collapse supernova ( sn ) , or in a less energetic type of sn expected for star with small cores . the latter can take place either as an electron - capture sn or as an iron core - collapse sn with a small iron - core mass ( @xcite ; @xcite ; @xcite ; @xcite ) . for the case of bhs , observations are rather scarce and patchy , thus it is not yet possible to discriminate between different models of bh formation ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) . in this paper , one of our goals is to investigate whether the observed galactic distribution of xrbs hosting a bh ( bh - xrbs ) can reveal something about how bhs are formed . the main underlying idea is that any offset of a bh - xrb from the galactic plane ( assumed as birth place ) is a signature of some peculiar velocity of the system with respect to the circular galactic motion . the magnitude of such velocity gives clues on the sn mechanism , in particular on the magnitude of the nk at birth ( @xcite ; @xcite ) . the idea of using the galactic position and/or line of sight velocities of a population of xrbs to investigate the formation of compact objects was employed previously for the ns case ( see e.g. @xcite ; @xcite ) . + we covered the topic of bh formation in two previous works . in @xcite , we followed the galactic trajectories of a simulated population of bh - xrbs , and investigated which nk distribution gives rise to the observed @xmath0-distribution of bh - xrbs ( where @xmath0 is the height above the galactic plane ) . the aim was to discriminate between _ high _ and _ reduced _ nks for bhs . high nks are larger than the nk expected in a standard formation scenario for bhs , in which the bh forms via fallback of material on to the proto - ns and the nk is caused by asymmetries in the sn ejecta . in the standard scenario , the nk would conserve the linear - momentum and roughly scale as the nk received by the ns multiplied by the ratio between the mass of the bh and the mass of the ns . we call these kicks as _ reduced _ or _ momentum - conserving _ nks . if the ns receives a nk of the order of @xmath1 km / s , a @xmath2 bh would get a nk of @xmath3 km / s . we define _ high _ nks as @xmath4 km / s . in @xcite , we found that high nks , comparable to ns nks , were required . in @xcite , we combined the information from the kinematics and binary evolution of a subset of bh - xrbs to find evidence both for low and high nks . in this paper , we aim at complementing and extending those previous studies . following up on the work by @xcite and @xcite , @xcite found that the root - mean - square ( rms ) value of the height above the galactic plane of bh - xrbs is similar to that of ns - xrbs , suggesting that bhs could also receive a _ high _ kick at formation , or even one as high as nss . in this work , we develop this idea further . we build synthetic populations of bh- and ns - xrbs and we model their binary evolution and their kinematics in the galaxy , to investigate whether different assumptions on compact object formation ( such as a different distribution for the nk and/or a different amount of mass ejected in the sn ) have an imprint on the observed galactic distribution of bh- and ns - xrbs , and we quantify these effects . furthermore , we will dedicate part of this work to discuss a method we previously employed to calculate the minimum peculiar velocity at birth of individual bh - xrbs ( @xcite ; @xcite ) . the difference of the galactic potential value between the observed position @xmath5 and its projection on to the galactic plane was used to analytically derive a lower limit for the peculiar velocity at birth . this method has been recently challenged by @xcite . we investigate how robust our estimate is , i.e. how close this estimate is to the true value of the minimal peculiar velocity at birth , how this estimate scales with the distance from the galactic centre , and how it varies for different choices of the galactic potential . + the paper is structured as follows . in section [ sec : estimate ] we study our estimate for the peculiar velocity at birth of individual bh - xrbs . in section [ sec : bps ] we build synthetic populations of bh- and ns - xrbs for different assumptions on the compact object formation . in section [ sec : results ] we look at the galactic distributions of these synthetic binaries while investigating how they differ , and inferring which nk distribution fits best the observed galactic position of ns- and bh - xrbs . in section [ sec : discussion ] we discuss our findings and in section [ sec : conclusion ] we draw our conclusions . 0.33 0.33 0.33 xrbs are thought to originate from binary progenitors born in the galactic plane , the birth - place of most massive stars ( @xcite ) . when the compact object forms , the binary typically acquires a peculiar velocity . the mass ejection in the sn imparts a recoil velocity to the binary ; the nk adds up vectorially to this velocity , giving the total peculiar velocity of the binary , @xmath6 . such a systemic velocity adds up vectorially to the local galactic rotation and probably has no preferential orientation . the full 3d velocity is measured only for a handful of bh - xrbs ( see @xcite ) . for these , the integration of the orbit backwards in time can in principle provide an estimate for @xmath6 at birth . however , uncertainty in the distance and differences in the galactic potential can prevent a unique determination of the initial position ( see e.g. @xcite ; @xcite ) . when the full 3d peculiar velocity is not known , one can estimate @xmath6 at birth using a simple model . for an object located at galactic height @xmath0 : the distance from the galactic centre , and @xmath0 : the height above the galactic plane . ] , we expect a trajectory purely perpendicular to the plane to be the one which minimises the initial @xmath6 . in our previous works @xcite and @xcite , we estimated the minimum peculiar velocity at birth of a bh - xrb employing energy conservation along such trajectory , and assuming that the maximum height @xmath0 from the plane is the observed one . we get : @xmath7 } , \label{eq : estimate}\ ] ] where @xmath8 is a model for the galactic potential , @xmath9 is the measured distance of the binary from the galactic centre projected on to the galactic plane , and @xmath0 is the current height above the plane . recently @xcite argued that the difference in the gravitational potential between the observed location and its projection on to the galactic plane is not an accurate estimate of the required minimum peculiar velocity at birth . he suggests that there are always possible trajectories different from a purely perpendicular one which require a lower @xmath10 at birth than the one estimated through equation [ eq : estimate ] to reach the same offset from the galactic plane . we check the validity of our estimate for the peculiar velocity at birth , @xmath11 , for high-@xmath0 sources , performing a monte carlo simulation using the python package for galactic dynamics ( @xcite ) . we simulate @xmath12 points , whose initial conditions are set as follows : 1 ) the initial position is at @xmath13 , where @xmath14 is uniformly distributed between @xmath15 and @xmath16 kpc ; 2 ) the initial peculiar velocity @xmath17 is uniform between @xmath15 and @xmath18 km / s ; 3 ) the orientation of this velocity is uniformly distributed over a sphere . we note that since we are only interested in the minimum value of @xmath17 , the shape of the assumed velocity distribution is not important . we add the circular motion in the galactic disc to the 3d peculiar velocity @xmath6 . we integrate the orbits in the galaxy for @xmath19 gyr , using a 4th - order runge - kutta integrator , and we check for conservation of energy over the trajectory making sure that the relative error on the energy is less than @xmath20 at the final step . we record the positions @xmath21 @xmath18 times over the orbit sampling from constant time steps , along with the initial peculiar velocity @xmath17 . from the simulated points , we select only those ones located at @xmath22 at the sampled times , to represent high-@xmath0 sources . we perform the simulation for three different choices of the galactic potential : model 2 of , @xcite , and the potential from @xcite , which are all multi - component potentials consisting of disc , bulge , and halo . the @xcite potential is made up of two miyamoto - nagai potentials for disc and bulge , and one pseudo - isothermal potential for the halo . the @xcite potential is made up of a power - law density profile with an exponential cut - off for the bulge , a miyamoto - nagai potential for the disc , and a navarro - frenk - white profile for the halo . the potential is composed of two miyamoto - nagai potentials and a wilkinson - evans potential for the halo . we show the rotation curve of each of the three potentials in figure [ fig : rotcurve ] . is the potential used by @xcite ; @xcite is the one we adopted in @xcite ; the is a realistic model for the milky way potential favoured by @xcite . we present the results of this simulation in figure [ fig : density ] . the red line is our estimate for the peculiar velocity taking @xmath23 kpc in equation [ eq : estimate ] and it follows the lower edge of the simulated points . figure [ fig : density ] shows that our analytical estimate ( eq . [ eq : estimate ] ) successfully describes the value and trend of the minimal peculiar velocity as a function of the galactocentric distance . in order to better quantify the goodness of our estimator @xmath11 , we compute the ratio @xmath24 using @xmath25 kpc - wide bins in @xmath26 , for those points which reach a height above the galactic plane along their orbit in the range @xmath27 kpc . the velocity @xmath17 is the actual initial peculiar velocity which we showed in figure [ fig : density ] . we plot @xmath28 in figures [ fig : ratio1 ] , [ fig : ratio2 ] , [ fig : ratio3 ] , for the three different potentials . @xmath11 is an excellent estimator for @xmath29 kpc , since at these radii @xmath28 is equal or greater than @xmath25 . it is less robust in the inner part of the bulge for the @xcite and potentials , but not in the potential , that is fit to the most recent dynamical constraints on the milky way and has a more realistic bulge model ( jo bovy , private communication ) . in the bulge region , our estimate is steeper than the real minimal peculiar velocity for the first two potentials , i.e. , it varies strongly for small variation in @xmath26 . this can be seen in figure [ fig : andrei ] , where for every position @xmath21 we show as a density map the real minimal peculiar velocity at birth necessary to reach that position . we integrated @xmath30 orbits for @xmath19 gyr and using as potential the one in . the contour lines show our analytical estimate @xmath11 ; the discrepancy between the two velocities is evident in the inner region of the galaxy . figures [ fig : ratio1 ] , [ fig : ratio2 ] , [ fig : ratio3 ] also show an increase of the average value of @xmath28 with larger distances @xmath26 . this is an artefact caused by our choice of the @xmath17 initial distribution ( uniform between @xmath31 km / s ) , as the numerator in the ratio @xmath28 can take all the values between @xmath32 and @xmath18 km / s . from our extensive analysis , we find that the estimate @xmath33 accurately represents the real minimal value for the peculiar velocity at distances from the galactic centre @xmath34 kpc , and can be safely applied to estimate the peculiar velocity at birth of xrbs born in the galactic plane . of simulated points which reach that position @xmath21 . the contour lines show our analytical estimate @xmath11 at that position . the estimate differs strongly from the real value in the bulge region ( each solid line differs by @xmath35 km / s from the closest - neighbouring one ) . the potential used is from . ] the estimate @xmath11 is a function of the potential used , in particular in the bulge , as can be seen in figure [ fig : difference ] , where we show @xmath11 for the @xcite , and @xcite potentials , and assuming @xmath23 kpc in eq . [ eq : estimate ] . additionally , from figures [ fig : ratio1 ] , [ fig : ratio2 ] , [ fig : ratio3 ] , we note that the fraction of systems with @xmath36 in the region @xmath37 $ ] kpc also strongly depends on the potential . the minimum values @xmath38 are : @xmath39 , @xmath40 , @xmath41 for @xcite , , @xcite potential respectively , where these lower limits are defined such that @xmath42 of the points in the same bin have a value larger than the lower limit . figure [ fig : density ] also shows that the galactic bulge ( @xmath43 kpc ) is much less populated ( an order of magnitude fewer systems than in regions at larger distance from the galactic center ) . there are two reasons for this : i ) the bulge volume is small ; ii ) it is unlikely for a binary born in the galactic disc to overcome the strong potential well in its motion towards the galactic bulge . the inaccuracy of our analytical estimate in the bulge region affects only the source h1705 - 250 , which is the only bh - xrb located close enough to the galactic centre ( see table [ tab : systems ] ) , at @xmath44 kpc ( @xcite ) . without a measurement of its 3d peculiar velocity , it is impossible to discriminate between a birth in the disc or a birth in the bulge ( hence close to its observed position ) . more in general , bulge sources are not suitable for estimating the peculiar velocities at birth , since the current view on bulge formation is that it was not formed in situ . the bulge population is thought to come from the disc through dynamical instabilities ( @xcite ) , with most of its mass coming from major and minor merger events with satellite galaxies ( @xcite ) . + we compute the minimum peculiar velocity at birth for the seven short - period bh - xrbs studied by @xcite , using the three galactic potentials ( see table [ tab : minvpec ] ) . we add to this sample two other short - period bh - xrbs which we did not consider in @xcite ( xte j1650 - 500 and xte j1859 + 226 ) , due to the lack of a strong constraint on the bh mass ( @xcite ) . for h 1705 - 250 , we put in parenthesis the velocity @xmath11 multiplied by the factor @xmath28 found above . c c c c + source & bovy & pac . & irrgang & repetto et al . + & & & & 2015 + xte j1118 + 480 & 62 & 70 & 68 & 72 + gro j0422 + 32 & 20 & 25 & 22 & 25 + grs 1009 - 45 & 34 & 40 & 37 & 41 + 1a 0620 - 00 & 8 & 10 & 8 & 10 + gs 2000 + 251 & 12 & 15 & 12 & 15 + nova mus 91 & 44 & 51 & 46 & 52 + h 1705 - 250 & 259 ( 262 ) & 363 ( 158 ) & 350 ( 186 ) & 402 + xte j1650 - 500 & 17 & 21 & 16 & - + xte j1859 + 226 & 61 & 68 & 68 & - + we have found an error in the halo component of the @xcite potential that we used for the computation of @xmath11 in @xcite . this mostly affects the bulge source h 1705 - 250 , whereas the other six sources are not greatly affected ( compare third and last column in table [ tab : minvpec ] ) . accounting for the thickness of the galactic disc instead of assuming a birth place at @xmath45 does not significantly affect the minimal peculiar velocity ( see @xcite ) . @xcite used the source h 1705 - 250 to conclude that the difference in the galactic potential between the observed position and the projection of this position on to the galactic plane is not a conservative estimate of the minimal initial velocity of the binary . they show an example of a trajectory for h 1705 - 250 which starts from the galactic plane and ends at the observed position for an initial velocity of @xmath46 km / s , lower than the value provided by eq . [ eq : estimate ] ( see table [ tab : minvpec ] ) . we agree with his conclusion , but only as far as sources close ( or in ) the bulge are concerned . on the contrary , for sources located at @xmath47 kpc , our analytical estimate perfectly matches the real minimal velocity . in @xcite we used the high minimal velocity at birth for xte j1118 + 480 and h 1705 - 250 to claim that at least two out of the seven bh - xrbs we considered were consistent with a high ( or relatively high ) nk . this holds true with our current revision of the minimal velocities at birth , and we find another bh - xrb that is potentially consistent with a relatively high nk : xte j1859 + 226 . the velocities we have been dealing so far with are _ minimal _ velocities , and do not necessarily correspond to _ expected _ ( _ realistic _ ) velocities . in what follows , we study the latter . for the peculiar velocity at birth as a function of the distance from the galactic centre @xmath26 ( projected on to the galactic plane ) for the three different galactic potentials used in this work : @xcite ( dashed line ) , @xcite ( dotted line ) , ( solid line ) . we assumed @xmath23 kpc . ] in this part of the work , instead of dealing with the minimal peculiar velocities , we deal with the expected peculiar velocities . we perform a binary population synthesis study of bh- and ns - xrbs , starting just before the bh / ns formation , varying the conditions at the formation of the compact object . the goal is to investigate the impact that different bh and ns formation assumptions have on the galactic distribution of xrbs containing a ns or a bh . we assume that the binaries are formed in the galactic thin disc , where most of the massive stars reside ( @xcite ) . in this study , we do not account for the possibility that a few systems could have been formed in the halo ( i.e. in star clusters that have now been dissolved ) , and neither of the possibility that a few systems could have been ejected from globular clusters ( gcs ) via n - body interactions . gcs seem to be very efficient in producing ns low - mass x - ray binaries ( ns - lmxbs ) , as @xmath48 of all ns - lmxbs are found in globular clusters , which contain only @xmath49 of all the stars in the galaxy ( @xcite ) . such an investigation is , however , outside the scope of this paper . we take different models for the formation of the compact object . the nk is drawn either from a maxwellian distribution peaked at @xmath50 km / s ( with @xmath51 km / s ) representing a low - nk , or from a maxwellian distribution peaked at @xmath52 km / s ( with @xmath53 km / s ) representing a high - nk . we assume a certain amount of mass ejection in the sn , @xmath54 . bhs are thought to form either via prompt collapse of the progenitor star or via partial fallback of the sn ejecta onto the proto - ns ( see @xcite ) . in our models , the progenitors of bhs either do not eject any mass at collapse , or they eject @xmath55 . stars with a zams - mass larger than @xmath56 are thought to leave a bh behind ( see e.g. @xcite ; @xcite ) . for a progenitor of mass @xmath57 , the helium core mass ( which collapses into a bh ) is between @xmath58 ( @xcite ) , which motivates our ( conservative ) choice for @xmath54 . for the previous models , we assume a bh mass of @xmath59 ( which is the typical mass for bhs in our galaxy ; @xcite ) . we also picture a higher - mass helium star ( @xmath60 ) which directly collapses into a bh with no mass ejection . for nss , the ejected mass is calculated as : @xmath61 , where @xmath62 is the helium core mass ( @xmath63~m_\odot$ ] , see @xcite ) , and @xmath64 . for the bh case the models are : * model 1 : high nk , @xmath65 , @xmath66 ; * model 2 : low nk , @xmath65 , @xmath66 ; * model 3 : high nk , @xmath65 , @xmath67 ; * model 4 : low nk , @xmath60 , @xmath66 . for the ns case the models are : * model 5 : high nk , @xmath54 uniform between @xmath68~m_\odot$ ] ; * model 6 : low nk , @xmath54 uniform between @xmath68 ~m_\odot$ ] . for all the models , we simulate @xmath69 binaries composed of the helium star ( which core - collapses ) and a companion star of @xmath70 . the pre - sn orbital separation is uniformly drawn in the range @xmath71 with zero initial eccentricity , where @xmath72 is the minimal orbital separation such that either one of the two components fills its roche lobe . we calculate the effect of the compact object formation on the orbital properties and on the kinematics of the binary ( for more details on the method , see @xcite ) . in particular , the effect of the mass ejection together with the nk impart a peculiar velocity to the binary : @xmath73 where @xmath74 is the total mass of the binary after the sn , @xmath75 is the magnitude of the nk , @xmath76 its component along the orbital speed of the bh progenitor , and @xmath77 is the _ mass - loss kick _ : @xmath78 the recoil the binary gets because of the instantaneous mass ejection @xmath54 ( @xmath79 is the initial mass of the binary ; @xmath80 is the mass of the companion ; @xmath81 is the initial orbital separation ) . we follow the evolution of the binaries under the coupling between tides and magnetic braking using the method developed in @xcite , and select those systems that start mass transfer ( mt ) , i.e. become x - ray sources , while the donor is on the main sequence . we choose the radial distribution of the binaries to follow the surface density of stars in the thin disc : @xmath82 , with @xmath83 kpc ( @xcite ; @xcite ) , and with a maximum distance from the galactic centre of @xmath84 kpc . concerning the height above the plane , we model it as an exponential with scale height @xmath85 equal to the scale height of the thin disc ( @xmath86 kpc ; @xcite ) . this is a conservative choice for the scale height , being the scale height of massive stars in the disc typically smaller ( @xmath87 pc ; see table 4 in @xcite ) . we assume that the stars follow the galactic rotation , with no additional component . various mechanisms can heat up the stars in the disc , increasing their dispersion velocity , such as encounters with spiral density waves , giant molecular clouds , and various other forms of stochastic heating ( @xcite ; @xcite ; ; @xcite ) . , using a large sample of late - type dwarfs in the milky way disc , measured a dispersion in the three velocity - components of @xmath88 km / s , @xmath89 km / s , @xmath90 km / s at @xmath91 gyr ( see also ) . we neglect this influence , as we expect that for low - mass stars hosted in ( massive ) binaries these velocity would be significantly lower . we integrate the orbit of the binaries for @xmath19 gyr using the potential from @xcite , which is a realistic model for the milky way potential . we record the position along the orbit every @xmath19 myr after @xmath25 gyr . using the catalogue of , we classify the systems into three main groups : 1 . short - period , dynamically confirmed bh - xrbs ( 9 systems ) ; 2 . short - period , dynamically confirmed bh - xrbs + short - period bh candidates ( 15 systems ) ; 3 . short- and long - period , dynamically confirmed bh - xrbs ( 12 systems ) , which we list in table [ tab : systems ] , along with their galactic position @xmath5 derived from their sky - position and distance . dynamically - confirmed bhs are those for which a dynamical measurement of the bh mass is available ( see e.g. @xcite ) . * 3|l| name & r & @xmath92 + & & + + xte j1118 + 480 & @xmath93 @xmath94 & @xmath95 @xmath96 + gro j0422 + 32 & @xmath97 @xmath98 & @xmath99 @xmath100 + grs 1009 - 45 & @xmath101 @xmath102 & @xmath103 @xmath104 + 1a 0620 - 00 & @xmath105 @xmath106 & @xmath107 @xmath108 + gs 2000 + 251 & @xmath109 @xmath110 & @xmath111 @xmath112 + nova mus 91 & @xmath113 @xmath96 & @xmath114 @xmath104 + h 1705 - 250 & @xmath115 @xmath116 & @xmath117 @xmath118 + xte j1650 - 500 & @xmath119 @xmath120 & @xmath121 @xmath122 + xte j1859 + 226 & @xmath123 @xmath124 & @xmath125 @xmath126 + + xte j1550 - 564 & @xmath127 @xmath128 & @xmath129 @xmath130 + grs 1915 + 105 & @xmath131 @xmath132 & @xmath133 @xmath134 + gs 2023 + 338 ( v404 cyg ) & @xmath135 @xmath136 & @xmath137 @xmath138 + + maxi j1836 - 194 & @xmath139 @xmath140 & @xmath141 @xmath142 + maxi j1659 - 152 & @xmath143 @xmath144 & @xmath145 @xmath146 + xte j1752 - 223 & @xmath147 @xmath148 & @xmath149 @xmath150 + swift j1753.5 - 0127 & @xmath151 @xmath126 & @xmath152 @xmath153 + 4u 1755 - 338 & @xmath154 @xmath155 & @xmath156 @xmath102 + grs 1716 - 249 & @xmath157 @xmath158 & @xmath159 @xmath130 + the observed bh - xrbs are both long ( @xmath160 day ) and short orbital period ( @xmath161 day ) , thereby originating from different evolutionary paths . hence , in order to compare the observed systems with the simulated binaries , we need to produce two separate synthetic population of binaries , one population with short - period and one population with long - period , to which we compare the observed binaries according to their type . for the short - period binaries , we follow the binary evolution of simulated binaries using the method we explained in section [ sec : bps ] . for the long period ones , which are driven by the nuclear evolution of the donor , we model them assuming the post supernova orbital separation to be such that @xmath162 , where @xmath163 is the circularised orbital separation and @xmath164 is the eccentricity in the post - sn configuration . this assumption is based on the fact that long - period binaries evolve to longer and longer period during the mt phase , hence : @xmath165 , where @xmath166 is the orbital separation at the onset of mt , and @xmath167 is the observed orbital separation . the assumptions on the compact object formation are the same as for the short - period binaries , as well as the masses of the binary components . since our simulated binaries have a companion mass of @xmath70 , we exclude from the observed sample those binaries with a companion mass : @xmath168 ( gro j1655 - 40 , 4u 1543 - 475 , and sax j1819.3 - 2525 ) . we account for a possible observational bias on the dynamically - confirmed bh - xrbs . in order to get a dynamical measurement of the bh mass , hence fully confirming the nature of the source , high signal - to - noise optical spectra are required ; this might be prevented in regions of high extinctions , i.e. in and close to the galactic plane . we then remove from our simulated populations those binaries which are located at @xmath169 kpc . we note that the lowest @xmath0 in the sample of short - period dynamically confirmed bh binaries is for 1a 0620 - 00 ( @xmath170 kpc ; see table [ tab : systems ] ) . for the long - period binaries , we exclude from the study the sources grs 1915 + 105 ( donor spectral type : k1/5 iii ) and v404 cyg ( donor spectral type : k0 iv ) , which are located at @xmath171 kpc and @xmath172 kpc respectively ( see table [ tab : systems ] ) . these two systems do have a dynamical measurement of the bh mass ( see @xcite ) . in figure [ fig : spectype ] we plot the absolute value of the height @xmath0 versus the spectral type and luminosity class of the @xmath173 to which we add the three bh - xrbs with an intermediate - mass companion . ] dynamically - confirmed bh - xrbs ( the spectral types are from ) . at small @xmath0 , stars have an earlier spectral type and/or are giants or sub - giants . whereas ms / dwarf stars tend to be seen at larger distances above the plane . and the spectral type and luminosity class of the 15 dynamically - confirmed bh - xrbs . when the spectral type of the donor star in the system is not univocally identified , we indicate the range of possible types . ] the only long - period binary in our sample , after removing those sources close to the galactic plane , is xte j1550 - 564 , which has a current orbital separation of @xmath174 , consistent with our assumption on @xmath163 . the galactic population of ns - xrbs consists of more than @xmath175 objects ( see @xcite and references therein ) . for our study , we select the @xmath176 ones with a short - orbital period ( @xmath177 day ; see table 2 in @xcite ) . the identification of a ns - xrb typically occurs via the detection of x - ray bursts which ignite on the surface of the ns . therefore , unlike for bhs , there are potentially no biases against the identification of such systems . the scale height of bh- and ns - xrbs is a proxy of the effect of different compact object formation mechanisms on to the galactic distribution of the binaries . we quantity the scale height of the binaries as the rms of their height @xmath0 as a function of @xmath26 for all points . to plot the results , we bin the systems into @xmath25 kpc - wide bins in the @xmath26-direction . we show the results in figure [ fig : zrms ] for the six models . the monotonic rise of @xmath178 is expected , since the galactic potential becomes weaker further away from the galactic centre , and the binary moves further up for the same initial velocity it is interesting to note that if bhs and nss receive the same nk , they would still show a different scale height , with nss reaching larger distances from the galactic plane ( compare black solid line with grey solid line , and black dashed line with grey dashed line ) . this is due to the fact that for the same linear momentum , a binary with a larger mass receives a lower @xmath17 ( as is shown in figure [ fig : vsys ] ) . if the progenitor of the bh ejects mass at core - collapse as in model 3 ( see black dashed - dotted line in figure [ fig : zrms ] ) , it will move further out from the plane than when no mass is ejected , since the mass ejection adds an extra contribution to @xmath17 . furthermore , @xmath17 does not depend on the mass of the bh when no mass is ejected at bh formation ( black dashed and black dotted lines in figure [ fig : zrms ] ) , since it scales as @xmath179 , for low - mass companion stars ( see equation [ eq : pec ] ) . in figure [ fig : vsys ] we also show as arrows the lower limits on the peculiar velocity at birth of the 9 bh - xrbs we studied in section [ sec : effect ] . it is clear that a high - nk distribution ( darker - grey solid line ) more easily accounts for the higher - velocity systems , as @xmath180 systems lie in or beyond the high - velocity tail of the distribution corresponding to the low - nk model . @xcite found a similar @xmath178 between ns- and bh - xrbs and deduced that bhs should receive nks too , unless differences in the binary evolution and observational biases were strong . we confirm that accounting for binary evolution does not strongly change the galactic distributions of bh- and ns - xrbs . however , the scale height does strongly depend on the position in the disc . we compute the @xmath178 of the observed bh - xrbs , both of the whole sample and of the dynamically - confirmed systems only . we find a rms of @xmath181 kpc and @xmath182 kpc respectively . for the short - period ns systems , we calculate a @xmath178 of @xmath183 kpc , when excluding the source xte j2123 - 058 since its velocity is consistent with being a halo source , as @xcite noted . the error on these @xmath178-values accounts for the uncertainty on the distance to the sources . in figure [ fig : space ] we show the galactic distribution of ns and bh systems ( the lines account for the uncertainty in the distance to the source ) . the result that ns systems should have a larger scale height than bh systems is consistent with what the observed populations show . above the galactic plane of simulated bh- and ns - xrbs for the different models used . ] ( after the formation of the compact object ) of bh - xrbs in model 1 ( black solid line ) and model 2 ( black dashed line ) , and of ns - xrbs in model 5 ( grey solid line ) and model 6 ( grey dashed line ) . the dotted and dotted - dashed dark - grey lines are variations of model 1 ( see section [ sec : discussion ] for details ) . the arrows represent the lower limits on the peculiar velocity at birth for the 9 short - period bh - xrbs using the potential from @xcite . ] is the distance from the galactic center projected on to the plane , and @xmath0 is the height above the plane . one ns - xrb falls off the figure : xte j2123 - 058 . for each source , the line accounts for the uncertainty on the distance . we also show the results from the population study in terms of @xmath178 as a function of @xmath26 : model 1 ( grey lines ) , model 2 ( grey - dashed lines ) , model 3 ( grey - dotted lines ) . ] in the models we used in section [ sec : bps ] , the orbit of the binary progenitors of bh- and ns - xrbs was chosen to be uniformly distributed in the range @xmath184~r_\odot$ ] . it could be that this choice biases our results towards certain values for @xmath17 . to test this , we check how the distribution of the initial orbital separation of the binaries ( i.e. prior to the formation of the compact object ) varies with the magnitude of the nk and of @xmath17 . from figure [ fig : presn ] , it is clear that the majority of the initial orbital separations are constrained to lie within a small range ( @xmath185 ) both for ns and bh systems , and both for high and low nks . furthermore , there is no clear trend of @xmath17 with respect to @xmath186 . we hence conclude that it is unlikely that the peculiar velocities @xmath17 would be very much influenced if the pre - sn orbits had a distribution different from the uniform one we use in our study , or if they were drawn from a smaller range . and the orbital separation @xmath186 of bh- and ns - xrbs prior to the formation of the compact object . each panel corresponds to different assumptions on the nk . the fraction of systems in each 2-dimensional bin is shown ; darker colours correspond to a larger fraction of systems.,title="fig : " ] and the orbital separation @xmath186 of bh- and ns - xrbs prior to the formation of the compact object . each panel corresponds to different assumptions on the nk . the fraction of systems in each 2-dimensional bin is shown ; darker colours correspond to a larger fraction of systems.,title="fig : " ] we now turn to the comparison of the different models with the observed bh - xrbs . in order to compare the simulations with the observed systems , we note that every subgroup of bh binaries of table [ tab : systems ] gives rise to a certain 2d distribution in @xmath26 and @xmath0 . one way of proceeding would be to compare the 2d simulated distribution with the 2d observed one . we compare the data with the simulated populations dividing the galaxy into @xmath25 kpc - wide bins along the @xmath26-direction . this allows to account for the fact that the galactic potential is a strong function of the position in the disk , as we showed in section [ sec : effect ] . for every @xmath26-bin , we compute the cumulative distribution function ( cdf ) of the height @xmath0 above the galactic plane based on the population synthesis results within model 1 and model 2 ( see as an example black and grey lines in figure [ fig : intersection ] , for the bin : @xmath187 $ ] kpc ) . then we calculate where in the cumulative distribution the observed systems lie ( see as an example the intersection between the blue vertical lines and the cdfs in figure [ fig : intersection ] ) . in such a way , we obtain a list of percentiles . if the model is correct , we expect these percentiles to be drawn from the uniform distribution . we note that we have removed from our comparison those sources located in the bulge of the galaxy ( i.e. h 1705 - 250 and maxi j1659 - 152 ) , which could have had a different origin rather than having formed in the plane ( see section [ sec : effect ] ) . we plot the cumulative distribution of these percentiles in figures [ fig : obs1 ] ( short - period confirmed bh - xrbs ) , [ fig : obs2 ] ( short - period confirmed + candidates ) , and [ fig : obs4 ] ( whole sample ) . in the figures , the solid lines correspond to model 1 and the dashed lines correspond to model 2 . we also consider a model which consists of a superposition of model 1 and model 2 in equal parts ( see thin solid in figure [ fig : obs1 ] , in the case of the short - period confirmed bh - xrbs ) . the model which fits best is the one which comes closer to the diagonal line ( that represents the cumulative of a uniform distribution ) . in all three cases , a high nk distribution is the most preferable one . we perform a kolmogorov - smirnov ( ks ) test to measure how close is the distribution of percentiles to the diagonal line of figures [ fig : obs1 ] , [ fig : obs2 ] , [ fig : obs4 ] . we summarise the @xmath188-values and their corresponding probabilities in table [ tab : dvalues ] for every subgroups of bh - xrbs . for each of the sub - groups the high - nk model fits the data best , although in the group with confirmed bhs only , the low - nk is also consistent with the data . for the confirmed+candidate short - period systems as well as for the whole sample , the low - nk model is inconsistent . interestingly , the model in which the bhs receive both low and high nks , fits the data best for the confirmed systems . in these results , we have excluded all the systems in the plane ( both observed and simulated ) . an accurate modelling of the obscured systems would require a model for the galactic extinction in and out of the plane combined with a model for the optical / nir magnitudes of bh - xrbs in their quiescent state . as a first step , we simplistically model the observational effects near the galactic plane including a certain fraction of those simulated points which end up in the galactic disc ( at @xmath189 kpc ) : either @xmath190 , or @xmath191 , or @xmath192 . we compare the galactic distribution of these simulated binaries with the distribution of the whole sample of binaries , including this time the obscured sources grs 1915 + 105 and v404 cyg as well . the results are presented in figure [ fig : obs5 ] and table [ tab : dvalues ] . also when including the obscured systems , the high - kick model is the most successful in reproducing the observed binaries . [ cols="<,^,^,^,^,^ " , ] the distance @xmath193 to a bh - xrb is typically estimated by measuring the apparent magnitude of the companion star in a certain colour band , and computing its absolute magnitude . once an estimate of the reddening towards the source is known and the spectral type of the donor star is clearly identified , the distance can be calculated . in the best case scenario , one would have the apparent magnitude of the source in different bands , and then would compute the scatter between the derived distances as estimate of the distance uncertainty . we expect such uncertainties to follow a gaussian distribution . however , in case a range of spectral types is equally probable , we expect the errors on the distance to be distributed more uniformly . to investigate the influence of the uncertainty in the distance , and since for most of the literature there is no easy way of determining the type of error distribution , we randomly generate 100 values for the distance to each bh - xrb , either distributed as a gaussian ( with @xmath194 equal to the distance uncertainty @xmath195 ) or as a uniform distribution in the range @xmath196 . such errors can cause a binary to move from one @xmath26-bin to the adjacent one , affecting the percentile values . however , we find that there is no systematic shift that would make low nks fit best the observed data , @xmath195 being smaller than the discrepancy between the two distributions . for model 1 ( black lines ) and model 2 ( grey lines ) , in the bin @xmath187 $ ] kpc . solid lines correspond to the whole sample of simulated binaries ; dashed lines correspond to the remaining part of the sample after the exclusion of systems close to the galactic plane , i.e. @xmath169 kpc . the blue vertical lines represent the observed @xmath92 of 3 bh - xrbs ( from left to right : 1a 0620 - 00 ; grs 1009 - 45 ; xte j1118 + 480 ) . ] ) . the model which fits best the observed data is the one closer to the diagonal line . ] ( black lines ) , @xmath191 ( darker grey lines ) , or @xmath192 ( lighter grey lines ) . the model which fits best the observed data is the one closer to the diagonal line . ] we compare the observed @xmath0 distribution of ns systems with the distribution of the two simulated population of ns - xrbs in the context of model 5 and model 6 . we perform the comparison in the same way we did for bh - systems in section [ s : rules_for_cumul ] . from figure [ fig : obsns ] we see that none of the distributions ( solid and dashed lines ) fits the data . our goal is not to calibrate the ns nk distribution from the ns - xrb population , nor from a population model of radio pulsars ( cf . ) . nevertheless , we can note that the observed population of ns - xrbs seems to be consistent with nks larger than @xmath197 km / s . this is in line with the catalogue of pulsar proper motions by @xcite , who inferred a mean pulsar birth velocity of @xmath198 km / s . however , the derivation of pulsar velocities from the measured proper motions has to be taken with caution , because of the possible uncertainties in the proper motion measurements as well as in the distance measurements . more in general , underestimating proper motion measurement errors can lead to an overestimate of pulsar velocities , as noted by . the distance to a pulsar is typically estimated through parallax . showed that a more proper bayesian approach to calculate the distance probability function from a single parallax measurement has to be used . such method has not been applied yet to the whole population of pulsars . we show the results of the ks - test for ns systems in table [ tab : dvalues ] : both models have large @xmath188-values . for an illustrative purpose , we also compare the observed population of ns - xrbs to a simulated one in which the nk is drawn from a maxwellian distribution with @xmath199 km / s ( @xcite ) . the results of the ks test favours this distribution : @xmath200 ; see dotted line in figure [ fig : obsns ] . we note that we did not include the long - period ns - xrbs to our study as in the sample of ns - xrbs from @xcite that we are using , there is only one long - period system with a low - mass companion , cygnus x-2 . 1 . in our models of section [ sec : bps ] , we have assumed an ejected mass at bh formation of @xmath15 or @xmath55 . taking @xmath201 would not greatly affect the scale height of the binaries in the case of a high nk distribution , since the typical peculiar velocities are comparable to the case when @xmath55 are ejected ( see dotted line in figure [ fig : vsys ] ) . this is due to the fact that a high ejected - mass is compatible only with the lower - velocity tail of the nk distribution in order for the binary to stay bound . in the case of a low nk distribution , the higher ejected - mass has a greater effect on the average peculiar velocity ( see dashed - dotted line in figure [ fig : vsys ] ) . the ns systems velocities are still larger . in order for bh systems to have the same peculiar velocity as ns systems , we would need that @xmath202 , which follows from the expression for @xmath17 ( equation [ eq : pec ] ) , imposing that @xmath203 and assuming that @xmath204 . this would constraint @xmath54 to be much larger than what is allowed for the binary to stay bound . 2 . in the modelling of the progenitors of bh- and ns - xrbs , we have assumed a flat distribution of the initial orbital separation in the range @xmath184~r_\odot$ ] ( section [ sec : bps ] ) . with this choice , we are including all possible pre - sn orbital separations . in @xcite we verified that larger separations do not contribute to the final separation of the binary ( see their figure 13 ) . this is due to the fact that the strength of the coupling between tides and magnetic braking , which is responsible for the shrinking of a binary to short orbital periods , decreases strongly with larger orbital separations . for long - period bh - xrbs , this choice is also acceptable , as none of the observed binaries in our sample have an orbital separation larger than @xmath205 . there could be of course cases in which the orbit in the post - sn phase is highly eccentric and very wide , but these cases would contribute only to a minority of the systems . more importantly , we have found that the nk distribution does not in fact depend on the pre - sn orbital period ( see figure [ fig : presn ] , where there is no trend of the orbital separation depending on the nk ) . we can also compare the circularised orbital period distribution after the sn in our models , @xmath206 , with the one in @xcite , who did detailed evolutionary calculations of ns - lmxbs . in our models , @xmath206 ranges from @xmath207 days to @xmath208 days , which is compatible with the range shown by @xcite in their figure 1 . we have assumed that the companion of bhs and nss are stars with an initial mass of @xmath70 . @xcite argued that the majority of lmxbs have likely originated from binaries with an intermediate mass companion ( @xmath209 ) . we have checked how our results would be affected when taking a companion of initial mass : @xmath210 . the peculiar velocity right after the bh formation ( see figure [ fig : vsys ] ) would decrease due to the larger binary mass : by a factor of @xmath211 on average . this implies that the nk would need to be even larger in order for the simulated systems to match the observed ones . 4 . the two maxwellian distributions used in section [ sec : bps ] do not correspond to the real physical distributions , but rather are representative of two complementary distributions , one generating large kicks , the other generating low kicks . the choice of two distributions peaked at two different velocities serves the purpose of analysing how close in magnitude are the velocities received by bhs with respect to the velocities received by nss . the nk distribution of nss can be estimated via proper motion studies of pulsars . the pulsar birth speed distribution has been estimated as a maxwellian distribution with @xmath212 km / s by @xcite ; however , one should bear in mind the caveats discussed in section [ sec : obsns ] . for bhs , the number of sources with measured 3d space velocity ( @xmath19 ; see @xcite ) is not sufficient to allow for a calibration of their nk distribution . when comparing the observed bh - xrbs with the synthetic bh - xrbs which result from our population synthesis , we found that the population in which bhs receive high nks best fit the observed data . this conclusion gains strength when including in the comparison between observations and simulated population the sources located in the bulge of the galaxy ( i.e. h 1705 - 250 and maxi j1659 - 152 ) . in this case , the ks - values for the short - period confirmed bh - xrb sample are : @xmath213 , @xmath214 , and for the short - period confirmed + candidates bh - xrb sample are : @xmath215 , @xmath216 . we did not include the long - period ns - xrbs in our study . we can still investigate how much long- and short - period ns - xrbs differ when it comes to the peculiar velocity after the sn , and hence how they differ in terms of the scale height above the galactic plane . we build a population of binaries which evolve into long - period ns - xrbs along the lines of the computational method used for long - period bh - xrbs ( see section [ sec : bps ] ) . we find that the long - period systems have slightly lower peculiar velocities , which results in a slightly lower scale height ( by a factor of @xmath217 , for every radial distance ) . this is due to the fact that the binary , having lower binding energy , can only survive lower kicks . as it was mentioned in the introduction , there is evidence for some nss receiving low kicks at birth : nss residing in double - ns systems ( @xcite ; @xcite ; @xcite ) and nss hosted in a subset of high - mass x - ray binaries ( hmxbs ; @xcite ) . it was suggested that electron - capture sne are too fast for large asymmetries to develop , resulting in a modest kick , of a few @xmath176 km / s at maximum . such kicks are of the order of the low natal kick model we took in our binary population synthesis of section [ sec : bps ] through which we found ( section [ sec : obsns ] ) that high - kick models best fit the galactic distribution of ns - lmxbs : i.e. we did not find any evidence for nss born in an electron - capture sn . this might be due to a different evolutionary path for the progenitors of ns - lmxbs . several studies have investigated the type of sn event ( either a standard collapse of the iron core or an electron capture sn ) as a function of the evolutionary state of the helium star progenitor of the ns as well as the characteristics of the binary orbit . @xcite performed detailed evolutionary sequences of binaries hosting an ns and a helium star . they found that helium stars with an initial mass of @xmath218 could be stripped of a significant fraction of their mass through case bb mass transfer towards an ns companion when the helium star expands as a giant . the low - mass core would then form an ns via electron - capture sn , provided the initial orbital period is sufficiently wide . shorter orbital period would instead result in a white dwarf ( wd ) . we can extrapolate this finding to the short - period systems formed by an ns and a low - mass companion studied in our work . however , we must bear in mind that the stripping effect is less understood for a giant helium star experiencing mass transfer to a low - mass main - sequence star , as indeed mentioned by @xcite . thus we conclude that , heretofore , there is no strong theoretical support for a preference for electron - capture sn in the progenitors of ns - lmxbs . we also wish to note that the study by @xcite highlighted the impossibility of forming short - period ( less than one day ) ns - lmxbs without large natal kicks at birth ( where for large they chose a maxwellian distribution with an average kick of 300 km / s ) . in assessing the quality of the fit of our simulations , we used the classical application of the ks - test . in order to test its validity and determine the _ power _ of the test in distinguishing the two hypothesis we draw samples of various sizes from the simulated populations , and we calculate the @xmath188-value distribution of each of these samples , according to the rules described in section [ s : rules_for_cumul ] . we find that the probabilities follow the classical ks test ( as expected ) to an accuracy ( @xmath219 ) that is comparable to the poisson noise in our simulations ( @xmath220 ) . more interesting is the measurement of how often we obtain @xmath188-values smaller then the ones we measured for our observed samples when _ testing the wrong hypothesis _ - i.e. when using a sample drawn from the high ( low ) nk synthetic population and testing the low ( high ) nk hypothesis ( also known as false negative rate ) . for the bh case , we find @xmath188-values smaller than the ones in table [ tab : dvalues ] in less than @xmath221 of the cases , for the high nk hypothesis , and in more than @xmath222 for the low nk hypothesis . a particularly interesting fact is that @xmath223 for short - period confirmed + candidate bh - xrbs . if we accept the @xmath224 confidence level and use the standard 4-to-1 weighting ( i.e. @xmath225 ) , the test we developed has enough power to distinguish between the high- and low - nk hypothesis in this case , and that the high - nk hypothesis is clearly preferable . this is a _ non_-expected result given the small number of object , and it can potentially dissolve if some of the bh candidates turn out to be ns - xrbs . to calculate what is the optimal number of observed systems to decrease such rate @xmath226 , we draw samples of various sizes from the population synthesis results of model 1 and we test the low - nk hypothesis , and vice - versa . to decrease this rate to the level that in @xmath42 of cases we obtain @xmath227 , we find that it is necessary to increase the size of the observed sample to @xmath3 systems , both in the bh and ns case . in this work we performed a binary population synthesis study of bh- and ns - xrbs , tracing their binary evolution from the moment of compact object formation until the observed phase of mass transfer , and integrated their orbits in the galaxy . the main goal was to investigate whether different assumptions on compact object formation manifest themselves in the galactic distribution of the binaries . we found that these assumptions do affect the scale height of the binaries , which we quantified through their @xmath178 . in particular , we found that if bhs and nss receive the same nk at birth , nss would still have a larger scale height above the galactic plane , due to the fact that their systemic velocities acquired when the compact object is formed are typically larger , their total binary mass being smaller . the larger scale height of ns - xrbs with respect to bh - xrbs is clearly seen also in the observed populations . we also found a clear trend for both populations of increasing scale height for larger galactocentric radii , which should manifest itself , but which is not clearly observed in the current populations ( see figure [ fig : space ] ) . the main outcome of this study is that when analysing the @xmath0-distribution of the observed bh systems as a function of @xmath26 , the simulated population in which at least some bhs receive a ( relatively ) high nk ( @xmath228 km / s ) fits the data best . this is in agreement with previous findings by @xcite , who compared the observed and simulated populations of bh - xrbs only in the @xmath0-direction , whereas we compare the 2d distributions , accounting for how the binaries are distributed along the @xmath26-direction as well . furthermore , we increased the sample of sources adding @xmath229 bh candidates , updated their distances according to the recently published bh catalogue of , and followed the binary evolution of the binaries in a detailed way ( accounting in particular for magnetic braking and tides ) . in this work we also checked numerically the validity of a simple one - dimensional analytical estimate for the peculiar velocity at birth of bh - xrbs which we used in our previous works @xcite and @xcite . we found that this estimate is less reliable for some gravitational potentials for sources in the bulge of the galaxy , i.e at @xmath230 kpc . this was also shown by @xcite , who studied the kinematics of h 1705 - 250 , a bh - xrb close to the galactic bulge . however , the estimate is robust for systems at galactocentric radii larger than @xmath25 kpc . @xcite followed the binary evolution of seven short - period bh - xrbs and estimated their minimal peculiar velocity at birth , to conclude that two out of the seven sources were consistent with a high ( or relatively high ) nk at birth . this conclusion remains valid even in view of the current analysis . @xcite found that the rms - value of the distance to the galactic plane for bh - xrbs was similar to that of ns - xrbs . this was suggestive for bhs receiving a kick - velocity at formation . we revised the distances and updated the sample of bh - xrbs using the catalogue from and we found that ns systems have a larger scale height than bh systems , a trait which is also present in the simulated populations . finally , we found that the comparison of the data to our simulations is limited by the small number of observed bh - xrbs , and thus that more systems should be found to determine in more detail the nk that bhs receive . in this respect , the possible future discovery of new bh transients with gaia ( @xcite ) , and through dedicated surveys such as the galactic bulge survey ( @xcite ) , are promising . sr is thankful to jo bovy for answering her questions concerning galpy and for useful comments on the manuscript . sr wishes to warmly thank her office mates ( pim van oirschot , martha irene saladino , laura rossetto , thomas wevers ) at radboud university for their great moral support during the hectic times of finishing up her thesis . sr wishes to acknowledge melvyn b. davies with whom she started her research on black hole natal kicks . this research has made use of nasas astrophysics data system . o. , 2015 , in points s. , kunder a. , eds , fifty years of wide field studies in the southern hemisphere : resolved stellar populations of the galactic bulge and magellanic clouds vol . 491 of astronomical society of the pacific conference series , formation models of the galactic bulge . p. 169 h. m. , 1996 , in wijers r. a. m. j. , davies m. b. , tout c. a. , eds , nato advanced science institutes ( asi ) series c vol . 477 of nato advanced science institutes ( asi ) series c , an observational approach to binary population synthesis . p. 385 g. , 2007 , in st .- louis n. , moffat a. f. j. , eds , massive stars in interactive binaries vol . 367 of astronomical society of the pacific conference series , what can we learn about black - hole formation from black - hole x - ray binaries ? . p. 533 j. a. , preto m. , 2002 , in athanassoula e. , bosma a. , mujica r. , eds , disks of galaxies : kinematics , dynamics and peturbations vol . 275 of astronomical society of the pacific conference series , scattering of stars by transient spiral waves .

two - dimensional ( 2d ) billiards long have been studied as model systems with hamiltonian dynamics . this is in great part due to their seeming simplicity that contrasts the wealth of different dynamical behaviors that they can exhibit , including integrable , chaotic , pseudointegrable , and mixed dynamics . one interesting class of 2d billiards are polygons , of which triangles are the simplest case . while some classes of triangular billiards are well understood , many unsolved problems remain for triangles of asymmetric shape . these open questions concern , for example , the existence , number , and stability with respect to geometric perturbations of their periodic orbits ( pos ) @xcite . while the existence of at least one po , the so - called fagnano s orbit , is assured for acute triangles , and the existence of pos for obtuse triangles with no angle greater than @xmath0 has also been proven @xcite , it is not known whether any po exists at all in an arbitrary obtuse triangular billiard . but even for triangles that are known to have one or more pos , their actual construction is often nontrivial , and even the shortest pos can be quite complicated . the search for pos in triangular billiards hence stays a field of active research . two - dimensional billiards are also studied in the context of quantum and wave - dynamical chaos @xcite to understand the manifestation of ray dynamics in the properties of the corresponding wave - dynamical systems . early experiments concentrated on microwave and acoustic resonators @xcite , and new interest has arisen with the advent of applications like optical microcavities and -lasers @xcite . in particular , the influence of pos on the spectral and emission properties of microlasers is important in view both of a fundamental understanding of these devices and their applications @xcite . while many microcavities have circular or deformed circular shape , different types of polygonal microresonators also have been investigated . examples of such structures include semiconductor @xcite , organic @xcite and crystal microlasers @xcite , silicon and silica microresonators @xcite , vertical - cavity surface - emitting lasers ( vcsels ) @xcite , and hexagonal zinc oxide nanocavities and -rods @xcite . these studies , however , examined only equilateral polygons , while very few experimental investigations of nonequilateral polygonal resonators have been reported @xcite . furthermore the scattering properties of triangular and other polygonal structures have been studied @xcite . thus the properties of nonequilateral triangular and polygonal microcavities are poorly understood . in this article we study the lasing characteristics of triangular organic microlasers of different shapes with an emphasis on the role of symmetries and their absence . a laser resonator is usually designed to confine light on a specific periodic ray trajectory and the active medium is positioned to provide optimal overlap with the corresponding resonant states . in the experiments reported here , the point of view was reversed : the geometry of the cavity was given and defined the various possible pos and the distribution of the active medium , but it was not known _ a priori _ on which po , if any , the lasing modes would be based . the laser was essentially left free to decide which po was being favored . while that po could often be guessed in advance for simple , symmetric triangles , the question of which specific po is chosen was particularly interesting in cases where even the shortest pos are long and complicated or the pos are even not known at all . furthermore , the lasing modes need not necessarily be based on a particular po . the aim was therefore to understand which parameters determine the dominant po , to what extent the properties of the lasing modes can be explained by that orbit , and which features of triangular microlasers are beyond simple ray - optical explanations . the article is organized as follows . section [ sec : triangledynamics ] summarizes the key characteristics of classical triangular billiards and their pos , and treats their implications for dielectric resonators . section [ sec : exptechnique ] explains the fabrication of our microlasers and the experimental setup for their characterization . the experimental results for different triangular microlasers are presented in , starting with highly symmetric and well - understood cases and going all the way to completely asymmetric cavities . section [ sec : conclusions ] concludes with a summary of the results . triangles are the simplest type of polygons and hence can serve in many respects as paradigms for more general polygons . the dynamics of a classical polygonal billiard with @xmath1 vertices depends on the internal angles @xmath2 at the vertices , where @xmath3 is the index of the vertex . a polygonal billiard is called rational if all of its angles can be written as a rational multiple of @xmath4 , that is , @xmath5 , where the @xmath6 and @xmath7 are coprime integer numbers . all other polygons are called irrational . the topology of the phase space of a rational polygon is determined by its genus @xmath8 , which is a function of the @xmath6 and @xmath7 @xcite . for @xmath9 the phase space has the topology of a torus and the billiard is integrable . the three cases of integrable triangle billiards are the equilateral triangle , the right isosceles triangle , and the triangle with angles @xmath10 , @xmath11 , and @xmath12 , which is an equilateral triangle cut in half @xcite . for @xmath13 , the phase space resembles a @xmath8-handled sphere and its dynamics are said to be pseudointegrable @xcite . irrational polygons have ergodic dynamics @xcite . it should be noted that the subset of rational polygons is dense in the set of all polygons , that is , each irrational polygon can be approximated by a rational triangle to arbitrary precision . one type of trajectory that plays an important role in the dynamics of both classical and wave - dynamical billiards are periodic orbits , that is , orbits that retrace themselves after a finite number of reflections . one example is shown in . while it has been proven that pos exist in any billiard with a smooth ( @xmath14 ) contour @xcite , there is no such theorem for polygonal billiards : it is not known whether any po exists in an arbitrary polygonal billiard . there are , however , several results for specific cases , one of the oldest concerning acute triangles . it was proven back in the 18th century by fagnano that the po connecting the feet of the three altitudes ( thick solid red line in ) is the shortest of all possible pos in such triangles @xcite . it is hence called fagnano s orbit . the existence of other pos in irrational acute triangles is , however , not evident . the pos in rational and several other types of triangles can be constructed with the so - called unfolding technique as demonstrated in ( a ) for an isosceles triangle with a top angle of @xmath15 . we follow a trajectory [ red ( dark gray ) line ] that starts perpendicularly to the height of the triangle by reflecting the triangle each time that the boundary is encountered so the trajectory unfolds into a straight line . the trajectory returns to its starting point after a finite number of reflections . the actual po is obtained by folding the red ( dark gray ) line back into the triangle as shown in ( b ) . furthermore , there are estimates for the number of pos up to a given length in rational triangles , and it has been shown that the pos are dense in their phase space @xcite . one particular class of rational triangles for which even stronger theorems concerning the pos are known are the so - called veech triangles @xcite . much less is known for irrational triangles . the pos in right and isosceles triangles can also be constructed by unfolding @xcite . it should be noted , though , that even in cases where pos can be constructed with the unfolding technique , the shortest po can be long and complicated compared to the simple examples shown in figs . [ fig : fagnano ] and [ fig : dbowtiepo ] . furthermore , the existence of pos in triangles with all angles smaller than @xmath0 has been proven @xcite . however , no general theorems are known concerning the existence of pos in obtuse triangles with one angle larger than @xmath0 . all trajectories in polygonal billiards are marginally stable with respect to perturbations of their initial conditions . pos with an odd number of reflections like fagnano s orbit or the double bow - tie orbit in are isolated , while pos with an even number of reflections are part of a family of pos with parallel trajectories @xcite . hence , repeating an isolated po an even number of times yields a nonisolated po . for example , the po indicated as dotted red line in belongs to the family of the twice - repeated fagnano s orbit . the complete family of a po can be found by unfolding since its other members cover a strip parallel to the po . this strip is called the po channel . the po channel of the double bow - tie orbit in the isosceles @xmath15 billiard is indicated as a gray strip in . the po channel is bounded by two trajectories ( thin black lines ) that touch the corners of the triangle . these lines are the optical boundaries of the po channel . parallel trajectories beyond these line are not members of that family as can be verified by unfolding them . depending on the po and the triangle , the po channel can cover either a part of the billiard like in or the complete billiard . for example , the pos of the equilateral and right isosceles triangles cover them completely since these triangles tessellate the plane when unfolding . the flat organic microlasers studied in this article are treated as 2d passive open dielectric resonators since the lasing modes close to threshold can usually be well described by the modes of the passive cavity . a passive resonator is governed by the scalar helmholtz equation @xmath16 \psi(x , y ) = 0\ ] ] where @xmath17 is the free - space wave number and @xmath18 is the effective refractive index @xmath19 for @xmath20 inside the resonator and the refractive index of the surrounding medium ( air with @xmath21 ) on the outside . the wave function @xmath22 corresponds either to the @xmath23 component of the electric field , @xmath24 , for transverse magnetic ( tm ) modes or to that of the magnetic field , @xmath25 , for transverse electric ( te ) modes . the wave functions inside and outside of the resonator are connected by the usual boundary conditions for dielectric interfaces @xcite . since the typical size of the cavities considered here is in the range of several hundred wavelengths , the resonators are in the so - called semiclassical regime which is the transition regime from classical physics ( @xmath26 ray optics ) to quantum mechanics ( @xmath26 wave optics ) . semiclassical methods permit to explain various properties of the resonators using concepts and quantities from the dynamics of the corresponding classical billiard systems @xcite , the pos playing an important role in such approximations . two well - known cases are trace formulas that connect the density of state with the pos @xcite and resonant states localized on pos , so - called scars @xcite and superscars @xcite . trace formulas and scars were mainly investigated in the context of closed resonators with dirichlet boundary conditions , but the underlying principles can be extended to dielectric resonators . in fact , a trace formula for dielectric resonators has been developed @xcite , and modes of dielectric resonators localized on classical trajectories are often observed . this includes gaussian modes that are localized on stable pos @xcite , scar states localized on unstable pos @xcite , and superscar states localized on families of marginally stable pos @xcite or classical tori @xcite . it has been shown in ref . @xcite for resonators with dirichlet boundary conditions that superscar states are localized inside the po channel ( see ) due to repeated diffraction at the corners that define the optical boundaries . a similar effect has been proposed in ref . @xcite for dielectric resonators even though the diffraction at dielectric corners is not understood @xcite . therefore we expect to find superscarred lasing modes in triangular dielectric resonators with pseudointegrable classical dynamics . the influence of a po on the properties of a resonator depends on several factors , among them its length , its stability or , in the case of nonisolated orbits , the area covered by its family and its refractive losses @xcite . the losses depend on the refractive index and the angles of incidence of a po since a ray traveling in a dielectric resonator is reflected and refracted at the side walls according to the fresnel formulas . the emission directions of the refracted rays are determined by snell s law . the most long - lived modes of passive dielectric cavities hence mainly exhibit the influence of the shortest and best - confined pos @xcite . on the other hand , modes based on pos that are not confined by total internal reflection can be observed for laser cavities . the threshold condition for a ray traveling along a po with length @xmath27 in an active medium with linear gain @xmath28 is given by @xmath29 where @xmath30 is the fresnel reflection coefficient for the reflection at the @xmath31th vertex and the product runs over all vertices of the po . the threshold gain @xmath32 is hence @xmath33 we use as a simple estimate for the threshold of a mode localized on a po . in practice , however , also other parameters can be of importance like the overlap between the gain region and the mode profiles @xcite or the coupling between the molecules of the gain medium and the electric field of a mode @xcite . therefore , we do not expect quantitative agreement of the measured lasing thresholds with . more sophisticated approaches are necessary for a quantitative understanding of the lasing thresholds . it should be noted that even though the pos of a classical billiard can explain many properties of the corresponding resonators , wave effects such as tunneling and diffraction can also have a significant influence . one example is the existence of so - called diffractive orbits that have one vertex at a diffractive corner of the billiard . diffractive corners are corners with an angle that is not equal to @xmath34 , where @xmath35 is an integer . the reflection of a ray at such a corner is not defined in classical mechanics . in contrast , a wave impinging on it is diffracted and hence scattered into various directions . this enables to close ray trajectories that impinge on a diffractive corner and thus diffractive orbits can appear in wave - dynamical billiards ( i.e. , resonators ) . for example , they contribute to the trace formula for resonators with diffractive corners or point scatterers @xcite . another important point is that the dynamics of wave systems is less sensitive to geometric perturbations than that of classical systems . therefore the influence of a po on a wave system can survive geometric perturbations even though the perturbation completely eliminates that po in the classical dynamics @xcite . the organic microlasers consisted of poly(methyl methacrylate ) ( pmma ) doped with @xmath36 wt% of the laser dye dcm . a solution of pmma and dcm was spin coated on a silicon wafer with a @xmath37-@xmath38m - thick layer of silica . the thickness of the pmma layer was about @xmath39 nm . the desired cavity shapes were written by @xmath40-kv electron - beam lithography . this process allowed us to define the cavity boundaries with nanometric precision and achieve vertical side walls and sharp corners and edges @xcite . scanning electron microscope ( sem ) images of an equilateral triangle cavity are presented in . the triangular microlasers considered here had typical side lengths @xmath41 in the range of @xmath42 to @xmath43 m , i.e. , several hundred times larger than the wavelength @xmath44 nm . they are considered two - dimensional ( 2d ) systems with an effective refractive index of @xmath45 since they are only about one wavelength thick and support only a single vertical excitation for each polarization @xcite . the experimental setup was similar to the one described in ref . the microlasers were pumped by a pulsed frequency - doubled nd : yag laser ( @xmath46 nm , @xmath47 ps , @xmath48 hz , teem photonics png-002025 - 140 ) that impinged perpendicularly to the cavity plane . the intensity and the polarization of the pump beam were controlled independently using half- and quarter - wavelength plates and polarizers . a circularly polarized pump beam was used , and the pump intensity was normally chosen only slightly above threshold . the pump beam had an approximately gaussian intensity profile and its diameter was adjusted to cover the complete area of a single microlaser . the lasing emission in the plane of the microlasers was collected in the far field by a lens @xmath49 cm away from the sample and transferred by a fiber to a spectrometer ( spectra pro 2500i , acton research ) and a cooled ccd camera ( pixis 110b , princeton instruments ) . the spectra were integrated over @xmath48 pump pulses . the samples could be rotated to record the spectra in all possible directions in the plane of the cavity and thus measure the azimuthal far - field distributions . the polarization of the lasing emission was determined using a linear polarization filter @xcite . all of the microlasers presented in the following emitted transverse electrically ( te ) polarized light , i.e. , the lasing emission had an electric field parallel to the plane of the cavities . furthermore , a complementary metal - oxide semiconductor sensor camera ( , ids imaging ) with a zoom lens ( zoom 6000 , navitar ) was used to take photographs of the lasing cavities . the observation angle of the camera was chosen at a @xmath50 tilt angle above the plane so the whole cavities could be surveyed . the pos that the lasing modes might be localized on can be deduced from the various experimental observables . it should be noted , however , that not all lasing modes are in fact localized on specific classical trajectories . therefore , a careful analysis of all available data is needed to determine the nature of the observed resonant states . the lasing thresholds are related to the lifetime of the cavity modes and hence to the losses of a possible underlying po [ cf . ] . the far - field distributions are often concentrated around a few specific directions . from these directions one can infer the possible trajectories within the resonator via snell s law . the photographs indicate from which parts of the cavities the light is emitted . the spectra typically exhibit multimode lasing with several tens of resonances . they are often organized in sequences of equidistant resonances . if a set of lasing modes is localized on a certain po , their resonance wave numbers are given by @xmath51 where @xmath35 is an integer and @xmath52 a constant phase shift . the resonance spacing @xmath53 is hence inversely proportional to the optical length @xmath54 of this po , which can be conveniently obtained from the fourier transform ( ft ) of the spectrum that exhibits peaks at @xmath54 and its multiples @xcite . it can also be deduced from the free spectral range ( fsr ) , @xmath55 , via the relation @xmath56 where @xmath57 is the wavelength of the lasing emission . the geometric length of the po , @xmath27 , is related to the optical length by @xmath58 , where @xmath59 is the group refractive index . the latter differs from the effective refractive index since it also takes into account dispersion . it has a value in the range of @xmath60 to @xmath61 depending on the sample @xcite . the precise value for each sample can be determined from calibration measurements with ribbon - shaped fabry - prot cavities since they sustain only a single type of po , the well - known bouncing ball or fabry - prot orbits . seven different triangles with varying degrees of symmetry and different types of classical dynamics were investigated . we start with the simplest and most symmetric triangles and gradually pass to less symmetric and accordingly more complicated ones . the first examples are the equilateral and the right isosceles triangle that have both integrable classical dynamics . the next two triangles are isosceles triangles with top angles @xmath0 and @xmath15 . they are rational triangles with pseudointegrable classical dynamics . since these four ( pseudo- ) integrable triangles feature short and simple pos it is expected to find lasing modes localized on some of these orbits . the fifth triangle is an irrational right triangle . it also exhibits a short and simple po due to its right angle but has ergodic classical dynamics in contrast to the previous examples . since the rational triangles are a dense subset of all triangles , it is interesting to see if the properties of rational and irrational triangle microlasers significantly differ . the final two triangles are perturbations of the equilateral triangle and the @xmath0 isosceles triangle , respectively . they are irrational and hence have ergodic dynamics . they were chosen because no simple pos are known for them except for fagnano s orbit in the quasiequilateral triangle . in addition , they permit us to study the influence of geometric perturbations on the microlasers properties and in particular the effect of breaking their mirror symmetry . the equilateral triangle ( et ) is the triangle with the highest degree of symmetry , and microlasers with equilateral triangular shape have been intensely studied @xcite . the equilateral triangle has integrable classical dynamics , and the corresponding cavity problem with dirichlet or neumann boundary conditions can be solved analytically @xcite . this problem was already investigated in the context of vibrating membranes by lam in the 19th century @xcite . there is , however , no analytical solution in the case of the dielectric boundary conditions considered here . all pos of the et are known , but none of them is confined by total internal reflection . this is due to the relatively low value of @xmath62 that corresponds to a critical angle of @xmath63 . the same is true for the other triangular microlasers . the lack of good confinement required comparably large cavities to provide sufficient gain . the side length of the et microlaser was @xmath64 m . figure [ fig : tequispectrum](a ) shows the lasing spectrum @xmath65 of the et microlaser for a pump energy just above the threshold in the direction @xmath66 [ see ( a ) for the definition of the azimuthal angle ] . the spectrum exhibits a clear structure of equidistant peaks . the fourier transform of the spectrum , @xmath67 , is plotted with respect to the optical length @xmath54 in ( b ) . it features several equidistant peaks with decreasing amplitude as expected for a series of equidistant resonances . the first peak at @xmath68 m corresponds to the fsr , @xmath69 nm , of the lasing spectrum , and the further peaks are harmonics . the two shortest types of pos in the et billiard are fagnano s orbit with @xmath70 m [ see ( a ) ] and the so - called quasi - fabry - prot ( qfp ) orbits shown in ( b ) . a qfp orbit has two reflections with perpendicular incidence and two reflections with an angle of incidence of @xmath71 with respect to the surface normal . it has a length of @xmath72 m . the family of the qfp orbits in the et is constructed as follows : first , one qfp orbit [ e.g. , the solid red line in ( b ) ] is shifted perpendicularly to its trajectory ( yielding , for example , the dashed blue lines ) and , second , the qfp orbits can be rotated by @xmath73 ( yielding , for example , the dotted green lines ) . the family of qfp orbits covers the whole surface of the et . the calculated optical lengths corresponding to fagnano s orbit and the qfp orbit are @xmath74 m and @xmath75 m , respectively . they are indicated by the arrows in ( b ) . obviously , the observed optical length corresponds to the qfp orbit while fagnano s orbit is too short . all other pos are significantly too long . further evidence is gained from the azimuthal far - field distribution in ( c ) . the lasing emission is concentrated in the three directions perpendicular to the cavity side walls . this is the behavior expected from modes localized on the qfp orbit as shown in ( b ) . note that there is no emission from the reflections with angle of incidence @xmath71 since this angle is larger than the critical angle . on the other hand , fagnano s orbit would correspond to six emission directions with an angle of @xmath76 with respect to the surface normals as indicated in ( a ) , but no emission was found in these directions . it should be furthermore noted that the lasing threshold of the et microlaser is only about @xmath77 higher than that of a fabry - prot ( fp ) cavity of corresponding width . altogether this proves that the observed lasing modes are localized on the qfp orbit . finally , a photo taken from the direction @xmath66 and presented in shows that the whole side wall of the cavity is lasing . this was expected since the family of the qfp orbit covers the whole triangle , though also all other po families do so . photos taken from @xmath78 and @xmath79 show the same behavior , whereas no lasing light was observed with the camera in all other directions . in summary , the qfp orbit was identified unambiguously from the experimental data as the orbit supporting the lasing modes . it is at first surprising that the dominant lasing modes are localized on the qfp orbit and not on fagnano s orbit like in refs . first , however , the refractive index of the semiconductor materials used in refs . @xcite was significantly higher so fagnano s orbit was confined by total internal reflection , which is not the case here . second , the lasing modes that we observed were te polarized , i.e. , their electric field was parallel to the plane of the resonator . in fact , te polarized modes are favored by the properties of the lasing dye and the pumping scheme that is used here @xcite . since the angle of incidence of fagnano s orbit , @xmath80 , is close to the brewster angle @xmath81 , a te mode localized on this po would suffer from very high losses . from we calculate @xmath82 as the threshold of fagnano s orbit and @xmath83 for the qfp orbit . thus , the dominance of the qfp modes can be well explained by taking into account the peculiarities of the organic microlasers . it should be noted that the modes of the other triangular microlasers considered in the following were all te polarized as well . the second triangle that was investigated is the right isosceles triangle ( rit ) , which is essentially a square cut in half along a diagonal . the rit billiard is integrable like the square and equilateral triangle billiards . both classical and quantum right triangle billiards have been studied and their pos investigated @xcite . a property well known by opticians is the fact that a corner with a right angle sends a ray back parallel to its initial direction regardless of the angle of incidence . from this follows directly the existence of a family of pos that impinge perpendicularly on the hypotenuse as indicated by the thick solid red line in . we call this orbit quasidiamond orbit in analogy to the diamond po in the square billiard , indicated by the dashed red line in . another important po is indicated by the dash - dotted blue line . it is reflected perpendicularly at the two short sides of the rit and with an angle of incidence of @xmath84 at the hypotenuse . since it corresponds to the fabry - prot orbit of the square billiard ( indicated by the dotted blue lines in ) , it is also called the qfp orbit . these are the two shortest pos of the rit , and both po families cover the whole area of the billiard . the qfp orbit exists in all isosceles triangles , with the angle of incidence on the long side and the area covered by its family depending on the top angle of the triangle . two spectra measured in the directions perpendicular to the hypotenuse and one of the short sides , respectively , are shown in . both spectra consist of families of equidistant resonances that have , however , a different fsr each . the largest peak in the ft of the spectrum for @xmath85 is at @xmath86 m . as expected for this direction , the corresponding po is the quasidiamond orbit with a geometric length of @xmath87 m , where @xmath88 m is the length of the hypotenuse , and a calculated optical length of @xmath89 m . the ft also features a smaller peak at half this optical length . it stems from the slight modulation of the resonance amplitudes [ see ( a ) ] , i.e. , the fact that every second resonance has a somewhat smaller amplitude than its neighbors . it is interesting to note that there is an isolated po along the height of the triangle , indicated as thin solid red line in , that has half the length of the quasidiamond orbit . the physical origin of the modulation of the resonance amplitudes and whether it is connected to this po remains , however , unclear . in contrast , the ft of the spectrum at @xmath90 shows a peak at @xmath91 m . this corresponds to the qfp orbit with a geometric length of @xmath92 m and an optical length of @xmath93 m . an unexpected find is a small peak at @xmath86 m [ indicated by the black arrow in the inset of ( b ) ] that corresponds to the quasidiamond orbit . it originates from another family of barely visible resonances in the spectrum . the far - field distribution shown in features several emission lobes with differing amplitudes . the strongest emission lobe is in the direction of @xmath85 and is due to the quasidiamond orbit . it was cut off in the inset of since its amplitude of @xmath94 counts far exceeds that of the other emission lobes . it should be noted that the emission lobe at @xmath95 is about @xmath36 times larger than that at @xmath84 , whereas the two lobes are expected to have equal amplitudes due to the mirror symmetry of the triangle . a significant asymmetry of the microcavity itself was excluded . further experiments demonstrated that the ratio between the amplitudes at @xmath96 and @xmath95 depended sensitively on how precisely the cavity was pumped , that is , for example , on the size and position of the pump beam . according to , modes localized on the qfp orbit should emit in the directions of @xmath84 and @xmath95 , and indeed the spectra measured in these directions exhibit a fsr corresponding to its optical length . however , small contributions also of the quasidiamond orbit were found in the spectra at @xmath84 and @xmath95 as demonstrated in ( b ) . in addition , two small lobes at @xmath97 and @xmath98 were found that are also related to the quasidiamond orbit . the origin of these can be elucidated by the photos shown in . the photos taken from @xmath96 show a strong emission from the side wall perpendicular to the camera perspective as predicted for the qfp orbit . classically , it is expected that the whole side wall emits like in since the qfp orbit family covers the whole triangle . why this is not observed experimentally remains unclear . it could also be related to the strong sensitivity of the emission to the pumping conditions . in addition , a weak emission from the side wall parallel to the camera perspective was observed . this is best seen in the top right panel of . a similar grazing emission was also observed at @xmath90 , @xmath99 , and @xmath98 . it is not expected classically since the quasidiamond orbit is totally reflected at the two short sides . the same kind of grazing emission is also observed for the diamond orbit modes of square organic microlasers and will be discussed elsewhere @xcite . the photo taken at @xmath85 finally shows emission from the whole hypotenuse as expected for modes localized on the quasidiamond orbit . the key characteristics of the rit microlaser can be well explained by simple pos of the corresponding billiard as in the case of the equilateral triangle . but in contrast to the equilateral triangle , the rit microlaser features two families of modes localized on different pos that coexist . a calculation of the thresholds according to reveals that they are very close to each other since both pos have the same losses and nearly the same lengths . this prediction agrees qualitatively with the measured thresholds . another interesting observation is the grazing emission of the modes localized on the quasidiamond orbit that is not expected from the ray dynamics . it demonstrates that some properties of the triangular microlasers need a more careful treatment taking into account wave - dynamical effects . the most general class of triangles with a symmetry are isosceles triangles . microlasers with an isosceles triangle shape have been investigated , for example , in ref . the pos of isosceles triangles can be constructed by the unfolding technique @xcite . the simplest po that exists in all isosceles triangles is the qfp orbit already known from the right isosceles triangle . another , more complicated example is the double bow - tie orbit shown in . it exists for top angles between @xmath100 and @xmath101 . the first of two obtuse isosceles triangles that are discussed here is the one with top angle @xmath102 ( abbreviated it100 in the following ) . its classical dynamics is pseudointegrable . the spectrum of the it100 microlaser measured in the direction @xmath103 , i.e. , perpendicular to one of the short side walls , is shown in . the fsr of the equidistant resonance family corresponds to an optical length of @xmath104 m . the underlying po is hence the qfp orbit shown in ( a ) that has a length of @xmath105 m , where the length of the long side is @xmath106 m , which corresponds to an optical length of @xmath107 m . the far - field distribution presented in ( b ) shows four major emission lobes . their directions , @xmath103 , @xmath108 , @xmath109 , and @xmath110 , are precisely the ones expected classically for the qfp orbit . there are also some smaller emission lobes at , e.g. , @xmath79 and @xmath111 , hardly visible in ( b ) . the spectra in these directions also have a fsr corresponding to the qfp orbit . these directions can not , however , be explained by the ray dynamics of the qfp orbit . figure [ fig : tiso100photos ] presents photographs of the it100 microlaser taken from two of the major emission directions . the photographs taken from @xmath112 in ( a ) show that the whole side wall is emitting with the exception of a small part next to the corner with the long side to the right . this is in fact the part of the boundary that is expected classically to emit , indicated by the thick red ( dark gray ) line in the sketch in the right panel of ( a ) , because the family of the qfp orbit covers only a part of the billiard ( indicated as the gray area ) in contrast to the right isosceles triangle . the photographs taken from @xmath113 in ( b ) demonstrate that the most intense emission into that direction originates from the middle part of the long side ( to the left in the photo ) , again in good agreement with the classical prediction shown in the right panel . however , also the smaller side wall ( to the right in the photo ) emits light , though with lesser intensity . this emission can not be explained classically with the properties of the qfp orbit family . in summary , most of the observed lasing characteristics of the it100 microlaser are in very good agreement with the classical predictions for the qfp orbit , while only some details are beyond a simple ray - dynamical analysis as in the case of the right isosceles triangle microlaser . the second isosceles triangle that was investigated is the one with a top angle of @xmath15 ( abbreviated it110 in the following ) . it is also pseudointegrable . the spectrum of the it110 microlaser at @xmath114 is shown in ( a ) . it shows a single family of resonances , the fsr of which corresponds to @xmath115 m . the corresponding po is again the qfp orbit with geometric length @xmath116 m and optical length @xmath117 m , where the length of the long side is @xmath106 m . the spectrum measured at @xmath118 is presented in ( b ) . its structure is less clean than that of the one at @xmath119 , but its ft ( see inset ) shows clear peaks at @xmath120 m and multiples of this length . the corresponding fsr of @xmath121 nm is the fsr of the dominant family of modes [ see ( b ) ] . the spectrum also exhibits a second family of modes with smaller amplitude and the same fsr . this optical length as well as the emission direction correspond well to the double bow - tie orbit ( shown as inset ) with a geometric length of @xmath122 = 706.4~\mu$]m and an optical length of @xmath123 m . it should be noted that the measured threshold of the modes localized on the qfp orbit is almost @xmath124 times higher than that of the modes localized on the double bow - tie orbit . this agrees qualitatively with , which predicts a @xmath125 times higher threshold . the far - field distribution of the it110 microlaser is presented in . the six principal emission directions agree very well with those calculated for the double bow - tie orbit that are indicated by the black arrows [ see also inset of ( b ) ] . the amplitudes of the emission lobes lack , however , the expected symmetry as in the case of the right isosceles triangle microlaser . no emissions lobes corresponding to the qfp orbit were observed since the microlaser was pumped slightly above the threshold of the double bow - tie orbit modes but well below the threshold of the qfp orbit modes . the photos shown in were taken with the same pump intensity as for the measurement of the far - field distribution . the directions of observation correspond to the major emission directions . all photos show that the most intense part of the laser emission originates from those parts of the side walls that are covered by the family of the double bow - tie orbit [ indicated by the red ( dark gray ) lines in the drawing of the it110 ] . this confirms that the lasing modes with the lowest threshold are localized on the double bow - tie orbit . however , the photos also show weak emission from other parts of the side walls that are not covered by the family of the double bow - tie orbit . a possible explanation is that even though the field distributions of superscarred resonant states are strongly concentrated inside the po channel , they also have a nonvanishing field outside of the po channel due to coupling to nonscarred states @xcite . in conclusion , the it110 microlaser is another example like the right isosceles triangle where two families of modes localized on different pos coexist . in contrast to the case of the right isosceles triangle , however , there is a significant difference between the thresholds of the two mode families so one of them is easily selected by keeping the pump intensity sufficiently low . furthermore , it should be noted that the dominant po , the double bow - tie orbit , is selected because it has the lowest losses even though it is longer and more complicated than the qfp orbit . the double bow - tie orbit also exists in the it100 billiard but has a higher threshold than the qfp orbit in that case . lasing modes based on the double bow - tie orbit were not found experimentally for the it100 microlasers even at considerably higher pump energies . this demonstrates that the change of a single geometric parameter , the top angle in this case , can significantly modify the lasing characteristics . the fifth triangle that was investigated is a pythagorean triangle ( pt ) with side length ratio @xmath126 as shown in . its hypotenuse is @xmath127 m long . in contrast to the other triangles studied so far , it is irrational and exhibits no symmetry . it exhibits a qfp orbit that impinges perpendicularly on the hypotenuse as shown in in analogy to the right isosceles triangle . the pos of right triangles can also be constructed by unfolding @xcite even if they are irrational like the pt . the lasing spectrum of the corresponding pt microlaser observed in the direction perpendicular to the hypotenuse is presented in ( a ) . it exhibits a single family of equidistant resonances . its ft , shown in ( b ) , features a dominant peak at @xmath128 m that corresponds to the fsr of the spectrum . the length of the qfp orbit is @xmath129 m , which yields an optical length of @xmath130 m that is in good agreement with the optical length observed in the ft . in complete analogy to the case of the right isosceles triangle , there is an additional , smaller peak at half this optical length that stems from a modulation of the resonance amplitudes , and there is a po along the height of the triangle with half the length of qfp orbit ( indicated as dashed blue line in ) . it should be noted that the relative amplitude of the peaks at @xmath131 m and @xmath132 m depends sensitively on the pump beam position and other details of the excitation scheme . the far - field distribution presented in has three major emission directions , @xmath133 , @xmath134 , and @xmath135 . the lasing spectra in these directions exhibit the same modes as the one shown in ( a ) . the emission directions predicted for the qfp orbit according to snell s law for @xmath62 are @xmath133 , @xmath136 , and @xmath137 , respectively ( see ) . they are indicated in by the black arrows and agree quite well with the observed ones . it should be noted that no emission is expected from the left ( smallest ) side of the pt since the angle of incidence of the qfp orbit on it is larger than the critical angle . the amplitude of the emission lobe at @xmath135 is considerably smaller than that of the lobe at @xmath134 while an equal amplitude is expected classically . in fact , the relative amplitude of the lobes strongly depended on the position and size of the pump beam as in the case of the right isosceles triangle microlaser . furthermore , the directions of these two lobes are not symmetric as well . according to the geometry of the qfp orbit , their angles with respect to the surface normal should be equal , but in reality they are @xmath138 and @xmath139 , respectively , and thus differ by @xmath140 . this is a significant deviation that is within the resolution of the setup . the third emission lobe on the contrary has precisely the expected direction perpendicular to the hypotenuse . in contrast to the previously considered triangles , however , the pt cavity itself exhibits no symmetry , and symmetric emission directions are only expected due to the properties of the underlying qfp orbit . a completely unexpected experimental result are the three smaller and broader emission lobes around @xmath141 , @xmath142 , and @xmath143 . the spectra in these directions feature the same structure and fsr as those in the three major emission lobes ; however , these three emission directions can not be related to the dynamics of the qfp orbit . photographs of the pt microlaser taken from the three major emission directions are shown in the left panels of . the sketches in the right panels indicate the parts of the side walls that are expected to emit in these directions according to the geometry of the qfp orbits . indeed , the brightest areas of emission in the photos correspond well to these classical predictions . in contrast , also some weak emissions are observed from the side wall on the left side at @xmath144 and from the side wall at the right side at @xmath145 , which are not expected classically . in summary , the spectra , the far - field distribution , and the photos clearly evidence that the observed lasing modes are localized on the qfp orbit family . in detail , however , there are deviations from the ray - dynamical predictions . first , the directions of the two emission lobes at @xmath134 and @xmath135 do not exhibit the expected symmetry ; second , there are three additional emission lobes the directions of which have no apparent connection to the qfp orbits ; and , third , parts of the lasing emission stem from sections of the cavity boundary that are not expected to emit classically . so even though the underlying po could be identified , the properties of the pt triangle microlaser can not be explained with the same precision and completeness as in the previous cases . this might be related to the fact that the pt billiard is neither rational nor symmetric , and in fact its pos can be constructed easily only due to its right angle . while all the triangles considered so far had symmetries or other properties that enabled an easy construction of their pos , the case considered in the following has none of these . it is a deformation of the equilateral triangle and hence called quasiequilateral triangle ( qet ) in the following . its side lengths are @xmath146 m , @xmath147 m , and @xmath148 m , respectively , as indicated in . it is an irrational triangle and we hence can not easily construct any po besides fagnano s orbit . in particular , the qfp orbits no longer exist due to the lack of symmetry . a typical spectrum , observed at @xmath149 , is shown in ( a ) . the structure of the spectrum is not as clear as for the previously shown ones ; nonetheless , a small sequence of equidistant resonances can be identified . the ft of the spectrum shows a peak at @xmath150 m and its approximate multiples . the corresponding fsr of @xmath69 nm matches that of the resonance sequence in the spectrum . this optical length is close to that of the qfp orbit in an equilateral triangle with side length @xmath41 , but the qfp orbit no longer exists in the qet . there are , however , three diffractive pos along the heights of the triangle that have nearly the same lengths . they are depicted in and called @xmath151 , @xmath152 , and @xmath153 in the following . they are called diffractive orbits because one of their vertices is at a diffractive corner of the billiard ( cf . ) . it should be noted that for an equilateral triangle , the orbits along the heights belong to the family of the qfp orbit . thus , these three diffractive pos can be considered as the remnants of the qfp orbit family . their lengths are @xmath154 m , @xmath155 m , and @xmath156 m . the corresponding optical lengths are indicated by arrows in ( b ) and are indeed close to the peaks observed in the ft of the spectrum . the far - field distribution of the qet microlaser is plotted in ( a ) . it features three broad bundles of emission lobes the centers of which are roughly perpendicular to the cavity side walls . the ft of the spectra measured at different azimuthal angles is shown in ( b ) . the dominant optical lengths vary somewhat with @xmath157 , but they always stay close to the lengths of the three diffractive pos , the optical lengths of which are indicated by the vertical white lines . this corresponds to the fact that the spectra always feature a similar fsr even though their structure and quality varies significantly , leading to a relatively high noise level in the ft . no evidence of fagnano s orbit was found like in the case of the equilateral triangle since its angles of incidence are close to the brewster angle . so while the fsrs approximately match those corresponding to the diffractive orbits , other observations do not indicate that the modes are localized on them . for example , the directions of maximal emission are not exactly perpendicular to the side walls as one would naively expect . the photos shown in enable a better understanding of the nature of the resonant modes . they show the lasing qet from the directions approximately perpendicular to the side walls . all of them show that the points of origin of the lasing emission are more or less broadly distributed over the side walls . the expectation for modes localized on the diffractive pos , in contrast , would be that the origin of emission is strongly concentrated around the feet of the heights that are roughly in the middle of the side walls . another expectation for this kind of mode would be strong emission coming from the corners of the triangle . this is , however , not observed in any direction . in fact , the images shown in are typical also for other directions in which the qet microlaser emits . in conclusion , the lasing characteristics of the qet microlaser can not be explained by any po of the qet billiard . the observed fsrs correspond to an optical length similar to that of the qfp orbits of an equilateral triangle . a possible explanation is that the observed modes are the perturbed modes of the equilateral triangle microlaser that were localized on the qfp orbit . due to the small perturbation , their fsr stays approximately the same , but the far - field distribution broadens around the emission directions of the qfp orbit that are perpendicular to the side walls . the same effect was observed in ref . @xcite for triangular resonators with dirichlet boundary conditions . some modes were shown to be localized on so - called ghost pos , i.e. , the pos of a geometrically different , but similar , triangle . an analogous case is the persistence of the influence of the bouncing ball orbits in a quantum stadium billiard when the originally parallel side walls of the stadium are slightly tilted @xcite . the reason for these effects is that while the classical dynamics can exhibit singular behavior with respect to perturbations of the billiard geometry , e.g. , pos suddenly vanishing completely , wave - dynamical systems react in a continuous manner to geometric perturbations , essentially smoothing out the singularities of classical mechanics . the data presented here lead us to believe that the modes of the quasiequilateral triangle microlasers are localized on the ghost qfp orbit , but numerical investigations of the wave functions will be necessary to confirm this notion . the last triangle is again a deformation of one of the previous triangles . the triangle was constructed by moving the right vertex of the isosceles @xmath0 triangle by @xmath158 m to the left while keeping the other two vertices fixed as demonstrated in ( a ) . hence it is called quasi - isosceles triangle ( qit ) . it is irrational and therefore no simple pos are known since the qfp and the double bow - tie orbit of the isosceles @xmath0 triangle are destroyed by the geometric perturbation . there are , however , three diffractive pos [ see ( b ) ] similar to the case of the quasiequilateral triangle . one is along the height of the triangle ( called the height orbit in the following ) , and the other two have one perpendicular reflection at a short side wall and one at the top vertex . they can be considered as the remnants of the qfp orbits of the isosceles @xmath0 triangle and are called diffractive quasi - fabry - prot orbits in the following . two spectra of the qit microlaser observed in different directions are plotted in . both spectra show a relatively clear structure of equidistant modes . however , their fsrs and the corresponding optical lengths differ as can be seen in the fts shown as insets . the optical length for the spectrum at @xmath159 is @xmath160 m . this can only correspond to the height orbit with @xmath161 m and @xmath162 m . the optical length for the spectrum at @xmath163 is @xmath164 m and corresponds approximately to the lengths of the two diffractive qfp orbits . the geometric and optical length of the left diffractive qfp orbit [ solid red line in ( b ) ] are @xmath165 m and @xmath166 m , respectively , and are identical to those of the qfp orbit in the isosceles @xmath0 triangle , while the geometric and optical length of the right diffractive qfp orbit [ dashed green line in ( b ) ] are @xmath167 m and @xmath168 m , respectively . so the qit microlaser exhibits ( at least ) two different families of modes that have similar thresholds but different fsrs . next , we investigated what family of modes emitted in which directions . the far - field distribution in shows a large number of emission lobes with varying amplitudes , though the emission is not as broadly distributed as in the case of the quasiequilateral triangle . the directions in which the spectra exhibit the optical length of the diffractive qfp orbits and the height orbit are indicated by the solid red and dotted blue arrows , respectively . note that for some of the smaller emission lobes the corresponding optical length could not be clearly determined due to indistinct spectra . in fact , some of the most prominent emission lobes point approximately in the directions expected for a diffractive qfp orbit which are indicated by the black double arrows [ see also ( b ) ] . there remain , however , significant deviations between the emission directions of the presumed diffractive qfp modes and the directions predicted by ray optics . these deviations can not be consistently explained by a different refractive index either . regarding the presumed height orbit modes , there are several emission lobes around , though not precisely in , the direction of @xmath169 predicted classically . furthermore , the two emission lobes close to @xmath100 seem reasonable for such modes , too . the strong emission in the direction of @xmath170 from the presumed height orbit modes and in the direction of @xmath171 from both families of modes , however , defy any simple ray - dynamical explanation . thus , the dynamics of the diffractive pos can explain the observed emission directions at best on a qualitative level . the photos taken in the main emission directions of the presumed diffractive qfp modes are presented in . they show that the emission originates from almost the whole side walls and not just small portions of it as would be expected for the isolated diffractive qfp orbits . the photos rather resemble those of the equilateral and isosceles triangle microlasers . in addition , a part of the emission originates from those side walls that are not directly facing the camera . emission from these side walls is not at all expected from any fabry - prot - like modes . the photos in were analogously taken in the main emission directions of the presumed height orbit modes . the lasing emission is not as broadly distributed along the side walls as in the cases presented in but shows nonetheless no single points of concentration . in particular , the photos at @xmath172 and @xmath173 demonstrate that the emission in these directions is not predominantly originating from the vertex of the triangle as expected for a mode localized on the diffractive height orbit . in summary , the photos evidence that both families of lasing modes cover more or less the whole microlaser and are not strongly concentrated along the isolated diffractive orbits shown in . we are hence led to believe that one family of modes is localized on the ghost qfp orbits , i.e. , they are the perturbed qfp modes of the isosceles @xmath0 triangle , analogously to the case of the quasiequilateral triangle . the situation for the other family of modes with an fsr corresponding to the height orbit is less clear since a ( nondiffractive ) height po does not exist in the isosceles @xmath0 triangle , and no family of modes with a similar fsr was observed for the isosceles @xmath0 triangle microlaser . in any case , further investigations are necessary for a full understanding of the qit microlaser s behavior . it is interesting to note that even though a po is known to exist in the qit since its angles are all less than @xmath0 , the dominant lasing modes are clearly not localized on any classical po . it can be presumed that even the shortest po in the qit is too long or has too high losses to support lasing modes with a reasonably low threshold and that therefore other types of lasing modes are predominant . while there are still open questions regarding the existence and properties of pos in classical triangular billiards and this remains a domain of active research , it is also very interesting to explore the influence of the pos on the properties of the corresponding wave - dynamical billiards . of particular interest are billiards for which the pos can not be easily constructed or are not even known because they are irrational and lack symmetries . one of the objectives of the experiments presented here was therefore to see whether the resonant states in these cases are localized on pos , and , if yes , on which ones or if , on the contrary , they have no relation to specific pos . we investigated this ray - wave correspondence in experiments with organic microlasers of triangular shape . the shapes that were chosen correspond to different types of classical dynamics featuring diverse types of pos and were hence expected to exhibit very different lasing characteristics . even though each triangle microlaser had properties distinct from the others and had to be treated separately , it is possible to draw some general conclusions from the experiments . several examples of triangles with well - understood classical dynamics and relatively simple pos were studied , namely the equilateral triangle and several rational isosceles triangles . they all exhibit lasing modes that are clearly localized on pos . the most important observables were the fsr of the spectra and the directions of the emission lobes in the far field that could be measured with high precision and showed excellent agreement with ray - optical predictions . in contrast , the amplitudes of the emission lobes proved to be very sensitive to small changes in the experimental setup . photographs of the lasing cavities allowed further insight into the characteristics of the lasing modes and showed good qualitative agreement with the classical predictions . it seems to be a general rule for polygonal billiards having simple , short pos with not too high losses that the modes of the corresponding microlasers are localized on these pos @xcite . nonetheless , even for these simple and symmetric triangles some details of the observed spectra and far - field distributions were beyond simple ray - dynamical explanations . this underlines the need to further refine the different models of modal localization in dielectric resonators to better account for wave - dynamical effects like diffraction . the agreement with the ray - optical calculations was less good for the case of an irrational right triangle , namely the pythagorean triangle . due to its right angle , it still features a simple po , and the fsr of the spectra corresponded to its optical length . the directions of the emission lobes , however , showed deviations from the classical expectations , and several emission lobes in addition to the predicted ones were found . also the photographs of the pt microlaser only partly agreed with the classical expectations . finally , two examples of irrational triangles lacking any symmetry and hence any simple pos ( with the exception of fagnano s orbit for the quasiequilateral triangle ) were investigated . both cases were deformations of previously investigated triangles . their modes seemed to be localized on ghost pos @xcite , i.e. , they resembled the modes localized on pos of the undeformed triangles . but even though their modes retained some of the features of the unperturbed triangles , their properties like the major emission directions could only be explained qualitatively , if at all . this means that , contrary to expectations , the lasing modes of triangular microlasers are not based on a po if no simple po exists in the corresponding billiard . it also demonstrates that there seems to be a significant , qualitative difference between microlasers with the shape of rational and irrational triangles which is surprising since any irrational triangle can be approximated by a rational one . the examples of irrational triangles demonstrate that the predictive power of ray optics decreases for irrational triangles like the pythagorean triangle , and quantitative predictions are no longer possible for irrational triangles without symmetries and simple pos . there is apparently a transition from symmetric triangles with simple classical dynamics for which the lasing properties can be very well explained by ray optics to less symmetric triangles with increasingly complicated classical dynamics for which ray - optical methods no longer yield a good description of the lasing properties . so the complexity of the classical dynamics is directly reflected in the complexity of the corresponding microlasers . it remains an interesting challenge to better understand the latter cases by further experimental and numerical studies . the authors thank e. bogomolny and a. nosich for illuminating and fruitful discussions . s. b. gratefully acknowledges funding from the european union seventh framework programme ( fp7/2007 - 2013 ) under grant this work was supported by a public grant from the laboratoire dexcellence physics atom light matter ( labex palm ) overseen by the french national research agency ( anr ) as part of the investissements davenir program ( reference no . anr-10-labx-0039 ) . 81ifxundefined [ 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 10.1007/bf01388890 [ * * , ( ) ] link:\doibase 10.1103/physreve.52.2066 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1090/s0002 - 9947 - 98 - 02089 - 3 [ * * , ( ) ] link:\doibase 10.1007/s000140050113 [ * * , ( ) ] link:\doibase 10.1080/10586458.2006.10128961 [ * * , ( ) ] link:\doibase 10.1007/s10711 - 007 - 9129 - 9 [ * * , ( ) ] @noop _ _ ( , , ) @noop _ _ ( , , ) link:\doibase 10.1103/physrevlett.67.785 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.64.2215 [ * * , ( ) ] in @noop _ _ , , vol . , ( , , ) pp . link:\doibase 10.1103/physrevlett.83.2171 [ * * , ( ) ] @noop `` , '' ( ) , ed . , @noop _ _ ( , , ) link:\doibase 10.1364/oe.10.000752 [ * * , ( ) ] link:\doibase 10.1364/ol.27.000824 [ * * , ( ) ] link:\doibase 10.1103/physreve.67.015207 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.023830 [ * * , ( ) ] link:\doibase 10.1103/physreva.62.013816 [ * * , ( ) ] link:\doibase 10.1364/ao.46.002969 [ * * , ( ) ] link:\doibase 10.1103/physreve.83.036208 [ * * , ( ) ] link:\doibase 10.1007/s003400000210 [ * * , ( ) ] link:\doibase 10.1364/ol.29.000005 [ * * , ( ) ] link:\doibase 10.1109/jstqe.2006.883150 [ * * , ( ) ] \doibase doi:10.1364/oe.16.016516 [ * * , ( ) ] link:\doibase 10.1364/ol.33.000509 [ * * , ( ) ] link:\doibase 10.1364/ol.34.001810 [ * * , ( ) ] link:\doibase 10.1364/ol.35.002723 [ * * , ( ) ] link:\doibase 10.1103/physreve.83.016208 [ * * , ( ) ] link:\doibase 10.1103/physreva.72.063806 [ * * , ( ) ] link:\doibase 10.1063/1.2779094 [ * * , ( ) ] link:\doibase 10.1063/1.3276069 [ * * , ( ) ] link:\doibase 10.1364/ol.29.002758 [ * * , ( ) ] link:\doibase 10.1140/epjp / i2012 - 12008 - 0 [ * * , ( ) ] link:\doibase 10.1109/aps.1991.174903 [ * * , ( ) ] link:\doibase 10.1364/josaa.24.002822 [ * * , ( ) ] link:\doibase 10.1109/tap.2009.2038181 [ * * , ( ) ] link:\doibase 10.1016/0167 - 2789(81)90024 - 5 [ * * , ( ) ] link:\doibase 10.2307/1971280 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1090/s0002 - 9939 - 2012 - 11379 - 6 [ * * , ( ) ] link:\doibase 10.1063/1.532762 [ * * , ( ) ] @noop _ _ ( , , ) @noop _ _ ( , , ) link:\doibase 10.1103/physreva.77.013804 [ * * , ( ) ] link:\doibase 10.1063/1.1665328 [ * * , ( ) ] link:\doibase 10.1063/1.1665596 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.53.1515 [ * * , ( ) ] @noop `` , '' ( ) , link:\doibase 10.1103/physrevlett.97.254102 [ * * , ( ) ] link:\doibase 10.1103/physreve.78.056202 [ * * , ( ) ] link:\doibase 10.1103/physreve.86.026202 [ * * , ( ) ] link:\doibase 10.1103/physreve.81.066215 [ * * , ( ) ] link:\doibase 10.1088/1751 - 8113/44/15/155305 [ * * , ( ) ] link:\doibase 10.1364/ol.36.001023 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.023834 [ * * , ( ) ] link:\doibase 10.1103/physreve.88.062906 [ * * , ( ) ] link:\doibase 10.1088/0951 - 7715/16/6/310 [ * * , ( ) ] link:\doibase 10.1109/tap.2010.2103031 [ * * , ( ) ] link:\doibase 10.1016/0167 - 2789(88)90075 - 9 [ * * , ( ) ] link:\doibase 10.1109/jqe.2010.2055836 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.033903 [ * * , ( ) ] link:\doibase 10.1103/physreva.86.043817 [ * * , ( ) ] link:\doibase 10.1088/0951 - 7715/9/4/012 [ * * , ( ) ] link:\doibase 10.1103/physreve.61.3689 [ * * , ( ) ] link:\doibase 10.1103/physreve.50.1886 [ * * , ( ) ] link:\doibase 10.1088/0305 - 4470/27/13/018 [ * * , ( ) ] link:\doibase 10.1063/1.4720474 [ * * , ( ) ] link:\doibase 10.1364/oe.22.012316 [ * * , ( ) ] link:\doibase 10.1103/physreve.68.066207 [ * * , ( ) ] link:\doibase 10.1088/1464 - 4258/7/9/009 [ * * , ( ) ] link:\doibase 10.1364/josab.23.001586 [ * * , ( ) ] link:\doibase 10.1103/physreva.76.023822 [ * * , ( ) ] link:\doibase 10.1109/jqe.2009.2024006 [ * * , ( ) ] link:\doibase 10.1364/josab.26.002449 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1088/0305 - 4470/34/40/306 [ * * , ( ) ] @noop link:\doibase 10.1103/physrevlett.100.204101 [ * * , ( ) ] link:\doibase 10.1103/physreve.78.045201 [ * * , ( ) ]

since last few decades one can observe a common pursuit towards miniaturization of electronic systems . according to the empirical moore s law , the number of transistors per microchip is doubled every two years . it is however obvious that the possibility of further miniaturization will be stopped in the near future due to the loss of chips stability when the device components achieve critical dimensions in the nanometer range . thus , the most challenging task facing contemporary science and technology is to implement structures , alternative to the silicon - based devices , whose size could be reduced further . the same tendency can also be observed in the development of hard discs , whose memory cells become smaller and smaller every year . this generates new challenges related to information reading / writing . an important step towards further miniaturization was the discovery of the giant magnetoresistance ( gmr ) effect . the gmr was discovered in 1988 in artificially layered metallic structures consisting of ferromagnetic 3d films separated by nonmagnetic metallic layers @xcite . it turned out that electrical resistance of magnetic metallic multilayers depends on their magnetic state , and usually drops when magnetic configuration varies from antiparallel alignment to the parallel one @xcite . the gmr effect occurs for current flowing in the film plane as well as perpendicularly to it . this phenomenon turned out to be very useful for applications in highly sensitive read heads , and allowed reading information from smaller memory cells using their weak magnetic field . the discovery of gmr initiated broad interest in spin polarized electronic transport in nanoscopic systems . it turned out that electron spin provides an additional degree of freedom , which considerably broadens the range of applications of mesoscopic systems in novel electronic devices . the spin - based nanoelectronics called now spin electronics or shortly spintronics is a relatively new area of mesoscopic physics dealing with the interplay of charge and spin degrees of freedom @xcite . although a lot of theoretical and experimental works on the spintronic properties of mesoscopic systems have been carried out , this field is still in an early stage of development . an effect similar to the aforementioned gmr also occurs when nonmagnetic metallic layer in a trilayer structure is replaced with a nonmagnetic insulating barrier , and the current flows owing to the phenomenon of quantum - mechanical tunnelling through the barrier . this effect was discovered in 1975 in ferromagnetic planar junctions by jullire @xcite , and is of current interest due to applications in magnetic storage technology ( magnetic random access memories ) and in spin - electronics devices @xcite . as in the case of gmr , the tunnel magnetoresistance ( tmr ) consists in a decrease ( increase is also possible ) in the junction resistance when magnetic configuration of the junction changes from the antiparallel to parallel one . tunneling in complex junctions , particularly in mesoscopic ones , where charging effects become important , is still not fully explored . a specific kind of such systems are double - barrier junctions with a small central electrode ( called an island ) . such systems are known as single - electron transistors ( sets ) , mainly because electrons in a biased device flow one by one through the system and the transfer of single electrons can be controlled by a gate voltage . electronic transport in such devices was extensively studied in the past decade , but mainly in the nonmagnetic limit @xcite . recent experiments on magnetic nano - structured materials revealed new phenomena associated with the interplay of ferromagnetism and discrete charging effects . a typical example is a ferromagnetic single - electron transistor , i.e. , a small metallic nanoparticle ( semiconducting quantum dots or molecules are also possible ) coupled by tunnel barriers to ferromagnetic electrodes @xcite . first ferromagnetic single - electron transistors were fabricated by ono et al . @xcite and later by brckl et al . transport in ferromagnetic single - electron transistors with nonmagnetic metallic islands both normal and superconducting was also measured @xcite . one should bear in mind , that the interplay of spin and charge effects was already studied long time ago in granular systems , in which magnetic nanoparticles were randomly dispersed in a nonmagnetic matrix @xcite . recently granular films were investigated again by several groups @xcite , but both the size of the grains and also their location were strictly controlled . in the case of sufficiently large metallic islands ( but still in the nanometer range ) , discrete structure of the electronic states in the grain is not resolved and is irrelevant . to observe the discrete electronic states in transport characteristics one should either diminish size of the metallic nanoparticles @xcite , or use semiconducting quantum dots based on two - dimensional electron gas @xcite . an alternative strategy is to use ferromagnetic semiconducting materials instead of metallic ones as the electrodes @xcite . magnetic impurities in the middle of the tunnel barrier of a ferromagnetic tunnel junction @xcite also can be considered as quantum dots with a very strong coulomb interaction . another group of single - electron devices are molecular ferromagnetic transistors @xcite and , especially , ferromagnetically contacted carbon nanotubes @xcite . in the following we will review basic transport characteristics of ferromagnetic single - electron transistors in the sequential tunnelling , cotunneling , and strong coupling ( kondo ) regimes . in particular , we will consider such properties of the device like charge and spin currents , tunnel magnetoresistance , spin accumulation , shot noise , kondo effect , and others . in section 2 we review basic principles of single - electron transport in the case of a large metallic island attached to ferromagnetic leads . the size of the island is however small enough so that the charging energy is the dominant energy scale in the system . the limits of fast and slow spin relaxation in the island are also discussed . then we consider shot noise and the role of discrete electronic structure in such devices . in section 3 we consider devices based on double metallic islands . in turn , in section 4 we deal with electronic transport through single - level quantum dots ( qds ) in the sequential tunnelling and cotunneling regimes . transport through multi - level qds , including transport through two - level quantum dots and carbon nanotubes attached to ferromagnetic leads is discussed in section 5 . transport through quantum dots in the kondo regime is briefly addressed in section 6 . final conclusions are in section 7 . the systems considered in this review consist of a mesoscopic central part ( island ) coupled by tunnel barriers to external ferromagnetic leads . the central part is characterized by an addition energy , which corresponds to the energy needed for adding a single electron , and includes contributions from the electrostatic charging energy and the discrete single - particle level separation . if the charging energy is the most relevant energy scale , the systems exhibit the single - electron charging effects @xcite . an electron can tunnel to the central part only when the energy provided by the transport voltage surpasses the corresponding charging energy ; otherwise , the current is exponentially suppressed and the system is in the coulomb blockade regime @xcite . once the bias voltage is larger than the threshold voltage , the electrons can tunnel one by one through the system leading to the step - like current - voltage characteristics so - called coulomb staircase . the blockade can also be overcome by applying a gate voltage @xmath0 that leads to sawtooth - like variation of electric current with @xmath0 so - called coulomb oscillations @xcite . when the electrodes are made of a ferromagnetic material , the system exhibits further interesting phenomena resulting from the interplay of charge and spin degrees of freedom @xcite . in particular , the tunnelling current flowing through the system depends on the relative alignment of the magnetic moments of ferromagnetic electrodes , giving rise to the tmr effect @xcite , which is described quantitatively by the ratio @xmath1 where @xmath2 and @xmath3 denote the total system resistance in the parallel and antiparallel magnetic configurations , respectively , and @xmath4 , @xmath5 are the corresponding currents . a simple theoretical model of tmr was introduced by jullire @xcite , who considered a single planar ferromagnetic tunnel junction and showed that tmr in such a device is given by @xmath6 , where the spin polarization @xmath7 of the lead @xmath8 ( @xmath9 for the left / right lead ) is defined as @xmath10 with @xmath11 being the spin - dependent density of states of lead @xmath8 for the spin - majority @xmath12 and spin - minority @xmath13 electrons . a sketch illustrating ( a ) a single first - order ( sequential tunnelling ) process , and ( b ) a second - order ( cotunneling ) transport process . ] in the following considerations we will distinguish between the three different transport regimes : _ sequential tunnelling _ in the regime of weak coupling between the island and leads , and out of the coulomb blockade , electron transport is dominated by processes of the first order in the coupling parameter . electrons flow then consecutively one by one due to the tunnelling events , and the transition rate from an initial state @xmath14 to a final state @xmath15 can be determined from the fermi golden - rule , @xmath16 where @xmath17 is the relevant tunnelling hamiltonian , and @xmath18 ( @xmath19 ) is the energy of the initial ( final ) state . this transport regime is known as the sequential tunnelling regime @xcite . an example of a single first - order process is sketched in fig . [ sec1:fig1]a . _ cotunneling _ although the sequential tunnelling in the coulomb blockade regime is exponentially suppressed , the current still flows due to higher - order tunnelling processes involving tunnelling of a single , two or more electrons _ via _ intermediate virtual states @xcite . these processes are known as cotunneling . an exemplary cotunneling process is illustrated in fig . [ sec1:fig1]b . the second order perturbation theory gives the cotunneling rate from an initial state @xmath14 to a final state @xmath15 @xcite @xmath20 where the summation is over all virtual states @xmath21 , and @xmath22 is the energy od the state @xmath23 ( @xmath24 ) . in the coulomb blockade regime this tunnelling rate is only algebraically suppressed , contrary to the sequential transport processes which are then suppressed exponentially . because of that even at low temperatures and in the strong coulomb blockade regime the rates of cotunneling processes do not vanish . the second - order corrections become also important on resonance for intermediate coupling strengths . _ strong coupling _ for strong coupling of the metallic island to electrodes , the tunnelling processes lead to logarithmic corrections to conductance , and the perturbation theory fails at the degeneracy points of two consecutive charge states . in the case of quantum dots additionally the kondo effect appears at low temperatures ( below the kondo temperature @xmath25 , @xmath26 ) leading to an enhanced conductance in the linear response regime @xcite . in this section we review spin - polarized transport in a metallic ferromagnetic single - electron transistor ( fm set ) . the device consists of a metallic nanoparticle as the central electrode ( island ) , which is coupled through tunnel barriers to external reservoirs of spin polarized electrons . a gate voltage is attached capacitively to the island , which allows to control position of the corresponding fermi level . electronic transport in nonmagnetic sets was already extensively studied in the past two decades @xcite . recently , the attention was also drawn to electron tunnelling in magnetic systems @xcite , which was stimulated by recent progress in nanotechnology . it has been shown theoretically that some qualitatively new effects may arise from the interplay of charging effects and spin degrees of freedom . these include , for example , oscillations of tmr with increasing bias voltage , spin accumulation , enhancement of tmr in the coulomb blockade regime , etc . @xcite . the enhancement of tmr in the cotunneling regime and the oscillations of tmr as a function of the transport voltage have also been observed experimentally @xcite . the capacitance @xmath27 of few - nanometer - size particles is of the order of @xmath28 f @xcite . consequently , the corresponding charging energy , @xmath29 , establishes a new relevant energy scale . if the charging energy is larger than the thermal energy , @xmath30 , where @xmath31 denotes temperature and @xmath32 is the boltzmann constant , the effects due to discreteness of charge become observable in transport characteristics @xcite . the single - electron transistors considered in this section are illustrated schematically in fig . [ sec2:fig1 ] , where part ( a ) shows a device whose all three electrodes are ferromagnetic , whereas part ( b ) shows a system with ferromagnetic source and drain electrodes and a nonmagnetic island . generally , magnetic moments of the leads may form an arbitrary magnetic configuration . however , we focuss on systems whose moments form either parallel ( p ) or antiparallel ( ap ) magnetic configuration , as shown in fig . [ sec2:fig1 ] . in the system shown in part ( a ) the magnetic moments of external electrodes point in the same direction , while magnetic moment of the island is either parallel or antiparallel to them . in the system shown in part ( b ) , magnetic moments of external electrodes in the parallel configuration are aligned , while in the antiparallel configuration they are anti - aligned . the two magnetic alignments can be easily achieved by sweeping magnetic field through the hysteresis loop , provided the respective ferromagnetic components have different coercive fields . one may also make use of exchange anisotropy to fix magnetic moment of a particular layer and rotate magnetic moment of the second layer with a weak magnetic field . generally , there is a chance that electron tunnelling through a barrier will change its spin orientation . however , we consider only spin - conserving tunnelling processes through the two barriers . first , we assume that the islands are relatively large , so the effects due to quantization of the corresponding energy levels can be neglected . for smaller islands , however , the discrete energy spectrum may modify transport characteristics @xcite , and this will be considered later ( see section [ discreteset ] ) . schematics of ferromagnetic single - electron transistors . the parallel and antiparallel magnetic configurations of the system are also specified . the island is separated from external electrodes by tunnel barriers . each barrier is characterized by its spin - dependent resistance @xmath33 and capacitance @xmath34 ( @xmath35 ) . the system is symmetrically biased , @xmath36 , @xmath37 , and there is also a gate voltage @xmath38 applied to the island . ] to describe charge and spin transport we need to specify a model hamiltonian of the system . first , the electrostatic energy required to add @xmath39 excess electrons to the island , while keeping constant voltages @xmath40 and @xmath41 in the left and right electrodes and the gate voltage @xmath42 , is given by @xcite @xmath43 where a constant term ( independent of @xmath39 ) is irrelevant and has been dropped . here , the total island s capacitance @xmath27 is the sum of the capacitances of the left and right junctions and of the gate , @xmath44 . the external charge @xmath45 accounts for the effect of applied voltages , and can be continuously tuned . the total hamiltonian of the device takes then the general form @xmath46 here @xmath47 describes the decoupled leads and island , while @xmath48 takes into account the lead - island coupling . the ferromagnetic leads and the island are described by @xmath49 for @xmath50 , where @xmath51 are the fermi operators for electrons with a wavevector @xmath52 and spin @xmath53 in the electrodes and island ( @xmath50 ) , and @xmath54 is the corresponding single - particle energy . the last part of the hamiltonian , @xmath55 describes tunnelling processes between the leads and island , with @xmath56 being the relevant matrix elements . when the resistances of both tunnel barriers are much larger than the quantum resistance , @xmath57 , and the system is not in the coulomb blockade regime , transport is dominated by sequential tunnelling processes . as a consequence , the charge is well localized in the island and the _ orthodox _ tunnelling theory is applicable @xcite . in order to calculate electric current in the sequential transport regime , one may use the method based on the master equation which is a detailed balance of electrons tunnelling to and off the island . in the stationary case the master equation reads @xmath58 p(n , v)\nonumber \\ & & + \sum\limits_{\sigma } \left[\gamma _ { l\sigma}^+(n-1 ) + \gamma _ { r\sigma}^+(n-1)\right]p(n-1,v ) \nonumber\\ & & + \sum\limits_{\sigma } \left[\gamma _ { l\sigma}^-(n+1 ) + \gamma _ { r\sigma}^-(n+1)\right]p(n+1,v),\end{aligned}\ ] ] where @xmath59 is the probability to find the island in a state with @xmath39 additional electrons when a bias voltage @xmath60 ( @xmath61 ) is applied , and @xmath62 is the spin - dependent rate for tunnelling of electrons with spin @xmath53 from the lead @xmath8 to island ( upper sign ) and from the island to lead @xmath8 ( lower sign ) , when the island is occupied by @xmath39 excess electrons . these tunnelling rates depend on the bias voltage ( not indicated explicitly ) , and can be expressed by means of the fermi golden rule as @xmath63 -1}\,,\ ] ] where @xmath64 describes a change in the electrostatic energy of the system caused by the corresponding tunnelling event , when in the initial state there were @xmath39 additional electrons on the island . in the above equation @xmath33 denotes the spin - dependent tunnel resistance of the @xmath8-th barrier , given by @xmath65 where @xmath66 has been assumed for simplicity . the spin dependence of the resistance is a consequence of the spin - dependent density of electron states at the fermi level in the respective electrodes and the corresponding tunnelling matrix elements . in particular , in eq . ( [ eq : setbarrierr ] ) it is due to the spin - dependent density of states of the @xmath8-th lead , @xmath67 , spin - dependent density of states of the island , @xmath68 , as well as due to the spin - dependent tunnelling matrix elements @xmath69 . in a stationary state , the net transition rate between the charge states with @xmath39 and @xmath70 excess electrons on the island vanishes . the probability @xmath59 can be then determined from the following recursion relation @xcite @xmath71 where @xmath72 and @xmath73 , corresponding to transition rates for tunnelling to and off the island , respectively . generally , energy of an electron after tunnelling event is relaxed to the relevant fermi level in a short time scale . one can assume that the energy relaxation time is the shortest time scale , shorter than the time between two successive tunnelling events . however , such a restriction can not be imposed on the spin relaxation time which can be relatively long . in a general case , a nonequilibrium magnetic moment may accumulate on the island due to the spin dependence of tunnelling processes , which leads to spin splitting of the corresponding fermi level . for arbitrary spin relaxation times , the splitting of the fermi level can be determined from the spin balance @xcite @xmath74 where @xmath75 is the island s volume , @xmath76 is the electron charge ( @xmath77 ) , @xmath78 denotes the spin relaxation time on the island , while @xmath79 and @xmath80 are the currents flowing through the left and right junctions in the spin channel @xmath53 . the last term in eq . ( [ eq : setspinconser ] ) takes into account intrinsic spin - flip processes on the island . > from this condition it is possible to calculate self - consistently the shifts of the fermi levels due to spin accumulation for both spin orientations . the electric current flowing through the left junction can be calculated from the following formula : @xmath81 p(n , v ) \,.\ ] ] similar formula also holds for @xmath82 . in the stationary state the currents flowing through both junctions are equal , @xmath83 . in the following we discuss two limiting cases : the limit of fast spin relaxation and the limit of slow spin relaxation on the island . in the former case the spin of an electron tunnelling to the island relaxes before a next tunnelling event takes place . in the latter case , on the other hand , the electron spin is conserved for a time much longer than the time between successive tunnelling events . in the limit of fast spin relaxation , sets with nonmagnetic islands behave like nonmagnetic junctions , exhibiting no tmr effect . therefore , the following discussion in this subsection is limited to fm sets with ferromagnetic islands , see fig . [ sec2:fig1]a . generally , one may assume that the spin - dependent resistances for the parallel ( @xmath84 ) and antiparallel ( @xmath85 ) configurations fulfill the condition @xmath86 , where @xmath87 . this formula follows from eq . ( [ eq : setbarrierr ] ) when assuming that all spin effects are included into the spin dependent density of states ( transfer matrix element are independent of spin and magnetic configuration ) . moreover , since there is no spin accumulation on the island in the limit of fast spin relaxation , there is no associated spin splitting of the fermi level , @xmath88 . current in the parallel and antiparallel magnetic configurations ( a ) and the resulting tmr ( b ) as a function of the bias voltage for a fm set with a ferromagnetic island in the limit of fast spin relaxation . the parameters are : @xmath89 k , @xmath90 m@xmath91 , @xmath92 m@xmath91 , @xmath93 m@xmath91 , @xmath94 m@xmath91 , @xmath95 ( for @xmath35 ) , @xmath96 af , @xmath97 af , @xmath98 af , @xmath99 , and the offset charge @xmath100 . the dots in part ( b ) present the experimental data taken from ref . @xcite . ] the current flowing through the system in the parallel and antiparallel configurations is displayed in fig . [ sec2:fig2]a . for both magnetic configurations , the @xmath101 curves reveal the well - known coulomb steps . moreover , these two curves are different ; the current flowing in the parallel configuration is generally larger than the current flowing in the antiparallel configuration , see fig . [ sec2:fig2]a . this difference leads in turn to nonzero tmr effect , as shown in fig . [ sec2:fig2]b . the tmr effect has a component that oscillates as a function of the bias voltage . the amplitude of these oscillations , however , decreases as the transport voltage increases . for the parameters assumed here , tmr reaches local maxima at the voltages corresponding to the positions of coulomb steps . however , this is not a general rule , and for other parameters tmr can have local minima at the coulomb steps . the global maximum value of tmr in fig . [ sec2:fig2]b appears at the first step , i.e. , at the threshold voltage . when the temperature increases , the effects due to discrete charging , i.e. the coulomb steps and enhancement of tmr at the coulomb steps become diminished and disappear at @xmath102 @xcite . the oscillatory behavior of the tmr effect with increasing transport voltage was observed experimentally , for example by ernult _ et al . _ @xcite , see the dots in fig . [ sec2:fig2]b . the curves presented in fig . [ sec2:fig2 ] are calculated for the parameters corresponding to those in ref . @xcite . to get a good agreement with experimental observations , a nonzero offset charge @xmath103 on the island ( due to external charges ) has been assumed . as can be seen in fig . [ sec2:fig2]b , there is a satisfactory agreement between the theoretical curve and experimental data for voltages above the threshold , @xmath104mv , while in the coulomb blockade regime transport calculated using the sequential tunnelling approximation is not properly described . in the case of a fm set with a nonmagnetic island , see fig . [ sec2:fig1]b , a nonzero tmr in the sequential tunnelling regime can exist only when the spin relaxation time is sufficiently long , i.e. significantly longer than the time between successive tunnelling events . we note that the longest spin relaxation times were measured for aluminium and copper @xcite ( for example , the relaxation time for copper was estimated to be of the order of @xmath105 s ) . if this is the case , a nonequilibrium magnetic moment builds up on the island due to spin accumulation . this moment leads a nonvanishing tmr . in other words , the island becomes magnetized in a nonequilibrium situation , and the created moment depends on the bias and gate voltages . basic characteristics of a fm set with a nonmagnetic island as a function of the bias voltage in the limit of slow spin relaxation on the island : ( a ) spin splitting of the fermi level , ( b ) @xmath101 characteristics , and ( c ) tunnel magnetoresistance . the parameters are : @xmath106 , @xmath107 m@xmath108 , @xmath109 m@xmath91 , @xmath110 m@xmath108 , @xmath111 m@xmath91 , @xmath112 af , and @xmath99 . ] since the density of states at the fermi level in a nonmagnetic island is independent of the spin orientation , one finds @xmath113 . apart from this , the resistances in the antiparallel configuration are @xmath114 and @xmath115 , see also fig . [ sec2:fig1]b . typical transport characteristics of a fm set with nonmagnetic island in the limit of slow spin relaxation are shown in fig . [ sec2:fig3 ] as a function of the bias voltage . the splitting of the fermi level is displayed in fig . [ sec2:fig3]a for both magnetic configurations . as one could expect , the splitting takes place only in the antiparallel alignment , while in the parallel configuration there is no spin accumulation . in the former case the ratio of tunnelling rates for electrons with opposite spin orientations becomes the same for electrons tunnelling to and off the island only when a nonequilibrium magnetic moment is built on the island . in the parallel configuration , however , this condition is already fulfilled without any spin accumulation . however , this is true only in the case when both junctions are characterized by equal spin asymmetries . when the spin asymmetries of both electrodes are different , spin accumulation also occurs in the parallel configuration . as illustrated in fig . [ sec2:fig3]a , behavior of the fermi level splitting with increasing transport voltage can be decomposed into two components . one component monotonously increases , while the second one oscillates with increasing bias voltage . this oscillatory behavior can be accounted for in the following way . let us assume that the voltage is slightly above that corresponding to a certain coulomb step and begins to increase . then , the spin accumulation also increases , until a local maximum value is reached . the local maximum occurs at a voltage , at which the chemical potential of the depleted spin channel approaches the value which allows the next charge state in the island . this , in turn , enhances tunnelling rate ( onto the island ) of electrons corresponding to the depleted spin channel , and consequently reduces the spin splitting of the fermi level . when the voltage increases further , a local minimum in the spin accumulation is then reached at a voltage , where the chemical potential of the second ( accumulated ) spin channel approaches the value which allows the next charge state on the island . the same scenario repeats at each coulomb step leading to the oscillatory component in the spin accumulation . in the sequential tunnelling regime , where only the first - order tunnelling processes are taken into account , spin accumulation is exponentially small in the coulomb blockade regime , as can be seen in fig . [ sec2:fig3]a . the current as a function of the bias voltage for the parallel and antiparallel configurations is shown in fig . [ sec2:fig3]b . as before , characteristic coulomb steps are clearly visible . moreover , owing to different stationary spin accumulations in the parallel and antiparallel configurations , the current in the parallel configuration is also different from that in the antiparallel configuration . this gives rise to the tmr effect which is presented in fig . [ sec2:fig3]c . now , the bias dependence is more complex than it was in the case of fm sets with magnetic islands in the absence of spin accumulation . it is interesting to note , that tmr can change sign in the vicinity of the coulomb steps in current - voltage curves , which is a consequence of different spin accumulations in the two magnetic configurations . generally , it is more difficult to obey the slow spin relaxation limit in ferromagnetic islands than in nonmagnetic ones . anyway , some experimental data show that this is achievable . in such a case , the spin accumulation builds up in the magnetic island and has significant influence on transport characteristics . in refs . @xcite it has been shown that spin accumulation modifies the staircase-like variation of the electric current with the bias voltage . since the spin accumulation depends on the magnetic configuration of the junction , this can also lead to an enhanced tmr effect . moreover , in some voltage regions tmr can change sign . in addition , the coulomb steps for the two magnetic configurations become slightly shifted , as was also observed experimentally @xcite . the difference between current - voltage curves in the fast and slow spin relaxation limits varies continuously with spin relaxation rates . this difference was used very recently to evaluate spin relaxation time in co nanoparticles @xcite . the electric current and spin accumulation in the limit of sequential tunnelling ( see the preceding subsection ) are exponentially small in the coulomb blockade regime . however , charge transport may occur not only due to spin - dependent sequential tunnelling , but also due to spin - dependent cotunneling processes @xcite , which give dominant contribution in the coulomb blockade regime and also lead to spin accumulation @xcite . close to resonance ( in the vicinity of the threshold voltage ) both sequential and cotunneling currents may be comparable . the current @xmath116 is then equal to the sum of first , @xmath117 , and second , @xmath118 , order contributions , @xmath119 . to calculate cotunneling current far from the resonance one could use eq . ( [ sec1:eq : cotrate ] ) . this formula , however , can not be used when voltage approaches the threshold voltage ( at resonance ) . to calculate current and associated spin accumulation in the whole voltage range , the real - time diagrammatic formalism @xcite has been used . the spin accumulation on the island ( or equivalently spin splitting of the electrochemical potential ) is then determined from the spin balance equation , similarly as in the sequential tunnelling regime , see eq . ( [ eq : setspinconser ] ) . differential conductance in fm sets with nonmagnetic islands as a function of the bias voltage @xmath60 in the ( a ) parallel and ( b ) antiparallel configurations , calculated in the whole ( sequential and cotunneling ) transport regime in the limit of slow spin relaxation in the island . the curves have been calculated for @xmath120 and symmetric junctions with @xmath121 . the spin polarization is @xmath122 and @xmath123 for ni and fe electrodes , respectively . the dotted line corresponds to the sequential tunnelling limit in the set with fe electrodes . ( after ref . @xcite ) ] the differential conductance in the whole ( sequential and cotunneling ) transport regime is shown in fig . [ sec2:fig4 ] for sets with ni and fe electrodes and for nonmagnetic islands ( in the limit of slow spin relaxation ) . upper part corresponds to the parallel configuration while the lower one to the antiparallel one . first , we find well resolved splitting of the conductance peaks in the antiparallel alignment , while no splitting can be seen in the parallel configuration . this splitting in conductance peaks is a direct consequence of the spin splitting of the corresponding electrochemical potential of the island , and therefore can be used to detect and measure spin accumulation . the absence of conductance splitting in the parallel configuration is simply a consequence of the absence of spin accumulation in this configuration . such a splitting of the conductance peaks in the cotunneling regime was recently observed experimentally @xcite . generally , there are several experimental techniques by which the spin accumulation can be detected indirectly @xcite . the peculiarities of transport characteristics of fm sets offer new possibilities . ( color online ) the influence of discrete electronic structure on transport characteristics in a fm set with nonmagnetic island in the limit of slow spin relaxation . voltage dependence of the tunnel current @xmath116 ( a ) , the differential conductance @xmath124 ( b ) , spin accumulation @xmath125 ( c ) , standard deviation @xmath126^{1/2 } $ ] ( d ) , and tunnel magnetoresistance @xmath127 ( e ) , calculated at @xmath128 . the solid and dashed curves in ( a ) , ( c ) and ( d ) correspond to the antiparallel and parallel configurations , respectively . the other parameters are : @xmath129 , @xmath130 , @xmath131 , @xmath132 , @xmath133 and @xmath134 for the parallel alignment ( @xmath135 for the antiparallel alignment ) . ( after ref . @xcite ) ] discussion up to now was limited to the case when addition of a single electron costed only an electrostatic energy . when , however , the size of metallic island becomes reduced further , electron spectrum of the island can not be considered as a continuous one , and discrete structure of the energy levels plays an important role . addition of an electron costs then not only the electrostatic charging energy , but also the energy equal to the level spacing @xmath136 . in this subsection we describe briefly the effect of discreteness on the transport characteristics . the relevant approach was developed in refs . @xcite . the electron transport in the stationary state is then governed by the solution of the generalized master equation @xcite @xmath137 where @xmath138 denotes the probability to find @xmath139 and @xmath140 excess electrons on the island ( @xmath141 is the total number of excess electrons ) . the first term in eq . ( [ seq2eq:1 ] ) describes how the probability of a given charge and spin state decays due to electron tunnelling to and from the island , whereas other terms describe the rate at which this probability increases . the @xmath91-terms account for spin - flip relaxation processes . the coefficients entering eq . ( [ seq2eq:1 ] ) are defined as @xmath142 and @xmath143 $ ] , where @xmath144 are the tunnelling rates for electrons with spin @xmath53 , tunnelling to ( @xmath145 ) the island from the lead @xmath146 or backward ( @xmath147 ) . these coefficients are given by @xcite @xmath148 here , @xmath149 is the fermi function ( @xmath150 ) , whereas @xmath151 ( @xmath152 ) describes the probability that the energy level @xmath153 is occupied by an electron with spin @xmath53 for a particular configuration @xmath154 . the parameter @xmath155 is the bare tunnelling rate of electrons between the lead @xmath156 and the energy level @xmath153 of the island , and @xmath157 is the transition probability from the state @xmath158 to @xmath159 of the island due to the spin - flip processes . the energies @xmath160 and @xmath161 are given by @xmath162 and @xmath163 where @xmath164 $ ] and @xmath165 . > from the solution @xmath138 of the master equation ( eq . [ seq2eq:1 ] ) , one can determine current flowing through the island , @xmath166 for further discussion we assume that the discrete energy levels @xmath153 are spin degenerate ( nonmagnetic situation ) and equally separated with the level spacing @xmath136 . to emphasize the role of spin accumulation let us assume that the intrinsic spin relaxation time on the island is long enough to neglect all intrinsic spin - flip processes . the corresponding @xmath101 characteristics for the parallel and antiparallel alignments are shown in fig . [ sec2:fig6]a . in both cases the electric current is blocked below a threshold voltage , and a typical coulomb staircase appears above it , with additional small steps due to the discrete levels of the island . the effects due to discrete charging and discrete electronic structure are more clearly seen in the differential conductance shown in fig . [ sec2:fig6]b , where the small peaks correspond to new discrete levels taking part in transport . the difference between the @xmath116-@xmath60 characteristics for the parallel and antiparallel configurations is due to a different spin accumulation in both geometries . in fig . [ sec2:fig6]c we see the average value of the difference between the numbers of spin - up and spin - down excess electrons on the island , @xmath167 , i.e. , the spin accumulation . there is no significant spin accumulation in the parallel configuration . the number @xmath168 of spins accumulated on the island fluctuates in time around its average value @xmath169 , as shown in fig . [ sec2:fig6]d , where the standard deviation @xmath170^{1/2}$ ] is plotted against the voltage @xmath60 . it is worth to note that although there is almost no spin accumulation in the parallel configuration , the corresponding fluctuations are relatively large . the difference between the @xmath101 curves in the parallel and antiparallel configurations leads to the tunnel magnetoresistance , shown in fig . [ sec2:fig6]e . the broad peaks correspond to the coulomb steps , while the fine structure originates from the discrete structure of the density of states of the island . one can extend the presented formalism also to the case of a fm set with a ferromagnetic island @xcite and calculate nonequilibrium spin accumulation on the island ( when spin relaxation time is sufficiently long ) . using similar approach inoue and brataas @xcite analyzed current - induced magnetization reversal induced by spin accumulation rather than by spin torque . they found that the magnetization reversal is possible when a free energy change due to nonequilibrium spin accumulation is comparable to the anisotropy energy . > from the application point of view , an important transport characteristics of the system is the corresponding current noise . functionality of a system depends on the relevant noise to signal ratio , which should be as small as possible . however , the noise is also a source of additional information on the system properties like quantum and coulomb correlations @xcite . the shot noise in ferromagnetic single - electron transistors was studied theoretically in ref . @xcite , where the method developed for spinless electrons in single - electron transistors @xcite was extended to magnetic systems . the time correlation function of any two quantities @xmath171 and @xmath172 can be expressed as @xcite @xmath173 where , @xmath174 is the conditional probability to find the system in the final state with @xmath175 and @xmath176 excess electrons at time @xmath177 , if there was @xmath139 and @xmath140 excess electrons in the initial time t=0 . the probability @xmath178 can be determined from eq . ( [ seq2eq:1 ] ) . voltage dependence of the current shot noise at @xmath179 ( a ) in a fm set with nonmagnetic island and in the slow spin relaxation limit for @xmath180 , @xmath181 for the antiparallel alignment ( @xmath182 , for the parallel configuration ) ( other parameters as in fig . [ sec2:fig6 ] ) . in part ( b ) @xmath183 is split into two components : @xmath184 ( upper curves ) and @xmath185 ( lower curves ) . ( after ref . @xcite ) ] following this procedure , the current - current correlation function in a fm set considered in the preceding subsection has been calculated in ref . the corresponding fourier transform can be presented as @xmath186 where @xmath187 is the schottky value ( the frequency independent part ) , while the second term in eq . ( [ sec2eq:5 ] ) is the frequency dependent component . figure [ sec2:fig7]a shows the bias dependence of the zero - frequency current noise @xmath183 . the current noise is smaller in the antiparallel configuration than in the parallel one . this is because in the presence of spin accumulation ( which is significant only in the antiparallel alignment ) the amplitude of fluctuations is smaller . in fig . [ sec2:fig7]b @xmath183 is split into two parts ; the frequency independent component @xmath184 and the contribution @xmath185 arising from the frequency dependent part of the current noise , see eq . ( [ sec2eq:5 ] ) . the component @xmath184 is almost constant @xmath188 at the plateaux of the @xmath101 curve and increases with opening of new channels . dynamical correlations between the currents are described by @xmath189 . its value in the limit @xmath190 can be positive between the @xmath101 steps and negative when new channels become open . this is evident for the antiparallel alignment at @xmath191 , when opening a tunnelling channel for electrons with @xmath192 leads to negative dynamical correlations . this effect is almost compensated by an increase in @xmath184 , and therefore one gets only a small reduction of the current noise @xmath183 . in the power spectrum of the current @xmath189 one can distinguish two distinct relaxation times , one in the high and another one in the low frequency regions @xcite . in a wide voltage range the corresponding relaxation times are very close to the effective relaxation times for the charge and spin noise . the asymmetry between the tunnelling channels for electrons with the opposite spins leads to activation of the spin component in the current noise . in ref . @xcite the components @xmath193 and @xmath194 , corresponding to the charge and spin noise , respectively , have been extracted from @xmath195 . it has been shown that the charge component is almost constant whereas the spin component increases with spin polarization @xmath196 of the leads , and for @xmath197 can be much larger than the charge component . the analysis showed that both charge and spin fluctuations are relevant for the shot noise in fm sets . it has been also pointed out that a super - poissonian shot noise , the fano factor @xmath198 , can occur due to lifting of the spin degeneracy . some new features of transport characteristics appear in spin - polarized electronic transport through double - island structures @xcite . a typical double - island device , shown schematically in fig . [ sec3:fig1 ] , consists of two metallic islands separated from each other and from external reservoirs by tunnel barriers . a gate voltage is additionally attached capacitively to each island . generally , both islands and external electrodes can be either magnetic or nonmagnetic @xcite . in the following , we consider the situations when at least two electrodes ( external or central ) are ferromagnetic , and their magnetic moments form only collinear ( parallel and antiparallel ) configurations . apart from this , we limit discussion to continuous density of states in the islands . schematic of a double - island device . the first ( 1 ) and second ( 2 ) islands are capacitively coupled to two gate voltages , @xmath199 and @xmath200 , and separated from each other and from the left and right electrodes by tunnel barriers . ] hamiltonian of the double - island systems is similar to that used for single - island devices , but now the electrostatic energy is given by @xcite @xmath201 where @xmath202 is the number of excess electrons on the first ( second ) island and @xmath203 describes the respective charge induced by applied voltages , @xmath204 . apart from this , @xmath205 and @xmath206 denote the charging energies of the corresponding islands , whereas @xmath207 describes electrostatic coupling of the islands , @xmath208 with @xmath209 being the total capacitance of the first ( second ) island , @xmath210 , and @xmath211 denoting the capacitance of the middle junction . to find the stationary current flowing through the system one needs then to know the probabilities @xmath212 of finding the system in the charge state with @xmath213 and @xmath214 additional electrons on the first and second islands , respectively , when a bias voltage @xmath60 is applied . these probabilities can be calculated in a recursive way from the following steady - state master equation @xmath215 p(n_1,n_2,v)\nonumber \\ & & + \sum\limits_{\sigma } \gamma _ { l 1}^\sigma(n_{1}-1,n_{2})p(n_{1}-1,n_{2},v)\nonumber\\ & & + \sum\limits_{\sigma } \gamma_{1 l } ^\sigma(n_{1}+1,n_{2})p(n_{1}+1,n_{2},v ) \nonumber\\ & & + \sum\limits_{\sigma}\gamma _ { 1 2}^\sigma(n_{1}+1,n_{2}-1)p(n_{1}+1,n_{2}-1,v)\nonumber\\ & & + \sum\limits_{\sigma}\gamma _ { 2 1}^\sigma(n_{1}-1,n_{2}+1)p(n_{1}-1,n_{2}+1,v ) \nonumber\\ & & + \sum\limits_{\sigma}\gamma _ { r 2}^\sigma(n_{1},n_{2}-1)p(n_{1},n_{2}-1,v)\nonumber\\ & & + \sum\limits_{\sigma}\gamma _ { 2 r}^\sigma(n_{1},n_{2}+1)p(n_{1},n_{2}+1,v ) \,,\end{aligned}\ ] ] with the normalization condition @xmath216 . the corresponding transition rates are given by eq . ( [ eq : setrate ] ) , with the respective changes in the total system electrostatic energy . for example for tunnelling from the first island to the second one , the change in the electrostatic energy can be written as @xmath217 here , @xmath218 and @xmath219 denote the corresponding shifts of the chemical potentials for spin @xmath53 in the first and second islands , respectively . the spin asymmetry of tunnelling through the barrier @xmath8 will be characterized in this section by the asymmetry factor @xmath220 , defined as @xmath221 . this asymmetry factor corresponds to the ratio of the respective spin - dependent densities of states . in particular , for fm / nm junctions it is given by @xmath222 , whereas for fm / fm junctions @xmath223 , provided that the two ferromagnetic electrodes are built of the same material . we note that the relation between spin dependent barrier resistances in the parallel and antiparallel configurations of magnetic moments on the opposite sides of the barrier is the same as that described in sec.2 . when a bias voltage is applied to the system , a nonequilibrium spin accumulation may appear on the islands . generally , the shifts of the fermi level for both spin orientations are different . however , one can assume that the ratio of the fermi level shifts for the spin - up and spin - down electrons , defined as @xmath224 , fulfills the relation @xmath225 , for the first ( @xmath226 ) and second ( @xmath227 ) island , respectively , with @xmath228 being the spin - dependent density of states of the @xmath229-th island . as a consequence , for nonmagnetic islands one directly gets , @xmath230 . electric current flowing through the system can be determined from the following formula : @xmath231 p(n_{1},n_{2},v)\,,\end{aligned}\ ] ] which corresponds to the current flowing through the left junction , but @xmath232 in the stationary state . the average electron number on the islands ( a ) , currents ( b ) in the parallel and antiparallel configurations and the resulting tmr ( c ) as a function of the bias voltage . the parameters are : @xmath233 af , @xmath234 . the spin asymmetries of resistances in the parallel configuration are @xmath235 , @xmath236 , whereas the total junction resistances are @xmath237 m@xmath91 . in the antiparallel configuration , @xmath238 , @xmath239 , and @xmath240 . ] consider first transport characteristics of a system built of two ferromagnetic islands and nonmagnetic external electrodes , shown in fig . [ sec3:fig2 ] . different magnetic configurations of the system are specified in the inset of fig . [ sec3:fig2]c . when the spin relaxation time in the islands is much shorter than the time between two successive tunnelling events , no spin accumulation builds up on the islands . since there is an asymmetry between the two barrier resistances , some charge accumulates on the islands , as displayed in fig . [ sec3:fig2]a . first of all , the magnitude of excess charge on the islands increases with increasing voltage . for the barrier asymmetry assumed in fig . [ sec3:fig2 ] , the electrons easier tunnel to the second island from the right lead than out of the second island to the first one . as a consequence , the electrons accumulate on the second island . on the other hand , the electrons easier tunnel out of the first island to the left lead than from the second island to the first one . thus , the number of excess electrons on the first island decreases with increasing the bias voltage there are holes accumulated on the first island . this occurs in both magnetic configurations . the electric current flowing through the system in the parallel and antiparallel configurations is shown in fig . [ sec3:fig2]b . the coulomb steps due to discrete charging are clearly evident , and the difference in currents flowing in the parallel and antiparallel configurations gives rise to the tunnel magnetoresistance , shown in fig . [ sec3:fig2]c . the tmr effect oscillates as a function of the bias voltage and the amplitude of these oscillations decreases as the voltage is increased in a similar way as in the case of single - island fm sets , see fig . . moreover , some dips occur now in tmr at the voltages corresponding to the steps in the current - voltage characteristics . the shifts of the fermi levels for spin - up electrons ( a ) , currents in the parallel and antiparallel configurations ( b ) and tmr ( c ) as a function of the bias voltage . the parameters are : @xmath233 af , @xmath241 , @xmath230 , @xmath242 , @xmath243 , @xmath235 , @xmath244 , whereas @xmath245 m@xmath91 . in the antiparallel configuration , @xmath246 , @xmath247 . ] when the spin relaxation time is longer than the time between two successive tunnelling events , a nonequilibrium magnetic moment appears on each island . the corresponding shifts of the fermi level due to spin accumulation can be calculated from spin current conservation , see eq . ( [ eq : setspinconser ] ) , written for each island @xcite . figure [ sec3:fig3 ] shows the shifts of the fermi levels and currents in the parallel and antiparallel configurations , as well as the resulting tmr calculated in the limit of long spin relaxation time for the system built of ferromagnetic electrodes and nonmagnetic islands . the magnetic moments of external electrodes can form either parallel or antiparallel configurations , as illustrated in the inset of fig . [ sec3:fig3]c . first of all , the nonequilibrium spin accumulation , shown in fig . [ sec3:fig3]a , exists not only in the antiparallel configuration but also in the parallel one . in the antiparallel configuration the shifts of the fermi level for a given spin orientation are equal on both islands , whereas in the parallel configuration they are opposite . the effects due to discrete charging lead to an oscillatory behavior of the fermi level shift , in a similar way as in the case of single - island fm sets with nonmagnetic islands discussed above . the currents flowing through the system in both magnetic configurations are shown in fig . [ sec3:fig3]b . due to nonequilibrium spin accumulation induced on the islands , these currents are different , which leads to nonzero tunnel magnetoresistance , as displayed in fig . [ sec3:fig3]c . it is interesting to note , that now tmr changes sign in certain transport voltage regions . these effects are clearly due to magnetic moments accumulated on the islands . if the spin relaxation time becomes shorter than the time between successive tunnelling events , spin accumulation disappears and , consequently , tmr also vanishes . the shifts of the fermi levels for majority - spin electrons ( a ) , currents ( b ) and differential conductance ( c ) in the parallel and antiparallel configurations , and the resulting tmr ( d ) as a function of the bias voltage . the parameters are : @xmath248 k , @xmath249 af , @xmath250 af , @xmath251 af , @xmath252 , @xmath253 , and @xmath254 , whereas @xmath255 . the total junction resistances are @xmath256 m@xmath91 , @xmath257 m@xmath91 . in the antiparallel configuration , @xmath258 , and @xmath259 , for @xmath260 . ( after ref . @xcite ) ] recently , several experiments on spin - polarized transport through granular systems were reported @xcite . for example , in ref . @xcite , the tunnelling current was driven from a tip of scanning tunnelling microscope through ferromagnetic grains to ferromagnetic electrode . transport measurements of such devices showed pronounced coulomb steps in the current - voltage characteristics . moreover , negative differential conductance was observed . it was further proposed that the negative differential conductance could be a consequence of a nonequilibrium spin accumulation . such systems can be modelled theoretically by double - island devices whose two islands and the right electrode are ferromagnetic , whereas the left electrode is nonmagnetic , corresponding to the nonmagnetic tip of scanning tunnelling microscope , see the inset of fig . [ sec3:fig4]c . transport characteristics of such a device are displayed in fig . [ sec3:fig4 ] for the parameters taken from ref . the shifts of the fermi level for spin - up electrons are shown in fig . [ sec3:fig4]a . these shifts are different in both magnetic configurations . generally , spin accumulation on the first island is larger than the accumulation on the second island . the reason for this is the fact that the rate for electron tunnelling from the first island to the left lead is smaller than the rate for tunnelling of electrons to or from the second island , which is due to asymmetry of barriers . the currents flowing through the system in the parallel and antiparallel configurations are illustrated in fig . [ sec3:fig4]b . moreover , negative differential conductance occurs in both magnetic configurations , however , it is more pronounced in the antiparallel configuration , as shown in fig . [ sec3:fig4]c . it can be seen that negative differential conductance increases with increasing the bias voltage . the resulting tmr effect is displayed in fig . [ sec3:fig4]d . it is interesting to note that tmr oscillates between negative and positive values . ( color online ) differential conductance in the antiparallel configuration as a function of the bias voltage calculated for different values of the spin relaxation time @xmath261 and for @xmath262/ev . the other parameters as well as magnetic configuration of the system are the same as in fig . [ sec3:fig4 ] . ( after ref . @xcite ) ] ( color online ) tmr as a function of the bias voltage calculated for different values of spin relaxation time @xmath261 and for @xmath262/ev . the other parameters as well as magnetic configuration of the system are the same as in fig . [ sec3:fig4 ] . ( after ref . @xcite ) ] oscillations of the sign of tmr and differential conductance result from spin accumulations in both islands and are absent in the limit of fast spin relaxation ( no spin accumulation ) in the islands . the results presented in fig . [ sec3:fig4 ] correspond to the long spin relaxation limit . in such a limit , some spin accumulation may occur even for a very small current flowing through the system , giving rise to ndc and oscillations of the tmr sign for small bias voltages . however , both effects disappear when the spin relaxation time is shorter than the time between successive tunnelling events . thus , for a finite relaxation time , one may expect absence of ndc and tmr oscillations for small voltages and the onset of these effects at larger voltages . this is because at some voltage there is a crossover from the fast to slow spin relaxation limits . in fact such a behavior is consistent with experimental data of ref . @xcite . the disappearance of ndc with decreasing spin relaxation time @xmath263 is shown explicitly in fig . [ sec3fig : g_ap ] , where the bias dependence of differential conductance is presented for different values of the spin relaxation time . this figure clearly shows that ndc disappears when spin relaxation time decreases , in agreement with the above discussion . in the limit of fast spin relaxation time , the differential conductance is positive , although its periodic modulation still remains . similarly , periodic oscillations of the sign of tmr also disappear with decreasing spin relaxation time @xmath264 . this behavior is shown in fig . [ sec3fig : tmr ] , where the bias dependence of tmr is shown for several values of @xmath264 . first , the transitions to negative tmr disappear with decreasing @xmath264 . the tmr becomes then positive , although some periodic modulations survive . second , the phase of the modulations shifts by about @xmath265 when the spin relaxation varies from fast to slow limits . in the case of analyzed systems , the negative differential conductance occurs due to nonequilibrium spin accumulation on the islands . it is however worth noting that the negative differential conductance may also exist in single - electron devices built of nonmagnetic materials @xcite . in the previous two sections we considered spin - dependent transport through single - electron devices , whose central electrodes ( islands ) were described by continuous ( or discrete with small level separation ) energy spectrum , and the most relevant and dominant energy scale was the electrostatic charging energy . in ultra - small metallic islands ( nanoparticles ) or in semiconducting quantum dots , the level spacing is comparable with the charging energy or even larger . this also happens in the case of a molecule attached to metallic leads . in this limit one arrives at slightly more sophisticated single - electron devices @xcite . in this section we consider a fm set based on a semiconductor quantum dot coupled to ferromagnetic leads . this model also applies to molecules attached to ferromagnetic electrodes . single - electron transistors based on quantum dots are of current interest not only because of new and interesting physics emerging in those systems , but , more importantly , due to possible future applications and due to the possibility of manipulation of a single electron charge and a single electron spin @xcite . furthermore , quantum dots are also interesting for future applications in quantum computing @xcite . transport properties of quantum dots coupled to nonmagnetic leads have already been extensively studied both theoretically and experimentally @xcite . however , further interesting effects occur in the case of quantum dots coupled to ferromagnetic leads , e.g. spin accumulation , parity effect on tunnel magnetoresistance , zero - bias anomaly in the coulomb blockade regime , exchange field , splitting of the kondo anomaly , and others . most of the works concerned theoretical description of spin - polarized transport in the weak coupling regime , as well as in the strong coupling regime , where the kondo physics emerges @xcite . sequential transport through a single - level quantum dot coupled to ferromagnetic leads was studied for both collinear @xcite and non - collinear @xcite configurations of the electrodes magnetic moments . spin - polarized transport in the cotunneling regime has also been addressed for collinear systems @xcite , as well as for systems magnetized non - collinearly @xcite . furthermore , the resonant tunnelling was also considered @xcite . quantum dots coupled to ferromagnetic leads may be realized experimentally in various ways , including ultrasmall metallic ( e.g. aluminum ) nanoparticles @xcite , single molecules @xcite , granular structures @xcite , self - assembled dots in ferromagnetic semiconductors @xcite , carbon nanotubes @xcite , and magnetic tunnel junctions @xcite . quite recently , semiconductor quantum dots based on two - dimensional electron gas were successfully attached to ferromagnetic leads @xcite . let us begin our discussion with the case when the dot is weakly coupled to the leads . one can then use a perturbative approach in which the coupling is considered as a small perturbation to the unperturbed quantum dot and the leads . transport is then dominated either by the first order or by the second order ( cotunneling ) contributions . later on we consider a more general situation when the coupling may be strong and a new kondo physics emerges at low temperatures . schematic of a quantum dot coupled to ferromagnetic leads . the magnetic moments of external electrodes can be aligned either in parallel or antiparallel . the system is symmetrically biased and there is a gate voltage attached to the dot . ] a schematic of a quantum dot coupled to ferromagnetic leads is presented in fig . [ sec4:fig1 ] . the magnetizations of the leads can be either parallel or antiparallel . there is also a gate voltage attached to the dot . the system is modelled by an anderson - like hamiltonian of the general form @xcite @xmath266 where the first and second terms describe the left and right reservoirs of noninteracting electrons , @xmath267 for @xmath268 [ see eq . ( [ eq : hamiltonian2 ] ) ] . the third term of the hamiltonian , @xmath269 , represents the dot and includes two components : one describes noninteracting electrons in the dot level and the other represents the coulomb interaction of two electrons residing in this level , @xmath270 where @xmath271 is the correlation energy , @xmath272 is the energy of an electron in the dot with spin @xmath53 , and @xmath273 ( @xmath274 ) is the corresponding creation ( annihilation ) operator . the position of the dot level can be tuned by the gate voltage , but is independent of the symmetrically applied transport voltage . generally , the dot level may be spin - split , for example due to a stray field of the electrodes or due to an external magnetic field . the corresponding level splitting is denoted by @xmath275 , whereas @xmath276 is the energy of the spin - degenerate dot level . in general , four different states of the dot are possible : empty dot ( @xmath277 ) , singly occupied dot with a spin - up ( @xmath278 ) or spin - down ( @xmath279 ) electron , and doubly occupied dot ( @xmath280 ) , where @xmath281 are the corresponding eigenfunctions . interaction between the leads and quantum dot is incorporated in the tunnelling hamiltonian , @xmath282 , given by @xmath283 where @xmath284 are the tunnel matrix elements . tunnelling gives rise to an intrinsic broadening @xmath285 of the dot levels , @xmath286 . the parameters @xmath287 and @xmath288 describe contributions to the level widths due to coupling of the dot to the lead @xmath8 . the respective contribution @xmath289 can be expressed in terms of the fermi golden rule as @xmath290 . assuming the tunnel matrix elements @xmath291 to be independent of the wave vector @xmath292 , one can write @xmath293 the coupling parameters are usually expressed in terms of the spin polarization @xmath7 of the lead @xmath8 , defined by eq . ( [ eq : polarization ] ) , as @xmath294 , where @xmath295 . as reported in @xcite , typical values of the dot - lead coupling strength @xmath296 in the weak coupling regime are of the order of tens of @xmath297ev . in order to investigate transport properties of the system in the whole range of parameters , one has to use a method which is more sophisticated than that based on the fermi golden rule . two such techniques are commonly used : the method based on the equation of motion for the electron green functions , and the real - time diagrammatic technique . discussion in this section is based mainly on the latter one @xcite . this technique is based on a systematic perturbation expansion of expectation value of the current operator and the density matrix order by order in the dot - lead coupling strength @xmath296 . even , if we limit further considerations based on this technique to the second order processes , this allows us to describe transport in the resonance regime , where the usual second order perturbation term diverges . one also could go beyond the second order theory and describe more subtle effects . within the real - time diagrammatic technique , the density matrix elements of the quantum dot , @xmath298 , are given by a kinetic equation in the liouville space @xmath299 where @xmath300 is the irreducible self - energy corresponding to transition forward in time from state @xmath301 to @xmath302 and then backward in time from state @xmath303 to @xmath304 . when the tunnelling processes are spin - conserving and magnetic moments of the leads are collinear , the density matrix is diagonal . after performing the perturbation expansion , one gets the following first - order ( sequential tunnelling ) and second - order ( cotunneling ) master equations : @xmath305 respectively , where the probabilities obey the normalization condition @xmath306 . systematic perturbation expansion of electric current with respect to the coupling strength @xmath296 can be performed in a similar way , and the first - order and second - order contributions to current are given by the expressions @xcite @xmath307 where the coefficients @xmath308 and @xmath309 are the first- and second - order self - energies , modified as compared to @xmath310 , to account for the number of electrons transferred through the barriers @xcite . transport characteristics of quantum dots are generally studied in both the linear and nonlinear response regimes @xcite . it is thus important to distinguish between different transport regimes of quantum dots , which are sketched in fig . [ sec4fig : regimes ] and labelled by the corresponding capital letters . a sketch illustrating different transport regimes . the respective regimes are separated by solid lines and labelled correspondingly . ] first of all , by changing position of the dot level ( by the gate voltage for instance ) or applying the bias voltage , one can cross over from one regime to another . the three regions ( a , b and a ) around zero bias correspond to the regime where sequential tunnelling is exponentially suppressed and current flows mainly due to cotunneling . the charge state of the dot is then fixed ( strictly at zero temperature ) to zero electrons in the regime a , one electron in the regime b , and two electrons in the regime a. the first - order tunnelling processes are possible once the bias voltage is increased above the threshold voltage , allowing for finite occupation of two adjacent charge states ( zero and one for regime c , and one and two for regime c ) . in the regime d all four dot states @xmath311 are possible . by performing a particle - hole transformation , the behavior in regime a and c can be mapped to that in regime a and c , respectively . to discuss the role of second - order processes we first briefly describe the case of a quantum dot coupled to nonmagnetic leads ( @xmath312 ) . linear conductance as a function of the level position . the dashed line corresponds to the first - order contribution @xmath313 , the dotted line presents the second - order conductance @xmath314 and the solid line shows the sum @xmath315 . the different transport regimes are also specified . the parameters are : @xmath316 , @xmath317 , and @xmath318 . ( after ref . @xcite ) ] figure [ sec4fig : nonmagncond ] shows the first - order ( dashed line ) and second - order ( dotted line ) contributions to the linear conductance as well as the total conductance ( solid line ) . the conductance is shown there as a function of the dot s level energy . when the dot level crosses the fermi level of electrodes , there is a resonance peak in the linear conductance . another resonance appears when @xmath319 crosses the fermi level . the resonance peaks acquire a certain width as a result of the level broadening due to coupling to the leads ( thermal fluctuations also contribute ) . it is interesting to note , that the second - order contribution becomes negative at resonances , which indicates that the second - order processes renormalize the first - order ( sequential ) contribution . except for resonances , the dot is either in the empty ( regime a ) or doubly occupied ( regime a ) state , or in the coulomb blockade regime ( regime b ) . in all these three cases the cotunneling contribution to electric current becomes dominant . it is also worth noting , that the second - order processes lead to renormalization of the dot level energy @xcite . when the external electrodes are ferromagnetic , qualitatively new features appear in the transport characteristics . these are particularly pronounced in the tunnel magnetoresistance , defined by eq . ( [ eq : tmr ] ) . below we describe some of them . ( color online ) the first - order ( a ) and first- plus second - order ( b ) tunnel magnetoresistance as a function of the bias and gate voltages for the parameters : @xmath320 , @xmath321 , and @xmath322 . ( after ref . @xcite),title="fig : " ] ( color online ) the first - order ( a ) and first- plus second - order ( b ) tunnel magnetoresistance as a function of the bias and gate voltages for the parameters : @xmath320 , @xmath321 , and @xmath322 . ( after ref . @xcite),title="fig : " ] tunnel magnetoresistance calculated in the first order of the dot - lead coupling strength @xmath296 ( sequential transport limit ) is shown in fig . [ sec4fig : tmr3d]a as a function of the bias and gate voltages . it is evident that the tmr generally acquires then two different values , depending on the transport region . for the regions a ( and a ) , b and d , the tmr value is @xmath323 while for the region c ( and c ) it is @xmath324 this behavior can be accounted for as follows . for the regions a ( a ) and b the first - order linear conductance in the parallel and antiparallel configurations is @xmath325 which leads to the magnetoresistance equal to @xmath326 . to account for the behavior of tmr in the regions c and d let us consider the zero temperature limit . in the region c there are then three dot states taking part in transport : @xmath327 ( because of the particle - hole symmetry , the results are also applicable to regime c ) . on finds then the first - order currents in the parallel and antiparallel configurations to be @xmath328 and @xmath329 , which leads to @xmath330 . similar analysis for the region d gives @xmath331 . as we already know from previous sections , first - order ( sequential ) transport does not describe properly the blockade regions ( the regions a , a and b ) . transport in these regions is expected to be strongly modified by the second - order ( cotunneling ) processes . accordingly , the above picture is also strongly modified . the corresponding tmr is displayed in fig . [ sec4fig : tmr3d]b . since the sequential tunnelling dominates transport above the threshold voltage ( outside the blockade regions , i.e. in the regions c and d ) , the total tmr is only slightly modified there as compared to that in the first - order . in the regions a , a and b , however , the first - order tunnelling processes are exponentially suppressed and the current is dominated by the second - order ( cotunneling ) processes . as follows from fig . [ sec4fig : tmr3d]b , cotunneling has then a significant influence on tmr . apart from this , tmr in the regions a and a behaves differently from that in the region b. the total linear tunnel magnetoresistance ( solid line ) as a function of the level position . the dashed line represents the first - order tunnel magnetoresistance , whereas the dotted line corresponds to the jullire s value . the parameters are @xmath332 , @xmath321 , and @xmath322 . ( after ref . @xcite ) ] let us look now in more details at the vertical cross - section of fig . [ sec4fig : tmr3d]b along the zero bias ( linear response ) . the corresponding linear tmr as a function of the level position ( gate voltage ) is shown in fig . [ sec4fig : regimeaba ] . by changing position of the dot level , one crosses over from the regime a through the regime b to the regime a. for comparison , the jullire s value of tmr , as well as the first - order contribution to tmr ( denoted as @xmath333 ) are shown there . note , that the first - order tmr is independent of the gate voltage and is equal to a half of the jullire s value . first of all , one can note that the second - order processes modify the tmr substantially . unlike the first - order term , the total tmr ( first - order plus second - order contributions ) does depend on the level position . the interesting feature of tmr shown in fig . [ sec4fig : regimeaba]b is a strong parity effect . the tmr reaches maximum when there is an even number of electrons on the dot ( zero for the regime a or two for the regime a ) , and has minimum for an odd ( one in the regime b ) number of electrons . a universal feature is that for even electron number , tmr exactly coincides with the jullire s value . the system behaves then like a single ferromagnetic tunnel junction . such a situation can take place when tunnelling processes are coherent . the only second - order processes that contribute to conductance in these regions are the non - spin - flip cotunneling processes , in which the electron spin is conserved . such processes indeed are fully coherent . the corresponding cotunneling rates are proportional to the product of the density of states of the left and right leads , thus , one can express the second - order linear conductance as @xmath334 for the parallel and antiparallel configuration , respectively . as a consequence , the tmr is then equal to that of a single planar ferromagnetic tunnel junction @xmath335 because of the particle - hole symmetry , the same result can be obtained for regime a. the situation becomes , however , more complex for an odd number of electrons on the dot . apart from the non - spin - flip processes , there are also spin - flip cotunneling processes that change spin of the dot . such second - order processes give rise to spin relaxation on the dot and lead to reduction of tmr in the regime b. the dependence of tmr on the position of the energy level in the regime b reflects the relative relation of the spin - flip to non - spin - flip cotunneling processes . the minimum value of tmr appears for @xmath336 , see fig . [ sec4fig : regimeaba]b , where one finds @xmath337 the total current ( a ) in the parallel ( solid line ) and antiparallel ( dashed line ) magnetic configurations as a function of the bias voltage in units of @xmath338 . part ( b ) shows the first - order contribution to the tmr ( dashed line ) and the total tmr ( solid line ) . the parameters are : @xmath332 , @xmath339 , @xmath321 , and @xmath322 . ( after ref . @xcite ) ] electric current flowing through the system in the nonlinear response regime is shown in fig . [ sec4fig : regimedbd]a for both magnetic configurations and for a symmetric anderson model ( @xmath336 ) . the corresponding tmr is also shown there , see fig . [ sec4fig : regimedbd]b . the contribution due to the second - order tunnelling processes is significant ( or even dominant ) in the coulomb blockade regime . the sequential tmr is shown by a dashed line in fig . [ sec4fig : regimedbd]b . one can see that sequential tmr is constant as a function of the bias voltage . however , this is not a universal behavior and occurs only for symmetric anderson model . generally , first - order tmr depends on the applied bias voltage , as illustrated in fig [ sec4fig : tmr3d]a . the second - order processes lead to a strong and nontrivial dependence of the total tmr on the bias voltage . for large values of the transport voltage ( regime d ) , the first - order tunnelling processes dominate , and therefore the total tmr is only slightly modified as compared to that in the sequential transport regime . however , for voltages below the threshold voltage ( regime b ) , the cotunneling processes lead to a strong bias dependence of tmr in the coulomb blockade regime , and to a deep minimum of tmr in the zero - bias limit . in the case of metallic islands , cotunneling processes usually lead to an enhancement of the tmr effect in the coulomb blockade regime @xcite . here , we have the opposite situation , i.e. , suppression of the effect . at low bias voltage , @xmath340 , the single - barrier spin - flip processes reduce the tmr . this is however no longer the case for nonlinear response regime , @xmath341 , where the spin accumulation diminishes the amount of spin - flip processes and the tmr increases . consequently , the tmr effect in regime b increases with rising the bias voltage within the limits @xmath342 . the minimal value is reached at @xmath343 and @xmath336 , whereas the maximal value is approached for bias voltages large as compared to thermal energy but still far away from the onset of sequential tunnelling . ( color online ) the total tunnel magnetoresistance as a function of the bias and gate voltages in the presence of external magnetic field for ( a ) @xmath344 and ( b ) @xmath345 . the parameters are : @xmath332 , @xmath321 , and @xmath322.,title="fig : " ] ( color online ) the total tunnel magnetoresistance as a function of the bias and gate voltages in the presence of external magnetic field for ( a ) @xmath344 and ( b ) @xmath345 . the parameters are : @xmath332 , @xmath321 , and @xmath322.,title="fig : " ] new features of transport characteristics appear when the spin degeneration of the dot level is lifted , @xmath346 , e.g. due to an external magnetic field applied to the system . it turns out that a finite zeeman splitting , @xmath347 , changes the transport characteristics substantially . figure [ sec4fig : tmr2dfield]a illustrates the gate and bias voltage dependence of tunnel magnetoresistance for @xmath348 . first of all , a finite zeeman splitting affects the tmr mainly in the regimes a ( a ) and b , where the second - order processes dominate . furthermore , tmr exhibits a distinctively different behavior in the regions a and a. it is enhanced in the case of empty dot , and reduced in the case of doubly occupied dot ( it may even become negative in the latter case ) . this effect depends on the orientation of applied magnetic field ; when the field is applied in the opposite direction , @xmath349 , there is an enhancement of tmr in regime a and suppression in regime a , as illustrated in fig . [ sec4fig : tmr2dfield]b . thus , by changing the sign of the zeeman splitting ( magnetic field orientation ) or applying a gate voltage , which changes the dot occupation , it is possible to reduce or enhance the tmr effect considerably . another interesting feature of tmr displayed in fig . [ sec4fig : tmr2dfield]a appears in the region b. there is a strong asymmetry of tmr with respect to the bias reversal . the tmr is decreased for positive and increased for negative bias voltage . this can also be seen in fig . [ sec4fig : tmr2dfield]b , but for positive transport voltage tmr is then larger than for negative bias voltage . the crossover between those two values of tmr takes place roughly at the zero bias . as a consequence , by changing the bias voltage in the small range one can substantially tune magnitude of the tmr effect . as we know from previous subsection , the most interesting features due to ferromagnetism of the electrodes , e.g. the zero - bias anomaly , occur in the blockade regime , where sequential tunnelling processes play a minor role . this anomaly appears only in the case when the dot is singly occupied at equilibrium ( @xmath350 , @xmath351 ) and the system is in a deep coulomb blockade regime ; latexmath:[$\gamma , k_{\rm b}t \ll sequential tunnelling is exponentially suppressed , and cotunneling gives the dominant contribution to electric current @xcite . therefore , the sequential tunnelling can be completely ignored . in this subsection we look more carefully at these features taking into account only second - order contribution . in order to calculate the cotunneling current in the deep blockade regime one can employ a simplified second - order perturbation theory and master equation for the occupation probabilities . the rate of a cotunneling processes from lead @xmath8 to lead @xmath353 , which change the dot state from @xmath281 to @xmath354 , can be written as @xmath355 with @xmath356 and @xmath357 denoting the energies of initial and final states , @xmath358 being the state of the system with an electron in the lead @xmath8 and the dot in state @xmath281 , whereas @xmath359 is a virtual state with @xmath360 denoting the corresponding energy . among different cotunneling processes one can distinguish the single - barrier ( @xmath361 ) and double - barrier ( @xmath362 ) cotunneling as well as spin - flip ( @xmath363 ) and non - spin - flip ( @xmath364 ) cotunneling . the spin - flip processes change the spin state of the dot , whereas the non - spin - flip processes do not change the dot state . the current flows through the system due to double - barrier cotunneling processes . on the other hand , the single - barrier processes do not contribute directly to electric current , however , they can change the dot occupations , and this way also influence the current . the cotunneling current flowing through the system from the left to right lead is given by @xmath365,\ ] ] where @xmath366 denotes the corresponding occupation probability . the probabilities @xmath366 can be found from the master equation , @xmath367 \,,\ ] ] together with the normalization condition @xmath368 . in order to elucidate and understand the anomalous behavior of tmr in the coulomb blockade regime , we show in fig . [ sec4fig : regimeb]a the differential conductance in the small bias regime for both parallel and antiparallel configurations and for several temperatures . figure [ sec4fig : regimeb]b displays the corresponding tmr effect . first of all , the tmr effect in the regime b for @xmath369 increases with lowering temperature and approaches the jullire s value , whereas the minimum at zero bias does not depend on temperature . the differential conductance in the parallel alignment has characteristics typical of the cotunneling regime , with a smooth parabolic dependence on the bias voltage . for antiparallel configuration , on the other hand , differential conductance has a local maximum at zero bias , followed by local minimum with increasing bias , as illustrated in fig . [ sec4fig : regimeb]a . this zero - bias anomaly stems from the interplay of the spin - flip and non - spin - flip single - barrier and double - barrier cotunneling processes @xcite . the minimum in the tmr effect is a direct consequence of this anomalous behavior of differential conductance in the antiparallel configuration . the differential conductance ( a ) for the parallel and antiparallel configurations and the tunnel magnetoresistance ( b ) as a function of the bias voltage for different values of temperature . the maximum in differential conductance for antiparallel configuration at zero bias is clearly demonstrated . the other parameters are the same as in fig . [ sec4fig : regimedbd ] . figure was generated using the scheme for the perturbation expansion in the coulomb blockade regime . ( after ref . @xcite ) ] the zero - bias anomaly in the cotunneling regime is qualitatively similar to the anomaly due to the kondo effect , which occurs in the strong coupling limit @xcite . there are , however , some distinct differences . first of all , processes responsible for the zero - bias anomaly in the cotunneling regime are of the second order in tunnelling processes , while these leading to the kondo effect are of higher order . the conductance in the cotunneling regime is much smaller than in the kondo regime , where almost perfect transmission ( @xmath370 ) through the dot is possible owing to the kondo peak in the density of states at the fermi level . furthermore , the kondo peak occurs at temperatures lower than the so - called kondo temperature , @xmath371 , and exists also in the parallel configuration @xcite . to understand the mechanism of the zero - bias anomaly it is crucial to distinguish between different types of cotunneling processes : the single - barrier cotunneling processes shown in fig . [ sec5fig : zba2]a and double - barrier cotunneling processes illustrated in fig . [ sec5fig : zba2]b . both single - barrier and double - barrier processes can be either spin - flip or non - spin - flip ones . the current flows due to double - barrier cotunneling , whereas the single - barrier cotunneling can influence the current in an indirect way , by changing the spin state of the dot . in the antiparallel configuration , there is a finite spin accumulation on the dot , as presented in fig . [ sec5fig : zba2]c . single - barrier ( a ) and double - barrier ( b ) cotunneling processes , and the occupation probabilities for spin - up and spin - down electrons in the antiparallel configuration ( c ) . the parameters are @xmath372 , @xmath373 , @xmath374 , and @xmath322 . ( after ref . @xcite ) ] the different occupation probabilities for spin - up and spin - down electrons appear due to spin asymmetry in tunnelling processes . in equilibrium , both rates are equal and there is no spin accumulation , @xmath375 . when a bias voltage is applied , and the system is in the antiparallel configuration , the relative amount of single - barrier cotunneling is diminished as compared to the double - barrier cotunneling . this is because the rate of single - barrier cotunneling is proportional to thermal energy , whereas that of double - barrier cotunneling is proportional to the bias voltage . in the nonlinear response regime , the magnetic state of the dot is mainly determined by the spin - flip processes that transfer an electron from the left to the right leads . the one shown in fig . [ sec5fig : zba2]b changes the dot spin from @xmath376 to @xmath377 . because the rate of this process is proportional to a product of densities of states for majority electrons , the corresponding rate is larger than that of the other process that changes the dot spin from @xmath377 to @xmath376 , where only the minority spins are involved . this results in a nonequilibrium spin accumulation , @xmath378 , that increases with increasing voltage , as shown in fig . [ sec5fig : zba2]c . the initial state for the dominant spin - flip cotunneling process that contributes to current is @xmath376 , as sketched in fig . [ sec5fig : zba2]b . thus , the conductance is diminished by spin accumulation . this is the mechanism by which spin accumulation gives rise to nonzero tunnel magnetoresistance effect , @xmath379 , see fig . [ sec4fig : regimeb]b . since the spin accumulation reduces electronic transport , any spin - flip process that reduces the spin accumulation should enhance the conductance . in particular , single - barrier spin - flip cotunneling is an example of such a process . as pointed above , the rate of single - barrier processes scales with @xmath380 while that of double - barrier cotunneling is proportional to @xmath381 , which explains the mechanism of the zero - bias anomaly . at low bias voltage , @xmath382 , single - barrier spin - flip processes play a significant role they decrease the spin accumulation opening this way the system for the fastest double - barrier cotunneling . as a consequence , the current increases relatively fast with applied bias , leading to a maximum in differential conductance . for @xmath383 , on the other hand , the relative role of single - barrier processes is negligible as compared to double - barrier cotunneling , and the conductance is reduced . thus , the interplay between the rates of double - barrier and single - barrier cotunneling processes leads to the maximum in the differential conductance at the zero bias . intrinsic spin relaxation in the dot can result , for instance , from spin - orbit interaction , coupling of the electron spin to nuclear spins , etc . however , we will not consider a particular microscopic mechanism of the intrinsic spin - flip processes , but simply assume that the spin - flip relaxation is described by a spin - relaxation time @xmath263 , and is taken into account _ via _ a relaxation term in the appropriate master equation for the occupation probabilities @xcite , @xmath384 where @xmath385 , @xmath386 denotes the probability that the dot is occupied by a spin-@xmath53 electron , and @xmath387 is the cotunneling rate from lead @xmath388 to lead @xmath389 with a change of the dot spin from @xmath53 to @xmath390 ( @xmath391 ) . the last term describes the spin relaxation processes , which in the case of spin - degenerate dot level reduces to @xmath392 . as before , ( see section 2 ) one can distinguish between the fast and slow spin relaxation limits . the former ( latter ) limit corresponds to the situation when the time between successive cotunneling events , @xmath393 , is significantly longer ( shorter ) than the intrinsic spin relaxation time @xmath263 . a typical spin relaxation time for quantum dots can be relatively long , up to @xmath297s @xcite . on the other hand , the time between successive cotunneling events can be estimated taking into account the fact that the rate of spin - flip cotunneling is generally larger than that of non - spin - flip cotunneling ( for a finite parameter @xmath271 ) . assuming @xmath394 , one then finds @xmath395 with @xmath396 and @xmath397 . assuming typical parameters @xcite , one can roughly estimate @xmath393 to range from @xmath398 ns to @xmath399 ns . the differential conductance in the antiparallel configurations ( a ) and tunnel magnetoresistance ( b ) as a function of the bias voltage for different spin relaxation @xmath400 . the parameters are : @xmath401 , @xmath402 , @xmath403 , and @xmath404 . ( after ref . @xcite ) ] the anomalous behavior of differential conductance in the antiparallel configuration results from spin asymmetry of the tunnelling processes , which gives rise to spin accumulation , @xmath405 . the intrinsic spin - flip processes in the dot reduce the spin accumulation and this way suppress the anomaly . this behavior is displayed in fig . [ sec5fig1](a ) , where different curves correspond to different values of the parameter @xmath8 defined as @xmath400 . thus , @xmath406 describes the case with no intrinsic spin relaxation , whereas the curves corresponding to nonzero @xmath8 describe the influence of intrinsic relaxation processes . first of all , one can note that small amount of intrinsic spin - flip processes enhances the zero - bias anomaly [ see the curve for @xmath407 in fig . [ sec5fig1](a ) ] . this is because such processes play then a role similar to that of single - barrier spin - flip cotunneling . in the case of fast spin relaxation , on the other hand , the spin accumulation is suppressed and the anomaly disappears , as can be seen in fig . [ sec5fig1](a ) for @xmath408 . it is also worth noting that spin - flip processes in the dot enhance the overall conductance in the antiparallel configuration . in the parallel configuration , however , the differential conductance does not depend on intrinsic relaxation . modifying the conductance in the antiparallel configuration , spin relaxation processes suppress the tunnel magnetoresistance , as shown in fig . [ sec5fig1](b ) . since the zero - bias maximum in conductance is suppressed in the fast spin relaxation limit , the corresponding dip in tmr at small voltages is also suppressed by the intrinsic relaxation processes . more specifically , the dip in tmr broadens with increasing @xmath8 and disappears in the limit of fast relaxation ( see the curve for @xmath408 ) . an interesting feature of tmr in the presence of spin - flip scattering in the dot is the crossover from positive to negative values when @xmath8 increases , as illustrated in fig . [ sec5fig1](b ) . thus , the difference between conductances in the parallel and antiparallel magnetic configurations persists even for fast spin relaxation in the dot , contrary to the sequential tunnelling regime , where such a difference disappears @xcite . this seemingly counterintuitive behavior can be understood by taking into account the following two facts : ( i ) absence of spin accumulation in the dot for fast spin relaxation ( @xmath409 ) , and ( ii ) difference in the fastest cotunneling processes contributing to the current in the two magnetic configurations . the fastest double - barrier cotunneling processes involve only the majority - spin electrons of the two leads thus , in the parallel configuration the fastest cotunneling processes are the non - spin - flip ones . they take place either _ via _ the empty - dot virtual state ( for one orientation of the dot spin ) or _ via _ the doubly occupied dot virtual state ( for the second orientation of the dot spin ) . the dominant contribution to the current is then proportional to @xmath410 . on the other hand , in the antiparallel magnetic configuration the fastest cotunneling processes are the spin - flip ones , which can occur only for one particular orientation of the dot spin . however , for this spin orientation cotunneling can take place _ via _ both empty and doubly occupied dot virtual states . the corresponding dominant contribution to electric current is then proportional to @xmath411 ^ 2=1/\varepsilon^2 + 1/(\varepsilon + u)^2 -2/[\varepsilon ( \varepsilon + u)]$ ] . it is thus clear that the difference in currents flowing through the system in the antiparallel and parallel configurations is equal to @xmath412 $ ] , which results from the interference term . since @xmath413 and @xmath414 , this interference contribution is positive . as a result , the current in the antiparallel configuration is larger than the current in the parallel configuration . the minimum in tmr at zero bias in the case of a symmetric anderson model can be expressed as @xmath415 whereas for @xmath416 and @xmath417 one finds @xmath418 } { 2(1-p^2)(3-p^2)k_{\rm b}t\gamma^2+(3+p^2)\varepsilon^2 h/\tau_{\rm sf}}.\ ] ] the latter formula approximates the value of tmr corresponding to the bias voltage at which the differential conductance has a local minimum . in the slow spin relaxation limit one finds @xmath419 , and @xmath420 . however , in the limit of fast spin relaxation tmr becomes negative and is given by @xmath421 the discussion up to now was limited to the case of degenerate dot level . the situation changes when @xmath422 , e.g. , due to an external magnetic field . the level splitting is described by the parameter @xmath423 , where the magnetic field is assumed to be along the magnetic moment of the left electrode . in fig . [ sec5fig4 ] we show the bias voltage dependence of the differential conductance in the parallel and antiparallel configurations for different values of parameter @xmath8 . the differential conductance in the nonlinear response regime for different spin relaxation in the parallel ( a ) and antiparallel ( b ) configurations . the parameters are : @xmath401 , @xmath424 , @xmath425 , @xmath403 and @xmath322 . ( after ref . @xcite ) ] in the limit of no intrinsic spin relaxation in the dot ( solid line in fig . [ sec5fig4 ] ) and at low bias voltage , the dot is occupied by a spin - up electron and the current flows mainly due to non - spin - flip cotunneling . the spin - flip cotunneling processes are suppressed for @xmath426 , which results in the steps in differential conductance at @xmath427 . the suppression of spin - flip inelastic cotunneling was recently used as a tool to determine the spectroscopic @xmath428-factor @xcite . when @xmath429 becomes larger than @xmath430 , spin - flip cotunneling is allowed , consequently the conductance increases . however , there is a large asymmetry of differential conductance in the antiparallel configuration with respect to the bias reversal . to understand this asymmetry , it is crucial to realize that when the splitting @xmath431 is larger than @xmath432 , the single - barrier spin - flip cotunneling processes can occur only when the dot is occupied by a spin - down electron thus , the single - barrier processes can assist the fastest double - barrier cotunneling processes , but only for positive bias . this is because the fastest processes can occur when the dot is occupied by a spin - down electron for negative bias and by a spin - up electron for positive bias , leading to larger conductance for positive than for negative bias voltage . no such asymmetry occurs in the parallel configuration [ see fig . [ sec5fig4](a ) ] , as now the system is fully symmetric with respect to bias reversal . the situation changes when intrinsic spin - flip relaxation processes occur in the dot . for the parameters assumed in fig . [ sec5fig4 ] , the spin relaxation in the dot affects the conductance only for @xmath433 , while for @xmath434 the conductance is basically independent of @xmath8 ( see fig . [ sec5fig4 ] ) . this is because for @xmath434 and @xmath435 , the dot is predominantly occupied by a spin - up electron and the transitions to the spin - down state due to relaxation processes are energetically forbidden . as a consequence , the current flows mainly due to non - spin - flip cotunneling , irrespective of spin relaxation time . this scenario holds for both magnetic configurations of the system . when @xmath433 , the spin - flip cotunneling processes can take place and the dot can be either in the spin - up or spin - down state . in the parallel configuration , fig . [ sec5fig4](a ) , the conductance is slightly reduced by the spin - flip relaxation processes . this can be understood by realizing the fact that the fastest non - spin - flip cotunneling processes in the parallel configuration are more probable when the dot is occupied by a spin - down electron than by a spin - up one [ due to smaller energy denominator , see eq . ( [ eq : cotunnelingrate ] ) ] . since the spin - down state ( as that of larger energy ) relaxes relatively fast to the spin - up state ( which has definitely smaller energy ) , this leads to a reduction in the conductance . on the other hand , the differential conductance in the antiparallel configuration is enhanced by the relaxation processes for positive bias and diminished for negative bias voltages . consider first the situation for positive bias . as already discussed above for @xmath406 , an important role in that transport regime is played by the single - barrier spin - flip cotunneling processes , which open the system for the fast double - barrier cotunneling by reversing spin of the dot from the spin - down to the spin - up state . the relaxation processes play a role similar to that of the single - barrier cotunneling , and lead to a certain increase in the conductance . for negative bias voltage , in turn , the fast double - barrier cotunneling processes occur when the dot is occupied by a spin - down electron . the probability of such events is decreased by spin relaxation , leading to a reduced conductance . an interesting consequence of the enhancement ( reduction ) of the differential conductance for positive ( negative ) bias voltage is an increase of the asymmetry with respect to the bias reversal see the curve for @xmath408 in fig . [ sec5fig4](b ) . an interesting situation occurs when the quantum dot is coupled asymmetrically to the left and right leads @xmath436 . the differential conductance for a system with one electrode nonmagnetic and the other one made of a ferromagnet with large spin polarization ( in the following referred to as strong ferromagnet ) is shown in fig . [ sec5fig5 ] for the case when the dot is described by an asymmetric anderson model ( @xmath437 ) . the differential conductance in the nonlinear response regime for the asymmetric anderson model , @xmath438 , for different spin relaxation . the parameters are : @xmath401 , @xmath439 , @xmath440 and @xmath441 . ( after ref . @xcite ) ] consider first the situation in the absence of intrinsic spin relaxation in the dot ( solid curves in fig . [ sec5fig5 ] ) . when @xmath442 , the influence of single - barrier cotunneling can be neglected @xcite . the cotunneling processes which transfer charge from one lead to another take place _ via _ two possible virtual states empty dot ( an electron residing in the dot tunnels to one of the leads and another electron from the second lead enters the dot ) and doubly occupied dot ( an electron of spin opposite to that in the dot enters the dot and then one of the two electrons leaves the dot ) . consider first positive bias ( @xmath443 , electrons flow from right to left , i.e. , from normal metal to strong ferromagnet ) , and assume for clarity of discussion that the strong ferromagnet is a half - metallic one with full spin polarization ( only spin - up electrons can then tunnel to the left lead ) . when a spin - down electron enters the dot , it has no possibility to leave the dot for a long time . the allowed cotunneling processes occur then _ via _ doubly occupied dot virtual states . in the absence of intrinsic spin relaxation in the dot , the only processes which can reverse the dot spin are the single - barrier cotunneling ones , which however play a minor role when @xmath444 . thus , the current flows due to non - spin - flip cotunneling _ via _ doubly - occupied dot virtual states , whereas cotunneling through empty - dot virtual states is suppressed . the situation is changed for negative bias ( electrons flow from strong ferromagnet to normal metal ) . now , the dot is mostly occupied by a spin - up electron , which suppresses cotunneling _ via _ doubly - occupied dot virtual state and the only contribution comes from cotunneling _ via _ empty - dot virtual state . the ratio of cotunneling rates through the empty dot and doubly occupied dot virtual states is approximately equal to @xmath445^{-2}$ ] . in the situation presented in fig . [ sec5fig5 ] one finds @xmath446 . accordingly , the conductance for negative bias is much larger than for positive bias voltage . when @xmath447 becomes of the order of @xmath432 or smaller , the rate of single - barrier cotunneling is of the order of the rate of double - barrier cotunneling . therefore , the single - barrier processes can play an important role in transport . more precisely , single - barrier cotunneling processes can reverse spin of an electron in the dot and thus can open the system for the fast cotunneling processes . intrinsic spin - flip processes in the dot have similar influence on electronic transport as in the case discussed in the previous section . as before , relaxation processes remove the asymmetry with respect to the bias reversal and suppress the zero - bias anomaly . thus , the diode - like behavior can appear only in the limit of slow spin relaxation , and is suppressed in the limit of fast spin relaxation , as shown in fig . [ sec5fig5 ] by the curves corresponding to @xmath408 . new features of transport characteristics occur when the leads magnetizations form an arbitrary noncollinear magnetic configuration @xcite . such a configuration can be controlled by a weak external magnetic field , weak enough to neglect the corresponding zeeman splitting of the dot level and weaker than the effective exchange field exerted on the dot by ferromagnetic leads . in particular , the differential conductance is significantly modified by the exchange field in noncollinear configurations . this also applies to the tmr effect , which for arbitrarily aligned leads magnetizations can be defined as , @xmath448/i(\varphi)$ ] , where @xmath449 is an angle between the leads magnetic moments . the system , whose spin moments @xmath450 and @xmath451 of the left and right lead , respectively , form an arbitrary configuration is shown in fig . [ sec6fig:1 ] . ( color online ) schematic of a quantum dot coupled to ferromagnetic leads with non - collinearly aligned magnetizations . the net spin moments of the left @xmath450 and right @xmath451 lead form an angle @xmath449 . there is a symmetric bias voltage applied to the system . ] the corresponding hamiltonian , written for the quantization axis of the dot equivalent to that of the left lead , is given by eq . ( [ sec4eq : qdhamiltonian ] ) , where the tunnel hamiltonian can be decomposed into two terms @xmath452 . the term @xmath453 describes tunnelling processes between the left electrode and the dot and takes the same form as for collinear configurations , while @xmath454 describes tunnelling processes between the dot and right lead , and acquires the form @xmath455 \,.\end{aligned}\ ] ] the real - time diagrammatic method described at the beginning of this section for collinear magnetic configurations can be extended to include the effects due to noncollinearity of the leads magnetic moments . the density matrix of the quantum dot for an arbitrary magnetic configuration is given by @xmath456 the diagonal elements of the density matrix correspond to the respective occupation probabilities , while the off - diagonal elements @xmath457 and @xmath458 describe the dot spin @xmath459 , with @xmath460 , @xmath461 , and @xmath462 . the density matrix elements can be determined from the corresponding kinetic equation , which in the steady state and for spin - degenerate dot level can be written as @xcite @xmath463 by expanding the self - energies and density matrix elements , one can calculate the sequential and cotunneling current . in the following we focus on transport in the coulomb blockade regime , @xmath464 , and when the dot is singly occupied , @xmath465 . > from the discussion above we know that there is an anomalous behavior of the differential conductance in the small bias regime for antiparallel magnetic configuration . now , we will consider how this anomaly changes when magnetic configuration varies continuously from antiparallel to parallel alignment . we consider first the symmetric anderson model , @xmath336 . the exchange field vanishes then for an arbitrary magnetic configuration , and both differential conductance and tmr change monotonically when going from antiparallel to parallel magnetic configurations . however , for @xmath466 , the effects of exchange field become important and lead to nontrivial behavior of transport characteristics , as will be discussed later . ( color online ) differential conductance @xmath467 ( a ) and tunnel magnetoresistance tmr ( b ) as a function of the bias voltage @xmath60 for indicated values of the angle @xmath449 between magnetic moments , calculated for symmetric anderson model . the parameters are : @xmath468 , @xmath469 , @xmath403 , and @xmath470 . ( after ref . @xcite ) , title="fig : " ] ( color online ) differential conductance @xmath467 ( a ) and tunnel magnetoresistance tmr ( b ) as a function of the bias voltage @xmath60 for indicated values of the angle @xmath449 between magnetic moments , calculated for symmetric anderson model . the parameters are : @xmath468 , @xmath469 , @xmath403 , and @xmath470 . ( after ref . @xcite ) , title="fig : " ] the differential conductance @xmath471 as a function of the bias voltage is shown in fig . [ sec6fig:2](a ) for several values of the angle @xmath449 . the conductance varies monotonically with the angle between leads magnetic moments . when the angle increases from zero to @xmath265 , the anomaly emerges at small values of @xmath449 and its relative height increases with increasing angle , reaching a maximum value at @xmath472 . the exchange field for a symmetric anderson model is negligible , and the average dot spin tends to zero in the linear response regime . the angular dependence of the linear conductance can be then expressed as @xmath473 \,.\ ] ] the variation of @xmath471 with @xmath449 at low bias is thus characterized by the factor @xmath474 , which leads to maximum ( minimum ) conductance in the parallel ( antiparallel ) magnetic configuration . such behavior is typical of a normal spin - valve effect . the bias dependence of the associated tmr is shown in fig . [ sec6fig:2](b ) for several values of the angle @xmath449 . the zero - bias anomaly in the differential conductance [ see fig . [ sec6fig:2](a ) ] leads to the corresponding anomaly ( dip ) in tmr at small bias voltages . the dip in tmr decreases when magnetic configuration departs from the antiparallel alignment , and eventually disappears in the parallel configuration . the variation of tmr with the angle is monotonic , similarly as the angular variation of the differential conductance . the dependence of tmr on the angle @xmath449 at zero bias is given approximately by the formula @xmath475 now , the angular dependence of tmr is governed by @xmath476 , which gives maximum tmr in the antiparallel configuration and zero tmr in the parallel one . ( color online ) the differential conductance @xmath467 ( a ) and tunnel magnetoresistance tmr ( b ) as a function of the bias voltage @xmath60 for indicated values of the angle @xmath449 for asymmetric anderson model @xmath477 and @xmath403 . the other parameters are the same as in fig . [ sec6fig:2 ] . ( after ref . @xcite ) , title="fig : " ] ( color online ) the differential conductance @xmath467 ( a ) and tunnel magnetoresistance tmr ( b ) as a function of the bias voltage @xmath60 for indicated values of the angle @xmath449 for asymmetric anderson model @xmath477 and @xmath403 . the other parameters are the same as in fig . [ sec6fig:2 ] . ( after ref . @xcite ) , title="fig : " ] the transport characteristics , described above for the symmetric anderson model , are significantly modified when the model becomes asymmetric , i.e. when @xmath466 . this can be realized by shifting the dot level position by a gate voltage applied to the dot . different contributions to the exchange field do not cancel then , and the resulting effective exchange field becomes nonzero . and this exchange field has a significant influence on transport properties . the strength of effective exchange field grows with deviation of the anderson model from the symmetric one , described quantitatively by @xmath478 ( with @xmath479 for the symmetric model ) . the leading contribution to the exchange field comes from the first - order diagrams of the perturbation expansion in terms of the real - time diagrammatic technique , whereas the current flows due to the second - order tunnelling processes . this leads to two different time scales which determine transport characteristics . transport properties are thus a result of the interplay of the first- and second - order processes . although the first - order processes do not contribute directly to current , they influence transport _ via _ modification of the dot spin . the differential conductance is shown in fig . [ sec6fig:3](a ) as a function of bias voltage for an asymmetric anderson model , @xmath480 , and for different values of @xmath449 . the low - bias differential conductance becomes enhanced in a certain range of the angle @xmath449 , leading to a nonmonotonic dependence of the conductance @xmath471 on the angle between the leads magnetizations . this nonmonotonic behavior is also visible in tmr , see fig . [ sec6fig:3](b ) . there is also a range of the angle @xmath449 , where tmr changes sign and becomes negative in the small bias region , i.e. the corresponding conductance is larger than that in the parallel configuration . the most characteristic features of transport characteristics in the presence of exchange field are the enhanced differential conductance at a non - collinear alignment and its rapid drop when the system approaches the antiparallel configuration . the key role in this behavior is played by the first - order processes giving rise to the exchange field . these processes lead to the precession of spin in the dot , which facilitates tunnelling processes and leads to an increase in the conductance as compared to that in the parallel configuration . when the configuration becomes close to the antiparallel one , the first - order processes become suppressed and the conductance drops to that for antiparallel alignment . the nonmonotonic behavior of the differential conductance with @xmath449 leads to a nonmonotonic dependence of tmr . the effects due to exchange field give rise to a local minimum in tmr at a non - collinear magnetic configuration . moreover , in this transport regime tmr changes sign and becomes negative . when the magnetic configuration is close to the antiparallel one , tmr starts to increase rapidly reaching maximum for @xmath481 . the negative tmr and its sudden increase when the configuration tends to the antiparallel one are a consequence of the processes leading to nonmonotonic behavior of the differential conductance , as described above . to understand more intuitively the above presented behavior of the differential conductance and tmr at low bias voltage and close to the antiparallel configuration , one should consider two different time scales . one time scale , @xmath482 , is established by the virtual first - order processes responsible for the spin precession due to exchange field , @xmath483 the second time scale , @xmath393 , is associated with second - order processes which drive the current through the system . at low temperature and low bias voltage , the cotunneling rate can be expressed as @xmath484 the rate @xmath393 depends linearly on the applied voltage , whereas @xmath482 is rather independent of @xmath60 . as a consequence , at low bias and for non - collinear configuration , the exchange field plays an important role leading to a nonmonotonic dependence of differential conductance on the angle between the leads magnetizations . when magnetic configuration is close to the antiparallel one , the spin precession rate is deceased ( @xmath485 ) and , at certain angle , the rate of spin precession becomes comparable to the cotunneling rate . this gives rise to a sudden drop ( increase ) in differential conductance ( tmr ) . we note that the nonmonotonic dependence of differential conductance and magnetoresistance has also been observed in quantum dots in the strong coupling limit @xcite . up to now we discussed theoretical aspects of electronic transport through the simplest quantum dots , i.e. the dots with only one orbital level . real dots are usually more complex and their electronic spectrum includes many orbital levels taking part in electronic transport . this may significantly change transport characteristics and also may lead to qualitatively new features @xcite . in this section we will consider some of the new effects in transport through multi - dot systems . the schematic of a two - level quantum dot coupled to ferromagnetic leads with collinear magnetizations is shown in fig . [ sec7fig:1 ] . ( color online ) energy diagram of a two - level quantum dot coupled to ferromagnetic leads . for clarity reasons the energy diagram is shown here for @xmath486 . the leads magnetizations can form either parallel or antiparallel configurations . the arrows indicate the net spin of the leads . ] the quantum dot is described by the following hamiltonian @xmath487 where @xmath488 is the particle number operator , @xmath489 , @xmath490 ( @xmath491 ) is the creation ( annihilation ) operator of an electron with spin @xmath53 on the @xmath229th level ( @xmath492 ) , and @xmath493 is the corresponding single - particle energy . the on - level coulomb repulsion between two electrons of opposite spins is described by @xmath271 , whereas the inter - level repulsion energy is denoted by @xmath494 . the forth term in eq . ( [ sec7eq : hamiltonian ] ) describes the zeeman energy , with @xmath495 being the zeeman splitting of the energy levels ( @xmath496 is an external magnetic field along the magnetic moment of the left electrode ) . to present the main features of transport characteristics , it is convenient to introduce the level spacing @xmath497 and also define @xmath498 . in the following we analyze the current @xmath116 , differential conductance @xmath471 , and the fano factor @xmath499 in the parallel and antiparallel magnetic configurations , as well as the corresponding tmr for two - level quantum dots . the fano factor , @xmath500 , describes the deviation of the zero - frequency shot noise @xmath501 from the poissonian shot noise @xmath502 . the presented results have been obtained within the real - time diagrammatic technique @xcite . ( color online ) the differential conductance @xmath467 as a function of the bias voltage and level position in the parallel magnetic configuration ( a ) and tmr ( b ) for the parameters : @xmath503 , @xmath504 , @xmath505 , @xmath506 , @xmath507 , and @xmath508 ( @xmath509 , @xmath492 ) . ( after ref . @xcite ) ] typical variation of the differential conductance with the bias voltage @xmath60 and level position @xmath276 ( gate voltage ) is shown in fig . [ sec7fig:2](a ) for the parallel magnetic configuration . conductance in the antiparallel configuration is qualitatively similar to that in the parallel configuration , but generally smaller , which leads to nonzero tmr effect . the diamonds in fig . [ sec7fig:2](a ) around @xmath343 correspond to the coulomb blockade regions . when lowering position of the dot levels , the charge of the dot changes successively . the dot is empty for @xmath510 , occupied by one electron for @xmath511 , doubly occupied for @xmath512 , occupied by three electrons for @xmath513 , and the two orbital levels of the dot are fully occupied for @xmath514 . in all these transport regions the dot is in a well - defined charge state , and the sequential tunnelling is exponentially suppressed . if the bias voltage is increased above a certain threshold voltage , the current flows due to first - order ( sequential ) tunnelling processes . when the thermal energy is low enough , one observes then a well - resolved step in the current as a function of the bias voltage . in the density plots shown in fig . [ sec7fig:2]a , this can be seen in the form of lines that clearly separate the coulomb blockade regions from transport regions associated with consecutive charge states taking part in transport . when the bias voltage increases further , additional steps [ and consequently lines in fig . [ sec7fig:2](a ) ] arise at voltages where new states becomes active in transport . the corresponding tmr as a function of the bias and gate voltages is shown in fig . [ sec7fig:2](b ) . it is worth noting that tmr takes now several well - defined values . such behavior of tmr is significantly different from that for a single - level quantum dot , where tmr in the sequential tunnelling regime acquires only two values @xcite . as in single - level dots , tmr in the linear response regime is independent of the gate voltage and is given by @xmath515 , i.e. half of the tmr in the julliere model @xcite . figure [ sec7fig:2 ] also shows that , when increasing the bias voltage @xmath60 and keeping constant position of the dot levels , the current and tmr acquire some specific and well - defined values in different transport regions . as shown in ref . @xcite , the current and tmr at these plateaus can be approximated by simple analytical formulas . ( color online ) current ( a ) in the units of @xmath516 and fano factor ( b ) in the parallel ( solid line ) and antiparallel ( dashed line ) configurations as well as tunnel magnetoresistance ( a ) as a function of the bias voltage for @xmath517 . the other parameters are the same as in fig . [ sec7fig:2 ] . ( after ref . @xcite),height=309 ] this behavior is presented in more details in fig . [ sec7fig:4](a ) , where the bias voltage dependence of the current and tmr is shown explicitly for the case when the dot level is above ( @xmath518 ) the fermi level of the leads at equilibrium . the current in both magnetic configurations and the associated tmr exhibit characteristic plateaus which correspond to different transport regions . the corresponding fano factors @xmath519 and @xmath520 in the parallel and antiparallel magnetic configurations are shown in fig . [ sec7fig:4]b . similarly as current and tmr , the fano factor acquires roughly constant values , different in different transport regions . due to the spin asymmetry in the coupling of the dot to external leads , the bias dependence of the fano factor is significantly different from that in the corresponding nonmagnetic situations @xcite . for both magnetic configurations of the system , the fano factor depends on the polarization factor @xmath196 ( differently in the two configurations , in general ) . if @xmath521 , the fano factor becomes divergent due to the thermal noise , which dominates in this transport regime ; in the case of @xmath343 , the noise is given by @xmath522 , with @xmath523 being the linear conductance , leading to a divergency of the fano factor @xcite . apart from this , in some transport regions we find @xmath524 , while in the other ones @xmath525 . the ratio @xmath526 depends generally on the spin polarization of the leads @xmath196 @xcite . furthermore , we note that if the leads are half - metallic , the fano factor in the parallel configuration diverges as @xmath527 . this increase of the fano factor is due to the enhanced spin asymmetry in transport processes through the dot @xcite . on the other hand , in the antiparallel configuration the fano factor tends to unity for @xmath527 , except for the coulomb blockade regime with two electrons trapped in the dot , where the shot noise is super - poissonian . when one of the leads is half - metallic ( @xmath528 ) and the other one is nonmagnetic ( @xmath318 ) , transport characteristics become asymmetric with respect to the bias reversal . furthermore , the current can be suppressed in certain bias regions , and this suppression is accompanied by the occurrence of ndc . this basically happens when the electrons residing in the dot have spin opposite to that of electrons in the half - metallic drain electrode . in fig . [ sec7fig:5 ] we show the current and fano factor for a quantum dot coupled to half - metallic ( left ) and nonmagnetic ( right ) lead as a function of the bias voltage . for the situation shown in fig . [ sec7fig:5 ] , current is suppressed in certain regions of positive bias voltage . as follows from fig . [ sec7fig:5](a ) , there are three such blockade regions , labelled with the consecutive numbers . on the other hand , for negative bias voltage , the current changes monotonically with the transport voltage , as shown in the inset of fig . [ sec7fig:5](a ) . the current ( a ) in units of @xmath516 and fano factor ( b ) as a function of the bias voltage for @xmath529 , @xmath441 , while the other parameters are the same as in fig . [ sec7fig:2 ] . the inset in part ( a ) shows the current in the whole range of the bias voltage . ( after ref . @xcite ) ] the mechanism of the blockade can be described as follows . in the blockade region ( 1 ) , @xmath530 , the dot is in the state @xmath531 and the spin - down electron residing in the dot has no possibility to tunnel further to the left lead , which leads to suppression of the current . here , the first ( second ) _ ket _ corresponds to the first ( second ) orbital level of the dot . the blockade region ( 2 ) , @xmath532 , is associated with the occupation of the states @xmath531 and @xmath533 . the current is then prohibited due to the full occupation of the single - particle spin - down states . when increasing the bias voltage further , @xmath534 , the blockade of the current becomes suppressed [ see the plateau between regions ( 2 ) and ( 3 ) in fig . [ sec7fig:5](a ) ] , which is due to a finite occupation of state @xmath535 . although tunnelling of spin - down electrons is then blocked , the current is still carried by spin - up electrons . in turn , the blockade region ( 3 ) occurs for @xmath536 , where the dot is in the triplet state @xmath537 and tunnelling is also suppressed . thus , the current is blocked when the total dot spin @xmath538 is either @xmath539 or @xmath540 , i.e. , the spin of electrons on the dot is opposite to that of electrons in the half - metallic lead . there is no suppression of current for negative voltage . this is because the electrons residing in the dot can always tunnel to the nonmagnetic drain electrode . it is also interesting to note that a pure triplet state is formed in the region ( 3 ) @xcite . the current blockade in the regions ( 1 ) to ( 3 ) in fig . [ sec7fig:5](a ) for positive bias is not due to the charging effects as in the coulomb blockade regime , but due to a particular occupation of the dot spin state . such blockade is frequently referred to as the pauli spin blockade , and has already been found in single - dot and double - dot systems @xcite . it is worth noting that the current blockade in fig . [ sec7fig:5](a ) is not complete there is a small leakage current in each blockade region , which results from the fact that the assumed spin polarization of the half - metallic lead is not exactly equal to unity . when spin polarization is increased , the current in the blockade regions decreases ( compare the curves for @xmath541 and @xmath542 ) . the spin blockade of charge current leads to the super - poissonian shot noise , i.e. , the corresponding fano factor is larger than unity . on the other hand , the fano factor outside the spin blockade regions is sub - poissonian ( smaller than unity ) , as shown in fig . [ sec7fig:5](b ) . the enhancement of the shot noise in the pauli blockade regions is a consequence of large spin asymmetry in the tunnelling processes . the occurrence of a spin - down electron on the dot prevents further tunnelling processes for a longer time , while spin - up electrons on the dot can escape much faster , allowing further tunnelling processes . this bunching of fast tunnelling processes gives rise to large current fluctuations , and consequently also to fano factors much larger than unity . recently an interacting three - terminal quantum dot with ferromagnetic leads was considered in ref . the dot operated as a beam splitter - one contact was a source and the other two acted as drains . the authors found a dynamical spin blockade ( spin - dependent bunching of tunnelling events ) and positive zero - frequency cross - correlations of the current in the drain electrodes . in all transport characteristics discussed above only sequential tunnelling processes were taken into account , while higher order processes , in particular the cotunneling ones , have been neglected . from our discussion on single - level dots we already know that higher order contributions are particularly pronounced in the blockade regions , where they have a significant influence on tmr and fano factor . the role of cotunneling processes in two - level dots was studied in a recent paper @xcite . it has been shown there that the cotunneling processes lead to a significant enhancement ( or reduction ) of tmr in the blockade regime . outside the blockade regions , however , tmr is determined mainly by sequential transport , so the contribution from cotunneling processes is rather minor . similarly , the cotunneling processes significantly modify the fano factor in the blockade regions , while outside these regions the fano factor is only weakly sensitive to the cotunneling . more specifically , the fano factor in the blockade regions is reduced in comparison to that obtained in the sequential transport limit . however , it is still larger than one , indicating super - poissonian shot noise in the blockade regime due to bunching of the fastest cotunneling processes . there is also another interesting phenomenon in the blockade regime , which resembles the zero bias anomaly observed in single - level dots in the antiparallel magnetic configuration . this takes place when each orbital level of the dot at equilibrium is occupied by a single electron , and the system is in the coulomb blockade regime @xcite . the hamiltonian @xmath543 of the system is given by eq . ( [ sec4eq : qdhamiltonian ] ) , where @xmath544 additionally includes the exchange term and can be expressed by the formula @xcite @xmath545 the last term in @xmath546 corresponds to the exchange energy according to the hund s rule , with @xmath547 being the respective exchange coupling and @xmath548 denoting a vector of pauli spin matrices . six different two - particle states of the dot are possible , these are : three singlets @xmath549 , @xmath550 , @xmath551 , and three triplets @xmath552 , @xmath553 and @xmath554 . in the case of finite level spacing , @xmath555 , and @xmath556 , the lowest singlet state is @xmath557 . transport characteristics strongly depend on the ground state of the system . it is therefore useful to introduce the difference between the energy of the lowest lying singlet @xmath558 and triplet @xmath559 states , @xmath560 . for @xmath561 , the ground state of the dot is a singlet , @xmath557 , whereas for @xmath562 , the ground state is a triplet , which is three - fold degenerate , @xmath563 , @xmath564 , @xmath565 . on the other hand , for @xmath566 , the dot is in a mixed state and the occupation of singlet and each triplet is equal at equilibrium and given by @xmath567 . ( color online ) the differential conductance in the parallel ( solid line ) and antiparallel ( dashed line ) magnetic configuration of the system as a function of the bias voltage for several values of @xmath568 as indicated : ( a ) @xmath569 , ( b ) @xmath570 , ( c ) @xmath571 , ( d ) @xmath572 . the other parameters are : @xmath573 , @xmath574 , @xmath575 , @xmath576 , and @xmath322 . ( after ref . @xcite ) ] figure [ sec7fig:8 ] presents the bias voltage dependence of differential conductance @xmath471 in the parallel and antiparallel magnetic configuration and for several values of @xmath577 . let us consider the situation in fig [ sec7fig:8](c ) , which corresponds to degenerate singlet and triplet states , @xmath571 . the differential conductance in the parallel configuration exhibits a smooth parabolic dependence on the bias voltage , whereas in the antiparallel configuration there is a maximum at the zero bias voltage . this effect bears a resemblance to that found in the case of singly occupied one - level quantum dots discussed in previous sections @xcite . here , however , the mechanism leading to the maximum is different the zero - bias peak appears due to cotunneling through the singlet and triplet states of the dot . in the case of @xmath571 and at low bias voltages , all the four dot states , i.e. @xmath557 , @xmath563 , @xmath565 , @xmath564 , participate in transport on an equal footing . consequently , the current flows due to both spin - flip and non - spin - flip cotunneling processes . to understand the mechanism leading to the zero - bias peak , one should bear in mind that in the antiparallel configuration the spin - majority electrons of one lead tunnel to the spin - minority electron band of the other lead . for example , for positive bias voltage ( electrons tunnel then from the right to left lead ) , the spin-@xmath578 electrons can easily tunnel to the left lead ( the spin-@xmath578 electrons are the majority ones ) , while this is more difficult for the spin-@xmath579 electrons ( they tunnel to the minority electron band ) . thus , the occupation of state @xmath565 ( @xmath537 ) is increased with increasing bias voltage , while the occupation of state @xmath564 ( @xmath580 ) decreases . the unequal occupations of these triplet states lead to a nonequilibrium spin accumulation in the dot , @xmath581 . it is further interesting to note that in the antiparallel configuration the possible non - spin - flip cotunneling processes are proportional to @xmath582 and @xmath583 , whereas the spin - flip cotunneling is proportional to @xmath584 and @xmath585 . it is clear that the fastest cotunneling processes are the ones involving only the majority spins , i.e. @xmath584 . however , because of nonequilibrium spin accumulation , with increasing the bias ( @xmath586 ) , the dot becomes dominantly occupied by majority electrons of the right lead , @xmath587 , and the processes proportional to @xmath584 are suppressed . as a consequence , the differential conductance drops with the bias voltage , leading to the zero - bias peak . this is thus the nonequilibrium accumulation of spin @xmath588 which is responsible for the maximum in @xmath471 at low bias voltage . because the dot is coupled symmetrically to the leads , there is no spin accumulation in the parallel configuration and the differential conductance exhibits a smooth parabolic dependence . the maximum in the conductance at zero bias disappears for @xmath561 as well as for @xmath589 . instead of the narrow maximum , a relatively broad minimum develops then in the conductance for both magnetic configurations , as shown in fig [ sec7fig:8](a , b ) or for one magnetic configuration as in fig [ sec7fig:8](d ) . consider for instance the case @xmath561 , shown in fig . [ sec7fig:8](a , b ) . at low bias voltage the dot is occupied by two electrons on the lowest energy level and the ground state is singlet , @xmath590 . current is mediated then only by non - spin - flip cotunneling processes . once @xmath591 , the triplet states start participating in transport leading to an increase in the differential conductance at @xmath592 . thus , the suppression of cotunneling through @xmath588 states gives rise to a broad minimum in the differential conductance . the width of this minimum is determined by the splitting between the singlet and triplet states , @xmath593 . ( color online ) tunnel magnetoresistance as a function of the bias voltage for @xmath594 , as indicated in the figure . the other parameters are the same as in fig . [ sec7fig:8 ] . ( after ref . @xcite ) ] the bias voltage dependence of tmr is shown in fig . [ sec7fig:10 ] . generally , tmr displays a nontrivial dependence on the ground state of the dot . when the dot is occupied by a singlet , @xmath561 , tmr displays a maximum plateau at low bias , which can be approximated by , @xmath595 @xcite . this is the julliere value of tmr . here , it results from the fact that in this transport regime current flows due to non - spin - flip cotunneling . on the other hand , for @xmath596 , the tmr exhibits a minimum at zero bias , which can be seen in fig . [ sec7fig:10 ] for @xmath571 and @xmath597 . when the number of discrete levels in the dot ( relevant for electronic transport ) becomes larger than two , transport characteristics of the dots attached to ferromagnetic leads become more complex and reveal further interesting features . such features were observed experimentally mainly in molecule - based quantum dots . when a large natural molecule is weakly attached to metallic leads , it can be treated simply as a multi - level qd @xcite . as a specific case , we discuss transport properties of a single - wall metallic carbon nanotube ( cnt ) weakly coupled to ferromagnetic leads @xcite . sequential transport in cnts weakly coupled to nonmagnetic and ferromagnetic leads was considered in a recent paper @xcite , where tmr and the fano factor have been calculated within the real time diagrammatic technique . the system was modelled by the hamiltonian in the form introduced by oreg _ @xcite , @xmath598 ^ 2 \nonumber \\ + \delta u \sum_{\mu j } n_{\mu j\uparrow } n_{\mu j\downarrow } + j \sum_{\mu j , \mu^\prime j^\prime } n_{\mu j\uparrow } n_{\mu^\prime j^\prime\downarrow } \,,\end{aligned}\ ] ] where @xmath599 , and @xmath600 ( @xmath601 ) is the creation ( annihilation ) operator of an electron with spin @xmath53 on the @xmath229th level in the subband @xmath297 ( @xmath602 ) . the corresponding energy @xmath603 of the @xmath229th discrete level in the subband @xmath297 is given by @xmath604 , where @xmath275 is the mean level spacing and @xmath605 describes the energy mismatch between the discrete levels corresponding to the two subbands . the second term in eq . ( [ eq : hnt ] ) stands for the electrostatic energy of a charged cnt , with @xmath271 denoting the charging energy and @xmath606 being the charge on the nanotube induced by gate voltages . the next term corresponds to the on - level coulomb interaction with @xmath607 being the relevant on - site coulomb parameter . finally , the last term in eq . ( [ eq : hnt ] ) describes the exchange energy , with @xmath547 being the relevant exchange parameter . ( color online ) the differential conductance in the parallel configuration ( a ) and tunnel magnetoresistance ( b ) as a function of bias and gate voltages . the parameters are : @xmath608 mev , @xmath609 , @xmath610 , @xmath611 , @xmath612 , @xmath613 , @xmath614 , @xmath615 , and @xmath616 mev . ] in fig . [ sec7fig:11 ] we show the differential conductance in the parallel configuration and associated tmr , calculated in the first order approximation . the results are for a specific choice of parameters , namely @xmath617 . the latter condition indicates that the following sequence of the ground states is realized when filling the cnt with electrons while sweeping the gate voltage ( for @xmath343 ) : @xmath618 , where @xmath501 is the total spin of the nanotube . this means that in a certain transport regions , the systems is in the triplet state at equilibrium . [ sec7fig:11](a ) clearly reveals the fourfold pattern of the conductance spectra , associated with filling of the consecutive levels in the two electron subbands of the cnt . the corresponding tmr is shown in fig . [ sec7fig:11](b ) , and reveals a complex variation with both gate and bias voltages . as shown in ref . @xcite , shot noise calculated in the first order approximation is super - poissonian in the blockade regions , and sub - poissonian in the transport regions were sequential tunnelling dominates . the corresponding fano factor in the latter case is close to 1/2 , while in the former one is above 1 . detailed calculations taking into account second order ( cotunneling ) processes show that the fano factor in the blockade regions is generally larger than 1 , although it is reduced in comparison to that derived in the first order approximation , similarly as in the case of two - level dot . when a quantum dot or a large molecule is strongly coupled to external leads , low - temperature transport characteristics reveal features that are typical of the kondo phenomenon @xcite . to analyze the kondo effect @xcite , one may use various theoretical techniques . for instance , one can use the perturbative methods , like the real - time diagrammatic technique discussed in previous sections . however , one has to go beyond the approximations used in the discussion of the sequential and cotunneling transport . another approximate method which is frequently used to study electronic transport through quantum dots , including the kondo regime , is the equation of motion method for the relevant green function . alternatively , one may employ nonperturbative methods , such as for example the numerical renormalization group method @xcite . the kondo effect @xcite in electronic transport through a quantum dot has been first predicted theoretically in refs . since then , it has become well documented experimentally @xcite . the dot with an odd number of electrons has a local spin which at low temperatures , @xmath619 , and in the presence of strong coupling to the electrodes behaves effectively like a magnetic kondo impurity . exchange coupling with the leads electrons gives rise to spin fluctuations in the dot and screening of the dot s spin . this , in turn , leads to the formation of a singlet state and kondo resonance peak at the fermi level in the dot s density of states . this kondo - correlated state leads to an increased transmission through the dot , and also gives rise to a sharp zero - bias anomaly in the current - voltage characteristics . the successful observation of the kondo effect in semiconductor - based quantum dots , as well as in molecular quantum dots based for instance on carbon nanotubes @xcite or other single molecules @xcite attached to metallic electrodes opened new possibility to study the influence of ferromagnetism on the kondo phenomenon @xcite . in this section we briefly discuss how the kondo effect is modified by ferromagnetism of the electrodes . first , we note that in the extreme case of half - metallic leads , i.e. in the absence of minority - spin electrons , the screening of the dot spin is not possible in the parallel magnetic configuration . consequently , no kondo - correlated state can be then formed in this particular magnetic configuration . the possibility of the kondo effect in a quantum dot attached to ferromagnetic electrodes was discussed in a number of papers @xcite . it was shown , that the kondo resonance in the parallel configuration is split and suppressed in the presence of ferromagnetic leads @xcite . however , it was also demonstrated that this splitting can be compensated by an appropriate external magnetic field which restores the kondo effect @xcite . let us begin with the collinear ( parallel and antiparallel ) magnetic configurations . for the antiparallel configuration and vanishing magnetic field and bias voltage , the model is equivalent to a quantum dot coupled to a single lead with density of states @xmath620 @xcite . thus , the usual kondo resonance forms in such a case , which is the same as for nonmagnetic electrodes @xcite . the situation changes in the parallel configuration , where there is an overall asymmetry for up and down spins , say @xmath621 . this significantly reduces ( or even suppresses ) the kondo effect . some basic relations concerning the kondo effect , and particularly those concerning the kondo temperature , can be derived from a simple poor man s scaling @xcite performed in two stages @xcite . first one reduces the energy scale of the effective electron bandwidth @xmath622 starting from @xmath623 . this leads to a renormalization of the level position @xmath624 according to the scaling equations @xcite @xmath625 this leads to a spin splitting of the level , which in the presence of a magnetic field simply adds to the initial zeeman splitting @xmath626 . as a result , one finds @xmath627 the scaling of eq . ( [ eq : hal_scaling ] ) is terminated at @xmath628 @xcite . the strong - coupling limit can be reached by tuning the external magnetic field @xmath496 in such a way that the total effective zeeman splitting vanishes , @xmath629 . in the second stage @xcite , spin fluctuations are integrated out . the schrieffer - wolff transformation @xcite allows to map the anderson model , with renormalized parameters @xmath630 and @xmath631 , to the effective kondo hamiltonian @xcite @xmath632 with @xmath633 in the large-@xmath271 limit . although initially identical , the three coupling constants @xmath634 , @xmath635 , and @xmath636 are renormalized differently during the second stage of scaling . the relevant scaling equations are @xmath637 all the coupling constants reach the stable strong - coupling fixed point @xmath638 at the kondo energy scale , @xmath639 . for the parallel configuration , the kondo temperature in leading order depends on the polarization @xmath196 in the leads following the formula @xmath640 the kondo temperature is maximal for nonmagnetic leads , @xmath641 , and vanishes for @xmath642 . the unitary limit for the parallel configuration can be reached by tuning the magnetic field , as discussed above . in this case , the maximum conductance through the quantum dot is the same as for nonmagnetic leads , i.e. @xmath643 per spin . the scaling procedure was extended in ref . @xcite to noncollinear configurations , where instead of eq . ( [ eq : kondo_temperature ] ) one finds @xmath644 this formula shows explicitly how the kondo temperature varies with the angle @xmath645 between magnetic moments of the leads . numerical renormalization - group ( nrg ) technique @xcite is one of the most powerful and accurate methods available currently to study strongly - correlated systems in the kondo regime . this technique has been adapted recently to the case of a quantum dot coupled to ferromagnetic leads @xcite . a simple way to model the ferromagnetic leads in the standard nrg procedure is to take the density of states in the leads to be constant and independent of electron spin , @xmath646 , the bandwidths to be equal @xmath647 , and lump all the spin - dependence into the spin - dependent coupling parameter , @xmath648 , which can be taken as independent of energy . the nrg method , with recent improvements related to high - energy features and finite magnetic field @xcite , is a well - established method to study the kondo impurity ( quantum dot ) physics . using this technique one can calculate the level occupation @xmath649 , and the spin - dependent single - particle spectral density @xmath650 for arbitrary temperature @xmath31 , magnetic field @xmath496 and polarization @xmath196 . using the spectral function one can find the spin - resolved linear conductance @xmath651 with @xmath652 denoting the fermi function . the kondo effect is usually observed as a zero bias anomaly in the differential conductance . this requires theoretical methods applicable to nonequilibrium situations . one of such techniques is based on the nonequilibrium green functions . the latter are usually determined from the relevant equation of motion and the appropriate decoupling scheme . to describe the main features of the kondo effect in transport through quantum dots connected to ferromagnetic leads ( at least qualitatively ) , one may use the decoupling scheme introduced in ref . @xcite , but generalized by a self - consistent determination of the level energy to account for the exchange field @xcite . alternatively , one may calculate the exchange field first and then include it in the formalism by hand @xcite . differential conductance in the p ( a , d ) and ap ( b , e ) configurations , and the corresponding tmr ( c , f ) . contributions to the conductance from spin - up ( dashed lines ) and spin - down ( dotted lines ) channels are also indicated . left column ( a , b , c ) corresponds to zero magnetic field , whereas the right one ( d , e , f ) to the case when the compensating magnetic field @xmath653 t is applied ( for the lande factor @xmath654 ) . the numerical results are for large @xmath271 limit , while the other parameters are : @xmath655 , @xmath656 , @xmath657 . ( after ref . @xcite ) ] figure [ sec8fig:3 ] shows the differential conductance as a function of the transport voltage . for zero magnetic field there is a pronounced splitting of the peak in the parallel configuration [ fig . [ sec8fig:3](a ) ] , which can be tuned away by an appropriate external magnetic field [ fig . [ sec8fig:3](b ) ] . in the antiparallel configuration , in turn , the opposite happens , i.e. there is no splitting at @xmath658 [ fig . [ sec8fig:3](d ) ] and a finite splitting at @xmath659 [ fig . [ sec8fig:3](e ) ] with an additional asymmetry in the peak amplitudes as a function of the bias voltage . figure [ sec8fig:3](c , f ) shows the corresponding tmr , which becomes negative without magnetic field and changes sign to positive when a compensating magnetic field is applied . within the eom approach the effect of spin - dependent quantum charge fluctuations is accounted in the self - consistent but intuitive manner . the real - time diagrammatic technique @xcite enables one to construct a systematic approach , where the effect of ferromagnetic electrodes can be analyzed without any additional assumptions . recently the resonant tunnelling approximation ( rta ) was extended by utsumi _ _ @xcite to account for influence of the electrodes ferromagnetism on the kondo phenomenon . this technique gives more reasonable results and allows for further systematic insight into the physics of the transport through quantum dots in the kondo regime . the kondo phenomenon in noncollinear magnetic configuration was considered in refs @xcite . the results show that in symmetric systems , the kondo anomaly gradually disappears when the magnetic configuration varies from antiparallel to parallel one . however , the main drop of the kondo peak appears already at small deviations from the antiparallel alignment . recently pasupathy et al . @xcite studied electronic transport through a single @xmath660 molecule attached to ferromagnetic nickel electrodes in the kondo regime . it was shown that the kondo correlations appear even in the presence of ferromagnetic leads . the zero - bias anomaly in the nonequilibrium conductance , however , was split for the parallel alignment of the leads magnetization in agreement with theoretical predictions . for the antiparallel alignment , on the other hand , no splitting of the zero - bias anomaly was observed . some residual splitting observed in this geometry for some samples can be interpreted as an effect of asymmetric coupling @xmath661 . similar behavior has also been observed in electronic transport through carbon nanotubes coupled to ferromagnetic leads @xcite . in this review we addressed certain aspects of electronic transport in mesoscopic tunnel junctions consisting of magnetic metallic nanoparticles , semiconducting quantum dots , or molecules attached to ferromagnetic electrodes . we are aware that the review does not cover all the problems considered in such systems and many references have not been included . there are some other aspects of electronic and spin transport which have been omitted . one of such aspects is spin torque acting on the island and/or electrodes in ferromagnetic single electron transistors based on magnetic metallic nanoparticles @xcite . another topic omitted in this review is the role of electron coupling to a phonon bath . such a coupling leads to new interesting phenomena in transport characteristics @xcite . for instance , the electron - phonon coupling leads to phonon satellite peaks in the differential conductance . however , the most important spin effects generated by spin - dependent tunnelling through the barriers separating the central part from the ferromagnetic electrodes have been addressed . we note that spin polarized transport through nanostructures , where the coulomb effects become important , is rapidly developing in recent years . this applies not only to theoretical part of the subject , but also to the experimental side , where recent progress in nanotechnology allows to attach external leads to individual nanoparticles , quantum dots , and single molecules . an interesting and new field within this topic is the spintronics based on molecular magnets . such molecules are good candidates for magnetic memory cells and also are considered as potential candidates for information processing devices . there is no doubt , that the rapid progress is the physics and technology of spin polarized electronic transport through nanostructures is stimulated by application possibilities in magnetoelectronics , spintronics , and information technology . we acknowledge discussions with j. knig , j. martinek , w. rudziski , r. wirkowicz , and m. wilczyski . this work was supported by funds of the ministry of science and higher education as a research project in years 2006 - 2009 . i. w. also acknowledges support from the foundation for polish science . 100 l. p. kouwenhoven , c. m. marcus , p. l. mceuen , s. tarucha , r. m. westervelt , and n. s. wingreen , in proceedings of the nato advanced study institute on mesoscopic electron transport , ed . by l. l. sohn , l. p. kouwenhoven , and g. schn ( kluwer series e345 , 1997 ) . a. jensen , j. nygrd and j. borggreen in _ proceedings of the international symposium on mesoscopic superconductivity and spintronics _ , edited by h. takayanagi and j. nitta , pp . 33 - 37 , ( world scientific 2003 ) .

possible electronic phase separation ( eps ) in the cuprates has usually been found in terms of stripe phases . thus , neutron diffraction measurements find evidence for fluctuating stripe order in la@xmath0sr@xmath1cuo@xmath2 ( lsco ) associated with incommensurate inelastic neutron scattering@xcite , which can be transformed into long range charged stripe order@xcite by co - doping with nd or eu . similar incommensurate peaks are found in other cuprates , and long range charge ordered states are also found@xcite in strongly underdoped samples of yba@xmath3cu@xmath4o@xmath5 ( ybco ) , with stripes parallel to the chains , while at higher doping short range stripe order is found at virtually all temperatures up to the pseudogap t@xmath6 . however , eps can also manifest itself in the form of _ larger domains _ , particularly if the dopant ions are somewhat mobile and can follow the hole motion . these domains , long known in la@xmath3cuo@xmath7 may recently have been found in scanning tunneling microscope ( stm)@xcite and microwave@xcite studies of bi@xmath3sr@xmath3cacu@xmath3o@xmath8 ( bi2212 ) . stripes are typically interpreted either in terms of charged domain walls in antiferromagnets or of restricted eps . while the models are virtually indistinguishible in the underdoped regime , they diverge as doping increases . a phase separation model implies the existence of a uniform end phase : at some doping antiferromagnetic ( afm ) stripes must disappear , leaving a uniform state similar to that found on the charged stripes . for slightly lower doping the afm stripes form a minority a situation not possible in a domain wall picture . ( a note of caution : more recent domain wall models find a lower doping on the charged stripes , and as the doping decreases , the distinction between the two models gradually blurs . ) evidence for the termination of the afm stripes at a fixed doping@xcite and recent observations of larger domain structures@xcite taken together provide extremely strong support for a phase separation scenario . at this point it is essential to better characterize the nature of the stripes , in particular the charged stripes , and to understand how their properties affect physical phenomena , in particular photoemission spectra.@xcite while fluctuations@xcite play an important role in the real cuprates , we have constructed ordered stripe arrays , for which detailed tight - binding calculations are possible@xcite . such calculations can aid in elucidating the structure of the charged stripes , both for wide stripes what stabilizes the preferred hole density and for narrow stripes how do quantum size effects ( qse ) modify properties of the stripes . the present paper provides an extended analysis of these issues ; some of these results were summarized recently@xcite . the paper is organized as follows . section ii enumerates key issues which must be addressed in any eps model of stripes , including determination of the hole density on charged stripes . it is found that the model can simultaneously account for a large number of experimental results . the charged stripes are probably stabilized by some competing order , either magnetic or paramagnetic ( including charge density waves ) . section iii shows that magnetic charged stripes can arise in a mean field hubbard model , and can be either ferromagnetic or a linear antiferromagnetic ( laf ) phase similar to the white - scalapino stripes@xcite . calculations on stripe arrays find that the charged stripes lead to midgap states near the fermi level ; section iv shows how these data can be interpreted in terms of quantum size effect ( qse ) on single stripes . section v presents the results of single stripe calculations : competing charge density wave ( cdw ) and superconducting orders can exist on paramagnetic stripes , but they are strongly modified by the qse . on laf stripes , d - wave superconductivity and an unusual form of cdw are both found to persist down to the narrowest ( 2 cells wide ) stripes . section vi includes results on stripe arrays : the laf stripes produce photoemission constant - energy maps in significantly better agreement with experiment . in the discussion , section vii , we summarize additional recent evidence which favors a phase separation model in particular , evidence that afm stripes terminate slightly beyond optimal doping , where superconducting properties remain strong , and that a second regime of phase separation exists above optimal doping and we show that experiments are consistent with our prediction that the superconducting gap grows in the underdoped regime . conclusions are given in section viii . a phase separation model of stripes is characterized by the two well - defined stable end phases between which phase separation takes place . the insulating stripes are generally understood to be antiferromagnetic ( afm ) essentially the same as the mott insulator found in undoped cuprates . the hole - doped stripes are assumed to have a finite doping , @xmath9 . this simple idea has three experimentally verifiable characteristic features : ( i ) _ the stripe phase must terminate _ when @xmath10 ; ( ii ) there will be a _ crossover _ at a lower doping , @xmath11 from magnetic - dominated ( @xmath12 ) to charge - dominated ( @xmath13 ) stripe arrays ; ( iii ) some _ interaction _ on the charged stripes stabilizes the end phase at @xmath9 . here we discuss current evidence for ( a ) the doping on the charged stripes , ( b ) evidence for a crossover , and ( c ) the nature of the dominant interaction stabilizing the charged stripes . in addition , we ask what constraints the yamada plot puts on the model , and how a model based on eps compares to a domain wall model . recent evidence suggests that the stripes and pseudogap terminate at the same doping @xmath9 while superconductivity persists to higher doping@xcite . however , the proper choice of @xmath9 requires some discussion . the neutron diffraction measurements of tranquada@xcite and yamada@xcite have established that charged stripes in la@xmath0sr@xmath1cuo@xmath14 ( lsco ) have an invariant topology over the doping range @xmath15 , acting as antiphase boundaries ( apbs ) for the afm stripes and having a net doping of 0.5 holes per stripe . however , there are two models for how this charge is distributed : either in one row with average hole density 0.5 hole per copper site or in two rows with 0.25 hole per copper . these two alternatives are often somewhat simplistically referred to as site order vs bond order . the strongest evidence distinguishing between the alternatives comes from x - ray data on the charge order@xcite at @xmath16=1/8 : non - observation of diffraction harmonics suggests a sinusoidal distribution of charge . for a 4-cu repeat distance , two insulating and two charged rows would be exactly sinusoidal , whereas one charged and three insulating rows should have signifigant harmonic content . a similar conclusion was reached by @xmath17sr lineshape analysis@xcite . however , the charge ordering peaks are weak , and it remains possible that fluctuations or disorder could wash out the harmonics . direct evidence for the density on the charged stripes is found from low temperature nqr measurements@xcite , which find values @xmath18 . while this is close to the lower value , the small difference can also be understood : this is a _ local _ measurement , and it is expected that some holes will be pushed off onto the magnetic stripes . indirect evidence favoring the lower hole density includes the fact that it is easier to understand the properties of afm stripes in terms of even - leg ladders ( e.g. , the afm stripes would be two coppers wide at @xmath16=1/8)@xcite , and that the stripe phase appears to terminate when the average doping approaches @xmath16=0.25 ( section vii.a.1 ) . the lower doping is also more consistent with the tj model simulations of white and scalapino@xcite . tallon@xcite finds an optimal doping at @xmath19 for all cuprates , with respect to which the stripes terminate at a doping @xmath20 . however , it is hard to reconcile a common optimal doping with muon spin resonance data@xcite , which find @xmath21 is optimized at very different values of @xmath22 ( @xmath23 is the pair density , which seems to scale with carrier concentration@xcite , and @xmath24 effective mass ) for lsco and ybco . we assume instead that @xmath25 scales with @xmath26 , so if @xmath19 for lsco , it is 0.21 for ybco , in good agreement with several estimates@xcite . this also resolves a problem with the thermopower . while the thermopower appears to be universal for most cuprates , and the best means of estimating the doping is from the room temperature thermopower , lsco is anomalous in that ` overdoped ' samples still have high thermopower@xcite . if the doping for ybco is rescaled as above , however , the thermopower data of lsco fall on the universal curve . hence , the anomaly for lsco is not in the thermopower , but in a too low value of @xmath21 , which is accompanied by a too low value of @xmath25 . it is likely that these features are associated with a competing ltt phase , which is most prominent in lsco , and which also leads to the most nearly static stripe correlations . taking @xmath25=0.21 for ybco gives @xmath27 , which we believe holds for _ all cuprates _ , including lsco , section vii.a.1 . this would lead to very wide charged stripes near optimal doping : the width of the charged stripes @xmath28 satisfies @xmath29 , or @xmath30 cu wide . hence , models of isolated quasi - one - dimensional charged stripes are likely to be valid only in the far underdoped regime , while for the good superconductors a better model would be a metal with intrinsic weak links@xcite . for @xmath31 , the crossover @xmath32 can be identified with the 1/8 anomaly , where both charged and afm stripes have their minimal width ( 2 cu atoms ) . there is considerable evidence that the doping 1/8 , in addition to its special stability , acts as a crossover in the properties of the stripes . thus , uchida , et al.@xcite , studying the hall coefficient @xmath33 , find a crossover from one - dimensional behavior ( @xmath34 as @xmath35 ) for @xmath36 to two dimensional behavior ( coupled charged stripes ) for @xmath13 . in ybco , the spin gap grows slowly with doping for @xmath12 , then more rapidly for @xmath13 ; this behavior can be understood in terms of coupled spin ladders , as the coupling changes with the width of the charged stripes@xcite . in eu substituted lsco@xcite , the meissner fraction is negligibly small for @xmath12 , then grows roughly linearly with doping until @xmath37 , staying large up to at least @xmath38 . finally , the two - magnon raman peak in lsco has a splitting at low temperatures which has been associated with stripes@xcite , on analogy with similar observations in la@xmath0sr@xmath1nio@xmath2 ( lsno)@xcite . for @xmath12 the ratio of the two peak frequencies is constant and consistent with a simple stripe model ; for @xmath39 the lower frequency starts decreasing with doping . moreover , the higher frequency loses intensity with doping ; near @xmath40 , the _ intensity _ of one mode approximately disappears , while the _ frequency _ of the other mode extrapolates to zero . = 0.25 0.5 cm = 0.48 0.5 cm the yamada@xcite plot , fig . [ fig : y1]a , provides a severe constraint on any model of stripes : in lsco the incommensurability @xmath41 is found to grow linearly with doping for @xmath12 but to saturate for @xmath42 . furthermore , the saturation value is just the incommensurability expected for @xmath43-doped stripes , @xmath44 . a similar saturation has been reported in ybco , but different groups find different values for @xmath45 : @xmath46 [ ] or @xmath47 [ ] . if one of these values proves correct , it would suggest some nonuniversality in @xmath9 , perhaps associated with bilayer splitting . in the domain wall model , the saturation is interpreted as evidence that the stripes stop changing in width , and additional holes leak into the antiferromagnetic background , washing out the stripes . for @xmath48 , the eps model agrees with the yamada plot , fig . [ fig : y1]b : increase in doping causes the afm stripes to narrow , with no changes in the charged stripes . note that , for concreteness we have assumed that the charged stripes have laf order ( section iii ) ; these stripes naturally act as antiphase boundaries ( apbs ) , consistent with the neutron evidence . however , the naive prediction of the eps model is in disagreement with the yamada plot for @xmath49 , fig . [ fig : y1]c : as the charge stripes get wider , the incommensurability should decrease , while the neutron peaks may broaden if the wider charge stripes do not act as apbs . this behavior is not observed . however , the model can be simply modified to explain the observed saturation , fig . [ fig : y1]d , e . this would be a commensurability effect , with part of the sample pinned at 1/8 doping while the rest forms a different phase ( e.g. , at 1/4 doping ) where no stripes are present . such behavior is well known in nickelates , where coexistence of 1/3 and 1/2 stripes is common . evidence for such commensurability effects can be found by comparing@xcite the chemical potential @xmath17 in lsco@xcite and lsno@xcite . in both materials , @xmath17 is approximately independent of doping between half filling ( @xmath50 ) and @xmath51 , with @xmath52 in lsco , 1/3 in lsno , which has long - range charge order . remarkably , the simple prediction of a macroscopic phase separation model ( @xmath17 constant throughout the phase separation regime ) is violated in lsno : @xmath17 is _ not _ constant for @xmath53 , although the stripe phase persists over the full doping range . this is presumably a commensurability effect , reflecting the coexistence of 1/3 and 1/2 stripe phases for intermediate dopings . the same doping dependence of @xmath17 is found in lsco : constant for @xmath54 , variable for @xmath49 . indeed , the lsno data can be scaled to that of the cuprates , giving a common doping dependence @xmath55 in both compounds@xcite . recently local charged domains have been directly visualized by stm studies in bi2212@xcite with gaps ranging from slightly overdoped to strongly underdoped . such behavior is extremely difficult to explain in the domain wall picture , but is easy to understand in terms of eps , and indeed is consistent with the commensurability effects discussed above . the gap distribution is found to be broad rather than bimodal , but that is expected since eps is very sensitive to charged impurities , and the local gap will correlate with the local density of dopant , presumably interstitial oxygen in bi2212 . this sensitivity to impurities leads to the question of which came first : is a preexisting eps sensitive to local impurities , or does oxygen clustering@xcite provide the driving mechanism for domain formation ? since the electronic inhomogeneity seems characteristic of most cuprates while there is considerable variety in the doping counterions , the simpler interpretation would appear to be that the eps is primary . thus , in la@xmath3cuo@xmath7 ( lco ) , the interstitial oxygens are highly mobile , allowing the domains to grow to macroscopic size . similar clusters form in ybco ( here associated with chain oxygens ) , but can be suppressed in fully oxygenated samples by quenching@xcite . on the other hand , well formed stripes appear when the doping counterions are least mobile , in lsco . the domains in bi2212 would seem to be consistent , but a number of questions remain about the role of annealing . while these domains are regularly found in stm studies when the samples are cleaved at low - temperature , they can be annealed out at high temperatures@xcite . this could be caused by a ` melting ' of the eps , as is found in lco near room temperature@xcite , and can be tested by careful annealing studies . alternatively , it may be a question of pinning the eps . certainly , it is known that a weak domain disorder is necessary to pin vortices , to observe the vortex lattice via stm.@xcite also , the role of the interstitial oxygen in the superlattice modulations in bi2212 needs to be better understood . the strength of this modulation suggests that it is associated with an ordering of the interstitial oxygen , but this should lead to a strong correlation between the domains and the superlattice , which seems not to be the case . further evidence for electronic inhomogeneity comes from microwave measurements : anomalies in bi2212 have been interpreted in terms of similar electronic domains@xcite , suggesting that they are representative of the bulk , while measurements on other cuprates@xcite find similar anomalous behavior which was interpreted in terms of pinned cdw s , possibly stripe related . a domain picture would also explain the persistence of nodal quasiparticles in the underdoped regime , at least down to 1/8 doping@xcite . table i summarizes the discussion of the previous subsections , and compares the predictions of eps models model & @xmath56&apb?&saturation & domains?&crossover & termination + & & & at @xmath57?&&_at _ @xmath58?&at @xmath59 ? + domain wall&fixed pattern&@xmath60&@xmath60&@xmath61&@xmath62&@xmath62 + & holes spread out & & & & & + eps ( simple)&charge stripes grow&@xmath63&@xmath61&@xmath61&@xmath60&@xmath60 + & in width & & & & & + eps ( commensurate & domain phase:&@xmath63&@xmath60&@xmath60&@xmath60&@xmath60 + pinning)&@xmath58 : stripes & & & & & + & @xmath64 : uniform & & & & & + with domain wall models of stripes . domain wall models arose in early unrestricted hartree - fock ( uhf ) calculations@xcite for the doped tj model , as heterogeneous ground states in which the holes are confined to domain walls between afm domains which act as apbs . these domain walls are not driven by phase separation@xcite ; the phase is realized as a long - period modulated afm , with holes doping the rows of spins where the moment changes sign@xcite . when @xmath65 , there is one hole per row , a large hole doping which is inconsistent both with experiments on stripes in the cuprates , and with density matrix renormalization group ( dmrg ) calculations@xcite . more recent calculations reduce the doping , in ways that bring the model close to the phase separation picture : ( 1 ) assuming cdw order along the charged stripes@xcite lowers @xmath9 to 0.5 , while ( 2 ) letting @xmath66 results in @xmath67.@xcite since the cdw order ( case 1 ) lowers the free energy of the charged stripes , it should be possible to dope the system all the way to a pure cdw at @xmath68 . in case 2 , it is not explained why a low density appears ; a good possibility is that this doping corresponds to the van hove singularity ( vhs ) , as in the phase separated stripes . ( this could be checked by varying @xmath69 and calculating @xmath70 . ) if such low - density domain walls do exist , the striped phase must somehow terminate when the average doping approached 0.2 , but how this happens is not explained . however , this calculation@xcite is not fully uhf , being restricted to periodic arrays , and uhf calculations@xcite find instead ferromagnetic charged stripes which can be understood within a phase separation model ( see below ) . since the stability of these @xmath71 domain wall stripes is unclear , in table i a comparison is made with the earlier , higher - hole - density domain wall models . for such high doping , stripe termination at @xmath72 , or a major crossover near 1/8 , are both difficult to interpret . stripes as apbs are more natural in domain wall models , but we shall show that they can also arise in eps models . note that the distinction between the models may be becoming moot , since in the domain wall models there is now a search for a ` hidden ' order parameter@xcite , and it is common for secondary order parameters to be expressed predominantly on domain walls of the primary order@xcite . in conclusion , assumption of a charged stripe doping @xmath73 reconciles the neutron diffraction data , evidence for a termination of the stripe phase near @xmath9 , and the 1/8 anomaly as a crossover effect near @xmath74 , while commensurability effects can explain the saturation in the yamada plot and the stm observation of charge domains . further supportive evidence will be presented in section vii.a . in the remainder of this paper , we will assume that @xmath75 . in any phase separation model , a key issue is understanding the nature of the charged stripes . indeed , since superconductivity seems to arise predominantly on these stripes , such understanding is likely to play an important role in elucidating the origin of the high superconducting transition temperatures . for the stripe phase to exist , the doping @xmath9 must be particularly _ stable_. this can arise via an electronic _ instability _ , which opens up a gap over much of the fermi surface , making the electronic phase nearly incompressible . this ` stability from instability ' is a fairly general feature , underlying , e.g. , hume - rothery alloys@xcite . [ this is a modification of an argument due to anderson@xcite . ] here , we explore a number of candidates for the predominant electronic instability . in a related paper@xcite , we will provide strong evidence that this ` hidden order ' is a form of cdw , which could include the flux phase . however , here we will explore a wider variety of possibilities . one issue is that in the low doping limit the charge stripes act as apbs for the afm stripes . such an effect arises naturally if the charged stripes have some residual magnetic interaction , and we will explore this possibility . however , in nickelates charged stripes coupled to a cdw are found to act as apbs@xcite . the large hubbard on - site repulsion @xmath76 plays an important role . strong correlation effects lead to two classes of charged stripe phases : either the constraint against double occupancy leads to magnetic order ( magnetic charged stripes ) or kinetic energy dominates , leading to a magnetically disordered phase ( paramagnetic charged stripes ) , with correlations leading to reduced hopping , as in tj@xcite and slave boson@xcite models . section iii will provide examples of both classes , denoted as class b and class a stripes , respectively . class a stripes could be simply correlated metals ( as in tj or slave boson calculations ) or could have additional , e.g. , cdw , order . a crossover from magnetic to correlated paramagnetic groundstate arises as a function of doping in models of itinerant ferromagnets@xcite . for completeness , class c stripes are defined as those arising not from phase separation but from long - range modulated afm order ( domain wall stripes ) . such stripes have been described in detail@xcite , and will not be further considered here . in the next section , we show that class b ( magnetic ) charge stripes can arise in a mean - field hubbard model@xcite , with the charged stripes displaying either ferromagnetic ( fm ) or linear antiferromagnetic ( laf ) ( ordering vector @xmath77 ) order . the laf stripes are very similar to white - scalapino stripes@xcite . the fm phase is stabilized by vhs nesting . this fm phase may be present in ruthenates@xcite , but is unlikely to be relevant for the cuprates ( for one thing , the fm stripes are likely to be diagonal , and do not form apbs@xcite , contrary to experiment ) . we have suggested that other vhs - stabilized phases are more likely@xcite ( see also ref . ) , and here we explore the properties of a class a cdw phase@xcite . at a doping @xmath78 , the effects of correlations are relatively weak , renormalizing the bandwidth by a factor of @xmath632 . thus for the present calculations on paramagnetic stripes , it will be assumed that renormalized parameters are used , and other effects of strong correlations will be neglected . while laf stripes are most stable when @xmath65 , we explore the possibility that they can be stabilized even when @xmath71 by on - stripe cdw or superconducting order . ordinarily , the cdw phase is believed to _ compete _ with strong correlation effects , but we find ( section v.b ) that an unusual form of cdw phase can _ coexist _ with laf order : the charge density varies between zero and one ( not two ) holes per atom . in a stripe array , the alternating stripes are associated with the two stable thermodynamic phases . here , we summarize the different ways these stripes are denoted in this paper . the stripe with lower hole doping is variously referred to as ` insulating ' or ` antiferromagnetic ' ( afm ) . ( these are also known as magnetic stripes , since the charged stripes have considerably weaker magnetic order , but we will avoid that notation here . ) the stripe with higher hole doping is generically referred to as ` charged ' or ` hole - doped ' . at low temperatures , these stripes are also ` superconducting ' , but at high temperatures , they are stabilized by some ` hidden order ' , and one purpose of this paper is to explore a number of possible orders . the orders fall into two groups : ` magnetic charged ' ( class b ) stripes could have fm or laf order ( the latter are white - scalapino - like stripes ) , while ` paramagnetic ' ( class a ) stripes could have cdw or flux - phase order . strong coupling models would seem to be natural for producing phase separated or striped ground states . any magnetic ordering avoids double occupancy , while changing from one form of magnetic order to another , via , e.g. , doping , requires highly collective spin rotations , as competing orders are orthogonal . while superexchange leads to antiferromagnetic ( afm ) insulators at half filling , doping tends to favor textures with parallel spins ( e.g. , ferromagnetic ( fm ) ) to maximize hole hopping . such ferron phases were introduced long before high @xmath21@xcite , but it remains controversial whether such states are ground states of the hubbard model@xcite . while the tj model does have phase separation for large @xmath79 , it is unclear whether such phases extend to the values @xmath80 expected for the cuprates@xcite . we have found phase separated solutions of the hubbard model at mean - field level@xcite . while these solutions are metastable in uhf calculations , they closely resemble the ws stripes , and provide an interesting example of phase - separation mediated stripe phases . we find a well defined surface tension for wide , isolated stripes , which decreases and changes sign as the stripes become narrower . when the surface tension becomes negative , the stripes no longer remain straight , and spontaneously meandering solutions are found . these solutions are found by considering only low - order commensurate phases , with wave vector @xmath81 0 or @xmath82 only . the bare dispersion is @xmath83 with @xmath84 . the hubbard interaction @xmath85 leads to magnetic order with a mean - field magnetization @xmath86 at wave vector @xmath87 , and the quasiparticle dispersion becomes @xmath88 where @xmath89 and @xmath90 for the cuprates , we expect@xcite @xmath91 , @xmath92 and @xmath93 . the magnetization is found self - consistently from @xmath94 with fermi function @xmath95 . the resulting free energy is @xmath96 with @xmath97 @xmath98 = 0.40 0.5 cm the competing phases include afm for @xmath99 , fm with @xmath100 , and a linear antiferromagnet ( laf ) with @xmath101 . when the laf stripes are two cells wide , this laf phase closely resembles the white - scalapino stripes , fig . [ fig : y1]e . the afm state has lowest free energy at half filling , but ( for @xmath65 ) is unstable for finite hole doping . for @xmath65 , there is phase separation between the afm and laf phases , while for finite @xmath69 the phase separation is between afm and fm phases , fig . [ fig:10 ] . ( when electron - phonon coupling is included , it is found that the fm phase is unstable with respect to a cdw phase@xcite . ) = 0.40 0.5 cm the laf stripes for @xmath65 are discussed in ref . . for @xmath71 electron - hole symmetry is absent ; for hole doping @xmath102 , there is phase separation to a fm phase , consistent with recent simulations by vozmediano , et al.@xcite , who find a uniform fm phase at @xmath103 for @xmath104 , @xmath105 . however , _ on the electron - doped side a uniform afm phase is stable over a large doping range _ , suggestive of the asymmetry found in the cuprates . the dot - dashed curve shows that in the phase separation regime , the low energy physics can be approximated by the form of free energy assumed in ref . , @xmath106 ( neglecting a term linear in @xmath16 ) with @xmath9 the hole doping of the uniform charged phase . the fm phase is stabilized by vhs nesting , inset to fig . [ fig:10 ] , as found previously@xcite . the tangent construction tends to select the fm phase at dopings somewhat away from optimal nesting ( fermi energy above midgap ) . both regimes of phase separation seem to be driven by hole delocalization : one - dimensional ( along the laf rows ) when @xmath65 , two - dimensional for finite @xmath107 . the resulting phase diagrams @xmath16 vs @xmath76 , fig . [ fig:6 ] , are strikingly different . for @xmath65 , fig . [ fig:6]a , the phase separation is between the afm and paramagnetic ( laf ) phase for @xmath108 ( @xmath109 ) , while for finite @xmath69 , fig . [ fig:6]b , there is generally a vhs - stabilized fm phase . for small @xmath76 and @xmath71 , there is a regime where simple spin - density wave theory works and a uniform afm phase is stable , but when @xmath65 phase separation persists for all finite @xmath76 . note that @xmath110 marks a crossover between class a ( paramagnetic ) and class b ( magnetic ) charged stripes . the value @xmath110 is close to the @xmath111 expected in the cuprates , although for finite @xmath69 @xmath110 decreases , @xmath112 for @xmath113 , and the range of @xmath76 for which paramagnetic stripes are stable becomes very small . while these stripes are metastable in uhf calculations , we will show below that the stability of phase separating stripes can be enhanced by _ additional interactions _ beyond the pure hubbard model . in ordered stripe arrays it is found@xcite that the afm stripes have a mott - hubbard gap , and the features near the fermi level are associated with the charged stripes . this charge stripe dispersion shows a series of quasi - one - dimensional features which qualitatively resemble the bands of an isolated stripe produced by qse . in this section , we make a quantitative comparison with isolated stripes , and explore the mechanism of qse - induced van hove splitting . for a single stripe @xmath28 cu atoms wide , the dispersion is still given by eq . [ eq:1 ] , but the allowed @xmath114 values are quantized , with @xmath115 , @xmath116 . these are in fact the usual quantized bloch bands , but for large @xmath28 the quantization is not noticeable . for small @xmath28 , the dispersion appears as a series of @xmath28 overlapping one - dimensional ( 1d ) dispersions . equation [ eq:1 ] can be rewritten as @xmath28 1d dispersions @xmath117 these are the qse , with corresponding density of states ( dos ) shown in fig . [ fig:1 ] . notice that for @xmath118 , the vhs is readily apparent in the dispersion . even down to @xmath119 , the vhs is clearly defined ( albeit only within a finite interval ) as the _ locus of energies where all subbands overlap_. in fact , _ the qse opens a gap at the vhs _ , effectively lowering the kinetic energy of the electrons just like a conventional ( e.g. , cdw ) gap . the vhs splitting can be found from eq . [ eq:1e ] : @xmath120 this splitting has two consequences : first , the vhs splitting enhances the stabilization of the striped phase ; but second , the qse gap competes with other gaps , such as cdw s and superconductivity . however , while the qse splits the vhs peak , substantial dos remains ungapped , so additional instabilities remain possible . this competition will be discussed further in the next section . = 0.40 0.5 cm it should be noted that this is the first direct demonstration that the vhs can be defined on a stripe only two atoms wide . the definition is quite analogous to the standard definition in two dimensions : the point at which the bands cross over from electron - like to hole - like . figure [ fig:1a ] compares the gaps of a single stripe with those found in the ordered stripe array@xcite ; the array is labelled @xmath121 when the magnetic stripes are @xmath24 coppers wide and the charged stripes are @xmath122 coppers wide . in the array calculation , no competing order was introduced on the charged stripes , so the qse provides the only gap . figure [ fig:1a ] shows that there is a very good match for both 2 cu wide and 6 cu wide stripes , although for the 2 cu stripe , the vhs gap is somewhat larger for the single stripe than in the array . from eq . [ eq:1e ] , the 1d bands have dos peaks at band bottom and top ; the band bottom corresponds to @xmath123 i.e. , the dispersion from @xmath124 is flat , at the energy corresponding to the lower dos peak . ( the intensity along @xmath125 is given by a structure factor , which does not directly come into a single stripe calculation . ) along @xmath126 one should see the 1d dispersion extrapolating to the band top at @xmath127 . given the good agreement , it should be possible to analyze competing orders on a single stripe , for which the calculations are simpler ( no need to self - consistently adjust doping on each row to account for charging effects ) . = 0.40 0.5 cm in this section , we will develop two closely related models of the competition of cdw order and superconductivity _ on a single charged stripe_. the first is a class a paramagnetic stripe , stabilized by electron - phonon coupling@xcite , while the second has dominant electron - electron coupling , with ( class b ) magnetic charged stripe order . the earlier calculation@xcite used a van hove stabilized cdw model to describe the doping dependence of the pseudogap . here we reapply the model for a single stripe , by introducing the following modifications . ( 1 ) the calculation is carried out on a single , finite width stripe . ( 2 ) pinning to the vhs arises naturally , since all stripes are at the same doping . ( 3 ) for closer approximation to experiment , d - wave superconductivity is assumed . ( 4 ) correlation effects due to hubbard @xmath76 are neglected : previous slave boson calculations suggest that the main effect is a bandwidth renormalization of a factor @xmath632.@xcite we briefly recall the energy dispersion and the gap equations of the model@xcite , generalized to d - wave . in terms of a function @xmath128 the gap functions are @xmath129 for superconductivity , and @xmath130 for the cdw . the energy eigenvalues are @xmath131 and their negatives , with @xmath132 @xmath133 , @xmath134 , @xmath135 , @xmath136 , and the nesting vector @xmath137 . the gap equations are @xmath138\over 2 e_{+,k}}\tanh{{e_{+,k}\over 2k_bt } } \nonumber \\ -{e_{-,k}^2-\epsilon _ { \vec k+\vec q}^2-\theta_{\vec k+\vec q}[\delta_{\vec k}^2+g_{\vec k}^2]\over 2 e_{-,k}}\tanh{{e_{-,k}\over 2k_bt } } \bigr ) , \label{eq:23}\end{aligned}\ ] ] @xmath139 with interaction energies @xmath140 and @xmath141 , and @xmath142 , @xmath143 . the cdw - superconducting competition was studied in bulk in ref . . the previous results are recovered for a sufficiently wide stripe ( @xmath63100 cu wide ) . for narrower stripes , it is found that the quantum confinement gap severely interferes with alternative gap formation . figure [ fig:19 ] illustrates how the various gaps vary with stripe width , near @xmath75 . the data are plotted vs average doping , assuming a regular stripe array with 2-cu wide afm stripes ( which do not contribute to the gaps ) and n - cu wide charged stripes of doping @xmath9 , giving an average hole doping @xmath144 . the cdw gap is most sensitive to stripe width , but in the narrowest stripes superconductivity is also suppressed ( the suppression is stronger for a d - wave gap ) . strong instabilities are possible on the stripes , but they are shifted in doping away from @xmath75 . thus , the cdw instability requires both @xmath145 and @xmath146 to be near the fermi level ; for a two - cell wide stripe , this is only possible near @xmath50 . on the other hand , superconductivity is possible anywhere , if the coupling is strong enough . for a two - cell wide stripe , the optimal superconductivity arises when the fermi level is at the _ one - dimensional vhs _ at the edge of one of the stripe subbands . this depends on @xmath69 , and for @xmath113 falls at @xmath147 ( a larger gap is found on the electron - doped side , @xmath148 ) . it should be noted that , even though the superconductivity is assumed to be d - wave , in general a finite minimum gap is found on the stripe , even when the cdw gap is zero . this is because the vanishing d - wave gap can be sampled only when the point @xmath149 is sufficiently close to the fermi surface , which in general requires @xmath28 to be odd ( recall that the allowed values of @xmath114 are integer multiples of @xmath150 ) or very large . = 0.40 0.5 cm in the above calculations , the @xmath151 s arose from electron - phonon coupling@xcite . similar contributions follow from electron - electron coupling , in an extended hubbard model . for instance , the near - neighbor coulomb repulsion has the following mean - field expansion : @xmath152 where @xmath153 and @xmath154 . the first two terms in eq . [ eq:24b ] renormalize the chemical potential and the hopping @xmath155 respectively , and can be neglected . the terms in @xmath156 comprise a pseudospin triplet of cdw - like distortions , with @xmath157 representing a cdw similar to the one discussed above , @xmath158 being related to the low - temperature tetragonal distortion , and @xmath159 an orbital antiferromagnet , closely related to the flux phase . the remaining terms are a d - wave superconductor . the coefficients of the terms must be found self - consistently by solving the gap equations : @xmath160 @xmath161 @xmath162 @xmath163 the terms @xmath164 and @xmath165 have recently been discussed by valenzuela and vozmediano@xcite . a detailed discussion of the competition between the three cdw - like modes is given in ref . . retaining only the superconducting term in eq . [ eq:24b ] , the interaction can be derived from a quartic term @xmath166 with @xmath167 . assuming @xmath168 , the gap equations can be written in the form @xmath169 ( @xmath170 ) , with @xmath171 and @xmath172 . for the uniform charged state ( infinitely wide stripe ) @xmath173 , and the gap symmetry can be simply analyzed . the symmetry can be either d - wave ( @xmath174 ) or extended s - wave ( @xmath175 ) , with the choice @xmath176 giving the largest gap . the @xmath177 term is always bcs - like , having the opposite sign from @xmath178 , while the @xmath179 term has the opposite sign from @xmath177 , the integral being dominated by the regions near the vhs s . hence , there are two possibilities : ( i ) attractive ( @xmath180 ) d - wave superconductivity or ( ii ) repulsive ( @xmath181 ) extended s - wave . however , the latter would require @xmath182 , which does not arise in the present model , so only case ( i ) is possible . these considerations readily generalize to a finite - width stripe , for which @xmath183 . agterberg , et al.@xcite recently introduced a model for ` exotic ' superconductivity in multiband superconductors . if the fermi surface consists of several inequivalent but degenerate pockets , the order parameter can consist of symmetry allowed superpositions of the order parameters of the individual pockets . equation [ eq:37 ] can be thought of as a form of exotic superconductivity , with the degenerate vhs s playing the role of hole pockets . = 0.40 0.5 cm in the above calculations , it was implicitly assumed that the doping is high enough that the only role of the on - site repulsion @xmath76 is to renormalize the band parameters . however baskaran@xcite recently estimated that near optimal doping correlation effects remain stronger than the kinetic energy associated with hopping . hence , it is important to look for stripe ground states which minimize this on - site repulsion ( class b stripes ) . the linear antiferromagnet ( laf ) stripes discussed in section iii are a good candidate for the cuprate charged stripes : they closely resemble the white - scalapino stripes@xcite , have an appropriate doping , close to @xmath75 , include strong correlations , and have the additional advantage that a two - cell wide laf charge stripe acts as a natural apb for the af stripes . in this section , we will explore these stripes , and show that they can be further stabilized by additional interactions . a special form of strongly correlated cdw is found to exist on a laf . the charge and spin distribution is shown in the insert to fig . [ fig:20b ] , with the corresponding dispersion in figure [ fig:20a]c . there is a strong antiferromagnetic ordering on one sublattice , while most of the holes are confined on the other , nonmagnetic sublattice . whereas in a conventional cdw the charge density is zero on one sublattice and two on the other , in this strong coupling case the hole density varies from @xmath184 to 1 , and there is no double occupancy , fig . [ fig:20b ] . whereas the paramagnetic stripes were extremely sensitive to quantum confinement , these magnetic charged stripes are much less so : this cdw is stable almost independently of the stripe width . from fig . [ fig:20a]c , it can be seen that the gapped fermi surface still has hole pockets near @xmath149 , which would lead to conducting stripes , consistent with optical properties@xcite . however , it is only found near a hole doping @xmath68 , and so does not appear to be relevant for stripe physics in the cuprates . = 0.40 0.5 cm away from this doping , cdw instabilities are relatively weak and it is possible to stabilize d - wave superconductivity , figs . [ fig:20a]b,[fig:20 ] . while the overall dispersion varies with stripe width , the superconducting gap is also relatively insensitive to the width , and actually increases for the narrowest stripes , fig . [ fig:20c ] . note that the order parameter is not a pure d - wave , the gap along the stripe being larger . such a large anisotropy is not consistent with tunneling measurements of the gap ; it is possible that the anisotropy is reduced by strong interstripe coupling . on the other hand , a large gap anisotropy has been found in ybco@xcite , where the stripes are aligned along the chain direction@xcite . = 0.40 0.5 cm = 0.40 0.5 cm note in figs . [ fig:20a]b , [ fig:20 ] that the combination of laf and d - wave order leads to a finite minimum gap over the full fermi surface . while the pure laf phase is not stabilized by the vhs , the d - wave superconductivity is optimized when the fermi level is at the @xmath77 vhs at essentially the _ same doping _ , @xmath185 , as the vhs on a paramagnetic stripe ! = 0.40 0.5 cm = 0.40 0.5 cm in ref . , the fermi surface was calculated for a series of ordered stripe arrays . these results can now be compared to experimental photoemission data@xcite . for this purpose we replot the data as integrated spectral weight over a finite energy cut within energy @xmath186 of the fermi surface . figure [ fig:21]a shows a cut with @xmath186 = 200mev , for the model of a 1/8 doped stripe array ( i.e. , @xmath187)@xcite . the pattern is readily understood : the stripe superlattice leads to a number of quasi - one - dimensional bands ; however , due to structure factor effects they have significant intensity only near the original fermi surface , solid line in fig . [ fig:21]a . for comparison with experiment , the calculated spectral weight is symmetrized @xmath188 in fig . [ fig:21]b to represent a sample with regions of stripes running along both @xmath189 and @xmath190 directions . finally , an empirical matrix element is included , fig . [ fig:21]c , which extinguishes spectral weight along the zone diagonal , @xmath191 , similar to the matrix element assumed in analyzing bi2212 , refs . . the resulting fermi surface maps for several values of @xmath186 are illustrated in ref . for paramagnetic charged stripes , and in figs . [ fig:22 ] ( @xmath43 ) and [ fig:23 ] ( @xmath192 ) for laf stripes . = 0.40 0.5 cm for both models , the stripe band nearest @xmath77 is in good agreement with experiment : there is little dispersion perpendicular to the stripe , while the intensity falls off toward @xmath193 due to the structure factor effect . in general , the laf stripes are in better agreement with experiment , since the additional subbands predicted for paramagnetic stripes ( moving from @xmath77 toward @xmath194 ) are not seen in the experiment . while the matrix element improves the agreement , theory suggests that this effect is present only for certain photon polarizations@xcite . one disagreement with experiment for both models is that for shallow energy cuts ( 30 , 100 mev ) the experiment still finds a smeared dispersion rather than a sharp fermi surface . this is presumably an effect of stripe fluctuations . it should be noted that all the spectral weight in figs . [ fig:21]-[fig:23 ] is associated with the charged stripes ; the lower hubbard band of the afm stripes lies below 0.5ev in lsco . it is somewhat surprising that the spectral weight nearest the fermi level is near @xmath77 , since this is where the pseudogap arises . nevertheless , our calculation reproduces both the ( quantum confinement ) pseudogap , fig . [ fig:1a ] , and the spectral weight distribution . the two most common interpretations for the pseudogap are in terms of either precursor pairing or a competing order parameter , which may be magnetic@xcite or charge - density wave ( cdw)@xcite . we will argue in the next subsection that the latter possibility is more likely . such competing instabilities arise naturally in a phase separation model ( e.g. , refs . ) , and we have long argued that the pseudogap is a manifestation of local phase separation@xcite . it should be noted that a pseudogap arises in the nickelates@xcite in conjunction with stripe fluctuations , and turns into a true gap at the charge ordering temperature . therefore , the fact that the pseudogap closes in the overdoped regime in the cuprates strongly suggests that the stripe phase terminates at the same doping . direct evidence for this has recently appeared@xcite . figure [ fig:3]a compares the intensity @xmath195 of the inelastic neutron peaks near @xmath196 as a function of doping in ybco@xcite and lsco@xcite . in a stripe picture , @xmath195 should be a measure of the fraction of material in afm stripes . remarkably , the intensity extrapolates to zero at nearly the same doping in both materials , even though @xmath197 peaks at substantially different dopings . for both materials , this is the doping at which the pseudogap closes . while in ybco , these peaks have been interpreted as a fermi surface nesting effect@xcite , this can not explain the charge order fluctuations , and in lsco the tranquada data@xcite is associated with _ elastic _ magnetic peaks . furthermore , a study of the neutron diffraction pair distribution function@xcite for lsco finds evidence for charge fluctuations , presumably associated with stripes . the excess fluctuations are maximal near @xmath103 , and terminate near @xmath198 . strong reductions of thermal conductivity associated with stripe scattering also terminate at a comparable doping@xcite , while a recent optical study@xcite finds evidence for a quantum critical point at a similar doping , @xmath199 . moreover , in bi2212 , tokunaga , et al.@xcite have introduced a new crossover temperature @xmath200 based on cu nmr , below which afm correlations develop ; they find @xmath201 near @xmath40 . a recent nqr study of the slowing of spin fluctuations in re substituted lsco@xcite finds that the effective spin stiffness @xmath202 ( or equivalently the effective exchange constant ) scales to zero at a comparable doping ; the inverted triangles in fig . [ fig:3 ] show @xmath203 . as might have been anticipated from sections ii , iv , the doping dependence of @xmath204 changes radically below @xmath205 . note that while the integrated neutron intensity scales approximately with the area fraction of charged stripes , @xmath206 scales to @xmath63460k as @xmath207 . this is only 1/4 of the actual spin stiffness , @xmath208 in the undoped afm . the change by nearly a factor of four is suggestive of a dimensional reduction ( lower coordination ) , but for an isolated straight spin ladder , a factor of two might have been expected . = 0.33 0.5 cm it is important to note the proximity of this termination of the phase separation regime to the vhs : the arrow in fig . [ fig:3]a shows the doping at which the pseudogap in the heat capacity@xcite closes , leaving an approximately logarithmic peak@xcite , while the @xmath61 indicates the point at which photoemission finds the vhs crossing the fermi level@xcite . termination of the stripe phase close to a vhs is an important prediction of our eps model of stripes . figure [ fig:3 ] has a bearing on the discussion given earlier of optimal doping in ybco , or equivalently of the doping @xmath16 at which the stripe phase terminates . setting optimal doping at @xmath209 for all cuprates puts the termination at doping @xmath210 . however , this value is clearly too low for lsco : when the stripes are gone , magnetic correlations should be weak , but there is evidence for long range magnetic order at @xmath211@xcite and @xmath212 , the latter coupled with a suppression of @xmath21@xcite . on the other hand , both facts are compatible with @xmath213 , fig . [ fig:3 ] . the termination of the pseudogap has also been interpreted in terms of a quantum critical point ( qcp)@xcite . at first glance , it looks straightforward to distinguish a model of eps , which is by nature first order , from a qcp , which is typically second order . however , the phases are actually quite similar . due to strong coulomb effects , the eps is restricted to nanoscopic scale , so that , e.g. , the superconducting critical temperature @xmath21 changes smoothly with doping . on the other hand , near a qcp with quenched disorder , long - time correlations can lead to effective spatial inhomogeneity@xcite . we suggested earlier@xcite that the peak and hump features seen in photoemission from bi2212 were associated with the charged and the afm stripes respectively . as such , the intensity of the peak should have the doping dependence predicted for charged stripes , with the intensity increasing from zero at half filling , approximately linearly with doping @xmath16 . this has now been verified experimentally@xcite . moreover , the maximum intensity of the spectral weight occurs at the same doping@xcite @xmath9 discussed above , where the stripe phase terminates . remarkably , the peak spectral weight closely tracks @xmath21 , suggesting that _ the superconducting pairs live on the charged stripes _ , as predicted by several models@xcite . consistent with this , a number of measures of the strength of superconductivity ( condensation energy , critical current ) are optimized at this same point@xcite where the charge stripe intensity is maximum and afm stripes vanish . the fact that @xmath21 itself is actually optimized at a slightly lower doping may be a hint that stripes can enhance the superconducting gap , as found above , fig . [ fig:20c ] . the sharp falloff of spectral weight at higher doping may be an indication for a _ second regime of phase separation _ in overdoped samples , where the new phase is not intrinsically superconducting . some additional evidence for such a phase separation is discussed below . it is important to note that as soon as the pseudogap closes , experiments find a peak in the density of states consistent with that expected for a vhs . this is discussed in ref . on p. 1203 21 ) ; more recent evidence is in ref . . for even higher dopings there are hints of a new regime of phase separation . this is an improtant prediction of the model : if stripes are stabilized by lowering the electronic energy at a vhs , then there could be a _ second range of phase separation _ in the overdoped regime , with the non - superconducting phase totally unrelated to the afm state at half filling . earlier experimental evidence for this second regime of phase separation in the cuprates has been presented in references ( section 11.6 ) and ( see also wen , et al@xcite ) . recently , loram@xcite has presented a detailed analysis of the heat capacity of la@xmath0sr@xmath1cuo@xmath2 over an extended doping range . in the range @xmath214 , the temperature dependence of the somerfeld constant @xmath215 is consistent with that expected for a uniform phase close to a vhs ( ref . , 21 ) . for lower doping , @xmath215 falls off too rapidly to be associated with rigid - band filling away from a vhs , but can be easily explained in terms of the opening of a pseudogap . for higher hole doping , @xmath215 again falls off@xcite too rapidly to be due to rigid band filling , although without an obvious pseudogap forming . the near symmetry of @xmath215 between the overdoped and underdoped sides is suggestive of two regimes of phase separation . further possible evidence for this second regime comes from tunneling studies@xcite which find a subdominant order parameter in optimally and over - doped ybco ( i.e. , in the same doping range where a split superconducting transition is found@xcite ) . this could arise from an allender - bray - bardeen mechanism@xcite , with the d - wave superconducting stripes inducing s - wave ( or @xmath216 ) superconductivity on adjacent normal - fermi - metal stripes . also , microwave measurements@xcite find a crossover near optimal doping from superconductor - insulator to superconductor - metal domains . uemura@xcite has also recently proposed phase separation in the overdoped cuprates . in extending the uemura plot to the overdoped regime , a ` boomerang ' effect is found : both @xmath21 and @xmath217 decrease in parallel@xcite . here @xmath218 is the superfluid density and @xmath24 is the effective mass , and their ratio is extracted from @xmath17sr measurements of the penetration depth . uemura@xcite suggests that this proportionality between @xmath21 and @xmath217 can most easily be accounted for in terms of ( nanoscale ) phase separation between the superconductor and a non - superconducting phase . thus for both overdoped and underdoped cuprates@xcite , the scaling between @xmath21 and @xmath23 can be understood in terms of phase separation . this has important implications for superconductivity : _ superconductivity arises only at a unique composition near optimal doping . _ the present results suggest that even at this doping , the superconductivity must compete with a second order parameter . above , we have presented evidence that the optimum @xmath21 occurs at a different doping in lsco than for the other cuprates , while the stripes terminate at approximately the same doping for all cuprates . further evidence for this can be found by comparing the superconducting fluctuations@xcite , fig . [ fig:38a ] . in all three materials , lsco , ybco , and bi2212 , some superconducting fluctuations appear to persist in the low doping regime , to temperatures in excess of optimal @xmath21 ( although considerably less than the pseudogap temperature ) . the fluctuating superconductivity in lsco is very similar to that found in the other cuprates . on the other hand , if one assumed a universal curve of superconductivity , with @xmath219 vs @xmath16 the same for all cuprates , then the scaled superconducting fluctuations in lsco would be at anomalously high temperatures . it seems simpler to assume that in lsco long range superconducting order is anomalously suppressed ( as also suggested by baskaran@xcite ) , in which case there is no reason for optimal @xmath21 to fall at the same doping . in bi2212 and ybco , the onset of superconducting fluctuations falls close to the ` strong pseudogap ' regime@xcite , and hence at temperatures substantially below the weak pseudogap temperature@xcite , making it unlikely that the weak pseudogap is associated with preformed pairs . in lsco , there is some confusion , since the pseudogap temperatures reported by batlogg , et al.@xcite are considerably higher than those found by ido , et al.@xcite . the ido @xmath220 s fall close to the fluctuation onset temperatures in fig . [ fig:38a ] . = 0.33 0.5 cm the fluctuations observed in fig . [ fig:38a ] are very suggestive of the theoretical results of fig . [ fig:20c ] : thus , quantum size effects enhance the average gap on a narrow stripe , which will lead to an enhanced mean field transition temperature in the underdoped regime . however , the coupling between stripes also weakens , leading to a loss of interstripe coherence , so the pairing is only evident as enhanced fluctuations.@xcite recent experiments have provided considerable evidence for the presence of stripes and eps in the cuprates , but there remain many questions of how universal these are , how they arise and vary with doping , and how they interact with superconductivity . we have here elaborated our earlier@xcite model of stripes driven by frustrated phase separation , in particular adducing evidence that the doping on the charged stripes is close to @xmath221 , and that when the average doping approaches this value eps terminates . moreover , near @xmath32 there is a crossover in stripe properties : for @xmath12 the charged stripes are quantum confined , for @xmath13 the afm stripes are so confined . this model can explain the 1/8 anomaly ( @xmath222 ) , the anomalous hall effect ( @xmath34)@xcite for @xmath54 ( charged stripes confined , hence one - dimensional ) , and the growing spin gap in ybco for @xmath49@xcite . on the important issue of the _ structure _ of a charged stripe , we have explored a number of possibilities without coming to any final conclusions . while there is evidence that superconductivity lives on the charged stripes , there also appears to be a second instability on these stripes , which stabilizes the stripe phase while competing with superconductivity . we have shown that a semiquantitative understanding can be achieved by looking at the properties of a single doped ladder , and we have discussed how a number of instabilities ( both cdw and cuperconducting ) vary with ladder width . we showed that strong correlation effects could lead to charged stripes with a residual magnetic order , introduced a simple model for white - scalapino - like stripes , and found novel superconducting and cdw instabilities associated with such stripes . we illustrated how stripe order would affect arpes spectra , both dispersions and fermi surface maps . future studies will apply the model to describing other properties of the cuprates . certain anomalous features of strong - correlation calculations may find an explanation in underlying phase separation . thus , the vanishing of the renormalized hopping parameter @xmath155 near half filling in slave boson calculations may reflect the vanishing of the charged stripes at half filling@xcite , while the frequently - observed pinning of the vhs near the fermi level@xcite is consistent with vhs - stabilized charged stripes . finally , the idea of a commensurability effect near 1/8 doping , leading to a coexistence of _ domains _ for @xmath49 , provides a simple explanation for a large variety of experimental findings , including the saturation of the yamada plot , the direct observation of domains in stm studies , and a variety of microwave anomalies . this may also lead to a resolution of the combined puzzle of magnetic neutron scattering incommensurability and the neutron resonance peak . a stripe model provides a natural explanation of the incommensurability for @xmath48 , including a stripe reorientation transition at the metal - superconductor transition near @xmath223 . however , a band picture@xcite ( with @xmath224 close to a vhs ) provides a superior model for the combined , frequency - dependent incommensurability _ cum _ resonance peak found near optimal doping in ybco and bi2212 . an eps crossover from stripes to domains near @xmath225 would provide a natural explanation of these phenomena . interestingly , the only calculation@xcite to approximately describe the doping dependence is based on a slave boson model , which as noted above tends to mimic eps behavior as @xmath207 . * acknowledgment : * these computations were carried out using the facilities of the advanced scientific computation center at northesatern university ( nu - ascc ) . their support is gratefully acknowledged . we had stimulating conversations with s. pan and j.c . ck s research was supported in part by nsf grant nsf-9711910 . publication 776 of the barnett institute . k. yamada , c.h . lee , k. kurahashi , j. wada , s. wakimoto , s. ueki , h. kimura , y. endoh , s. hosoya , g. shirane , r.j . birgeneau , m. greven , m.a . kastner , and y.j . kim , phys . b*57 * , 6165 ( 1998 ) . tranquada , b.j . sternlieb , j.d . axe , y. nakamura , and s. uchida , nature * 375 * , 561 ( 1995 ) ; j.m . tranquada , j.d . axe , n. ichikawa , a.r . moodenbaugh , y. nakamura , and s. uchida , phys . rev . lett * 78 * , 338 ( 1997 ) . mook , presented at the sixth international conference on the materials and mechanisms of superconductivity high - temperature superconductivity ( m@xmath226s - htsc - vi ) , houston , tx , feb . 20 - 25 , 2000 ; h.a . mook and f. doan , nature * 401 * , 145 ( 1999 ) , and cond - mat/0103037 . t. cren , d. roditchev , w. sacks , j. klein , j .- b . moussy , c. deville - cavellin , and m. lagus , phys . * 84 * , 147 ( 2000 ) ; c. howald , p. fournier , a. kapitulnik , cond - mat/0101251 ; s.h . pan , j.p . oneal , r.l . badzey , c. chamon , h. ding , j.r . engelbrecht , z. wang , h. eisaki , s. uchida , a.k . gupta , k .- w . hudson , k.m . lang , and j.c . davis , unpublished ; and j.c . davis , presented at the sixth internat . conf . on spectroscopies of novel superconductors ( sns01 ) , chicago , may 13 - 17 ( 2001 ) , unpublished . j. orenstein , presented at sns01 , unpublished . j.l . tallon , cond - mat/9911422 , j.l . tallon , j.w . loram , and g.v.m . williams , cond - mat/9911423 , j.l . tallon , j.w . loram , g.v.m . williams , j.r . cooper , i.r . fisher , j.d . johnson , m.p . staines , and c. bernhard , phys . b*215 * , 531 ( 1999 ) . salkola , v.j . emery , and s.a . kivelson , phys . * 77 * , 155 ( 1996 ) . s. caprara , c. di castro , m. grilli , a. perali , and m. sulpizi , physica c*317 - 8 * , 230 ( 1999 ) ; g. seibold , f. becca , f. bucci , c. castellani , c. di castro , and m. grilli , cond - mat/9906108 . markiewicz , phys . rev . b*62 * , 1252 ( 2000 ) . zacher , r. eder , e. arrigoni , and w. hanke , phys . rev . lett . * 85 * , 2585 ( 2000 ) . j. eroles , g. ortiz , a.v . balatsky , and a.r . bishop , cond - mat/0008341 . r.s . markiewicz and c. kusko , int . j. mod b*14 * , 3561 ( 2000 ) . white and d.j . scalapino , phys . lett . * 80 * , 1272 ( 1998 ) , _ ibid . _ * 81 * , 3227 ( 1998 ) . zimmermann , a. vigliante , t. niemller , n. ichikawa , t. frello , j. madsen , p. wochner , s. uchida , n.h . andersen , j.m . tranquada , d. gibbs , and j.r . schneider , europhys . lett . * 41 * , 629 ( 1998 ) ; t. niemoeller , n. ichikawa , t. frello , h. huennefeld , n.h . andersen , s. uchida , j.r . schneider , and j.m . tranquada , cond - mat/9904383 ; t. niemller , h. hnnefeld , t. frello , n.h . andersen , n. ichikawa , s. uchida , and j.r . schneider , j. low - t . phys . * 117 * , 455 ( 1999 ) . kojima , h. eisaki , s. uchida , y. fudamoto , i.m . gat , a. kinkhabwala , m.i . larkin , g.m . luke , and y.j . uemura , physica b*289 - 90 * , 343 ( 2000 ) . teitelbaum , b. bchner , and h. de gronkel , phys . * 84 * , 2949 ( 2000 ) . j. tworzydlo , c.n.a . van duin , and j. zaanen , proc . of 2nd intern . conf . on stripes and high tc superconductivity , rome 98 , to be published , j. supercond . ( cond - mat/9808034 ) . tallon , physica c*168 * , 85 ( 1990 ) . uemura , g.m . luke , b.j . sternlieb , j.h . brewer , j.f . carolan , w.n . hardy , r. kadano , j.r . kempton , r.f . kiefl , s.r . kreitzman , p. mulhern , t.m . riseman , d.ll . williams , b.x . yang , s. uchida , h. takagi , j. gopalakrishnan , a.w . sleight , m.a . subramanian , c.l . chien , m.z . cieplak , g. xiao , v.y . lee , b.w . statt , c.e . stronach , w.j . kossler , and x.h . yu , phys . lett . * 62 * , 2317 ( 1989 ) . optical evidence that @xmath16 and the superelectron density @xmath23 both scale with @xmath21 is found in h.l . liu , m.a . quijada , a. zibold , y .- d . yoon , d.b . tanner , g. cao , j.e . crow , h. berger , g. margaritondo , l. forr , beom - hoan o , j.t . markert , r.j . kelly , and m. onellion , j. phys . 11 * , 239 ( 1999 ) . consistent with @xmath25 = 0.21 [ y. tokura , j.b . torrance , t.c . huang , and a.i . nazzal , phys . rev . * b38 * , 7156 ( 1988 ) ] , 0.24 [ u. tutsch , p. schweiss , r. hauff , b. obst , th . wolf , and h. whl , j. low - t . phys . * 117 * , 951 ( 1999 ) ] , 0.2 [ m. merz , n. ncker , p. schweiss , s. schuppler , c.t . chen , v. chakarian , j. freeland , y.u . idzerta , m. klser , g. mller - vogt , and th . wolf , phys . lett . * 80 * , 5192 ( 1998 ) ] , 0.21 [ s.l . drechsler , j. malek , and h. eschrig , phys . rev . b*55 * , 606 ( 1997 ) ] . a recent critique of tallon s bond valence sum calculation@xcite also finds evidence for a larger @xmath25 , but concludes that the technique can not accurately determine the mobile hole fraction on the cuo@xmath3 planes [ o. chmaissem , y. eckstein , and c.g . kuper , phys . rev . b*63 * , 174510 ( 2001 ) . tallon , j.r . cooper , p.s.i.p.n . de silva , g.v.m . williams , and j.w . loram , phys . * 75 * , 4114 ( 1995 ) . jones , d.c . christen , j.r . thompson , r. feenstra , s. zhu , d.h . lowndes , j.m . phillips , m.p . siegal , and j.d . budai , phys . rev . b*47 * , 8986 ( 1993 ) . t. noda , h. eisaki , and s. uchida , science * 286 * , 265 ( 1999 ) . v. kataev , b. rameev , a. validov , b. bchner , m. hcker , and r. borowski , phys . rev . b*58 * , 11876 ( 1998 ) . s. sugai and n. hayamizu , j. phys . solids * 62 * , 177 ( 2001 ) . g. blumberg , m.v . klein , and s .- w . cheong , phys . lett . * 80 * , 564 ( 1998 ) ; k. yamamoto , t. katsufuji , t. tanabe , and y. tokura , phys . lett . * 80 * , 1493 ( 1998 ) ; and s. sugai , n. kitamori , s. hosoya , and k. yamada , j. phys . 67 * , 2992 ( 1998 ) . m. fujita , k. yamada , h. hiraka , p.m. gehring , s.h . lee , s. wakimoto , and g. shirane , cond - mat/0101320 . m. arai , t. nishijima , y. endoh , a.w . garrett , s. tajima , k. tomimoto , y. shiohara , c.d . frost , and s.m . bennington , cond - mat/9912233 ; m. arai , y. endoh , s. tajima , and s.m . bennington , int . b*14 * , 3312 ( 2000 ) . p. dai , h.a . mook , r.d . hunt , and f. dogan , phys . rev . b*63 * , 54525 ( 2001 ) . a. ino , t. mizokawa , a. fujimori , k. tamasaku , h. eisaki , s. uchida , t. kimura , t. sasagawa , and k. kishio , phys . lett . * 79 * , 2101 ( 1997 ) . m. satake , k. kobayashi , t. mizokawa , a. fujimori , t. tanabe , t. katsufuji , and y. tokura , phys . * , 15515 ( 2000 ) . balatsky , presented at sns01 , unpublished . a. erb , a.a . manuel , m. dhalle , f. marti , j .- y . genoud , b. revaz , a. junod , d. vasumathi , s. ishibashi , a. shukla , e. walker , o. fischer , r. fluekiger , r. pozzi , m. mali , and d. brinkmann , sol . * 112 * , 245 ( 1999 ) . o. fischer , presented at sns01 , unpublished . hammel , a.p . reyes , s - w . cheong , z. fisk , and j.e . schirber , phys . lett . * 71 * , 440 ( 1993 ) . c. kusko , z. zhai , n. hakim , r.s . markiewicz , s. sridhar , d. colson , viallet - guillet , a. forget , yu.a . nefyodov , m.r . trunin , n.n . kolesnikov , a. maignan , a. daignere , and a. erb , unpublished ; r.s . markiewicz , unpublished . x.j . zhou , t. yoshida , s.a . kellar , p.v . bogdanov , e.d . lu , a. lanzara , m. nakamura , t. noda , t. takeshita , h. eisaki , s. uchida , a. fujimori , z. hussain , and z .- x . shen , phys . * 86 * , 5578 ( 2001 ) . j. zaanen and a.m. oles , ann . physik * 5 * , 224 ( 1996 ) ; c. nayak and f. wilczek , phys . rev . lett . * 78 * , 2465 ( 1997 ) ; v.j . emery , s.a . kivelson , and o. zachar , phys . rev . b*56 * , 6120 ( 1997 ) ; s.a . kivelson , e. fradkin , and v.j . emery , nature * 393 * , 550 ( 1998 ) . m. ichioka and k. machida , j. phys . . jpn . * 68 * , 4020 ( 1999 ) . b. valenzuela , m. vozmediano , and f. guinea , phys . rev . b*62 * , 11312 ( 2000 ) , and m. vozmediano , personal communication . s. chakravarty , r.b . laughlin , d.k . morr , and c. nayak , phys . rev . b*63 * , 94503 ( 2001 ) ; j. zaanen , cond - mat/0103255 . kaplan and b.g . vekhter , cooperative phenomena in jahn - teller crystals " ( plenum , n.y . , 1995 ) , p. 216 . j. hafner , `` from hamiltonians to phase diagrams '' ( springer , berlin , 1987 ) . anderson , _ `` van hove singularities as a source of anomalies in the cuprates '' _ , pup preprint 414 ( 1994 ) , and `` the theory of high - t@xmath227 superconductivity '' , ( princeton , university press , 1997 ) . yi , z .- g . bishop , and j.t . gammel , phys . rev . b*58 * , 503 ( 1998 ) . t. tohyama , s. nagai , y. shibata , and s. maekawa , phys . lett . * 82 * , 4910 ( 1999 ) . markiewicz , phys . rev . b*56 * , 9091 ( 1997 ) . p. fazekas , foundations of physics * 30 * , 1999 ( 2000 ) . c. kusko and r.s . markiewicz , unpublished , cond - mat/0102440 . r. hlubina , s. sorella , and f. guinea , phys . lett . * 78 * , 1343 ( 1997 ) ; r. hlubina , phys . rev . b*59 * , 9600 ( 1999 ) . see also d. vollhardt , n. blmer , k. held , m. kollar , j. schlipf , and m. ulmke , z. phys . b*103 * , 283 ( 1997 ) . markiewicz , j. phys . sol . * 58 * , 1179 ( 1997 ) . j. fritzenktter and k. dichtel , in `` stripes and related phenomena '' , ed . by a. bianconi and n. saini , ( kluwer academic / plenum , n.y . , 2000 ) , p. 369 . visscher , phys . rev . b*10 * , 943 ( 1974 ) ; e.l . nagaev , physics of magnetic semiconductors " ( moscow , mir , 1983 ) . g. su , phys . rev . b*54 * , 8281 ( 1996 ) ; a.c . cosentini , m. capone , l. guidoni , and g.b . bachelet , phys . rev . b*58 * , 14685 ( 1998 ) . putikka , r.l . glenister , and r.r.p . singh , phys . lett . * 73 * , 170 ( 1994 ) ; c.t . shih , y.c . chen , and t.k . lee , phys . rev . b*57 * , 627 ( 1998 ) ; and m. calandra , f. becca , and s. sorella , phys . lett . * 81 * , 5185 ( 1998 ) . c.s . hellberg and e. manousakis , ( a ) lett . * 78 * , 4609 ( 1997 ) , ( b ) phys . rev . b*61 * , 11787 ( 2000 ) . l. p. pryadko , s. kivelson and d. w. hone , phys . lett , * 80 * , 5651 ( 1998 ) . h. schulz , j. phys . ( paris ) * 50 * , 2833 ( 1989 ) ; d. poilblanc and t.m . rice , phys . rev . b*39 * , 9749 ( 1989 ) ; j. zaanen and o. gunnarsson , _ ibid . _ b*40 * , 7391 ( 1989 ) ; m. inui and p.b . littlewood , _ ibid . _ b * 44 * , 4415 ( 1991 ) ; a. singh and z. teanovi , _ ibid . _ b*41 * , 614 ( 1990 ) ; k. yonemitsu , a.r . bishop , and j. lorenzana , phys . rev . b*47 * , 8065 , 12059 ( 1993 ) ; and d. gra , k. rociszewski , and a.m. ole , phys . rev . b*60 * , 7429 ( 1999 ) . vergs , e. louis , m.p . lpez - sancho , f. guinea , and a.r . bishop , phys . rev . b*43 * , 6099 ( 1991 ) . m. kato , k. machida , h. nakanishi , and m. fujita , j. phys . . jpn . * 59 * , 1047 ( 1990 ) . laughlin , cond - mat/9709195 ; j. zaanen , j. phys . solids * 59 * , 1769 ( 1998 ) . markiewicz , c. kusko and v. kidambi , phys . b.*60 * , 627 ( 1999 ) . c. balseiro and l. falicov , phys . rev . b*20 * , 4457 ( 1979 ) . b. valenzuela and m.a.h . vozmediano , phys . b*63 * , 153103 ( 2000 ) . markiewicz and c. kusko , unpublished , cond - mat/0102452 . agterberg , v. barzykin , and l.p . gorkov , europhys . lett . * 48 * , 449 ( 1999 ) and phys . rev . b*60 * , 14868 ( 2000 ) . g. baskaran , cond - mat/0007137 . markiewicz , unpublished . feng , n.p . armitage , k.m . shen , a. damascelli , c. kim , f. ronning , z .- x . shen , d.a . bonn , r. liang , w.n . hardy , a.i . rykov , and s. tajima , phys . lett . * 86 * , 4370 ( 2001 ) . zhou , p. bogdanov , s.a . kellar , t. noda , h. eisaki , s. uchida , z. hussain , and z .- x . shen , science * 286 * , 268 ( 1999 ) . d. orgad , s.a . kivelson , e.w . carlson , v.j . emery , x.j . zhou , and z .- x . shen , phys . lett . * 86 * , 4362 ( 2001 ) . m. lindroos and a. bansil , personal communication . a. kampf and j. schrieffer , phys . rev . b*41 * , 6399 ( 1990 ) . s .- c . zhang , science * 275 * , 1089 ( 1997 ) . klemm , unpublished , and physica c*341 - 8 * , 839 ( 2000 ) . c. castellani , c. di castro , and m. grilli , phys . lett . * 75 * , 4650 ( 1995 ) . markiewicz , physica c*169 * , 63 ( 1990 ) . t. katsufuji , t. tanabe , t. ishikawa , y. fukuda , t. arima , and y. tokura , phys . rev . b*54 * , 14230 ( 1996 ) . p. bourges , cond - mat/0009373 . p. bourges , l.p . regnault , j.y . henry , c. vettier , y. sidis , and p. burlet , physica b*215 * , 30 ( 1995 ) . j. brinckmann and p.a . lee , phys . * 82 * , 2915 ( 1999 ) . norman , cond - mat/9912203 ; y .- j . kao , q. si , and k. levin , phys . rev . b*61 * , 11898 ( 2000 ) . boin , g.h . kwei , h. takagi , and s.j.l . billinge , phys . lett . * 84 * , 5856 ( 2000 ) . o. baberski , a. lang , o. maldonado , m. hcker , b. bchner , and a. freimuth , europhys . lett . * 44 * , 335 ( 1998 ) . t. timusk , a.v . puchkov , s. doyle , a.m. hermann , and n.n . kolesnikov , unpublished . y. tokunaga , k. ishida , k. yoshida , t. mito , y. kitaoka , y. nakayama , j. shimoyama , k. kishio , o. narikiyo , and k. miyake , physica b * 284 - 88 * , 663 ( 2000 ) . a.w . hunt , p.m. singer , a.f . cederstrm , and t. imai , cond - mat/0011380 . loram , j.l . luo , j.r . cooper , w.y . liang , and j.l . tallon , physica c*341 - 8 * , 831 ( 2000 ) . shen , presented at m@xmath226s - htsc - vi , houston , tx , feb . 20 - 25 , 2000 ; d.l . feng , d.h . shen , c. kim , h. eisaki , a. damascelli , r. yoshizaki , j .- shimoyama , k. kishio , g. gu , s. oh , a. andrus , j. odonnell , j.n . eckstein , and z .- x . shen , science * 280 * , 277 ( 2000 ) . y. koike , m. akoshima , m. aoyama , k. nishimaki , t. kawamata , t. adachi , t. noji , i. watanabe , s. ohira , w. higemoto , and k. nagamine , int . j. mod b*14 * , 3520 ( 2000 ) . c. sire , c.m . varma , a.e . ruckenstein , and t. giamarchi , phys . * 72 * , 2478 ( 1994 ) . d. belitz and t.r . kirkpatrick , cond - mat/0106279 . h. ding , j.r . engelbrecht , z. wang , j.c . campuzano , s .- c . wang , h .- b . yang , r. rogan , t. takahashi , k. kadowaki , and d.g . hinks , cond - mat/0006143 . i. martin , g. ortiz , a.v . balatsky , and a.r . bishop , int . phys , b*14 * , 3567 ( 2000 ) . c. bernhard , j.l . tallon , th . blasius , a. golnik , and ch . niedermayer , phys . * 86 * , 1614 ( 2001 ) . tallon , j.w . loram , and g.v.m . williams , cond - mat/0008295 ; however , note j. bobroff , w.a . macfarlane , h. alloul , and p. mendels , cond - mat/0008396 . j.b . goodenough and j .- s . zhou , phys . rev . b*49 * , 4251 ( 1994 ) . wen , w.l . yang , and z.x . zhao , physica c*341 - 8 * , 1735 ( 2000 ) , and h.h . wen , x.h . chen , w.l . yang , and z.x . zhao , phys . lett . * 85 * , 2805 ( 2000 ) . cooper and j.w . loram , j. phys . i france * 6 * 2237 ( 1996 ) . r. krupke and g. deutscher , phys . lett . * 83 * , 4634 ( 1999 ) ; a. sharoni , g. koren , and o. millo , cond - mat/0103581 . s. rusiecki , e. kaldis , e. jilek , and c. rossel , j. less comm . metals * 164 * , 31 ( 1990 ) . d. allender , j. bray , and j. bardeen , phys . rev . b*7 * , 1020 ( 1973 ) . uemura , physica c*341 - 8 * , 2117 ( 2000 ) . uemura , a. keren , l.p . luke , w.d . wu , y. kubo , t. manako , y. shimakawa , m. subramanian , j.l . cobb , and j.t . markert , nature * 364 * , 605 ( 1993 ) . markiewicz and b.c . giessen , physica c*160 * , 497 ( 1989 ) . j. corson , r. mallozzi , j. orenstein , j.n . eckstein , and i. bozovic , nature * 398 * , 221 ( 1999 ) . ong , y. wang , t. kakeshita , and s. uchida , nature * 406 * , 486 ( 2000 ) . c. bernhard , d. munzar , a. golnik , c.t . lin , a. wittlin , j. humlek , and m. cardona , phys . rev . b*61 * , 618 ( 2000 ) . c. meingast , v. pasler , p. nagel , a. rykov , s. tajima , and p. olsson , phys . lett . * 86 * , 1606 ( 2001 ) . g. baskaran , mod . b*14 * , 377 ( 2000 ) . b. batlogg and v.j . enery , nature * 382 * , 20 ( 1996 ) . v. barzykin and d. pines , phys . b*52*,13585 ( 1995 ) ; j. schmalian , d. pines , and b. stojkovic , phys . lett . * 80 * , 3839 ( 1998 ) . b. batlogg , h.y . hwang , h. takagi , r.j . cava , h.l . kao , and j. kwo , physica c*235 - 240 * , 130 ( 1994 ) ; j.l . tallon , j.r . cooper , p.s.i.p.n . de silva , g.v.m . williams , and j.w . loram , phys . lett . * 75 * , 4114 ( 1995 ) . m. ido , n. momono , and m. oda , j. low . * 117 * , 329 ( 1999 ) . v.j . emery and s.a . kivelson , nature * 374 * , 6521 ( 1995 ) . markiewicz , j. phys . * 2 * , 665 ( 1990 ) ; a. himeda and m. ogata , phys . lett . * 85 * , 4345 ( 2000 ) .

nervous system is continuously confronted by megabytes of information , representing light , sound , smell , etc . this information is compiled by the brain into a set of decisions , representing behaviors of living organisms . the mechanisms involved in this reduction have been under investigation for many years ( glimcher , 2003 ; romo and salinas , 2003 ) . in this study we address a question complimentary to the issue of decision making ( dm ) mechanisms . we define neuronal units involved in making perceptual decisions . for this purpose we determine dm activity in surrogate networks , defined mathematically , in which a complete control is present over stimuli , mechanisms , and responses . such decision making analysis ( dma ) has practical significance , since once units involved in making particular decision are located , further efforts could be concentrated on uncovering the underlying mechanisms . in this study dm task is defined as evaluation of a function in the multidimensional stimulus space ( figure 1a ) . this function has a discrete set of values , representing the repertoire of responses available to the organism . the decisions may , of course , be stochastic , to reflect the uncertainty , pertinent to behavior . this definition is suitable for experiments where subjects perform poly - alternative forced - choice tasks , such as saccadic response to the direction of stimulus motion ( shadlen and newsome , 2001 ) . let us consider motion - discrimination task in more detail . figure 1b lists some visual areas , which are involved in this task . the areas are arranged along a rough sensory - motor axis , so that the areas on the left are more `` sensory '' , while those on the right are more `` motor '' . this implies that the responses in these areas are more correlated with stimulus or response respectively . where on this sensory - motor axis one should position the dm elements ? one could argue that the elements most correlated with the decision itself are the decision makers , following the analogy with the definition of sensory and motor elements . it is , however , difficult , if not impossible , to distinguish such definition from the definition of purely motor units ( shadlen and newsome , 2001 ) . the latter relay the results of decision making process , without involvement in the formation of the decision . an alternative approach is therefore needed to define the dm units . the dm components may be surmised to be located on the interface between sensory and motor areas . more precisely , the _ first _ element in the sensory - motor chain , which carries significant correlation with the response , may be identified as the decision maker . in this study we develop this idea into rigorous mathematical formulation and find a special correlation function , which determines contributions of units to dm . this formalism allows us to answer two questions pertaining to the identities of dm units . first , we consider the case when not one but _ several _ elements are involved in the same decision simultaneously . our approach allows us to evaluate relative importance of various units in such a distributed dm . second , we consider the systems with loops in connectivity . for such systems the concept of ` the first element ' becomes more arbitrary and one has to proceed more carefully in defining contributions to dm . we succeed in doing so for our surrogate networks and define dm units for recurrent networks in a way , which is consistent with the linear sensory - motor chains , thus satisfying the requirement of the correspondence principle . this paper is organized as follows . we first analyze simple linear chain models , and networks , such as trees , which have similar properties . we then use this analysis to define decision makers in networks of arbitrary connectivity . finally , we extend our study to the cases , when electric stimulation can be applied to units , and show that dm components can be identified in a way consistent with our preceeding analyses . the goal of this section is to formulate quantitative principles by which dm network elements can be identified . we approach this task by analyzing simple cases , which can be solved exactly without the use of computer , and in which the identities of dm elements are clear . these cases allow us to emphasize the properties of dm task we are attempting to describe . we proceed therefore to the analysis of the simplest network capable of making decisions . in this subsection we consider the network , which we call ` nematode ' , because of its resemblance to simple biological organisms , both in the layout and in the fundamental significance . we first define the model ; then show that it can make simple decisions ; and , finally , define the positions of decision makers in the network . consider a linear chain of units , whose response is characterized by a set of real numbers @xmath0 , where @xmath1 is the position of the element in the chain ( figure 2 ) . response of each element does not depend on time . this model is therefore static . this assumption is introduced here to simplify the analysis and can be relaxed as described below ( section [ trees ] ) . each unit performs a simple linear transformation between the unit s input and the output . thus , for element number @xmath2 @xmath3 here @xmath4 is noise associated with the element . in this work we assume that noise has zero mean , is individual to each unit , and , therefore , is uncorrelated between units , i.e. @xmath5 we further assume that noise has a gaussian distribution . the chain of linear elements is thus completely specified by a set of noise variances @xmath6 . the model described by ( [ eq1 ] ) and ( [ eq2 ] ) yields the following solution for the response of the last element in the chain @xmath7 thus , the response of the last element is just a sum of the input into network @xmath8 and noise contributions from all units , independently on the order of unit in the chain . the last element in the chain has non - linear response properties . its response is defined by @xmath9 where @xmath10 is the heaviside step function , which is equal to one / zero if the argument is positive / negative . it follows then that our ` nematode ' network is capable of making decisions based on the values of input variable @xmath8 . this is if we interpret variable @xmath11 , which is equal either 0 or 1 , as the result of dm process , as defined in figure 1a . the decisions are made stochastically and are dependent upon the instantiations of random variables @xmath4 , which vary from trial to trial . our model is completely defined by the set of noise variances , pertinent to each unit @xmath12 . although decisions made by this chain are quite simple , the identities of decision makers are not so easy to find . the distribution of impact to dm along the chain should depend upon the distribution of noise variables @xmath6 . our next goal is to develop a sensible definition of contributions to dm based on the vector of variances @xmath6 . before doing so we describe general input - output properties of the chain . since decision made by the network varies from trial to trial , one can define averaged over trials response of the system @xmath13 . as shown in figure 2b it has a sigmoid shape , smeared by the total amount of noise in the system . one can , therefore , consider two cases , depending on whether the signal - to - noise ratio for the chain is large or small . these two regimes are shown in figure 3a and b respectively . to analyze responses of units in these two cases we define their correlation with the decision . this correlation is defined for each element in the chain ( figure 3 , right ) . as a measure of correlation we choose mutual information ( mi ) between response of the @xmath14th unit , @xmath15 , and the decision , @xmath11 . mi has an advantage of being unitless ( it is measured in bits ) and having clear intuitive properties , as described below . we will also show below in this section that mi has limitations as a measure of dm . mi describes the information transmission from the @xmath14th unit to the output of the system . since the output can only have values 0 or 1 , mi can not exceed the value of one bit . we now consider two cases , depending on the network s signal - to - noise ratio . if network input @xmath16 is large , as in figure 3a , response of the system is well correlated with the input . hence , activities of all units are well correlated with both input and output , and @xmath17 for all of the units . in the opposite limit , when the signal - to - noise ratio is small , @xmath16 is smaller than noise , and the system s response is weakly correlated with the input ( figure 3b ) . in this case mi as a function of unit s position displays a structure , shown in figure 3b ( right ) . this structure , as shown below , has a key to the definition of dm components and is qualitatively discussed here . the units , which are close to the exit from the network , show strong correlation with the decision , similarly to the high signal - to - noise ratio case . their mi is therefore close to 1 bit . on the other hand , more ` sensory ' units , in the beginning of the chain are strongly correlated with the input . since input - output correlation is weak in low signal - to - noise ratio case , the ` sensory ' units display virtually _ no _ relation to the output and @xmath18 for such units ( figure 3b , right ) . thus , mi , as a function of @xmath2 displays a transition from 0 to 1 in the low signal - to - noise ratio case . how could one deduce identities of decision makers from these dependencies ( figure 3a and b ) ? one could suggest that the elements perfectly correlated with the output of the system , such as exit elements from the chain , are the ones that make the decision . however , such elements may be just the relay or ` motor ' units , in which case their contribution to dm is small . indeed , when we type , our decisions are perfectly correlated with activities of finger muscles ; but one could hardly blame our fingers for the content of the typing . thus , despite their high correlation with the output , exit elements could not be called decision makers . input elements , having no correlation with the decision , are responsible for dm in even lesser degree . we thus need to analyze the dependence of mi on position in more detail and suggest another scheme for defining dm units . our discarding of motor units as decision makers can be further extended onto the entire high signal - to - noise ratio case ( figure 3a ) . we suggest that the deterministic regime is not descriptive from the point of view of dm analysis . first , in this regime all units become indistinguishable from motor . the latter are not decision makers , as suggested above . second , the dependence shown in figure 3a ( right ) does not reveal the contributions of individual units to the decision . since all units have the same correlation , it is hard , if not impossible , to differentiate them and assign different contributions . third , the responses of units in this case are deterministically related to the input . hence , units act as relays , passively transmitting information along the chain . it can be argued that the external environment , providing the input variable @xmath8 , acts as the decision maker . we conclude that to find decision making activity one has to concentrate on the low signal - to - noise ratio case . we show below that the identities of decision making units can be deduced from the shape of transition in figure 3b ( right ) . to this end we analyze a set of examples of networks with various distributions of noise @xmath19 . we start from the simplest example of a single noisy unit . consider a chain in which noise is absent from all units but one , whose order number in the chain is @xmath20 ( figure 4 ) . since , according to our previous discussion , we need to consider the low signal - to - noise ratio case , we will assume that @xmath21 i.e. network receives no input . making the decision in this case is still possible , based on the values of noise inside the network . since noise is only present in one neuron , from ( [ eq3 ] ) we conclude that @xmath22 the decision made by the network is @xmath23 thus , decision is causally linked to the processes controlling unit number @xmath20 , which leads us to conclusion that this neuron is the decision maker . paradoxically , the noisiest unit in this simple formulation makes the largest impact . all noiseless elements , even nonlinear , are deterministic , and work as simple relays which transmit information from the previous node to the next one . the output of the circuit is linked to the processes controlling noise in neuron number @xmath20 , rather that in any other neuron in the network . one would be tempted to conclude that the non - linear element is actually the decision maker in this case . we deduce that the non - linear element does not have a causal effect on output from the circuit ; therefore its role is just to relay response from neuron @xmath20 to the output . in this respect the non - linear element is not different from other noiseless elements . to link this example to our previous discussion ( figure 3b ) we plot mi as a function of position in the chain in figure 4 ( top ) . as we discussed , mi is high for exit ( ` motor ' ) units and low for input ( ` sensory ' ) elements . figure 4 also shows the derivative of mi with respect to position in the chain . it is clear that this derivative represents the decision making element . thus , we conclude that not correlation with the decision but the _ rate of change _ of the latter along the network is the indicator of dm . our next example shows that the conclusion about derivative of mi is basically correct , but has to be slightly amended to be numerically precise . consider the chain in which all elements are noisy and the variance of noise is the same for each element . in this case @xmath24 i.e. all units contribute to decision _ equally_. this is because eq . ( [ eq8 ] ) does not distinguish the order in which contributions from the units are added , and all contributions are of equal strength on average . can this conclusion be confirmed by the derivative of mi ? figure 5a shows mi as a function of position in the chain for this case . this dependence is obtained in appendix a. it increases smoothly from 0 to 1 resulting in a non - zero derivative at all units . this is consistent with ( [ eq8 ] ) and the notion that all units participate in the decision . however , ( [ eq8 ] ) suggests that all units participate in decision _ equally_. the derivative of mi turns out to be slightly non - uniform , as seen in figure 5a . this can be corrected if not mi itself but a non - linear function of mi , denoted @xmath25 , is considered . this non - linear function is calculated in appendix a and is shown in figure 5b . the new correlator @xmath25 has the same basic properties as the mi . it rises from 0 to 1 monotonously when passing through the array ( figure 5c ) . but , in addition , its derivative turns out to be _ uniform _ , as shown in figure 5c ( bottom ) . this is consistent with equal participation of all units in dm in the uniformly distributed noise case and eq . ( [ eq8 ] ) . thus , we conclude that for this case the contributions to dm are given by the rate of increase of @xmath25 when moving through the array @xmath26 here @xmath2 is the index along the chain . ( [ eq9 ] ) is the main result of this paper . it represents our definition of contributions to dm for networks of simple connectivity , such as chains . three points should be made about the definition ( [ eq9 ] ) . first , it reproduces the result obtained in the previous example of noisy neuron . indeed , the mutual information rises from 0 to 1 on the noisy neuron in figure 4 . but @xmath25 coincides with mi at these values , as follows from its plot in figure 5b . thus , the derivative of @xmath25 is also given by a single spike at the position of noisy neuron , as in figure 4 ( bottom ) . second , eq . ( [ eq9 ] ) implies that , from point of view of dm , not mutual information , but another correlator , given by @xmath25 , is more relevant . function @xmath27 deviates from linear function only slightly ( figure 5b ) , and for practical purposes the distinction between the mi and @xmath25 could be ignored . however , we retain it throughout the manuscript to ensure mathematical rigor . third , when deducing ( [ eq9 ] ) we did not postulate that contributions to dm are proportional to the variance of noise . instead , we suggested that eq . ( [ eq8 ] ) implies that all units contribute equally , independently on the order in the chain . this simple qualitative statement is powerful enough to constrain our quantitative reasoning and lead to a measure of dm in form of function @xmath25 and definition ( [ eq9 ] ) . we do not know yet if the derivative of @xmath25 is proportional to the variance of noise , square root of this variance , or any other characteristic of noise in each element . all of these parameters give the same results in the uniform noise case . we need to have a difference between units to measure relative strength of their contributions . this is achieved by the next example . , is close to linear , shown by the dotted line . * c * , the new correlator @xmath25 ( top panel ) has a uniform derivative ( bottom panel ) . thus , derivative of @xmath25 is a sensible measure of decision making in the case of uniform noise . , width=288 ] , is close to linear , shown by the dotted line . * c * , the new correlator @xmath25 ( top panel ) has a uniform derivative ( bottom panel ) . thus , derivative of @xmath25 is a sensible measure of decision making in the case of uniform noise . , width=288 ] in this example the variances of noise on all neurons are the same , similarly to the previous case . however , here we amend the network definition given by ( [ eq1 ] ) . we do so for only one neuron . we assume that the link between units 5 and 6 is characterized by a very large strength @xmath28 . thus , for neuron number 6 ( figure 6 ) instead of ( [ eq1 ] ) we have @xmath29 therefore this example is the same as the previous , except that the single network connection is changed . what are the dm units in this case ? the network s output is given by @xmath30 thus , units 1 through 5 contribute equally to decision . in addition , their contributions are multiplied by a large factor @xmath31 . units 6 through 11 also contribute equally , but their contribution is much smaller than that of the former group . we conclude that units 1 through 5 are much stronger decision makers than units 6 through 11 . this conclusion is supported by the derivative of @xmath25 , as shown in figure 6 ( bottom ) . thus , changing one link in the chain produces large effect on the distribution of dm . the units downstream from the link contribute less to decisions , while the units upstream contribute a lot . what is the measure of decision making , which could differentiate these two types of units ? calculations in appendix a show that derivative of @xmath25 is proportional to @xmath32 for units 1 through 5 . this is easy to understand qualitatively , since mi increases along the chain even for negative ( @xmath33 links . this is _ not _ possible if contribution from units 1 to 5 are multiplied by @xmath31 for example . thus , an even power of @xmath31 is required , which is shown in appendix a to be @xmath32 . so far we have used definition ( [ eq9 ] ) , which is quite complex , since it involves calculation of a nonlinear function @xmath25 . is it possible to reproduce the results derived above in a simpler way ? it turns out that the role of given unit in dm is proportional to its contribution to the variability of the output @xmath34 . this leads us to an alternative to ( [ eq9 ] ) definition of dm . let us introduce the new definition using the examples , considered above . from ( [ eq11 ] ) in the ` loud ' neuron case we derive @xmath35 we could conjecture that the contributions to dm from different units are weighted proportionally to the corresponding summands in ( [ eq12 ] ) . indeed , if we assume @xmath36 @xmath37 by choosing appropriate values of variance of noise and gain , we can reproduce the results of all three of our previous examples . thus , in the case of noisy neuron the variance of noise is only present in one unit , rendering this unit decision maker , according to ( [ eq13 ] ) . in the case of uniform noise , when @xmath38 and all @xmath39 are the same , ( [ eq13 ] ) gives uniform contributions to dm . in the case of ` loud ' neuron , ( [ eq13 ] ) gives the correct factor @xmath32 describing the advantage of upstream neurons . thus , the contributions to dm are proportional to the variance of noise on given element , multiplied by the square of the gain from this element to the output . we can rewrite ( [ eq13 ] ) in a more compact form to emphasize this latter statement @xmath40 one could verify ( [ eq14 ] ) , by applying it to ( [ eq12 ] ) and obtaining relationships ( [ eq13 ] ) . this justifies ( [ eq14 ] ) in the three examples considered above . ( [ eq14 ] ) also applies to linear chains in general . in appendix a we derive ( [ eq14 ] ) from previous definition ( [ eq9 ] ) for arbitrary distribution of connection strengths and noise . thus , ( [ eq14 ] ) can be considered an alternative definition to ( [ eq9 ] ) . the equivalence between ( [ eq9 ] ) and ( [ eq14 ] ) is demonstrated graphically in figure 7 . why should one consider an alternative definition ? this is because ( [ eq9 ] ) can not be applied to networks of arbitrary connectivity , such as circuits containing loops . definition ( [ eq14 ] ) however applies to all topologies , including the linear chain examples , considered here . the bottom panel displays the derivative of @xmath25 , defined by ( [ eq9 ] ) . the derivative is numerically the same as the variance of noise . both can be used as measures of decision making . , width=288 ] let us review our findings . first , we arrived to the definition of dm activity using the information - theoretical approach ( [ eq9 ] ) . according to this definition , dm is the rate of change of correlation with decision along the chain . in other words , the _ first _ element or elements , which correlate with the decision , are the decision makers . this approach has its pros and contras . indeed , the viewpoint expressed by ( [ eq9 ] ) has a potential to be transferred to other systems , which contain non - linear elements . ( [ eq9 ] ) has an information - theoretical origin ; hence its applicability may be broader than our simple system . another advantage of ( [ eq9 ] ) is that it relies on the characteristics measurable in single - electrode recording experiments , such as response of single unit and its correlation with behavioral decision . thus , ( [ eq9 ] ) could be used experimentally . the disadvantage of the information - theoretical approach is that it is not clear how to apply it to the systems with loops , as we have mentioned above . since biological networks almost always contain loops this significantly limits the applicability of information - theoretical formula ( [ eq9 ] ) . our second step was to derive an alternative definition ( [ eq14 ] ) . the latter is _ equivalent _ to the former definition ( [ eq9 ] ) for linear - chain ( ` nematode ' ) example , as we have demonstrated on simple examples and have shown more rigorously in appendix a. the alternative definition ( [ eq14 ] ) can be understood on the basis of the following two observations . first , the example of ` noisy ' neuron shows that the variability is the source of decisions . thus , under fixed other conditions , an increase in variability and noise in a single unit leads to a larger contribution to dm from this unit . @xmath41 second , the example of ` loud ' neuron shows that not only variability and noise are important but also how much of this variability reaches the motor units . dm is hence a property of network connectivity too . thus , we arrive to the next rule the stronger is the pathway from given unit to the motor output , the larger is the contribution of this unit to dm . @xmath42 these two rules are combined into the definition ( [ eq14 ] ) . although ( [ eq14 ] ) and ( [ eq16 ] ) assume that the output element is unique , this requirement will be removed below , when we consider arbitrary topology networks . what are the features of ( [ eq14 ] ) ? it could be used for an arbitrary topology network , since it does not contain derivative along the chain , as ( [ eq9 ] ) does . definition ( [ eq14 ] ) can also be used operationally to measure the contribution of each neuron to the decision experimentally . to do that one needs to vary noise at the given unit and measure the variability of the responses . the details are discussed in section [ stimulation ] below . a special note should be made about normalization in ( [ eq14 ] ) . throughout this work we adopt the convention that dm contributions are evaluated for all units and then normalized proportionally to ( [ eq14 ] ) , so that the total sum of all contributions is equal to one ( or 100% ) . we will assume this to hold below without explicitly mentioning . finally , we give another definition of dm contributions , which could be useful when noise in the system is the same for all units . in this case the only difference between units is due to difference in their position in the network . we therefore call such quantity _ topological _ dm . @xmath43 as seen from e.g. ( [ eq12 ] ) it does not depend on the levels of noise , and can be obtained from ( [ eq14 ] ) by assuming that @xmath44 for all units . it therefore describes how strongly each elements of the circuit affects the output . this quantity is sometimes helpful in describing the network s topology . lastly , we discuss the notion of noise and variability in our approach . is this really noise , which leads networks to decisions ? not necessarily . imagine that we have studied a chain - like network ( figure 8a ) and performed the dm analysis , described above . we found that the network contains two decision makers , which are equally important . a more thorough investigation may suggest that these units are inputs from external network , which in effect is responsible for dm . for example , these hidden pathways may be inputs from other sensory modalities or regulatory inputs of other type . thus , dma may help identify entry points from other , less studied , parts of the network . our studies indicate that information - theoretical analysis [ definition ( [ eq9 ] ) ] can be further extended to tree - like topologies ( figure 9 ) . to this end we define column - vector @xmath45 such that @xmath46 . then ( [ eq9 ] ) is equivalent to @xmath47 here @xmath48 is the structure matrix defined as follows . an element @xmath49 of the structure matrix is equal to 1 if there is a connection from unit number @xmath2 to @xmath50 . matrix @xmath51 thus implements evaluating differences between connected elements in ( [ eq9 ] ) . structure matrix is related to connectivity matrix , containing network s weights through @xmath52 . connectivity matrices for some networks are shown in figures 9 and 10 . information - theoretical approach can be even further extended on the cases , when signals propagate along the network in time , therefore resulting in delays between signal and response . in this case by @xmath53 one should understand a sum of correlations over all times preceding the decision . this compensates for the presence of delays . so far there is no understanding if eq . ( [ eq18 ] ) [ or ( [ eq9 ] ) ] can be used for topologies other than trees . definition ( [ eq14 ] ) , however , can be used with networks of arbitrary connectivity . this is the topic of the next section . all previous examples , except the one mentioned at the end of the last session , were static , i.e. variables did not depend on time . the deficiency of this approach is that it is not clear how to treat networks with loops . to apply our analysis to the cases with loops , and , in general , to networks with _ arbitrary _ connectivity ( figure 10 ) , we consider time - dependent models here . this allows us to observe propagation of noise around the loop explicitly and to make accurate conclusions about contributions to dm . we limit ourselves to linear dynamical systems , where the single nonlinear element is the last one , transforming an analog system output to a binary response . as the first step we consider temporal dynamics in the discrete - time approximation , which contains all essential features of our approach . later in the section we extend discrete model to the continuous - time case and show their equivalence . in this section we consider a system of @xmath54 elements , whose activity at each instant is described by an @xmath54-dimensional column - vector @xmath55 . time has discrete values separated by an interval @xmath56 . therefore this model is called the discrete - time model . the values of activity at two neighboring time - slices are related by the connection matrix @xmath57 @xmath58 here @xmath59 is the vector describing noise added to activity vector on each time - slice . the variable @xmath60 describes sensory input into the system . the rules of temporal evolution of activities described by this equation are general enough to include almost all interesting phenomena and mimic modeling of real systems on digital computers . in appendix b we will prove that this model is equivalent to systems with continuously defined time . noise is specified by the parameter @xmath59 , which has a zero mean and is defined by the correlation matrix @xmath61 we assume here that neighboring in time values on noise are not correlated , implying that we consider a system with white noise . this assumption can be easily relaxed and is used here to simplify the analysis . it becomes rigorously valid when time - interval @xmath56 is longer than the correlation time of noise . further , if noise is specific to each neuron , the same - time correlation matrix @xmath62 is diagonal @xmath63 this takes place i.e. when stochasticity is induced by probabilistic nature of synaptic vesicle release , in which case every two neurons receive uncorrelated fluctuating inputs . some time after presentation of the stimulus [ @xmath64 the system is forced to make a decision through the following process first , a scalar quantity @xmath65 is evaluated . here time corresponds to the instant , when the choice is to be made . the output metrics vector @xmath66 describes the way in which system s activity affects motor response . in the simplest case , which was considered in the previous section , when a single element number @xmath20 evokes responses , @xmath67 . in a more complex situation , when multiple areas / neurons have direct influence on decision , vector @xmath66 has more than one non - zero element . on the second step , decision is made based on the sign of @xmath68 @xmath69 thus , this model describes a two - alternative forced - choice task . our system is completely defined by the following set of parameters : @xmath70 , @xmath60 , @xmath62 , and @xmath66 . as we have shown in the previous section , the presence of the stimulus is not required to define dm elements [ eq . ( [ eq5 ] ) ] . we therefore set @xmath60 to zero and are left with three parameters @xmath57 , @xmath62 , and @xmath66 . we now are ready to determine dm elements in our simple model . to find decision makers we will use eq . ( [ eq14 ] ) . in this case it becomes @xmath71 therefore , we need to evaluate the variability on the output from the system @xmath72 . this is accomplished if we notice that @xmath73 and @xmath74 here we introduced the cross - correlation matrix defined as follows @xmath75 we replace here the time variable by the integers , specifying the time - slice number . the averaging in ( [ eq25 ] ) and ( [ eq26 ] ) is assumed over different instantiations of noise ( trials ) . due to the properties of noise in our model , this correlator does not depend on the absolute values of time ( @xmath20 and @xmath76 , but only on the difference ( @xmath77 . as follows from ( [ eq25 ] ) , of particular interest is the same - time correlator @xmath78 , which determines fluctuations in @xmath68 . we now derive equation for same - time correlator @xmath79 . using ( [ eq19 ] ) we obtain @xmath80\left [ { \vec { x}^t(n)\hat { c}^t+\vec { \eta } ^t(n ) } \right ] } \\ \end{array}\ ] ] we then notice that the correlator @xmath81 is identically zero , since @xmath82 is a linear combination of values of noise at times @xmath83 [ see eq . ( [ eq19 ] ) ] . we thus deduce from eq . ( [ eq27 ] ) that @xmath84 which leads us , finally , to @xmath85 this equation allows us to determine the same - time correlator @xmath79 from connectivity and noise cross - correlogram , defined in ( [ eq20 ] ) , which is a diagonal matrix . we would like to pause here and describe the properties of this equation . first of all , in the most generic case ( [ eq28 ] ) allows us to determine @xmath86 from @xmath57 and @xmath62 uniquely . indeed , ( [ eq28 ] ) is a system of @xmath87 linear equations for @xmath87 unknowns @xmath79 , arranged in the matrix form . hence , this system , in most cases , can be solved uniquely . on the other hand , with one exception , @xmath79 can not be expressed explicitly in terms of matrices @xmath57 and @xmath88 . thus , one has to either appeal to the representation of @xmath79 in terms of eigenvectors and eigenvalues of @xmath57 , or use computer to arrange elements of matrix @xmath79 in vector form and solve resulting linear system . the contribution to dm from a given element can be determined from eq . ( [ eq25 ] ) @xmath89 the topological dm contributions are @xmath90 using eqs . ( [ eq28 ] ) and ( [ eq29 ] ) one can analyze a variety of network connectivities . some new effect emerging for non - tree systems are described next . we now consider network shown in figure 11a , in which all elements have the same variance of noise and all connections have unitary strength . figure 11a shows two pathways from unit * 2 * to the exit unit , * 6*. the resulting network gain from unit * 2 * to unit * 6 * is thus equal to two . all other units gain at the exit is one . the contribution to dm from unit * 2 * is thus four times larger that from other units . this is because noise at this unit is multiplied by a factor of two , and the variance of noise , by a factor of four . we conclude that there may be some special elements in network , which occupy hub - like positions , gaining large influence due to abundance of their outputs . it should be noted that fan - in hubs are not special from the point of view of dm in any way . let us now examine the network with a loop . figure 11b shows such an example with unitary link strength and uniform noise variance , as in previous case . the presence of loop affects dm drastically : our discrete - time model marks units belonging to the loop as decision makers ) and ( [ eq29 ] ) does not have a valid solution for the loop with all connection equal to unity . one needs to set one of the connection as a parameter , @xmath91 , solve the equations , and consider the limit @xmath92 . this is easy to understand , since noise , generated by each unit on each time - step , can not leave the loop and , therefore , builds up there without limits . therefore the variance of noise in the output of element number three grows proportionally to time @xmath93 ^ 2 } = \overline { \eta ^2 } t\to \infty $ ] . here averaging is assumed over instantiations of noise ( trials ) . thus , loop becomes the crucial decision maker . this case is somewhat analogous to our previous ` noisy ' neuron example . what is the possible role of loops in biological networks ? why would one introduce such unreliable components ? loops , similar to shown in figure 11b , have many useful properties . for instance , they can act as parametric memory systems . imagine that responses of all units in the loop have the same values , equal to @xmath94 . this could be accomplished by manipulating the sensory inputs . assume that no more inputs are received from the outside of the system . it follows that , in the absence of noise on each element , this value of response will reverberate around the loop forever . this is because all links have unitary strength . loops can thus memorize a graded value , such as @xmath94 , functioning as parametric memory elements . suppose , in addition , that a non - zero input @xmath95 is applied to element number * 1 * at all times . since this element acts as a summator , its response on the next step is @xmath96 . the signal @xmath95 propagates around the loop , and in four steps it reaches the first element again , at which time its response is @xmath97 . in four more steps @xmath98 . thus , not only noise , but also signal can build up in the system . therefore , a loop can operate as a temporal integrator . the integration is not perfect if one of the links has a non - unitary strength , in which case integrator becomes leaky ( robinson , 1989 ) . temporal integrators play special role in dm , since they act as accumulators of sensory information , which puts them into special position with respect to other areas ( gold and shadlen , 2002 ) . as an example , such is area lip in primate visual cortex , which is involved in dm in direction - discrimination task ( shadlen and newsome , 2001 ; roitman and shadlen , 2002 ; mazurek et al . , 2003 ) . we finally consider a model , in which time runs continuously . this model has potential relevance to real - life networks . the responses of units satisfy the following equation @xmath99 the network connectivity matrix @xmath100 can be related to connection matrix from the discrete - time model in ( [ eq19 ] ) through @xmath101 ( see appendix b ) . noise is defined by its cross - correlation @xmath102 where @xmath103 is a diagonal cross - correlogram of noise . ( [ eq31])-([eq33 ] ) are analogous to the discrete - time case ( [ eq19])-([eq21 ] ) . similarly , we define the output scalar and the decision variable @xmath104 @xmath105 here @xmath106 is time when the system makes the decision . our model is thus defined by eqs . ( [ eq31])-([eq34 ] ) . we will now use definition ( [ eq29 ] ) to find decision makers . as in discrete - time case we need to know the variance of the output variable , @xmath72 , after which ( [ eq29 ] ) leads to @xmath107 important for us is the time - dependent correlator @xmath108 which we now evaluate . solution of ( [ eq31 ] ) is obtained using matrix exponentials @xmath109}\ ] ] if external stimulus is zero or a constant in time , due to ( [ eq5 ] ) , the correlator at @xmath110 @xmath111 we seek @xmath112 in the form @xmath113 where @xmath79 is equal - time cross - correlation . to find equation for @xmath79 we differentiate ( [ eq38 ] ) as follows @xmath114 we arrive thus to the following equation for @xmath79 @xmath115 this equation is the central tool for the continuous - time theory . the contributions to dm from each unit are found by differentiating @xmath116 with respect to noise , as in eq . ( [ eq29 ] ) @xmath117 once the same - time correlation matrix @xmath79 is found from eq . ( [ eq41 ] ) , cross - correlation for arbitrary time is @xmath118 this equation suggests a helpful strategy for determining noise matrix @xmath119 . indeed , ( [ eq41 ] ) and ( [ eq43 ] ) imply that @xmath120 here @xmath121 is infinitesimally small positive number . in other words , noise matrix is equal to discontinuity in time - derivative of cross - correlation at @xmath122 . since noise correlation matrix is diagonal , the non - zero elements are @xmath123 two comments are in order here . first , noise term @xmath59 plays the role of input noise in ( [ eq31 ] ) . it can not be measured directly . equation ( [ eq44 ] ) provides a way to single it out . second , ( [ eq44 ] ) does not apply to the discrete - time model . indeed , in the latter we either have @xmath122 , or @xmath124 , etc . , i.e. the condition @xmath125 with @xmath121 infinitesimally small is hard to enforce . it may happen that @xmath126 is acceptable due to presence of slow components in the circuit , such as temporal integrators . however , in general case ( [ eq44 ] ) can not be applied to the discrete - time case . for instance , it fails dramatically for the case of ` nematode ' chain considered above . equations ( [ eq41 ] ) , ( [ eq42 ] ) , and ( [ eq44 ] ) represent a useful set of tools to find dm components for various connectivities . we present here two possible cases , in which decision makers can be found . they differ in what is known about the system . assume we know the network connectivity @xmath100 , output metrics vector @xmath66 , and autocorrelation for each unit @xmath127 . the steps below allow finding the decision makers . 1 . since noise matrix is diagonal , as per ( [ eq33 ] ) , it can be found from autocorrelation using ( [ eq44 ] ) . 2 . solving ( [ eq41 ] ) allows determining @xmath128 , the derivative of equal - time crosscorrelation with respect to noise in each element . decision makers are found from ( [ eq42 ] ) . 4 . normalize contributions to dm so that @xmath129 . scenario 1 does not require simultaneous measurements from all units . it requires the knowledge of the network connectivity however . the next scenario is complimentary in this respect . suppose we have measured the full crosscorrelation matrix @xmath112 by simultaneous recordings from all units . suppose also that we know how the output of the system is evaluated ( vector @xmath130 . these are the steps to determine dm units . 1 . use ( [ eq44 ] ) to find noise matrix @xmath62 . 2 . use ( [ eq41 ] ) to find the connection matrix @xmath100 . 3 . solve ( [ eq41 ] ) to calculate @xmath128 for each element . 4 . use ( [ eq42 ] ) to find decision makers . normalize contributions to dm so that @xmath129 . both scenarios use extensive knowledge about the system , which renders them useless in experimental conditions . in the next subsection we discuss a way to bypass these limitations . finally , we would like to provide solution to ( [ eq41 ] ) using eigenbasis of matrix @xmath100 . since @xmath100 is not necessarily symmetric , a distinction should be made between right and left eigenvectors . the latter turn out to be useful for our purposes . they are defined by @xmath131 here and below greek indexes denote numbers of eigenvalues , while latin ones label spatial components of vectors and matrices . solution of ( [ eq41 ] ) is @xmath132 where @xmath133 is the gram matrix of eigenvectors . ( [ eq46_5 ] ) is valid if the eigenvectors form a complete basis in the @xmath54-dimensional space . as follows from ( [ eq46_5 ] ) , eigenvalues of @xmath100 with small real part contribute to dm in a large degree . this justifies the use of principal component analysis when such eigenvalues are present . an example of such principal component is the temporal integrator loop in figure 11b , which has vanishing @xmath134 . in case if matrix @xmath100 is symmetric , its eigenvalues are real and eigenvectors are orthogonal . this leads to a unit gram matrix . then , eq . ( 3.29 ) becomes more compact @xmath135 similar equations , called kubo formulas , are obtained for various correlators in case of diffusion of particles in random media ( efetov , 1997 ) . the distinguishing feature of ( [ eq48 ] ) is that a product of four eigenvectors enters the expression . thus , propagation of noise in this case can be accompanied by interference between different pathways . an example of destructive interference of this kind is given below , in section [ stimulation ] . ( [ eq48 ] ) can be further simplified . indeed , our model uses diagonal noise matrices , i.e. @xmath136 in ( [ eq48 ] ) . suppose also that the output from the network occurs through one exit element number @xmath2 , which is specified by taking @xmath137 . in this case the use of eq . ( [ eq42 ] ) gives @xmath138 from this equation we conclude that for element @xmath20 to contribute to dm , an eigenvector should exist , which is non - zero on both unit number @xmath20 and exit unit @xmath2 . thus , we conclude that eigenvectors of @xmath100 should be delocalized for broader impact of elements on the decision . this is not surprising in view of the mentioned analogy with the diffusion problem . in case if matrix @xmath100 is not symmetric , the gram matrix may be non - diagonal and ( [ eq49 ] ) can not be used . however , the off - diagonal elements of @xmath139 are usually smaller than diagonal ones , due to uncorrelated sign changes , when ( [ eq47 ] ) is computed with @xmath140 . therefore , ( [ eq49 ] ) may apply approximately . stimulations with electric current add a new degree of freedom to dma , thus leading to more effective ways of finding decision makers . there are two great advantages of the stimulation method . first , it only involves stimulation of a single neuron , therefore no simultaneous multiple - electrode measurements are required . second , the knowledge of network connectivity is not needed to solve the problem . in this section we study our simple networks and find what stimulation strategies are consistent with our earlier definitions , such as eq . ( [ eq14 ] ) . we will use continuous model for concreteness ( section [ continuous ] ) . consider the output variable @xmath68 . it is a linear function of the inputs . it is also a function , which contains noise components , variable from trial to trial . the noise components were acquired from all units in different degree . since noise in each unit is gaussian , the output variable is described by gaussian distribution too @xmath141 in each trial a random value of @xmath68 is obtained , according to distribution ( [ eq50 ] ) . the response of the system is equal to 1 if @xmath68 is positive , and 0 otherwise . the probability to obtain response equal to 1 to given stimulus @xmath95 is given by the error function ( abramowitz and stegun , 1972 ) @xmath142,\ ] ] whereas the probability of zero response is @xmath143.\ ] ] both probabilities depend upon the mean response to stimulus @xmath144 and the standard deviation @xmath145 . therefore the electric stimulation strategies may be based on affecting either the former or the latter . we now consider both of these strategies and show that affecting the mean response may provide misleading results , while changing the variance of response allows estimating contributions to dm consistently with our previous definitions . thus , strategies of stimulation based on standard deviation of the output variable are _ always _ correct in our simple model , independently on the topology of the network . this may seem a trivial consequence of definition ( [ eq14 ] ) , but we will discuss it here for the sake of comparison of two strategies and optimizing them . we start with the strategies of stimulation , which affect the mean response @xmath144 . in our simple model this may be accomplished by injecting a tonic input current into a unit number @xmath2 . mathematically it is accomplished by adding extra stimulus @xmath146 to this unit in eq . ( [ eq31 ] ) . note that in biological systems the stimulating current is alternating with constant amplitude ( salzman et al . , 1992 ) . the mean response is shifted by the stimulation , i.e. @xmath147 where @xmath146 is the magnitude of injected tonic current . this leads to observable changes in the probability @xmath148 @xmath149 here @xmath150 is the change in probability of correct responses after unit number @xmath2 is electrically stimulated . can @xmath150 be a measure of dm ? we notice that @xmath150 can be either positive or negative . this depends on the sign of derivative @xmath151 , which is positive for excitatory pathway from unit @xmath2 to the output and negative for inhibitory pathway . since contribution to dm ought to be positive , we can not assume simply that @xmath152 . the correct expression , which we provide here without derivation is @xmath153 ^ 2.\ ] ] this equation is understood in proportional sense , since @xmath154 should be normalized to ensure that @xmath155 . our investigations show that this expression is accurate for trees and is consistent with both our earlier definitions ( [ eq9 ] ) or ( [ eq14 ] ) . remarkably , it employs quantities , which can be measured in a single - electrode experiment . indeed , the amplitude of noise @xmath6 can be found from autocorrelation of unit s response , using ( [ eq44 ] ) ; and @xmath150 is determined from behavioral changes in response to single - unit stimulation . this equation thus provides an approach potentially useful in practice . does this relationship work for networks of arbitrary connectivity ? figure 12b shows a counterexample , in which a unit is stimulated , which results in _ no _ change in probability of correct response ( we consider trials in which 1 is the correct response throughout this section ) . this is because there are two pathways , leading from this unit to the exit , one positive and one negative . they have equal strength , and , therefore , compensate each other . on the other hand , unit number one _ does _ participate is dm , because if a non - stationary stimulation / stimulus is applied , its effect on the decision is not zero . thus , ( [ eq54 ] ) and tonic stimulation method can not be applied to arbitrary circuits , such as shown in figure 12b , to accurately reveal decision makers . . , width=288 ] . , width=288 ] is there a stimulation method for finding decision making components in arbitrary networks ? the method follows directly from the definition ( [ eq14 ] ) [ or ( [ eq35 ] ) , which is equivalent ] . indeed , when stimulating current is a temporal white noise , the output variable @xmath68 acquires a larger variance ( figure 12c ) . hence , the derivative of output variance , entering ( [ eq35 ] ) can be calculated operationally , by injecting a distracter current . more precisely , if the variance of stimulating current applied to unit @xmath2 is @xmath156 the derivative entering definition ( [ eq35 ] ) is @xmath157 in practice one has no access to the variable @xmath68 , so one can not measure directly the change in variance @xmath158 . instead , one could measure the change in the probability of correct responses under the influence of distracting current . indeed , from ( [ eq50_5 ] ) we obtain @xmath159 combining the last two equations we obtain for the important derivative @xmath160 since the probability of correct responses always decreases under the influence of distracters , the derivative @xmath161 is a negative constant . it is the same for all units . we arrive therefore to the expression for contributions to dm , which follows from ( [ eq35 ] ) @xmath162 here @xmath150 is the decrease in probability of correct responses produced by electric stimulation with variance of the random current equal to @xmath156 . the variance of noise on each unit @xmath163 can be found from autocorrelation using ( [ eq45 ] ) . this procedure works for any topology in our simplified model . it should be noted here that if noise is not entirely white or can not be considered white , ( [ eq45 ] ) can not be used directly and should be replaced by an expression reflecting the spectral characteristics of noise appropriate for the system under investigation . thus , if noise is provided by other parts of the network , its dynamic features may be more complex . therefore , ( [ eq45 ] ) may not apply directly to the ` hidden pathway ' example given in the end of section [ conclusions ] . the procedure , which we just described , permits further optimization . indeed , imagine that the probability of correct responses is exactly @xmath164 . adding distracting stimulation current will not change this probability , i.e. @xmath165 no matter what unit is stimulated . in the opposite limiting case when @xmath166 , the effect of distracter on performance is exponentially small . hence , behavioral response to stimulation has an optimum between @xmath167 and @xmath168 . to find the optimum we observe from ( [ eq56 ] ) that @xmath169 is maximum for the same variation in @xmath170 when @xmath171 is maximum . we therefore plot the latter derivative as a function of @xmath148 in figure 12d . we indeed observe a maximum at the value of probability of correct responses close to @xmath172 to summarize , the following scenario describes algorithm for finding contributions to dm using random stimulation . assume that we _ do not know _ the network connectivity @xmath100 and output metrics vector @xmath66 ; but we know autocorrelation for each unit @xmath127 . the steps below allow finding the decision makers . 1 . prepare stimulus so that the probability of correct responses is close to the value given by ( [ eq59 ] ) . 2 . stimulate one unit with random current , whose variance is @xmath156 , and measure the decrease in probability of correct responses @xmath169 . 3 . record autocorrelation and evaluate noise variance @xmath6 for this unit using ( [ eq45 ] ) . 4 . find contribution to dm for this unit using equation ( [ eq58 ] ) . repeat steps 1 through 4 for all units in the system . normalize contributions to dm so that @xmath129 . in this work we defined decision makers in networks , which behave in a well - defined fashion . as with any definition , there is certain degree of arbitrariness in our study , since this is the first mathematical study of this sort . we had to make choices about the features of decision making we were attempting to describe as well as about the way they were quantified . we demonstrated these features in a set of examples . future studies will show if these features can be used as a basis of a more complete model - independent theory . in this study we postulated that variability and noise , causally linked to decisions , are the chief descriptors of dm . although this point may seem paradoxical we suggest three arguments in its favor . first , variability may reflect additional information needed to make a decision in case of uncertainty . such may be inputs from other modalities , memories , or some other relevant modulatory inputs , supplying e.g. emotional condition of the subject or changing utility values ( figure 8) . second , many behaviors , such as c - start escape responses in fish ( eaton and emberley , 1991 ) and other organisms ( glimcher , 2003 ) , have stochastic character . this makes the task of pursuer more difficult . such unpredictable behaviors are reproduced in our model if the sensory input is weak or in the small signal - to - noise ratio case . third , the goal of dm is to dissipate sensory information , as suggested in the introduction ( figure 1a ) , whereby an analog multi - dimensional stimulus space is reduced to a discrete space of several decisions . we argue that this transformation is facilitated by noise . we have studied the problem of finding decision making units in networks of various connectivities . this path took us from simple linear chains , for which the information - theoretical ( it ) approach was found to be effective , to trees , and , finally to an alternative definition of decision makers , based on propagation of noise in networks . this latter definition is valid in networks of arbitrary topology . all these approaches are equivalent , when they can be compared , but include progressively broader classes of networks . as a practical application for the alternative definition we considered the problem of electric stimulation in the surrogate networks and showed a way of determining dm contributions for arbitrary networks using stimulation with random current . our findings are summarized in figures 13 and 14 . although we studied networks of complex connectivity , the model describing a single network element was quite simple . not all of the units are linear , of course , since dm is a non - linear task ( figure 1a ) . however , our model is essentially based on linear elements . the motivation for this model is that it is easy to analyze . the study of simple models is a necessary step before analysis proceeds any further . once the methodological issues are resolved for simpler models , complex non - linear systems can be studied in the same paradigm . one of important questions resolved here is that a completely linear element can be a decision maker , despite the presence of non - linear units in the network . thus , nonlinearity is not a necessary attribute of dm . this question would be impossible to answer for more realistic system , since in practice all units contain nonlinearities . decision making task , as formulated in figure 1a , is similar to general object discrimination task . representation of motor response in our model is not distinguishable mathematically from the representation of abstract object / decision category ( horwitz and newsome , 1998 ; shadlen and newsome , 2001 ) . the latter does not necessarily lead to a motor command . thus , our analysis may uncover the identities of units responsible for categorization of sensory inputs . in terms of this analysis we emphasize the distinction between units representing the object category and the units in which this representation is actually formed . the former are analogous to motor units in the decision task , while the latter are similar to decision makers . as follows from this study , the analysis is dependent upon the topology of the network involved . for simple linear sensory chains our conclusion is that the _ first _ unit , spatially or temporally , in which the representation of the object is correlated with final outcome of the discrimination process , is responsible for casting the stimulus in one of the abstract classes . in case of recurrent networks a more detailed quantitative analysis is needed to draw conclusions about identities of categorizing units . thus , dma may find a broader use in identifying units representing abstract object s percepts . a special care should be taken in distinguishing the dm task from the sensory discrimination task . it may occur that in the same experiment these tasks are performed by different populations of neurons . an example is given by ( salinas and romo , 1998 ) . they discovered a population of m1 neurons responding differentially to two categories of tactile stimuli . some of these neurons did not respond , when the same behavior was guided by visual cues . this observation is consistent with these neurons performing sensory discrimination of tactile stimuli , while some other population making decisions about the actual motor response . our mathematical analysis is general enough to include both of these functions . thus , if correlations with motor response are studied , it will result in the decision makers ; while when the correlations with percepts are investigated , dma should provide the identities of discriminating elements . we suggest that dma may be relevant to other biological systems . possible applications may include the analysis of molecular networks , such as genetic regulatory or protein binding networks ; finding decision makers in compartmental models of dentritic trees ( poirazi and mel , 2001 ) ; studies of neural networks and structural networks of connectivities between different brain areas ; and analysis of social networks . in this study we define network elements responsible for making decisions . we obtain two equivalent definitions . according to one , decisions are made by elements , in which correlations with the decision are first formed . according to the second definition , decision making activity is measured by the impact of variability in given unit on the response . we give examples of network motifs , especially potent from decision making prospective , such as fan - out hubs and recurrent loops . the latter can function as temporal integrators of sensory inputs . we also study how electric stimulations can reveal decision making components . we conclude that stimulations with time - varying random current produce correct results for all network topologies . here we solve a more general version of linear chain model than considered in the text . the responses of neighboring neurons are related linearly @xmath173 this is a generalization of ( [ eq1 ] ) . the response of the @xmath20th unit is @xmath174 where coefficients @xmath175 , @xmath176 . the external signal @xmath177 is assumed to be zero in this appendix , due to ( [ eq5 ] ) . for the last element in the chain we have @xmath178 comparing ( [ eqa2 ] ) and ( [ eqa3 ] ) we conclude that @xmath179 where @xmath180 is a variable , which describes noise in the networks downstream from unit @xmath20 . it is , thus , uncorrelated with @xmath181 . this is where tree - like topology enters our solution , since in case of loops , @xmath181 and @xmath180 are correlated . our goal now is to calculate mi between the decision variable @xmath182 and @xmath181 . we will use the definition for mi @xmath183 } } \label{eqa6}\ ] ] here @xmath184 , since there is no signal ; @xmath185 and @xmath186 . \label{eqa9}\ ] ] the upper / lower sign is assumed for @xmath187 or @xmath168 in ( [ eqa9 ] ) ; @xmath188 is the standard deviation of gaussian variable @xmath180 defined in ( [ eqa4 ] ) . the expression for mi ( [ eqa6 ] ) results in @xmath189 } \log _ 2 \left [ { 1+\mbox{erf}\left ( { zs_n } \right ) } \right ] \\ s_n = \sigma ( \alpha _ { nn } x_n ) /\sigma \\ \end{array } \label{eqa11}\ ] ] mi is therefore a function of signal - to - noise ratio @xmath190 . inversely , @xmath191 ^ 2 \label{eqa12}\ ] ] on the other hand , ( [ eqa4 ] ) leads to @xmath192 solving ( [ eqa12 ] ) and ( [ eqa13 ] ) with respect to @xmath193 we have @xmath194 ^ 2}{1+\left [ { m^{-1}(mi_n ) } \right]^2}\equiv f(mi_n ) \label{eqa14}\ ] ] function @xmath195 here is inverse to @xmath196 defined in ( [ eqa11 ] ) . function @xmath25 numerically calculated from ( [ eqa11 ] ) and ( [ eqa14 ] ) is shown in figure 5 . lastly , we recall that variances @xmath197 are related to the strength of noise @xmath6 through ( [ eqa2 ] ) . we have @xmath198 eqs . ( [ eqa14 ] ) and ( [ eqa15 ] ) are used below to prove a variety of statements about function @xmath25 used in the main text . in this case @xmath199 , and , consequently , @xmath200 . noise variance is the same on every node , i.e. @xmath201 . as follows from ( [ eqa15 ] ) @xmath202 , which results in @xmath203 it follows that contributions to dm defined by ( [ eq9 ] ) are the same for all units . in this case @xmath204 , while @xmath205 , assuming that the link from unit @xmath206 to @xmath207 is strengthened . in the example in the text @xmath208 [ cf . ( [ eq10 ] ) ] . eq . ( [ eqa15 ] ) leads us to the values for variances of responses @xmath209 applying ( [ eqa14 ] ) we obtain the expression for @xmath25 @xmath210 which is a piece - wise linear function of @xmath20 . ( [ eq9 ] ) determines contributions to dm as @xmath211 this confirms that the upstream units ( @xmath212 are @xmath32 times more potent than the downstream ones ( @xmath213 . let us start by determining decision makers from definition ( [ eq14 ] ) . according to ( [ eqa3 ] ) @xmath214 definition ( [ eq14 ] ) gives @xmath215 after normalization we obtain @xmath216 let us derive the same result from ( [ eq9 ] ) . as follows from ( [ eqa14 ] ) @xmath217 this proves the equivalence of ( [ eq9 ] ) and ( [ eq14 ] ) , since the result is identical to ( [ eqa22 ] ) . in this section we show that the discrete - time model can be derived from continuous - time model . starting from equation ( [ eq37 ] ) for the unit responses in the continuous case we obtain the relation for solutions at two different time points separated by the time - interval @xmath218 , analogous to ( [ eq19 ] ) in the discrete - time description . then we show that in the limiting case @xmath219 two descriptions are equivalent . from ( [ eq37 ] ) we obtain @xmath220 this equation can be rewritten as @xmath221 , where @xmath222 thus it has the same form as ( [ eq19 ] ) . using ( [ eq32 ] ) we obtain that the new noise cross - correlation matrix @xmath223 the solution of the continuous - time problem satisfies the equations of the discrete - time model for an arbitrarily large time interval @xmath56 , but the new noise cross - correlation matrix @xmath224 is non - diagonal in this case . in the limiting case @xmath219 it becomes diagonal . indeed ( [ eqb3 ] ) implies that in this limit @xmath225 which is diagonal by the definition of the continuous - time model . here we kept only terms linear in @xmath56 . thus , in this limit the matrix @xmath224 is diagonal as needed in our formulation of discrete - time model . one can also derive ( [ eq41 ] ) from ( [ eq28 ] ) using ( [ eqb2 ] ) and ( [ eqb4 ] ) and taking the limit @xmath219 .

the importance of atomic diffusion @xcite inside stars is now well established : not only can it modify the atmospheric abundances , as observed in the so - called `` chemically peculiar stars '' , but it can also have strong implications for the stellar internal structure . the variations with depth of the radiative accelerations on specific elements can lead to their accumulation or depletion in various layers inside the stars . in this framework , the accumulation of iron ( together with less important metals like nickel ) may have strong consequences , which are different according to the stellar type . in main sequence a stars , it may lead to the creation of an extra convective zone due to an iron - induced increase of opacity @xcite . if overshooting leads to complete mixing between this extra convective zone and the one due to helium , helium settling may be slowed down in the outer stellar layers . such an effect could explain the existence of evolved oscillating am stars @xcite . the occurrence of an iron convective zone may also help explaining the excitation of g - modes in @xmath2 doradus stars , through the `` convective blocking '' process , as suggested by @xcite . furthermore , in some cases iron accumulation can trigger stellar pulsations through the iron - induced @xmath0-mechanism . this process has been invoked to explain main sequence b - type pulsators like spb and @xmath3 cephei stars @xcite . @xcite predicted that the extreme horizontal branch stars referred to as sdb stars could also be destabilized due to iron - induced @xmath0-mechanism . the oscillations of these stars were independently observed ( e.g. * ? ? ? * ) . the qualitative success of the iron accumulation theory is impressive . however when quantitatively comparing models and observations , some difficulties remain . we will further discuss these difficulties in section [ discussion ] . in previous stellar modelling done in this framework , the question of the stability of a heavy iron layer lying above layers with smaller mean molecular weights was not addressed . this represents a serious problem , because of the existence of local inverse @xmath1-gradients , which are highly unstable against thermohaline ( double diffusive ) convection ( see * ? ? ? the induced mixing has to be taken into account in the computations , otherwise the results which are obtained do not correctly represent the real situation found inside the stars . the importance of thermohaline convection has been invoked in a few astrophysical cases . the first case which was pointed out concerned the @xmath4he burning regions in giant stars @xcite . situations with @xmath5he enhancement in outer stellar layers were also discussed in several studies . this enhancement could be due to mass transfer @xcite or to helium diffusion in a stellar wind @xcite . the observations of main sequence helium - rich stars , with effective temperatures around 20 000k , show that helium is enhanced in average by a factor of two . this demonstrates that thermohaline mixing is important , otherwise the helium overabundance would be much larger . however , thermohaline convection mixing can not be extremely efficient since this would lead to no helium overabundance . in this case , the reason for this lack of efficiency may be due to the presence of a magnetic field . the same kind of processes may occur inside roap stars , as discussed by @xcite . thermohaline convection has also been studied for the case of accretion of metal - rich matter onto main sequence stars . this situation may occur for exoplanet host stars , if they accrete hydrogen poor matter in the process of planetary formation . because of thermohaline convection , and contrarily to preliminary conclusions found in early papers @xcite , the accreted matter does not remain inside the convective zone , but diffuses downwards @xcite . another case concerns carbon enhanced main - sequence stars ( cemps ) , which are assumed to have suffered some accretion of material coming from an agb companion . this heavy matter falls down inside the star due to thermohaline convection @xcite . owing to the stabilizing @xmath1-gradient induced by helium diffusion , some of it can remain in the outer layers and may account for the observed abundances @xcite . in this paper , we first give a general discussion of thermohaline convection in stars ( section [ thermoconv ] ) and show how this physical process can be introduced in the stellar evolution computations ( section [ computations ] ) . then results are presented for heavy elements accumulation inside a - type stars . we mainly concentrate on the evolution of a 1.7 @xmath6 star , but also give results for 1.5 @xmath6 and 1.9 @xmath6 stars ( section [ results ] ) . for these computations , the tgec code ( cf . * ? ? ? * ; * ? ? ? * ) was modified to include the computations of radiative accelerations using the svp ( standing for single valued parameters ) approximation proposed by @xcite . this modified version of the tgec code was first used to compute stellar evolution models ( without introducing thermohaline convection @xcite ) which were compared to the results obtained by the montreal group . these computations were repeated while including thermohaline convection below the iron accumulation layer . due to the stabilizing effect of helium settling , which fortunately occurs in the same stellar region , iron can still accumulate , but the abundance profiles are strongly modified . we also discuss the potential important consequences of this process for other types of stars in section [ discussion ] . a short conclusion will then follow . thermohaline convection is a well - known process in oceanography : salted water layers lying above fresh water ones are rapidly mixed downwards even in the presence of stabilizing temperature gradients , due to the different diffusivities of heat and salt . when a warm salted blob falls down in cool fresh water , the heat diffuses out more quickly than the salt . the blob goes on falling due to its weight until it mixes with the surroundings . this leads to the formation of so - called `` salt fingers '' . a similar kind of convection occurs in stellar radiative zones in the presence of positive @xmath1-gradients , that is @xmath1-values increasing upwards . an``iron layer '' lying above hydrogen rich medium will lead to `` iron fingers '' , because when a blob falls down heat diffuses between the blob and its surroundings more quickly than iron ions . with reference to oceanography , this process is called `` thermohaline '' convection . it is also sometimes referred to as double diffusive convection , because it is directly related to the presence of two `` scalars '' which diffuse on very different time scales . another kind of double diffusive convection occurs in the reversed case , in the presence of destabilizing temperature gradients and stabilizing @xmath1-gradients like , for example , at the edge of convective cores . this process , quite different from thermohaline convection , is generally referred to as `` semi convection '' . it will not be considered here . in the case of positive @xmath1-gradients , the medium can become dynamically unstable if ( ledoux criterion ) : @xmath7 where @xmath8 ln @xmath9 ln @xmath10 and @xmath11 ln @xmath9 ln @xmath12 . when @xmath13 vanishes , marginal stability is achieved . for positive or null values of @xmath13 , the medium is stable against dynamical convection . then thermohaline convection occurs on a time scale which is typically of a few thousand years , long compared to the dynamical time scales but short compared to stellar lifetimes . fingers may form if the following condition is verified : @xmath14 with @xmath15 = @xmath16 where @xmath17 and @xmath18 are the thermal and molecular diffusion coefficients while @xmath19 and @xmath20 are the corresponding time scales . in stars the value of the @xmath21 ratio is typically on the order of @xmath22 if one assumes that @xmath18 is the `` atomic '' diffusion coefficient . however , it can become larger when the shear flow instabilities which induce mixing between the edges of the fingers and the surroundings are taken into account ( see * ? ? ? the effects of thermohaline convection as a mixing process in stars are far from trivial . many detailed studies related to the occurence of this physical process in water have been published . for example , @xcite gave precise comparisons between numerical simulations and laboratory experiments . however the stellar case is different since mixing occurs in a compressible fluid . two different parametrisation recipes , given respectively by @xcite and @xcite can differ by up to two orders of magnitude : this illustrates that treating thermohaline convection as a simple diffusion process may lead to inaccurate results . the basic problem here concerns the vertical shear flow instability which occurs between the fingers and the inter - fingers medium . this instability leads to local turbulence which increases the mixing at the edge of the fingers . consequently , a process of self destruction appears for the blobs , so that fingers eventually reach a regime where they can not form anymore : this effect is taken into account in the kippenhahn et al . procedure . in the present paper , we take into account the fact that the thermohaline mixing time scales are short compared to the main sequence evolution time scales , and that mixing stops when the positive @xmath1-gradients vanish . in our computations ( see discussion below ) , we model rapid mixing so that the stellar medium always adjusts to keep either flat @xmath1-profiles or @xmath1-values increasing downwards . as helium settles downwards , it creates a stabilizing @xmath1-gradient which in turn may allow some heavy elements accumulation . we computed stellar models to describe the evolution of a - type stars while using the toulouse - geneva evolution code ( tgec ) . the code is described in detail in @xcite and @xcite , except for the computations of opacities and atomic diffusion which have undergone major improvements @xcite . in the present computations , important improvements were introduced in the tgec code concerning the opacities and radiative accelerations . the opacities were computed using the opcd v3.3 codes and data @xcite . it allows to compute self consistent rosseland opacities taking into account the detailed composition of the chemical mixture . the opacities were recalculated at each time step . to more realistically treat atomic diffusion in the tgec code , the radiative accelerations on c , n , o , ca and fe were included following the improved version @xcite of the semi - analytical prescription proposed by @xcite . this method allows for very fast computations of radiative accelerations with a reasonable accuracy . the svp approximation may be implemented in existing codes in a simple way . there are much less data to process than for detailed radiative acceleration calculation , because complete and detailed monochromatic opacites for each ion are not needed . a new grid of svp - parameters , well fitted to the stellar mass range considered in this work , was computed following the procedure described in @xcite . the diffusion of h , he and the five metals mentioned above were included in the stellar models . the abundance of the other elements were assumed to be constant throughout the star . the radiative accelerations of these five heavy elements as well as the detailed abundances for all seven elements were computed at each diffusion time step , chosen to be about 50 times smaller than the evolution time step . for our modelling , the convective regions were instantaneously homogenised . the hi and heii convective regions were supposed close enough to be connected by overshooting and mixed together . on the other hand , the iron convective region which may appear in some models in much deeper regions , was supposed disconnected from the surface convective zone . the models were evolved from pre - main sequence up to hydrogen core exhaustion . they were assumed homogeneous on the pre - main sequence and atomic diffusion was introduced at the beginning of the main sequence . we first computed models without thermohaline convection . to avoid the appearance of steep and unrealistic abundance gradients at the transition between radiative and convective regions , we introduced mild mixing at the bottom of each convective zone . this mixing was modeled as a diffusion process through a coefficient d@xmath23 introduced in the chemical transport equation : @xmath24 @xmath25 represents the mean concentration of a species i , @xmath26 is the density , r is the radius , d@xmath23 is given by : @xmath27 d@xmath28 and r@xmath28 are respectively the value of d@xmath23 and the value of the radius at the base of the convective zone , @xmath29 is the half width of the mixing region . d@xmath28 and @xmath29 are free parameters , choosen to produce a mild mixing on a small extent . the value of d@xmath28 is taken equal to @xmath30@xmath31.s@xmath32 and @xmath29 is fixed to 0.5% of the stellar radius below the surface convective region and to 0.05% of the stellar radius below the iron convective zone . as discussed in the previous sections , thermohaline convection arises in regions where the mean molecular weight increases towards the surface . it leads to mixing on short time scales , compared to those of evolution , and it disappears when the mean molecular weight gradient vanishes . to simulate this phenomenon , we found that the most efficient and simple way was to model the mixing as usual , provided that the chosen diffusion coefficient is large enough to flatten the @xmath1-profile any time a destabilizing @xmath1-gradient is created ( at each time step ) . the diffusion coefficient proposed by @xcite is convenient in this framework : @xmath33 where @xmath34 is the pressure scale height , a the radiative pressure , c the speed of light , t the temperature , @xmath35 the adiabatic gradient , @xmath36 the real temperature gradient , @xmath37 the specific heat at constant pressure , @xmath0 the opacity and @xmath1 the mean molecular weight of the material . we used this prescription in our computations and checked that the @xmath1-gradient remained always null or negative at each time step in all our models . we computed models for 3 different masses : 1.5 , 1.7 and 1.9m@xmath38 . each set of models includes a first series of computations without thermohaline convection and a second series including it . the initial metal mixture used in these computations is the solar mixture presented by @xcite with the following values as h and he initial mass fractions : @xmath39 and @xmath40 . the initial iron mass fraction was consequently @xmath41 . figure [ diaghr ] presents the evolutionary tracks for the 3 sets of models . inclusion of thermohaline convection does not modify these tracks in a sensible way : the two curves coincide . the crosses show the positions of the models which are discussed below . to [ dlnmu ] . those situated on the 1.7 m@xmath42 track locate models at 62 myrs , 86 myrs , 299 myrs , 403 myrs , 588 myrs , 788 myrs , 988 myrs , 1188 myrs and 1388 myrs . on the 1.5 and 1.9 m@xmath42 track , only two models are presented , respectively at 838 myrs and 632 myrs.,width=302 ] for the discussion here , we specifically focus on the case of 1.7m@xmath38 models . figures [ fe ] to [ dlnmu ] compare the results obtained with ( solid lines ) and without ( dashed lines ) thermohaline mixing , at various epochs during the h burning phase ( crosses in fig . [ diaghr ] ) . figure [ fe ] displays the ratio of the iron mass fraction at various times to that of its initial value . figure [ logk ] shows the opacity profiles at these same evolution times . figure [ conv ] presents the differences between the radiative and adiabatic gradients , which allows to locate the convective regions . figure [ mu ] displays the mean molecular weight profiles and figure [ dlnmu ] displays the mean molecular weight gradient , for the interval @xmath43 . the models reach the main sequence with chemically homogeneous envelopes . then , due to atomic diffusion , helium sinks and creates a stable ( negative ) @xmath1-gradient below the convective envelope , whereas the radiative acceleration on iron induces an increase of its abundance in the upper stellar layers , and more specifically in the iron - peak elements opacity bump centered at @xmath44 . for the 1.7m@xmath38 case , the models with and without thermohaline mixing begin to diverge at about 400 myrs . at the age of 62 myrs , in both cases the iron accumulation reaches a factor of 7 below the surface , which does not alter the opacity , and a factor 4 in the opacity bump region , which leads locally to a small opacity increase . consequently , the radiative gradient ( @xmath45 ) increases , but it remains smaller than the adiabatic gradient so that the corresponding region is still convectively stable . at this time ( 62 myrs ) the model only has a single surface convective zone , due to the ionization of hydrogen . as can be seen in figure [ conv ] , from the close - to - zero value of @xmath46 around @xmath47 , the heii convective zone has just disappeared , due to he settling . at this point , the iron accumulation is not large enough to significantly alter the @xmath1-profile . here , the effect of the he - induced @xmath1-gradient is indeed much more important than the iron contribution . during stellar evolution , as the atomic diffusion proceeds , the region where the helium - induced stable @xmath1-gradient is the steepest slowly deepens inside the star . as soon as it occurs below the iron opacity bump , the effects of the iron accumulation on the @xmath1-profiles become important . at 86 myrs , iron accumulation exceeds a factor of 19 and affects the @xmath1-profile . it is responsible for the change in the slope of the @xmath1-profile at @xmath48 . the iron enrichment leads to a significant opacity increase and therefore to an increase of the radiative gradient which now exceeds the adiabatic one . in both cases , with or without thermohaline mixing , a new very narrow convective region appears around @xmath44 . at 299 myrs , this iron convective region becomes significant and leads to chemical homogenization between @xmath44 and 5.4 as seen in the @xmath1- and @xmath49-profiles shown . the homogenization of the iron convective zone leads to an expansion of the iron accumulation region and reduces the iron maximum value from a factor of 19 to a factor of 12.5 . from the beginning of the main - sequence until 299 myrs , the @xmath1-gradient remains null or negative in most models , except below the surface convective zone where it sometimes takes a slightly positive value , due to iron enrichment in the outer regions . after 400myrs , the @xmath1-gradient induced by he - settling is then located below the iron accumulation region and the effects of iron accumulation become visible . from now on , models including thermohaline convection below the iron accumulation layers significantly differ from those without thermohaline convection . during the subsequent stellar evolution , radiative levitation leads to iron accumulation in the iron opacity bump region . in models without thermohaline convection , this iron accumulation drastically increases with time , reaching a factor of 20 at 400 myrs and up to a factor of 95 at 1388 myrs ( fig . [ fe ] , dashed lines ) . the consequences on the opacity profiles can be seen in figure [ logk ] , and those on the radiative gradients in figure [ conv ] . it also leads to a new convective zone which persists during most of the main sequence lifetime . iron accumulation in these specific layers leads to a spectacular increase of the @xmath1-values ( fig . [ mu ] , dashed lines ) . after 400 myrs , `` @xmath1-bumps '' develop with a rapid increase slightly above @xmath44 and a steep decrease below @xmath50 , which is highly unstable . the corresponding @xmath1-gradients are shown in figure [ dlnmu ] . in the models including thermohaline convection induced by these unstable @xmath1-gradients , iron accumulation in the opacity bump region is drastically reduced and never exceeds a factor of 15 . it is never completely suppressed however , due to the stabilizing effect of helium settling . the consequences on the opacity profiles , radiative gradients , @xmath1-values and their gradients can be seen in figure [ logk ] to [ dlnmu ] , solid lines . the evolution of the internal stellar structure in the two cases studied here is sensibly different . contrarily to the models without thermohaline mixing , those which include it do not have persistent iron convective zones , although marginal convection appears and disappears several times during stellar evolution . actually , in models including thermohaline mixing , the iron abundance in the opacity bump always stays close to the critical value for the onset of convection . when iron accumulation exceeds this critical value , convective mixing occurs and reduces the iron accumulation , making this region stable again . note that when this occasional convective zone appears , it has a smaller extent than the iron bump itself , so that only part of the iron enriched zone is homogenized this is the reason why a small iron peak remains around @xmath44 in the 58 myrs , 788 myrs and 988 myrs models . the results obtained for 1.5 and 1.9m@xmath38 models are qualitatively similar to those of the 1.7m@xmath38 models . to illustrate the efficiency of thermohaline convection in these models , we present in figure [ compar3 ] the molecular weight and the iron profiles obtained with and without thermohaline mixing for one model for each mass , located at the middle of the main - sequence ( cf . [ diaghr ] ) . the radiative acceleration on iron increases with the stellar mass , so that , when no thermohaline convection is taken into account , the iron accumulation is larger in more massive models . the consequent `` @xmath1-bump '' is also more prominent , as can be seen in figure [ compar3 ] ( dashed lines ) . in the results obtained including thermohaline mixing ( solid lines ) , we can see that iron accumulation is again drastically reduced . in the 1.9m@xmath38 model , it does not exceed a factor of 10 whereas it goes up to 150 when thermohaline mixing is neglected . figure [ zconv ] presents the position of the boundaries of the convective zones inside the models with and without thermohaline mixing for the three masses studied here . the crosses locate the boundaries of the iron convective zone during phases with mixing episodes . when the convective zone settles durably , its boundaries are represented with solid lines . for the three stellar masses considered , an iron convective zone resulting from a diffusion - induced iron accumulation rapidly appears after the beginning of the main sequence . in models without thermohaline mixing ( fig . [ zconv ] , upper panels ) , a first phase occurs during which convective episodes are followed by convectively stable periods . when convection appears in the opacity bump region , the iron peak decreases due to convective mixing . consequently the opacity also decreases and the radiative gradient becomes smaller than the adiabatic gradient . the region becomes radiatively stable again . in the 1.5m@xmath38 models , this succession of convective and radiative episodes goes on until the end of the main sequence . in the 1.7 and 1.9m@xmath38 models , after a few hundred megayears , iron accumulation has become large enough ( more than a factor of 15 ) , so that convective mixing can not reduce the iron abundance enough to stop convection . the iron convective zone then persists during the rest of the main sequence lifetime . in models including thermohaline convection , the situation is different . here again , there is a first phase during which mixing episodes alternate with convectively stable periods . afterwards the onset of thermohaline mixing strongly modifies the convective ( un)stability of the iron opacity bump region . iron accumulation is reduced and leads , most of the time , to a radiative gradient slightly below the adiabatic gradient . when , during stellar evolution , the accumulation of iron exceeds , somewhere in the opacity bump region , the critical value for the onset of convection , the induced mixing rapidly reduces the iron abundance and the radiative stability is restored . as a result the iron convective zone does not persist in these models . thermohaline convection is expected to take place inside stars any time inverted @xmath1-gradients are built . this happens in all types of stars where heavy element accumulation occurs in specific stellar layers , either due to radiative levitation or to nuclear reactions during stellar evolution . it also occurs in case of accretion of hydrogen poor material onto stellar surfaces , due to planet swallowing in early phases , or to accretion of matter from a companion . here we presented how introducing thermohaline convection modifies the results obtained for the accumulation of iron due to diffusion inside a - type stars . some other cases where this process has to be introduced were already discussed in previous papers ( e.g. @xcite ) . before giving our general conclusions in the next section , we first propose a discussion of three other cases where thermohaline convection should be introduced and for which we expect important consequences . the spb ( slowly pulsating b stars ) and the @xmath3 cephei stars are main sequence b type pulsators . the spb stars oscillate with periods between 0.5 to 5 days , while the @xmath3 cephei stars possess 2 to 8 hours - oscillations . the former are interpreted as high - order g - modes , the latter as low order p and/or g - modes . in the early nineties , the new opal opacity determinations @xcite allowed to solve the long standing enigma of b type pulsators . rosseland mean opacities computated from these tables were substancially increased in the z - bump region compared to the previously computed opacities . this opacity enhancement ( as large as a factor of 3 ) provided a boost for the @xmath0-mechanism potentially activated in the metal opacity bump region and could then explain the excitation of pulsations in @xmath3 cephei and spb theoretical standard models @xcite . however , as observational capabilities progress , b type pulsators offer new challenges to the theorists . some of them are now found in low metallicity environments ( e.g. * ? ? ? * and references therein ) . a large number of pulsation modes are indeed detected in b stars , with frequency spectra sometimes revealing new discrepancies between theory and observations . as an example , the two @xmath3 cephei stars , @xmath51 eri and 12 lac ( * ? ? ? * ; * ? ? ? * and references therein ) present frequency spectra which can not be reproduced with b type stars standard models ( e.g. for @xmath51 eri , * ? ? ? @xcite showed that an `` ad hoc '' factor 4 enhancement of the iron group elements , in the z - bump region could help reproducing the observed pulsation spectrum of @xmath51 eri . the combined effects of radiative levitation and gravitional settling provides a natural explanation for this local metal enrichment and could then explain the observation of more than 60 @xmath3 cephei stars in low metallicity environments . following the results of @xcite , @xcite presented preliminary evolutionary models of @xmath3 cephei stars evolved from the zams up to 1 myr : their evolutionary models did not include consistent diffusion computations but they included , at each timestep , a rough estimation of the diffusion - induced iron accumulation in the z - bump region . after one million years , their inferred iron enrichment is increased by a factor of 12 in the iron opacity bump . the stability analysis of these models showed that such iron enrichments makes the theoretical frequency spectra denser , which could help reducing the discrepancies between asteroseismic observations and theoretical stability analysis . following the same idea , @xcite carried out a parametric study of the effects of local iron enhancement on the stability of spb and @xmath3 cephei stars . the parametric iron accumulation profile introduced in their models is described by a gaussian function centered at @xmath44 , calibrated to be consistent with the iron accumulation profile deduced by @xcite in their a and f star models . as @xcite , @xcite showed that such accumulation profiles widen the theoretical spb and @xmath3 cephei instability strips and increases the number of excited modes . the computations by @xcite and @xcite reduced the discrepancies between theoretical and observed pulsations . however they did not discuss the credibility of the iron profiles introduced in their models , nor the stability of the induced @xmath1-gradients . on the other hand , @xcite analysed the effects of uncertainties in the opacity computations on the excitation of pulsation modes in b type stars . they compared models computed with opal opacities @xcite with models computed with the recently updated op opacity tables @xcite . the op updated data lead to an enhancement of 18% of the opacity in the z - bump . @xcite showed that these differences considerably affect the theoretical instability strips of @xmath3 cephei and spb stars . using the op opacities drastically increases the number of excited modes and significantly extends the theoretical instability strips . these new opacity computations could help solve , at least partly , the discrepancies between theoretical predictions and observations in @xmath3 cephei , spb and hybrid stars . up to now , no evolutionary models have been computed for b - type stars including consistent diffusion computations and the new op opacities . however in the light of @xcite results , we expect the iron accumulation derived by @xcite to recouncile observations and theory ( by a factor of 4 ) to be overestimated . in this context the iron enrichment inferred by @xcite ( by a factor of 12 ) for a 1 myr - model should also be too large , which suggests the presence of a `` forgotten '' process ( such as the thermohaline convection ) which could reduce the diffusion - induced iron accumulation . contrarily to spb stars , subdwarf b stars ( sdb stars ) are evolved , compact objects with low - mass cores and outer h - rich envelopes . they are experiencing the core helium burning phase on the extended horizontal branch . they all present chemical peculiarities attributed to gravitational settling and radiative levitation in the presence of weak stellar winds . subdwarf b stars host two groups of pulsators : the first one , composed of hot sdb stars , is characterised by rapid ocillations , with periods ranging from 80 to 600s , caused by low - order , low - degree p - modes . the second group , composed of the coolest sdb stars , oscillates with periods in the 2000 - 9000s range , due to high - order , low - degree g - modes . both types of pulsators are driven by the @xmath0-mechanism acting in the iron - peak element opacity bump . the relation between the envelope metal content and the pulsation driving mechanism was first established by @xcite for the short - period b subdwarf pulsators . they showed that a solar metal content in the envelope of their models is unable to excite modes but that models with envelopes containing an overabundance of metals can globally destabilize modes such as observed . they argued that this metal enrichment needs only to occur in the driving region itself and not necessarily in the whole envelope as assumed in their models . a natural explanation for the required local metal enhancement is the radiative levitation process , which is expected to act efficiently in the radiative envelopes of subdwarf b stars . in these stars , as in a and f stars , the competition between gravitational settling and radiative aceleration is expected to produce local accumulations of heavy elements . @xcite presented more sophisticated and realistic models of sdb stars , the so - called `` second generation models '' , including non uniform iron abundances . these envelope models rely on the assumption that a state of diffusive equilibrium is reached between radiative levitation and gravitational settling operating on iron ( disregarding other potentially competing processes ) , leading to stable non uniform iron profiles . @xcite demonstrated that such local iron accumulations in the z - bump region may produce efficient excitation for pulsation modes . similar models were successfully used for detailed asteroseismic studies of several sdb stars : they globally nicely reproduced the global properties of the pulsating sdb stars and were able to closely reproduce their observed periods @xcite . as emphasized by @xcite , these `` second generation models '' however suffer from several shortcomings . the theoretical instability strip predicted by these models is larger than the observed one and the observed period ranges in individual stars are also usually narrower than predicted . these discrepancies are likely due to an overestimate of the iron abundance in the z - bump region . in the `` second generation models '' of charpinet the opacities are computed using the opal opacity tables @xcite . as discussed previously for spb and @xmath3 cephei stars , the use of the most recent op opacities @xcite may also have a direct impact on the properties of the sdb pulsation driving mechanism . similarly to other types of stars @xcite , using op opacities instead of opal data , would increase the opacity in the z - bump , resulting in a more efficient pulsation driving mechanism . here again we find evidences that the iron accumulation needed to account for the observations must be lower than obtained by the atomic diffusion alone . introducing thermohaline convection could certainly help reducing the discrepancies . the @xmath2 doradus stars are a - f main sequence pulsators . their oscillations are interpreted as high - order g - modes with periods ranging from 0.35 to 3 days . they are driven by a flux blocking mechanism at the base of their convective envelope , as shown by @xcite with frozen convection models and by @xcite and @xcite with time - dependent convection models . it has been demonstrated that the `` convection blocking '' of radiation can drive high order g - modes only in stellar models for which the temperature at the bottom of the convective envelope , ranges between @xmath52k ( which corresponds nearly to the z - bump region central temperature ) and @xmath53k . the depth of the convective envelope plays a crucial role in the driving mechanism proposed by @xcite to explain the @xmath2 doradus pulsations . as a consequence , the seismic properties of @xmath2 doradus stars are quite sensitive to atomic diffusion including radiative levitation because of the way they influence the convective zones . as emphasized by @xcite , the diffusion effects on convection are of various kinds . the he settling which leads to h - enrichment in the external stellar layers induces an opacity increase which affects the depth of the surface convective zone . on the other hand , as discussed in previous paragraphs , radiative levitation produces reservoirs of iron - peak elements which may induce the appearance of an iron convective zone near @xmath52k . consequently the driving mechanism of @xmath2 doradus stars is expected to be sensitive to the thermohaline convection which occurs below the heavy elements accumulation zone . in the present paper , we tested the effect of thermohaline convection on the amplitude of iron accumulation which occurs inside a - f type stars due to atomic diffusion . we computed two series of models for three stellar masses : 1.5 , 1.7 and 1.9m@xmath38 . in both series , atomic diffusion was consistently computed for the elements h , he , c , n , o , ca and fe . the computations included the radiative accelerations on c , n , o , ca and fe , by using the semi - analytical prescription proposed by @xcite and improved by @xcite . no thermohaline convection was introduced in the first series of models . we verified that our results were similar to those obtained by @xcite and @xcite , with a large accumulation of iron in the iron peak elements opacity bump region . thermohaline convection was then introduced for the second series of models . considering the short time scales of this mixing process compared to the evolution time scales , and the fact that mixing occurs until the unstable @xmath1-gradients disappear , we introduced it in such a way that the @xmath1-gradients were kept close to zero or negative all along the evolutionary tracks . in practice , we used an extra diffusion coefficient , as proposed by @xcite , which is large enough to reach this purpose . we presented in detail the results obtained for the 1.7m@xmath38 case , during the main sequence phase . we also showed some results for the two other stellar masses considered here . the effect of thermohaline convection is very efficient in strongly reducing the heavy element accumulation in the opacity bump region . whereas the iron abundance can increase by up to factors larger than 100 when thermohaline convection is neglected , it never increases by more than a factor of 15 when it is taken into account . it is important to note that some iron accumulation always remains , even in the presence of thermohaline convection . the reason why this happens is directly related to helium settling , which leads to a stabilizing @xmath1-gradient . this stabilizing effect allows in turn some heavy material abundance increase up to the point where the @xmath1-profile becomes flat . as a consequence , the cases where iron accumulation has been invoked to account for specific observations in stars remain . indeed , in many cases the observations would be better explained with a smaller iron accumulation than that given by atomic diffusion alone . all these cases ( am stars , spb stas , sdb stars , @xmath3 cephei stars and @xmath2 doradus stars as discussed in section 5 ) will have to be studied individually for a detailed comparison with the observations . in the present computations , nickel was not included . although less important than iron , it may significantly contribute to the opacity and will have to be included in future work . no mass loss nor extra turbulence were taken into account . including such macroscopic motion may reduce the heavy element accumulation even without thermohaline convection . however , thermohaline convection is a physical process which must occur in any case every time the @xmath1-gradient is unstable . the final accumulation of elements thus strongly depends on this process , even in the presence of other macroscopic motion . in this paper , we have demonstrated the importance of thermohaline convection which has to be taken into account in the computation of the abundance variations induced by atomic diffusion . precise applications to specific stellar cases still have to be undertaken . they will lead to important consequences on the abundance anomalies observed in stars , as well as their oscillating properties through several processes like @xmath0-mechanism , convective blocking , etc . the possible applications of thermohaline convection in stellar astrophysics are vast and will certainly be a topic of interest in years to come . alecian , g. , & leblanc , f. 2002 , , 332 , 891 ausseloos , m. , scuflaire , r. , thoul , a. , & aerts , c. 2004 , , 355 , 352 badnell , n. r. , bautista , m. a. , butler , k. , delahaye , f. , mendoza , c. , palmeri , p. , zeippen , c. j. , & seaton , m. j. 2005 , , 360 , 458 balmforth , n. j. , cunha , m. s. , dolez , n. , gough , d. o. , & vauclair , s. 2001 , , 323 , 362 bourge , p .- o . , & alecian , g. 2006 , asp conference series , vol . 349 , 201 brassard , p. , billres , m. , charpinet s. , liebert , j. , & saffer , r. a. 2001 , , 563 , l1013 charbonnel , c. , zahn , j.p . , 2007 , , 476 , l29 charpinet s. , fontaine , g. , brassard , p. , & dorman , b. 1996 , , 471 , l103 charpinet s. , fontaine , g. , brassard , p. , chayer , p. , rogers , f. j. , iglesias , c. a. , & dorman , b. 1997 , , 483 , l123 charpinet , s. , fontaine , g. , & brassard , p. 2001 , , 113 , 775 charpinet s. , fontaine , g. , brassard , p. , green , e. m. , & chayer , p. 2005a , , 437 , 575 charpinet s. , fontaine , g. , brassard , p. , billres , m. , green , e. m. , & chayer , p. 2005b , , 443 , 251 charpinet s. , silvotti , r. , bonanno , a. , fontaine , g. , brassard , p. , chayer , p. , green , e. m. , bergeron , p. , et al . 2006 , , 459 , 565 charpinet s. , van grootel , v. , reese , d. , fontaine , g. , green , e. m. , brassard , p. , & chayer , p. 2008 , , 489 , 377 charpinet s. , fontaine , g. , & brassard , p. 2009 , , 493 , 595 cox , a. n. , morgan , siobahn , m. , rogers , f. j. , iglesias , & c. a. 1992 , , 393 272 dupret , m .- a . , grigahcne , a. , garrido , r. , gabriel , m. , & scuflaire , r. 2004 , , 414 , l17 dupret , m .- a . , grigahcne , a. , garrido , r. , gabriel , m. , & scuflaire , r. 2004 , , 435 , 927 dziembowski , w. a. , & pamyatnykh , a. a. 1993a , , 262 , 204 dziembowski , w. a. , moskalik , p. , & pamyatnykh , a. a. 1993b , , 265 588 gargett , a. , ruddick , b. , eds . 2003 , double - diffusion in oceanography ( oxford : pergamon ) , 381 grevesse , n. , & noels , a. 1993 , in origin and evolution of the elements , eds kubono , s. and kajino , t. , 14 guzik , j. a. , kaye , a. b. , bradley , p. a. , cox , a. n. , & neuforge , c. 2000 , , 542 , l57 handler , g. , jerzykiewicz , m. , rodrguez , e. , uytterhoeven , k. , amado , p. j. , dorokhova , t. n. , dorokhov , n. i. , poretti , e. , sareyan , j .- p . , parrao , l. , and 21 coauthors , 2006 , , 365 , 327 hui - bon - hoa , a. 2008 , , 316 , 55 iglesias , c. a. , & rogers , f. j. 1996 , , 464 , 943 jeffery , c. s. , & saio , h. 2007 , , 378 , 379 jerzykiewicz , m. , handler , g. , shobbrook , r. r. , pigulski , a. , medupe , r. , mokgwetsi , t. , tlhagwane , p. , & rodrguez , e. 2005 , , 360 , 619 kippenhahn , r. , ruschenplatt , g. , & thomas , h .- c . 1980 , , 91 , 175 kiriakidis , m. , el eid , m. f. , & glatzel , w. 1992 , , 255 , 1 koaczkowski , z. , pigulski , a. , soszyski , i. , udalski , a. , kubiak , m. , szymaski , m. , ebru , k. , pietrzyski , g. , woniak , p. r. , szewczyk , o. , & wyrzykowski , . 2006 , memorie della societa astronomica italiana , vol . 77 , 336 leblanc , f. , & alecian , g. 2004 , , 352 , 1329 michaud , g. 1970 , , 160 , 641 miglio , a. , montalbn , j. , & dupret , m .- a . 2007 , , 375 , 21 miglio , a. , bourge , p .- o . , montalbn , j. , & dupret , m .- a . 2007 , coast , 150 , 209 montalbn , j. , miglio , a. , & thado , s. 2007 , coast , 150 , 137 moskalik , p. , & dziembowski , w. a. 1992 , , 256 , l5 odonoghue , d. , lynas - gray , a. e. , kilkenny , d. , stobie , r. s. , & koen , c. 1997 , , 285 , 657 pamyatnykh , a. a. , handler , g. , & dziembowski , w. a. 2004 , , 350 , 1022 richard , o. , michaud , g. , & richer , j. 2001 , , 558 , 377 richard , o. , thado , s. , & vauclair , s. 2004 , soph , 220 , 243 richer , j. , michaud , g. , & turcotte , s. 2000 , , 529 , 338 rogers , f. j. , & iglesias , c. a. 1992 , , 79 , 507 santos , n.c . , israelian , g. , & mayor , m. 2001 , , 373 , 1019 seaton , m. j. 2005 , , 362 , l1 stancliffe , r.j . 2009 , , 394 , 1051 stancliffe , r.j . , glebbeek , e. , izzard , r.g . , & pols , o.r . 2007 , , 464 , 57 stancliffe , r.j . , & glebbeek , e. 2008 , , 389 , 1828 stankov , a. , & handler , g. 2005 , , 158 , 193 stothers , r. , & simon , n.r . 1969 , , 157 , 673 thado , s , alecian , g. , & leblanc , f. 2009 , in prep . thompson , i.b . , ivans , i.i , bisterzo , s. , sneden , c. , gallino , r. , vauclair , s. , burley , g.s . , shectman , s.a . , & preston , g.w . 2008 , , 677 , 556 turcotte , s. , richer , j. , michaud , g. , & christensen - dalsgaard , j. 2000 , , 360 , 603 ulrich , r. k. 1972 , , 172 , 165 vauclair , s. 1975 , , 45 , 233 vauclair , s. 2004 , , 605 , 874

it is well established that a number of early - type galaxies contain small amounts of cold interstellar gas and dust ( eg . knapp _ et al . _ 1989 ; colbert , mulchaey & zabludoff 2001 ; sadler _ et al . _ 2000 ) presumably obtained through mergers / accretion or stellar mass loss . what is not well known is the ultimate fate of this cold gas and dust . van gorkom et al . ( 1989 ) estimate that several hundredths of a solar mass per year may be falling onto and feeding the nuclear black hole in some early - type galaxies . it is also possible that the cold gas may be destroyed due to the presence of a hostile environment . on the other hand , if the neutral gas does not get stripped , destroyed , or fall into a central agn , it may settle into a disk and form stars . it has been argued by okuda _ et al . _ ( 2005 ) that gas disks in co - rich early - types are stabilized by their large dynamical masses , and so star formation is much less likely to occur in these galaxies than in late - type galaxies . however , many authors ( e.g. kauffman 1996 ; scorza & bender 1995 , 1998 ) agree that if a significant episode of star formation does occur in an early - type galaxy , then it may be transformed into something resembling a late - type galaxy . indeed , many early - types are known to contain small stellar disks at their centers ( scorza & bender 1998 ; emsellem _ et al . _ the possibility of star formation in early - type galaxies is important since it challenges the long held belief that early - type galaxies are passive evolvers . recently , kaviraj _ et al._(2006 ) used the galaxy evolution explorer ( galex ) to observe a large sample of low redshift early - type galaxies in the near ultraviolet ( nuv ) . they combine these observations with optical data from the sloan digital sky survey ( sdss ) and find that at least @xmath030 percent of the sample galaxies have uv to optical colors consistent with some recent ( @xmath4gyr ) star formation . their findings are in agreement with semi - analytical @xmath5cdm hierarchical merger models which predict that low redshift early - types will contain @xmath01 to 3 percent of their total stellar mass in stars less than 1 gyr old . they also determine that the observed blue colors in the star forming fraction of their sample can not be produced by assuming a `` monolithic '' model of evolution . furthermore , their results show that star formation is equally likely to occur in either high density or field environments . although these results strongly suggest that a large fraction of early - type galaxies are currently forming stars , no conclusions can be drawn as to the actual star formation rates ( which could show some environmental dependence ) . whether a galaxy is forming stars is sometimes traced by its placement on the fir - radio correlation known to exist in normal spiral galaxies ( young & scoville 1991 ; condon 1992 ) . the radio continuum emission in these galaxies arises in hii regions , type ii supernovae and the synchrotron emission from the electrons accelerated by those supernovae . dust heated by these same massive stars is thought to produce the fir emission from spiral galaxies . studies of interstellar medium of early - type galaxies have shown that their cold gas is physically similar to that of spirals ( rupen 1997 ; young & lo 1996 ; young 2002 ) . based on this , it is easy to imagine that star formation might be traced in early - types in the same manner as normal spiral galaxies . however , many early - types contain agn , and so they may have multiple sources of radio continuum . clearly , the total radio fluxes combined with a knowledge of the radio morphology are needed in order to disentangle these sources of radio emission . wrobel & heeschen ( 1988 ) were the first to suggest that several early - type galaxies might be undergoing significant star formation , based on both placement on the fir - radio correlation and the fact that their 6 cm continuum detections are extended on scales of 10 kpc or more . cotton and condon ( 1998 ) find support for this idea in the radio and fir flux ratios of e / s0s detected in the nrao vla sky survey ( nvss ) . about a quarter of them have flux ratios consistent with star formation activity . the other three quarters are presumably powered by agn . while studies of this kind can rule out the presence of a luminous agn , they can not definitively prove that the radio continuum is associated with recent star formation for several reasons . firstly , early - type galaxies are fundamentally different from spirals , with different stellar populations and different stellar dynamics , hence different types of energy input into the ism . secondly , there is evidence to suggest that post asymptotic giant branch ( agb ) stars can provide enough uv radiation to heat the dust and ionize the gas and account for the observed ir luminosities ( ferrari _ et al . _ 1999 ) . in other words , perhaps both the ir and the radio continuum could be traced back to an old stellar population . one way to determine if the radio continuum emission is associated with star formation is to compare high resolution continuum images with molecular gas maps . since molecular gas is the raw material for star formation , it constrains the location and the amount of star formation in the galaxy . if the radio continuum emission is extended and coincident with the molecular gas , that is strong evidence for star formation activity . on the other hand , if the continuum emission is well described by a point source , whereas the co is extended , then the radio continuum is more likely associated with an agn or perhaps a nuclear starburst than with normal disk - type star formation . if the radio emission is extended but in a very different location from the molecular gas , the situation is similar : the radio continuum is more likely to be powered by an agn or the evolved stellar population . up until now there has been very little knowledge of the co morphology in early - type galaxies . recently , studies by young ( 2000 , 2001 , 2002 , 2005 ) have produced the largest sample of high resolution molecular gas maps of early - type galaxies to date . in this paper we present a high resolution search for radio continuum emission from young s ( 2002 ) seven galaxy sample plus two dwarf early - types . the radio maps are then compared to high resolution co maps . this information is then combined with radio morphology and fir - radio correlation placement to trace recent star formation activity . we also evaluate the toomre q parameter in order to address the question of gravitational stability . this project allows us the opportunity to independently corroborate previous optical and radio studies that suggest star formation activity in early - type galaxies and to probe the origin of the radio - fir correlation in those galaxies . the layout of the paper is as follows : a sample description is given in section 2 . existing and new observations are discussed in section 3 and 4 , respectively . the results of the new observations are given in section 5 . radio continuum emission mechanisms are discussed in section 6 . the origin of fir emission is discussed in section 7 . in sections 9 and 10 , we discuss the implications of the results . in section 11 we present the paper conclusions . the iras satellite detected many early - type galaxies at 60 @xmath6 m and 100 @xmath6 m ( knapp et al . 1989 ) . the brightest ones ( 100 @xmath6 m flux @xmath7 1 jy ) have been searched for co using single dish telescopes ( wiklind , combes , and henkel 1995 ( hence forth wch95 ) ; lees et al . 1991 ; knapp & rupen 1996 ) . young ( 2002 ) recently completed a high resolution survey of the molecular gas from seven bright co detections of wch95 , using the berkeley illinois maryland association ( bima ) and the owens valley radio observatory ( ovro ) interferometers . of the seven early - types observed , five were detected and resolved in co. the molecular gas in these five galaxies is located in very regular , symmetric rotating disk with diameters ranging from 2 to 12 kpc . we chose the same seven - galaxy sample in which to image the radio continuum . we removed ngc 4649 from young s ( 2002 ) original sample since it has recently been determined that the fir from this galaxy is actually associated with its spiral companion ngc 4647 ( temi _ et al . _ 2004 ) , and also no co was detected by young ( 2002 ) . two dwarf elliptical companions of m31 , ngc 205 and ngc 185 , are added to the sample in order to explore the possibility of star formation in very low mass early - type galaxies . these two galaxies have been imaged and detected in co using the bima millimeter interferometer ( young 2000 , 2001 ) . all of the galaxies in our final sample have an r@xmath8 profile or `` a consistent classification as e or e / so in several catalogs '' ( wch95 ) . in some cases , these classifications are based on photographic evidence . recently , ccd imaging suggests that one of our sample galaxies may have a more complicated morphology ( ngc 5666 , donzelli & davoust 2003 ) so we prefer to refer to the galaxies in our sample as `` early - type '' galaxies . the sample galaxies have a variety of properties and are found in a variety of environments . some of the galaxies are isolated ; some are in loose groups , and some are deep within the virgo cluster . some have very smooth , regular morphologies , whereas others are classified as `` peculiar '' because of dust lanes , stellar shells and ripples , etc . the sample galaxies have optical luminosities that range from 10@xmath9 to 10@xmath10 solar luminosities . table 1 contains ephemeris information for our galaxy sample . vla data with matching resolution and signal to noise at 20 cm to that of young s ( 2000 , 2001 , 2002 ) co maps exist for only three of the sample galaxies ( ngc 3656 , ngc 7468 , and ngc 4476 ) . ngc 4476 suffers from confusion in the nvss but is located within a deep multiconfiguration ( d , c , and b ) vla map of m87 at 20 centimeters ( f. owen , priv . comm . 2001 ) . we obtained new vla 1.4 ghz c configuration data for the galaxies that are not detected in the nvss . for the purposes of spatial comparisons with young s ( 2002 ) co maps , new 20 cm observations were also obtained for ngc 807 , ngc 5666 , and ugc 1503 using the vla b array . we also re - observed ngc 3656 using the vla b array in order to produce a map with similar noise level to the observations just mentioned . both the c and b array observations were made using two if pairs , each consisting of 50 mhz each in left and right circular polarizations centered on two different frequencies . this gives a total bandwidth of 200 mhz . the requested rms noise for these observations is 0.01 mjy beam@xmath2 . the main parameters of the observed fields are listed in table 3 . the data were edited , calibrated , and imaged using standard tasks in the nrao astronomical image processing system ( aips ; see van moorsel , kemball , & greisen 1996 for an up - to - date description ) . in the case of ngc 807 it was necessary to combine the c and b array data to minimize confusion due to the bright sidelobes of a nearby bright source . the integrated radio fluxes , position angles , major axes , and minor axes for our detected sources ( see table 4 ) were determined using the miriad software package ( sault , teuben , & wright 1995 ) . position angles and deconvolved source sizes were determined by fitting elliptical gaussians to the radio emission . integrated fluxes were summed in polygonal regions and in concentric ellipses . great care was taken to insure that no confusing sources contributed to the integrated flux . a comparison of the co and radio is discussed for each galaxy below : * ngc 7468 * - ngc 7468 is a member of a small group of galaxies . it has been characterized as a very peculiar galaxy because of the presence of elliptical isophotes surrounding a nucleus that has been resolved into at least two sources of emission aligned along the major axis of the galaxy ( deeg , duric , & brinks 1997 ) . another extended optical feature connected with this galaxy is visible approximately 30 to the southwest ( nordgren _ et al . _ 1995 ) . et al . _ ( 1990 ) consider this galaxy to be a possible polar - ring galaxy , with these structures being part of the polar ring . recent broad band optical and h@xmath11 images reveal the presence of several hii regions coincident with the two central optical components as well as h@xmath11 emission associated with the polar ring ( cair@xmath12s _ et al . _ 2001 ; paz , madore , & pevunova 2003 ) . duprie & schneider ( 1996 ) detect a large hi disk ( d@xmath0 11@xmath13 ) using the 300 meter arecibo telescope . ngc 7468 has been observed in co at several single dish telescopes with varying results . this galaxy has not been detected in the 2 - 1 line of co ( lees _ et al . _ 1991 ) , but there are several tentative detections from the iram 30 meter telescope in the 1 - 0 line ( wch95 ; sage , welch & young 2006 ) . interferometric observations at 3 mm produced a non - detection ( young 2002 ) . however , an upper limit ( derived from summing over the same spatial scale and velocity area as that of the single dish results ) produces a flux consistent with that measured in the most recent single dish observations . this is highly suggestive that the single dish detections may in fact be real , and that there is a high probability that ngc 7468 contains some molecular gas . ngc 7468 already has sufficiently high resolution data in the vla archive ( cox & sparke 2004 ) . in both sets of archive data the image quality suffers due to confusion from a bright source to the north . the poor image quality is more pronounced in the 20 cm c array data and is perhaps due to its poorer uv coverage . the emission in this image is elongated in the same manner as the beam . both images show that the radio emission in this galaxy is resolved into two large regions ( see figure 1 ) . the northern blob is much larger , and contains more than 80@xmath14 of the total flux . the radio emission is coincident with the two central features observed in the aforementioned optical and h@xmath11 images . no radio continuum is detected along the polar ring . both sets of data show some low level radio emission that appears to extend outside the eastern part of the galaxy . the image quality is too poor to say whether this emission is real or just noise . perhaps deeper vla observations can alleviate this difficulty in the future . * ngc 3656 * - this is a field early - type that shows indications of a gravitational interaction sometime in the near past ( balcells _ et al . _ 2001 ) . there are several small spherical objects , presumed to be dwarf galaxies , surrounding ngc 3656 ( wch95 ) . young ( 2002 ) favors the explanation that the molecular gas was acquired during a merger event . both the co and radio emission peak at the optical center , are elongated in the same position angle and have similar minor axis widths . however , the radio emission is less extended in radius ( by about 10 ) than the co in the northern parts of the galaxy . the radio continuum also does not follow the co extension curving off the south east side of the molecular disk . the extension probably traces gas that has not yet reached dynamical equilibrium and so has not settled into the disk ; we suspect that if the gas is not yet settled , star formation is unlikely to occur there . a comparison of the molecular gas and the radio continuum can be found in fig 2 . * ngc 5666 * - this is an isolated , low - luminosity early - type galaxy . this galaxy is known to have an extended hi disk ( lake , schommer , & van gorkom 1987 ) , and extended diffuse radio emission ( wrobel & heeschen 1988 ) . the radio continuum in ngc 5666 is extended on scales of a few kpc , and matches the co extent along both axes rather well . the co is centrally peaked whereas the continuum intensity peak ( @xmath02 mjy beam@xmath2 ) is not located at the center , but to the southeast . a slice through the major and minor axes of the radio emission shows that the radio emission drops to @xmath01.5 mjy beam@xmath2 inside a roughly circular region , @xmath0 7 in diameter , centered on the optical nucleus . the co and radio continuum data are at matched resolution and so the absence of a central hole in the co emission suggests that the underlying emitting structures are different , and so the mismatch at the center is not the result of a resolution effect . a comparison of the molecular gas and the radio continuum can be found in fig 3 . a recent @xmath15-band image of ngc 5666 from donzelli & davoust ( 2003 ) shows a ring of star formation , 10 in diameter and coincident with the ring of radio continuum emission in figure 3 . this irregular ring contains some fragments which are suggestive of being spiral arm segments , all within the region where co emission is found . the galaxy s spectrum is dominated by an old stellar population , however ( donzelli & davoust 2003 ) . thus it appears that some flocculent spiral structure may be developing in the center of this gas rich early type galaxy . * ugc 1503 * - this is a an isolated early - type galaxy . the radio emission in this galaxy is extended on scales of @xmath05 kpc , and is centered on the optical center of the galaxy . all of the emission is contained in a ring with a central circular depression ( @xmath05@xmath16 in diameter ) where the radio emission nearly drops to zero . the peak of the radio emission ( @xmath0 0.3 mjy beam@xmath2 ) occurs in several places along the ring whereas the co emission is centrally peaked . the extent of the co and radio emission are roughly coincident along both the galaxy s major and minor axes . a comparison of the molecular gas and the radio continuum can be found in fig 4 . * ngc 807 * - this is an isolated giant early - type galaxy . the co distribution is asymmetric with most of the molecular gas ( @xmath0 70@xmath14 ) located in the southeast section of the galaxy . young ( 2002 ) suggests that shear from the differential rotation in the disk would destroy the asymmetry on a time scale of 10@xmath17 years . since the asymmetry is still present , this may be an indication that the gas was acquired in an interaction less than 10@xmath17 years ago . the radio continuum in ngc 807 is extended on scales of 9 kpc , and matches the co extent along both axes rather well . there is an asymmetry in the distribution of the continuum much like that of the co distribution ( most of the emission , @xmath0 60@xmath14 , is located in the southeast ) . in fact the radio emission ( @xmath00.3 mjy beam@xmath2 ) peaks at roughly the same location as the co emission . however , it may be possible that the perceived asymmetry in the co is in reality due to differing co - to - h@xmath18 conversion factors existing on alternate sides of the galaxy . if this is the case , then the correlation of the two emission peaks may be nothing more than a coincidence . however , the high degree of morphological similarity between the co and radio continuum present in several of the other sample galaxies suggest it is more than a coincidence . a comparison of the molecular gas and the radio continuum can be found in fig 5 . * ngc 185 * - this is a dwarf elliptical companion of m31 . the molecular gas in this galaxy is found in large complexes similar to those observed in the milky way ( young 2000 , 2001 ) . an hi and optical study of the interstellar medium ( ism ) by young & lo ( 1996 ) shows that complex dust regions near the center of the galaxy are coincident with the detected co emission , and the hi emission exhibits a very clumpy distribution . we detect a small amount of radio emission ( peak intensity is 5@xmath19 ) in our image of ngc 185 . the small patch of emission is slightly offset from the center of the galaxy and has a flux of @xmath20jy ( see figure 6 ) . the emission is roughly coincident with a region of extended h@xmath21 emission detected by young & lo ( 1997 ) . the surface brightness , luminosity , diameter , and morphology of the h@xmath11 emission are within the ranges of known supernova remnants ( e.g. , long _ et al . _ 1990 supernova remnants in m31 have 20 cm radio fluxes that range from s@xmath22=150 - 2600 @xmath6jy ( braun & walterbos 1993 ) . since m31 and ngc 185 are at relatively the same distance we conclude that the small patch of radio emission probably originates from a supernova remnant inside the galaxy . the small spatial offsets between the co detections and the radio continuum emission would not be noticeable in the other sample galaxies at 1860 mpc . * ngc 4476 * - this galaxy is a small early - type 12@xmath13 in projection from m87 near the center of the virgo cluster . the stellar kinematics are normal for an elliptical galaxy with a central velocity dispersion of 132 km s@xmath2 . dust is clearly seen on optical photographs , showing the presence of a nuclear dust - ring ( tomita _ et al . _ 2000 ) . young ( 2002 ) favors an external origin for the detected co emission based on the fact that the specific angular momentum of the gas is three times the specific angular momentum of stars . vla 20 cm continuum observations of ngc 4476 in the b , c , and d configurations yield an upper limit of 5@xmath2310@xmath24 w m@xmath25 hz@xmath2 or 0.5 mjy ( f. owen , priv . comm . no previous radio detections have been reported for this galaxy . * ngc 205 * - this galaxy is another dwarf elliptical companion to m31 . similarly to ngc 185 , the molecular gas in this galaxy is found in large complexes similar to those observed in the milky way ( young 2000 , 2001 ) . an hi and optical study of the interstellar medium ( ism ) by young & lo ( 1996 ) shows that complex dust regions near the center of the galaxies are coincident with the detected co emission , and the hi emission exhibits a very clumpy distribution . no radio continuum emission was detected in the image for ngc 205 . an upper limit to the radio continuum brightness in this galaxy is determined from a 3 sigma upper limit of the rms noise ( @xmath0 0.02 mjy beam@xmath2 ) . the flux for ngc 205 is located in table 4 . no previous radio detections have been reported for this galaxy . there are two major components of the radio continuum emission in galaxies . these are synchrotron emission ( non - thermal ) and free - free emission ( thermal bremsstrahlung ) . in star - forming galaxies , synchrotron emission comes from cosmic ray ( cr ) electrons spiraling in the magnetic field of a galaxy . type ii and type ib supernovae and their remnants are thought to be the main electron accelerators ( kennicutt 1989 , condon 1992 ) . often , the synchrotron spectra can be approximated by a power law in some limited frequency range . when this occurs , the spectral index @xmath11 is defined as s@xmath26 , where s is the integrated flux density and @xmath27 is the frequency . thermal emission originates from shocked regions around supermassive stars where superheated electrons collide with the ism . usually the optically thin thermal emission dominates over the optically thick . optically thin thermal radiation has a characteristic spectrum ( @xmath280.5 ) , and so it should be distinguishable from the steeper - spectrum non - thermal emission . typically @xmath11 is determined by comparing the total flux densities obtained at two frequencies with closely matching resolutions . for example : @xmath29 the available data allow us to calculate the spectral index for three of our sample galaxies : ngc 5666 , ngc 7468 , and ngc 3656 ( see table 4 ) . the spectral indices determined from the total fluxes in the two frequency bands for these three galaxies range from 0.5 to 0.8 . pixel by pixel maps of the spectral index do not show any evidence of a central point source of significantly different spectral index . we note here that in order to avoid variability effects , this type of comparison should be done using observations made with identical visibility coverage on the same day . we have not done this , so the calculated spectral indices are only rough estimates . the high resolution radio images rule out nuclear monsters both spectroscopically ( @xmath30 1 for luminous agn ; zhang , lin , & fan 2003 ) and morphologically ( only ngc 3656 exhibits a centrally peaked radio distribution ) . it is possible that our galaxies contain very low radio power agn . we estimate the limit to the flux that an unresolved agn would contribute to the total radio power by measuring the radio flux contained inside one beam centered on the optical centers of our galaxies . we find that these limits are only @xmath010 percent of the total flux and so any contribution from a central point source to the total radio power is small . the correlation between the fir luminosity and the radio continuum emission observed for normal spiral galaxies is extremely tight and roughly linear ( @xmath4q@xmath312.3 at 1.4 ghz ; helou , soifer , @xmath32 rowan - robinson 1985 ) which implies that the process of star formation produces both types of emission . it is assumed that the fir and radio emission comes from dust heated by massive stars and cr from supernovae remnants of the same massive stars , respectively . the fir - radio correlation has been observed to be non - linear at low frequencies where nonthermal emission dominates ( @xmath335 ghz ) as well as for optically - selected samples containing low - luminosity ( @xmath34 ) galaxies ( fitt , alexander , & cox 1988 , devereux @xmath32 eales 1989 , condon 1992 ) . one explanation for this is that low - luminosity galaxies are less massive than normal galaxies , and are more likely to lose their cr to convection ( chi & wolfendale 1990 ) . subsequently , a small galaxy would appear under - luminous in the radio with respect to the fir emission , and the correlation would become non - linear even if the low luminosity galaxies are forming stars . it has also been shown that galaxies with low l@xmath35 also drop off the fir - radio correlation ( yun reddy & condon 2001 ) . deviations from the correlation may also be due to changes in the surrounding environment such as the presence of an intercluster medium ( icm ) . the magnetic fields in galaxies can become compressed as they move through a dense icm . this compression may significantly alter the amount of observed synchrotron emission ( reddy & yun 2003 ) and should enhance the radio continuum emission in the galaxy , making it appear over - luminous in the radio . the fir flux can be approximated from the iras 100 and 60 micron flux densities using the following relation : @xmath36 where s@xmath37 and s@xmath38 are the 100@xmath6 m and 60@xmath6 m fir fluxes in janskys ( condon 1992 ) . helou , soifer , & rowan - robinson ( 1985 ) defined the ratio between the far - infrared and radio continuum emission as follows : @xmath39 for @xmath27=1.4 ghz . an evaluation of the ratio q for the galaxies in our sample show that five ( ngc 3656 , ngc 5666 , ngc 7468 , ngc 807 , and ugc1503 ) of the eight galaxies sit within 3@xmath19 of the radio - fir correlation defined for star forming spirals . the median q value for the sample galaxies detected in radio continuum is @xmath4q@xmath312.7@xmath400.1 ( see figure 7 ) . we would also like to note that our sample galaxies with the lowest optical luminosities ( @xmath41 ) also have the lowest fir luminosities . condon , cotton , and broderick ( 2002 ) find that nearly all galaxies with spectroscopically or morphologically determined agns have a fir spectral index @xmath42 @xmath4 1.5 . the fir spectral index can be determined as follows : @xmath43 the galaxies in our sample have fir spectral indices greater than 1.5 ( see table 5 ) , which is another indication that these galaxies do not contain agn or at least the fir emission does not have a significant contribution from a dust enshrouded agn . this result is consistent with the radio morphologies of the sample galaxies , which also indicates that luminous agn are not present in these galaxies . the star formation rate ( sfr ) is calculated using the method derived by condon , cotton , and broderick ( 2002 ) for spiral galaxies . the fir luminosity is proportional to the sfr as : @xmath44\ ] ] this is the star formation rate for stars of m @xmath45 5 m@xmath1 . to calculate the star formation rate for all the stars between 0.1 to 100 m@xmath1 we simply multiply this result by a factor of 5.5 ( for a salpeter initial mass function ) . the average star formation rate for stars between 0.1 and 100 m@xmath1 for the sample galaxies detected in radio continuum is @xmath0 2 m@xmath1 year@xmath2 ( see table 5 ) . the gas depletion time scale can be determined by dividing the h@xmath46 masses found by young ( 2002 ) by the star formation rate . the h@xmath46 masses for the sample galaxies can be found in table 6 . the average depletion time scale for our sample galaxies detected in co is @xmath0 0.7 gyr ( see table 5 ) which is much smaller compared to typical gas depletion times scales found in spiral galaxies ( @xmath0 2 gyr ; kennicutt 1983 ) . the smaller depletion timescales may indicate that star formation in early - type galaxies is more efficient than in normal spiral galaxies . there are several other possibilities for the smaller depletion time scales . the star formation could be episodic with prolonged periods between star formation events either due to differing ism conditions throughout the galaxies or to the fact that some molecular gas is still in falling from outside of the galaxies ( pipino _ et al . _ 2005 ; ferreras & silk 2003 ; and many others ) or both . it is also possible that the gas masses are incorrectly estimated ( this will be discussed further in section 10 ) . the five best candidates for star formation in our sample are ngc 3656 , ngc 5666 , ngc 807 , ngc 7468 , and ugc 1503 . these galaxies sit within 3@xmath47 of the radio - fir correlation defined for star forming spirals . ngc 3656 , ngc 5666 , ngc 807 , and ugc 1503 have co emission that is roughly coincident with the observed radio continuum emission . in fact , the distribution of the continuum emission in these galaxies resembles those of normal spiral galaxies , and probably traces a ring of star formation . ngc 3656 has a star formation rate a factor of two greater than the average for our sample . we might expect this since ngc 3656 has undergone some sort of interaction in the past , which may have produced tidally - induced star formation . optical and h@xmath11 images show that ngc 7468 possess star forming regions much like those in our own galaxy ( cair@xmath12s _ et al . _ the radio emission is coincident with two large patches of h@xmath48 emission . ngc 7468 sits right on the radio - fir correlation . however , the star formation rate for ngc 7468 ( 0.8 m@xmath1 year@xmath2 ) is approximately a factor of three less than the average of our sample . this smaller sfr could indicate that most of the molecular gas has been used up in a recent star formation event , and the sfr has slowed down to its present value . the lower star formation rate is also consistent with the fact that this galaxy contains two orders of magnitude less molecular gas than the other galaxies in the sample . our results indicate that star formation is occurring in more than half of our sample galaxies . we note here that our galaxy sample is fir and co selected , and so it may not be representative of early - type galaxies . however , if we consider all of the early - type galaxies in the sample of cotton , condon & broderick ( 2002 ) we find that many of them do lie on the radio - fir correlation ( see figure 7 ) . their galaxies are both radio and fir selected , not co selected . full detections that sit within 3@xmath19 of @xmath4q@xmath7@xmath492.3 include @xmath015 galaxies , suggesting that our sample galaxies may not be that unusual . interferometric images of the molecular gas in ngc 185 and ngc 205 find objects much like the giant molecular clouds in our own galaxy , in terms of sizes and masses ( young 2000 , 2001 ) . strangely , these galaxies show significant deviations from the spiral galaxy radio - fir correlation in the sense that they are radio faint ( they are the two symbols at the bottom left of fig 7 ) . the radio nondetection found for ngc 205 and the small amount of radio emission detected in ngc 185 are consistent with studies of low surface brightness dwarfs ( hoeppe et al . ngc 205 and ngc 185 most likely deviate from the radio - fir correlation because they do not posses enough mass to retain all of their cr . the superbubbles created by supernovae are able to expand out of the galaxy , compressing the magnetic field . it is easier for the cr to escape if the magnetic fields are located in shocks produced by the expanding bubbles ( chi @xmath32 wolfendale 1990 ) . also , the sfr may be slow enough so that the cr from defunct snr have already long since escaped the galaxy ( klein _ et al . _ 1992 ) . ngc 4476 is the most puzzling galaxy in our sample . there is certainly enough molecular gas in this galaxy to form stars . why then is there no detected radio emission ? the star formation rate in this galaxy is low , but @xmath50 is similar to the star - forming ngc 807 . one explanation for the lack of radio continuum may be that most of the atomic hydrogen has been removed due to ram pressure as it travels through the icm surrounding m87 ( lucero , young , & van gorkom 2005 ) . perhaps the bulk of the magnetic field has been removed along with the hi in the stripping event , and consequentially the radio is not observed . a similar phenomenon appears to to be occuring in the virgo cluster spiral galaxy ngc 4522 . this galaxy has a large amount of extraplanar hi , and radio continuum is indeed associated with the extraplanar gas ( vollmer _ et al . _ 2007 ) . okuda _ et al . _ ( 2005 ) recently have shown that the giant radio galaxy 3c 31 ( ngc 383 ) contains a massive molecular gas disk , though little star formation activity has been found ( owen , odea , & keel 1990 ) . et al._(2005 ) argue that star formation is not occurring because the galaxy s deep potential stabilizes the gas disk against collapse , an idea which was also put forward by kennicutt ( 1989 ) for early type galaxies in general . the stability of a rotating gas disk can be assessed quantitatively by toomre s criterion as revised by sakamoto _ et al . _ ( 1999 ) , @xmath51 where @xmath52 and where @xmath19 is the velocity dispersion , g is the gravitational constant , @xmath53 is the gas surface density , @xmath54 is the angular velocity , v is the gas velocity , and @xmath55 is the epicyclic frequency . writing the @xmath3 parameter in this way elucidates the role of the galaxy s dynamical mass and the gas to dynamical mass ratio ; values of @xmath56 indicate gravitational stability . for the estimation of @xmath3 we assume our galaxies have an axisymmetric potential in which the velocity of the gas either rises linearly ( @xmath57 ) or is constant ( @xmath58 ) out to the edge of the gas disk ( binney & tremaine 1987 , pp . 120 - 122 ) . gas and dynamical masses are taken from young ( 2002 ) and can be found in table 6 . molecular hydrogen masses are multiplied up by a factor of 1.36 to account for the presence of helium . the true velocity dispersion of the gas is not known so a typical value of 10 km s@xmath2 is assumed . the ratios m@xmath59/m@xmath60 for this sample range from 0.02 to 0.1 , similar to the values found by koda _ et al . _ ( 2005 ) in early type spirals . the values of the @xmath3 parameter range from 0.8 to 4 under these assumptions . only one of the estimated @xmath3 values is less than 1 ( indicating instability ) , but the remainder are sufficiently close to 1 that they may also be consistent with gravitationally unstable disks . the simplified analysis described above neglects the destabilizing influence of the stars in the galaxy . jog & solomon ( 1984 ) show that the interactions between gas and stars can make the two - fluid system gravitationally unstable even if the gas is stable when considered by itself ( as is done above ) . thus the values of @xmath3 which are somewhat higher than 1 should not necessarily be interpreted as stability . we must also note that local surface densities are undoubtedly higher than average at some points in the galaxies and lower at others . a more detailed analysis is precluded by the fact that @xmath3 scales with the velocity dispersion , which might be a factor of two different from what we have assumed ( e.g. okuda _ et al . _ 2005 ; jog & solomon 1984 ) . we have shown that the existence of a toomre - type gravitational instability in these molecular disks can not be ruled out , and the moderate to low @xmath3 values found here may be consistent with the other evidence suggesting star formation activity in these galaxies . in particular , ngc 3656 has the lowest @xmath3 value in the sample , and it has one of the highest star formation rates and @xmath50 values . recently it has been suggested that that the star formation process itself may have an effect on a galaxies position on the radio - fir correlation . several studies comparing the spatial extent of radio continuum and fir emission in spiral galaxies suggest that the diffusion scale length of cr electrons is much larger than the mean free path of the dust heating photons ( bicay & helou 1990 ; murphy _ et al . _ this means that the continuum emission will become more and more smoothed out compared to the fir emission as the cr diffusion timescales increase . these studies also show that the cr diffusion timescales tend to increase as the disk averaged sfr in the galaxy decreases ( murphy _ et al . _ so if a galaxy has a sfr below some threshold value , its radio continuum emission could become so smoothed out that it may be hard to detect . this effect would be amplified if a galaxy is not massive enough to retain its cr , has had its magnetic field removed , or experiences episodic star formation . it is interesting to note that the galaxies in our sample with the smallest sfr ( @xmath61 0.2 m@xmath62 yr@xmath2 ) also appear to contain little to no radio continuum . it may be that these galaxies deviate from the radio - fir correlation simply because their sfr are too low . if the star formation proceeds at rates inferred from the far infrared emission , the transformation from gas to stellar disks will occur over a few billion years . according to young ( 2002 ) , these disks will be rotationally supported , on the order of a few kpc , will contain 1@xmath14 of the total stellar mass , and will be very similar to the stellar disks known to exist in many early - types ( scorza & bender 1995 ; emsellem _ et al . _ our results clearly show that stellar disks can grow out of gas disks in at least some cases . the gas in these disks could be produced by either internal processes ( temi _ et al . _ 2007 ) or it could be left over from mergers ( barnes 2002 ; naab , jesseit & burkert 2006 ) . in other words , the presence of a star forming gas disks in spheroidal galaxies is consistent with both the early - assembly ( monolithic collapse ) and the late assembly ( hierarchical growth ) pictures . while our results can neither confirm of rule out either scenario they do emphasize that the morphology of an early - type galaxy , parameterized by its bulge - to - disk ratio , may evolve somewhat through the process of star formation even while the galaxy remains on the red sequence . it would be very interesting to see if stellar disks are detectable in early - type galaxies at high redshift . the h@xmath46 mass in galaxies is estimated using co(1 - 0 ) line emission . the standard conversion factor used for molecular clouds in the milky way is given by n(h@xmath46)/i@xmath63 = 3@xmath2310@xmath64 cm@xmath25 k km s@xmath2)@xmath2 ( sanders , solomon , @xmath32 scoville 1984 ) . the use of this conversion factor assumes that the mean properties of the molecular gas in the early - type galaxies in the sample are similar to those in the milky way . the molecular clouds in early - types might not exhibit the same properties , such as temperature and density , as those found in the milky way ( georgakakis _ et al . _ it is expected that such differences will modify the co to h@xmath46 conversion factor . if the molecular gas densities in early - types are larger than in the milky way , the conversion will underestimate the the h@xmath46 mass . if instead the densities are the same , but the temperatures are lower than those in the milky way , the conversion factor will overestimate the h@xmath46 mass . a modified h@xmath46 mass will affect the depletion timescales as well as the gas mass to dynamical mass ratios . if gas masses are overestimated then the depletion timescales become even faster , and the gravitational stability greater . in order to achieve depletion time scales like those in spiral galaxies the gas mass would have to be underestimated by a factor of 4 . in this paper we present new high resolution radio continuum observations of eight fir and co selected early - type galaxies , and compare them to high resolution co maps in order to investigate the importance of star formation in these galaxies . in summary : + + @xmath66 five of the eight galaxies ( ngc 807 , ngc 3656 , ngc 5666 , ngc 7468 and ugc 1503 ) sit within 3@xmath47 of the radio - fir correlation defined for star forming spirals . + + @xmath66 the radio continuum is extended on scales of a few to @xmath09 kpc . good spatial correlations between the co and radio emission exist in four of the sample galaxies ( ngc 807 , ngc 5666 , ugc 1503 , and ngc 3656 ) . this morphological similarity confirms wrobel & heeschen s ( 1988 ) suggestion that early - type galaxies on the fir - radio correlation may be forming stars . many of the early - type galaxies in the nvss catalogue also sit on the radio - fir correlation . these results are in agreement with other recent work suggesting that star forming early - type galaxies are more common than has been thought in the past . + + @xmath66 the average star formation rate for the galaxies detected in radio continuum is @xmath0 2 m@xmath1 year@xmath2 . the average gas depletion time scale for the galaxies detected in co is @xmath0 0.7 gyr , which is much less than the gas depletion time scales observed in star forming spiral galaxies ( @xmath0 2 gyr ) . + + @xmath66 a simple gravitational stability estimate suggests that the molecular disks are likely to be unstable , driving the inferred star formation activity . + + @xmath66 the molecular gas in our star forming early - type sample galaxies will produce stellar disks containing @xmath0 1@xmath14 of the total stellar population , and radii of a few kpc . + + @xmath66 a few of the sample galaxies ( ngc 4476 , ngc 185 , and ngc 205 ) deviate significantly from the fir - radio correlation in that they are radio faint . in the case of the two dwarf early - types , their small masses may allow a significant number of cosmic rays to escape . in the case of ngc 4476 , the large scale magnetic field may have been removed along with hi gas due to the passage through a hot intercluster medium . we thank the anoymous referee for a very constructive report , and joan wrobel for her insightful comments which have made this work much stronger . 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 to the national aeronautics and space administration . last but not least , this work has been supported by nsf grants ast-0074709 and ast-0507432 to new mexico institute of mining and technology . lisa m. young thanks new mexico tech for a sabbatical leave and is grateful to the university of oxford sub - department of astrophysics for their hospitality . lcccccl + galaxy & type@xmath67 & ra & dec & l@xmath68 & distance & environment + & & & & ( 10@xmath17 l@xmath69 ) & ( mpc ) & + ngc185 & de3p & 00h38m57.5s & 48d20m12s & 0.057 & 0.63 & group + ngc205 & e5p & 00h40m22.0s & 41d41m07s & 0.30 & 0.85 & group + ugc1503 & e & 02h01m19.8s & 33d19m46s & 17 & 69 & field + ngc807 & e & 02h04m55.7s & 28d59m15s & 32 & 64 & field + ngc3656 & io & 11h23m38.4s & 53d50m31s & 16 & 45 & merger remnant + ngc4476 & e / so & 12h29m59.1s & 12d20m55s & 3.5 & 18 & virgo cluster + ngc5666 & ec & 14h33m09.2s & 10d30m39s & 6.3 & 35 & field + ngc7468 & e3p & 23h02m59.2s & 16s36m19s & 4.0 & 28 & group +

_ cosmic magnetism _ , the origin and nature of magnetic fields in our universe , is one of outstanding problems of modern astrophysics ( see * ? ? ? exploration of the galactic magnetic field ( gmf ) and the intergalactic magnetic field ( igmf ) is listed as one of key science projects for the square kilometer array ( ska ) , and is one of the important science projects for jansky very large array ( jvla ) , murchison widefield array ( mwa ) , the low frequency array ( lofar ) , the australian ska pathfinder ( askap ) , and the south african karoo array telescope ( meerkat ) ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? measuring faraday rotation , the rotation of the plane of linearly - polarized radio emission due to the birefringence of magneto - ionic medium , is a powerful method to observe the gmf as well as the igmf . up to now , there have been a number of studies of faraday rotation measure ( rm ) to elucidate the structure and statistical properties of the gmf and the igmf . for the igmf , there have been a number of observations of rm in clusters of galaxies ( see * ? ? ? * for a review ) . observations have been also extended to the outside of clusters , toward the cosmic web , and extragalactic contributions of rm due to the igmf or others have been discussed ( see , e.g. , * ? ? ? ? * ; * ? ? ? * ; * ? ? ? there are several theoretical works for rm in filaments of galaxies @xcite . for instance , @xcite investigated rm in filaments using a model igmf based on a turbulent dynamo @xcite . they found that the root - mean - square ( rms ) value of rms through a single filament is expected to be @xmath3 in the local universe . @xcite extended this work by using the redshift distribution of polarized background radio sources against which rm is observed . they found that the rms value of rms through filaments up to redshift @xmath4 would be @xmath5 . they also found that the second - order structure function ( sf ) has a nearly - flat profile in angular separations of @xmath6 , meaning that rms through filaments decorrelate on angles less than @xmath7 . recently , the rms toward the north and south galactic poles ( ngp and sgp , respectively ) have been investigated to study the gmf as well as the igmf . @xcite studied the rm data from the nrao vla sky survey ( nvss ) , and estimated non - zero vertical strengths of the gmf , about @xmath8 and @xmath9 toward the ngp and sgp , respectively . @xcite used rm data from the westerbork radio synthesis telescope ( wsrt ) and the australia telescope compact array ( atca ) , and found that the median value of rms toward the sgp is @xmath10 ( corresponding to a vertical gmf strength of @xmath11 ) , while that toward the ngp is @xmath12 ( @xmath13 ) . the standard deviations of rms were @xmath14 and @xmath15 toward the ngp and sgp , respectively . @xcite also put an upper limit of @xmath16 on the strength of random magnetic field at high galactic latitudes . @xcite examined the nvss data in detail , and found that the second - order sfs of rm at angular separations of @xmath17 have a value @xmath18 toward the ngp and @xmath19 toward the sgp . in addition , based on the latitude dependence of rm , @xcite examined the galactic and extragalactic contributions to rm in the nvss data . he estimated that the galactic contribution is @xmath20 and the extragalactic contribution ( including those intrinsic to the polarized background radio sources and due to the igmf ) is @xmath21 for the northern ( southern ) hemisphere . the separation of the galactic and extragalactic contributions in observed rms , however , is not trivial . it requires good understandings of the gmf and its contribution to observed rms . there have been a number of theoretical works to model the galactic rm ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ( see also works on cosmic - ray propagation : * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the galactic rm arises from the global , regular component as well as from the turbulent , random component of the gmf . while the regular component has been modeled with analytic fitting formulae based on observations , the random component has been modeled using power - law spectra with random phases in fourier space . for instance , @xcite used the publicly - available hammurabi code @xcite , and adopted a kolmogorov - like power spectrum with average amplitude @xmath22 in a box of 10 pc size . they found that at galactic latitudes @xmath23 , the second - order sf has a magnitude of up to a few @xmath24 at angular scales of @xmath25 . previous studies have successfully reproduced the observed properties of the galactic rm as well as those of the radio continuum emission toward low and mid - galactic latitudes ( e.g. , * ? ? ? there is , however , a lack of studies that can be compared with recent observations of rms toward the galactic poles . in this paper , we simulate the galactic rm toward high galactic latitudes and investigate its statistical properties . while we model the regular component of the gmf based on a number of observations , we use the data of three - dimensional magnetohydrodynamic ( mhd ) turbulence simulations to model the random component . we obtain simulated maps of the galactic rm toward the galactic poles in a field - of - view ( fov ) of @xmath26 to compare with recent observations and in a @xmath27 fov to predict future ska observations . we estimate how much the gmf can contribute to observed rms and discuss how the statistics of simulated rms compare with those of observed rms . the rest of the paper is organized as follows . in section 2 , we describe our models . the results are shown in section 3 . discussion and summary follow in sections 4 and 5 , respectively . we first briefly describe our models for the global , regular components of the electron density and the gmf . the details along with the definitions of the coordinate systems used in the paper are described in appendices . we also briefly describe models for the electron temperature and the rms speed of random flow motions , which are used in modeling the electron density fluctuations and the turbulent , random magnetic field in the next subsection . for the electron density , @xmath28 , we employ the ne2001 model @xcite displayed in figure [ figure1 ] . we take the original parameters of the model except for the scale height of the thick disk , @xmath29 , and the mid - plane electron density for the thick disk , @xmath30 ; we use @xmath31 and @xmath32 @xcite , which better reproduce both the dispersion measure ( dm ) and emission measure ( em ) toward high galactic latitude in our model ( appendices ) . the top - left panel of figure [ figure2 ] shows one - dimensional profiles of @xmath28 from the sun along the ngp and sgp . the electron density in the northern sky is smaller than that in the southern sky at low altitudes this is due to the presence of a local , hot bubble with @xmath33 , centered at @xmath34 in kpc @xcite . for the regular magnetic field , @xmath35 , we use combinations of the axi - symmetric spiral ( ass ) or bi - symmetric spiral ( bss ) field introduced by @xcite , the halo toroidal field of @xcite , and the halo poloidal fields of @xcite or @xcite . the top panels of figure [ figure3 ] show one - dimensional profiles of the strength of the regular field , @xmath36 , and the line - of - sight ( los ) field strength , @xmath37 , from the sun along the galactic poles . for instance , adpn indicates a model toward the ngp ( n ) including the ass field ( a ) and the dipole toroidal field ( d ) with the dipole poloidal field ( p ) . for the full list of models and the model name convention , refer to table 1 . we note that the dipole and quadrupole models are identical toward the ngp , but differ toward the sgp . the ds models tend to have larger @xmath36 than other models , since in other models the spiral and toroidal fields partly cancel each other . the spiral field dominates @xmath36 at @xmath38 , while the toroidal field dominates at @xmath39 . for the los field strength , the field strength is @xmath40 if the poloidal field exists . otherwise , @xmath41 . for the electron temperature distribution , we employ the analytic expression adopted by @xcite @xmath42 where @xmath43 is in units of k and the galactocentric cylindrical coordinates @xmath44 and @xmath45 are in kpc . the bottom - left panel of figure [ figure2 ] shows the one - dimensional profile from the sun along the galactic poles . for the rms speed of random flow motions , @xmath46 , @xmath47 observations provide a weak constraint on the plausible range of values , @xmath48 @xcite . @xcite studied the distribution of emission measure ( em ) , and found that the rms mach number @xmath49 for @xmath50 and @xmath51 is smaller at higher galactic latitudes . studies of polarization gradients also broadly constrain @xmath52 @xcite . we employ the simplest case , i.e. , a uniform distribution of @xmath53 ( @xmath54 ) or @xmath55 ( @xmath56 ) , as shown in the right panels of figure [ figure2 ] . the bottom panels of figure [ figure3 ] show one - dimensional profiles of @xmath57 , the ratio of the gas pressure to the magnetic pressure due to @xmath35 . our @xmath57 is in the range of @xmath58 . here , we assume that hydrogen is fully - ionized , helium is neutral , and that their mass fractions are @xmath59 and @xmath60 , respectively . observations suggest the presence of electron density fluctuations and turbulent magnetic fields , in addition to smooth components in the galaxy . for instance , the volume filling factor of electrons , @xmath61 , which quantifies the clumpiness , has been estimated from dispersion measures ( dms ) and emissions / absorptions ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . highly disturbed distributions of rm and polarization angle clearly indicate turbulent structures of the gmf ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the random , turbulent components of the electron density and the gmf could be modeled analytically with preassigned spectra and random phases in fourier space . as mentioned in section 1 , @xcite followed such an approach with a constant amplitude everywhere . @xcite and @xcite introduced an ordered " or striated " field component . these works successfully reproduced the galactic rm mainly toward low and mid - galactic latitudes . however , it is not clear whether such treatments are good enough for studies of the galactic rm toward high galactic latitudes , where the random components are the dominant contribution to the rm . and constant rms amplitudes for electron density fluctuations and turbulent magnetic field would not be justified for broad distributions of @xmath51 and @xmath57 toward high altitudes ( see figures [ figure2 ] and [ figure3 ] ) , since they depend on @xmath51 and @xmath57 ( see appendix ) . in addition , in turbulent flows , phases are not really random . an alternative approach would be to use the random electron density given by the ne2001 model for the electron density fluctuations . but we would then need to model the turbulent magnetic field separately . we therefore do not adopt this approach . as in @xcite , we model the random components with mhd turbulence simulations in a closed box , as described below . in principle , if we performed full mhd simulations of the galaxy , we could reproduce the electron density fluctuations and the turbulent magnetic field as well as the regular components in the galactic disk and halo . but currently available computational resources do not allow a numerical resolution high enough to simultaneously reproduce both the large - scale global components and the small - scale turbulent components . we embed the data of mhd turbulence simulations in the galactic halo described by the global distributions of @xmath28 and @xmath35 of the previous subsection . for this purpose , we carried out three - dimensional simulations of driven , isothermal , compressible mhd turbulence without self - gravity , using a multi - dimensional mhd code based on the total variation diminishing ( tvd ) scheme @xcite . this is an mhd extension of the explicit , second - order finite - difference , upwinded , conservative scheme of @xcite for hydrodynamics . the version of the code used includes a flux constraining scheme that maintains @xmath62 up to the machine accuracy @xcite . in uniform , static medium with @xmath35 assumed to be along the @xmath63-direction , turbulence was driven by imposing a solenoidal forcing with perturbations drawn from a gaussian random field at wavenumbers around @xmath64 in fourier space . the amplitude of the forcing was fixed in such a way that turbulence saturates at preassigned mach numbers . simulations were done in a periodic computational box with @xmath65 grid zones . the size of the simulation box , @xmath66 , and the outer scale or the driving scale of turbulence , @xmath67 , were determined as follows . observations suggest that @xmath68 is on the order of @xmath69 and @xmath70 in spiral arms and in interarm regions , respectively @xcite . these are ascribed to stellar sources such as stellar winds and protostellar outflows , or supernova and superbubble explosions . @xcite adopted @xmath71 for studies of rm in low and mid - galactic latitudes ( @xmath68 is not defined in their approach ) . on the other hand , @xcite studied the em distribution estimated with the wisconsin @xmath47 mapper ( wham ) . using mhd turbulence simulations , they found that the em distribution depends on @xmath66 and @xmath68 ; larger @xmath66 and @xmath68 results in smaller ems . they argued that the histogram of the wham data on the warm ionized medium that includes high galactic latitude data could be reproduced with @xmath72 pc and @xmath73 pc . adopting the results of @xcite , we set @xmath74 and @xmath75 . the grid of @xmath65 uniform zones for the box of @xmath74 corresponds to the spatial resolution of @xmath69 . we note that this resolution is enough to resolve the characteristic scales of turbulence ; for instance , the most energy containing scale and the integral scale , @xmath76 and @xmath77 , respectively ( see * ? ? ? * ) , are sufficiently large compared to the spatial resolution . representative simulations were performed for @xmath78 and @xmath79 , in order to cover the ranges of @xmath51 and @xmath57 toward the ngp and sgp ( figures [ figure2 ] and [ figure3 ] ) . table 2 lists the simulations along with the rms values of density and magnetic field strength . for the case of @xmath80 at high galactic altitudes ( @xmath81 kpc , figure [ figure3 ] ) , we use the data for @xmath82 . since the contribution from high altitudes to the integrated rm is at most several per cent , this approximation does not seriously affect our results . in turbulence simulations , @xmath46 saturates roughly in one flow - crossing time , @xmath83 @xcite . for the length scale of @xmath84 , the corresponding timescale of @xmath85 and @xmath86 for the rms flow speed of @xmath87 and @xmath88 is sufficiently short compared to the age of the galaxy . we thus use the data at the saturation stage of turbulence . the data of simulations are stacked ( or piled up ) from the galactic mid - plane up to the outer edge of the ne2001 model , @xmath89 ( figure [ figure1 ] ) . in the stacking , we try to align the direction of the regular field in simulation data with that of the analytic model as follows . as we already noted , the regular field is dominated by the spiral field at @xmath38 and by the toroidal field at @xmath39 ( see figure [ figure3 ] ) . in each domain , the regular field direction is close to either the @xmath90 or @xmath91 direction in our coordinate systems , except narrow transition regions with a width @xmath92 . we divide each of the northern and southern hemispheres into four blocks : @xmath93 where @xmath45 is in units of kpc . we calculate the average values of @xmath51 and @xmath57 in each block and seek the simulation of closest parameters . table 1 summarizes the direction of the regular magnetic field and the values of @xmath51 and @xmath57 adopted in each block . for example , in block 1 of adon15 , where 15 means @xmath53 , we choose the simulation with @xmath94 and @xmath95 , and align the regular field with the @xmath90-direction . the stacked data are renormalized in such a way that we match the average electron density and magnetic field in the computational box with those of the regular components of section 2.1 . after the renormalization , each grid keeps the fluctuations based on mhd turbulence simulations . examples of profiles of the resulting electron density , magnetic field , and rm ( see section 2.3 ) from the sun along the ngp , including both the regular and random components , are shown in figure [ figure4 ] . it is apparent that fluctuations in the case with @xmath96 @xmath97 are larger than those for @xmath98 @xmath97 , as expected . the strength of the turbulent magnetic field ( @xmath99 ) is at most a few @xmath100 , and mostly @xmath101 . this seems to be smaller than the strength of the random field , a few to several @xmath100 , in the galactic disk ( e.g. , * ? ? ? * ; * ? ? ? * ) , but consistent with @xmath102 a few kpc from the galactic plane estimated by assuming equipartition between the thermal and nonthermal pressures @xcite , as well as with the recent estimates of @xmath103 and @xmath104 toward the ngp and sgp , respectively @xcite . the sign of los magnetic field changes several times . @xmath105 increases with the path length , and then saturates around @xmath106 kpc where the electron density becomes small . we check our model for the electron density distribution by comparing the resulting dm and em with observed values @xcite . dm , which is determined only by the global , regular component , is 22 and 25 @xmath107 toward the ngp and sgp , respectively , while the observed value is @xmath108 @xmath107 . we note that the ne2001 model , from which we get the regular component , was designed to reproduce observations including dm . the em in our model is 1.5 ( 1.0 ) and 2.3 ( 1.6 ) @xmath109 toward the ngp and sgp , respectively , for @xmath110 30 ( 15 ) @xmath97 , which is in a good agreement with the observed value , @xmath111 @xmath109 . our model also reproduces the overall distribution of em at @xmath112 presented by @xcite . details are given in the appendix . the rm toward a source outside the galaxy is defined as @xmath113 where @xmath114 , @xmath115 , and the path length @xmath116 are in units of @xmath117 , @xmath100 , and pc , respectively . conventionally rm is positive when the los magnetic field points toward us , and @xmath118 is the maximum distance along the los up to the outer edge of our model space . in this paper , we calculate the rm only due to the gmf . we do not include other possible contributions , such as the rm due to the igmf , the intrinsic rm at sources , and observational errors . we discuss those in section 4 . we consider a square fov of @xmath26 with @xmath119 on a side centered on each of the galactic poles , which is comparable to the fovs of rm observations toward the ngp and sgp @xcite . in addition , we also consider a square fov of @xmath27 with @xmath120 on a side . the smaller fov is roughly that proposed for the dense aperture array of the ska @xcite . the smaller fov is also used to check the fov dependence of rm statistics . to see structures in angular separations of @xmath121 for which the second - order sf of observed rms is available @xcite , we choose the angular resolution of pixels to be @xmath122 . the numbers of pixels we use are @xmath123 and @xmath124 for @xmath26 and @xmath27 fovs , respectively . we calculate rms with equation ( [ eq : rm ] ) assuming one extragalactic radio source in each pixel , and construct mock rm maps . for each map , we randomly rotate stacked simulation boxes around the axis parallel to the regular magnetic field in table 1 ( keeping the direction of the regular magnetic field ) and randomly shift box centers to avoid the repeat of the same grid zones which would make artifacts in statistics . when the integration reaches the top of a simulation box , the integration proceeds into the next stacked box and the integration is continued . if the integration reaches the side of the simulation box , we apply the periodic boundary condition to the side , and replicate the box beyond the side . we obtain 200 mock maps for each model listed in table 1 and each fov , and calculate statistical quantities . the quantities shown below in sections 3.3 and 3.4 are the averages for 200 rm maps . figure [ figure5 ] shows sample maps over a @xmath125 fov . maps due to the regular components in the model adon ( leftmost panel ) and due to the random components in one of 200 simulations for adon30 ( second panel ) are shown . the combined map due to the regular and random components ( third panel ) as well as the binned map of median rms in @xmath126 pixels ( rightmost panel ) are also shown . we can see that the regular components of density and magnetic field alone produce large - scale structures of up to the fov size . the structures highlight @xmath37 as induced by the radial and azimuthal components of the spiral and toroidal fields ( the model shown does not include a poloidal field ) . the random components of density and magnetic field , on the other hand , produce complex structures , such as clump - like and filament - like features , on angular scales of @xmath127 . such structures persist in the combined map . @xcite showed binned maps of median rms in @xmath126 pixels toward the ngp and sgp , which were produced from observational data . the maps display structures of a few to several degrees . in our model , such structures are mostly the consequence of the random components of density and magnetic field ( rightmost panel ) . our results indicate that the random components mainly produce the observed rm structures on scales less than several degrees , while the regular components contribute to larger scale structures . we first examine the statistics of rms in maps due to the regular components alone . we calculated the average , @xmath128 , and the standard deviation , @xmath129 , in maps for the models and fovs we consider . the resulting values are shown in table 3 . the average is mainly determined by the existence of the poloidal component of the gmf ; @xmath130 without the poloidal component , or @xmath131 and @xmath132 toward the ngp and sgp , respectively , if the dipole poloidal component exists , and @xmath133 and @xmath134 toward the ngp and sgp , respectively , if the x - field poloidal component exists . the absolute values toward the ngp are a little smaller than those toward the sgp , due to the existence of a low density , local hot bubble centered in the northern sky ( figure [ figure2 ] ) . the averages for the @xmath27 fov are slightly smaller than those for the @xmath26 fov , since the smaller fov includes high latitude regions only where the density and magnetic field strength are both smaller . the standard deviations are @xmath135 and @xmath136 toward the ngp and sgp , respectively , for the @xmath26 fov . these are caused mostly by contributions from the radial and azimuthal components of the gmf ( see also the leftmost panel of figure [ figure5 ] ) . the ds models tend to have larger standard deviations than the dn and qs models , because the ds models have larger @xmath36 at mid and high altitudes as noted in section 2.1 . larger values toward the sgp than those toward the ngp are again due to the existence of a local hot bubble near the sun . the standard deviations for the @xmath27 fov are smaller by a factor of @xmath137 than those for the @xmath26 fov , since the smaller fov includes a narrower range of rm values . we next examine the statistics of rms due to both the regular and random components . figure [ figure6 ] shows the probability distribution functions ( pdfs ) of rms in maps of @xmath26 fov , averaged over 200 rm maps , for all models we consider . the pdfs roughly follow the gaussian distribution , as pointed out by @xcite . the figure also shows the observed pdfs of @xcite . the average and standard deviation of @xmath138 ( the average of rms for a map ) and @xmath139 ( the standard deviation of rms for a map ) over 200 maps were calculated . figure [ figure7 ] shows the resulting values for @xmath26 fov as well as for @xmath27 fov . the value of @xmath138 , which approximates the peak positions for the nearly - symmetric pdfs in figure [ figure6 ] , is determined mostly by the regular components . so the average of @xmath138 is close to @xmath138 of the regular components alone in section 3.1 ; @xmath140 @xmath141 , without the poloidal component of the gmf , or @xmath142 and @xmath143 toward the ngp and sgp , respectively , if the poloidal component exists . as noted in section 1 , the medians of observed rms toward the ngp and sgp are @xmath0 and @xmath1 , respectively @xcite . hence , the models with poloidal components better reproduce the observed average toward the sgp , but the models without poloidal components are preferred for the ngp . on the other hand , the fluctuation of @xmath138 over the 200 maps is rather small ; the standard deviations of @xmath138 for 200 maps are @xmath144 for models with @xmath145 and @xmath146 for models with @xmath147 , as shown with error bars in figure [ figure7 ] . these values are much smaller than the difference between the observed averages , @xmath148 @xmath141 , toward the ngp and sgp . this means that the difference can not be explained by the statistical fluctuation caused by the random components in our models . these results indicate that none of our models can simultaneously reproduce the observed averages of rms toward the ngp and sgp . in figure [ figure6 ] , the simulated pdfs have narrower and more sharply peaked profiles than the observed pdfs . the width of the pdfs is quantified by @xmath139 . in figure [ figure7 ] , the averages of @xmath139 are @xmath149 toward the ngp and @xmath150 toward the sgp for models with @xmath145 , and @xmath151 toward the ngp and @xmath152 toward the sgp for models with @xmath147 . the fluctuation of @xmath139 in 200 maps is again small ; the standard deviations of @xmath139 for 200 maps are @xmath153 for models with @xmath145 and @xmath154 for models with @xmath147 , as shown with error bars . on the other hand , the estimations of @xcite with their observed rms are @xmath155 and @xmath15 toward the ngp and sgp . so our estimations of @xmath139 with simulated rms are substantially smaller than the values of @xcite . we argue that all of our models fail to reproduce not only the peak positions but also the widths of the pdfs of observed rms toward the ngp and sgp ( see further discussions in sections 4 and 5 ) . from figure [ figure7 ] , we see the average and standard deviation of @xmath138 are not sensitive to the size of fov . but the values of @xmath139 for the @xmath27 fov are somewhat smaller than those for the @xmath26 fov , as expected . we also calculated the two - dimensional power spectrum ( ps ) and structure function ( sf ) with the sky map of rm . they tell us at which angular scales most power of rm resides and the spatial structure of rm decorrelates . figure [ figure8 ] shows the ps of rms due to both the regular and random components in maps of @xmath26 fov , averaged over 200 rm maps , for all models we consider . the ps toward the sgp are larger than those toward the ngp . this is again attributed to the presence of a local hot bubble in the northern sky . toward the sgp , the ps for the models with a dipole toroidal field ( d ) are larger than those for the models with quadrupole toroidal field ( q ) . this is because the dipole models have larger @xmath36 , as shown in figure [ figure3 ] . ( the dipole and quadrupole models are identical toward the ngp , as noted in section 2.1 . ) the amplitude of the ps depends on the rms flow speed ; the ps for @xmath156 are larger by up to a factor of @xmath157 than those for @xmath158 . the slope also depends on the rms flow speed . larger @xmath46 s result in shallower profiles . this is because the density and magnetic field ps of supersonic flows ( @xmath55 corresponds to @xmath159 ) are shallower than those of subsonic flows ( @xmath53 corresponds to @xmath160 ) ( e.g. , see * ? ? ? we quantified the slope of ps , @xmath161 , defined with @xmath162 in the angular range of @xmath127 , and show the resulting values in figure [ figure9 ] for @xmath26 fov as well as for @xmath27 fov . these are the averages for 200 maps . the figure confirms the dependence of @xmath161 on @xmath46 ; @xmath161 s for @xmath55 are larger by up to @xmath163 than those for @xmath53 for the models shown in figure [ figure8 ] ( with @xmath26 fov ) . some of the ps have slopes consistent with the kolmogorov slope @xmath164 , but others have steeper slopes , indicating that the gmf model influences the slope of the ps . from the observer s point of view , the sf of rms is a statistical quantity that is easier to obtain than the ps . the @xmath165-th order sf is defined as @xmath166 with @xmath167 , where the subscript indicates the averaging over the data domain of @xmath168 . figure [ figure10 ] shows the second - order sfs @xmath169 in maps of @xmath26 fov , averaged over 200 rm maps , for all models we consider . the sfs monotonically increase with angular separation , and reach up to @xmath170 toward the ngp and @xmath171 several @xmath172 toward the sgp , respectively , at the angular separation of @xmath173 . the sfs toward the sgp are larger than those toward the ngp . toward the sgp , the sfs for the models with dipole toroidal fields are larger than those for the models with quadrupole toroidal fields . as for the ps , the sfs for @xmath156 have amplitudes larger by up to a factor of @xmath157 than those for @xmath158 . figure [ figure10 ] also shows the observed sfs toward the ngp and sgp @xcite . it is clear that the sfs of our simulated rms do not match the observed sfs at any angular scales . at the largest observed scale of @xmath173 , the simulated sfs are smaller by an order of amplitude or so than the observed sfs . but the difference is even larger at smaller angular separations . not just the amplitudes , but also the slopes of the sfs of our simulated rms are quite different from the observed ones ; the slopes of simulated rms are much steeper . we quantify the slope of sf , @xmath174 , defined with @xmath175 over the angular separations of @xmath127 , and show the resulting values in figure [ figure9 ] for a @xmath26 fov as well as for a @xmath27 fov . again these are the averages for 200 maps . for the @xmath26 fov , @xmath176 for @xmath158 , and @xmath177 for @xmath156 . the slopes of observed sfs are @xmath178 @xcite . so our models fail to reproduce the observed sfs not only in the amplitude but also in the slope . the results suggest that the observed rms may contain contributions due to structures smaller than those typically found in the galactic halo . the ps and sfs for the @xmath27 fov ( not shown ) show behaviors similar to those for the @xmath26 fov , except that there are no data beyond a scale of @xmath179 . quantitatively , the slopes of ps and sf , @xmath161 and @xmath174 , for the @xmath27 fov are a bit smaller than those for the @xmath26 fov , as shown in figure [ figure9 ] . in our models , the average or median of the rm toward high galactic latitudes , @xmath138 , is determined mostly by the halo poloidal component of the gmf , while the standard deviation , @xmath139 , is determined mostly by the random component of the gmf . our models fail to simultaneously reproduce the medians of observed rms toward the ngp and sgp , @xmath0 and @xmath1 @xcite . the models that contain a poloidal component of the gmf have a vertical magnetic field of @xmath180 in the earth vicinity , which induces @xmath181 toward the ngp and @xmath182 toward the sgp . the models without a poloidal component have @xmath183 , so @xmath184 . the different values , @xmath185 toward the ngp and @xmath186 toward the sgp , which would explain the observed medians , however , is not accommodated in our models . the difficulties of mixed regular field geometries in the steady state are discussed by @xcite . the fluctuation ( standard deviation ) of @xmath138 in simulations , @xmath146 , is too small to account for the difference between the observed medians toward the ngp and sgp . to see whether larger fluctuations of @xmath138 are possible with different model parameters , simulations with larger values of @xmath68 and @xmath46 were carried out ( not shown ) . we found that the fluctuation of @xmath138 increases with increasing @xmath68 and becomes @xmath187 for @xmath188 . however , with such large value of @xmath68 , the em distribution estimated from wham data @xcite can not be explained ( see appendix ) . the fluctuation of @xmath138 increases with increasing @xmath46 also . we find a sufficiently large fluctuation for @xmath189 , which corresponds to @xmath190 . but such large @xmath51 is clearly inconsistent with the value constrained from the em distribution @xcite and from polarization gradients @xcite . our models also fail to reproduce the standard deviations of observed rms , @xmath155 and @xmath15 toward the ngp and sgp , respectively @xcite . our estimates , @xmath191 , are substantially smaller . the values are not consistent even with the galactic contributions estimated by @xcite , @xmath192 and @xmath193 toward the ngp and sgp , respectively . ( note that the data of @xcite and @xcite are different . ) we would get @xmath194 if we adopt @xmath195 . but again such a value is too large to explain the observed em distribution . we could explain the observed medians and standard deviations if there are additional global or local structures not represented in our models for the gmf ( e.g. , * ? ? ? * ; * ? ? ? such structures could have been produced , for instance , by the parker instability or supernova explosions . in addition , @xcite have shown that transient structures can be produced in three - dimensional mhd simulations of turbulent gaseous disks . we could also address the discrepancy between the simulations and the data if the regular field is stronger than that in our models . for instance , we would have simulated the observed @xmath139 s , if the regular field strength was increased by a factor of @xmath137 . but then , the corresponding regular field strength near the galactic plane would be @xmath196 ( see figure [ figure3 ] ) , which is larger than that adopted in recent observational studies @xcite . on the other hand , after investigating rms toward the perseus arm region , @xcite argued that observed rms are consistent with a toroidal field of strength @xmath102 toward the north , but @xmath197 toward the south ( note that the observation covers a rather thin region of @xmath198 and @xmath199 ) . this indicates that there may be room for additional modeling of the gmf , which we leave for future studies . even with additional structures and stronger regular fields , however , the observed sfs are difficult to reproduce @xcite . the second - order sfs of simulated rms are substantially smaller than the observed sfs , especially at small angular separations of @xmath200 . this means that we need structures that would provide significant powers on scales of a parsec or so , but there is no observational support for such structures ( e.g. , * ? ? ? * ; * ? ? ? the slope of observed sfs , @xmath201 @xcite , is much flatter than that of simulated sfs , @xmath202 . we note that two - dimensional , white noise results in a flat sf ; @xmath203 , where @xmath204 is the standard deviation of the noise . we would get @xmath205 for @xmath206 . observational errors should follow the distribution of white noise . but @xcite estimated that the observational errors in their data are @xmath4 and @xmath207 toward the ngp and sgp , respectively . @xcite estimated that the error in their data is @xmath208 . so observational errors may not be enough to explain the observed flat sfs of @xmath209 ( see also * ? ? ? it is expected that radio sources against which rms are observed have their own intrinsic rms . the rms of radio sources should also follow the distribution of white noise and contribute to a flat sf . @xcite and @xcite showed that the observed sf toward the ngp deviates from a flat behavior at very small scales ; the square value of rm difference between two sources at @xmath210 and @xmath211 , @xmath212 ( not the second - order sf ) , is around @xmath170 at angular separations of @xmath213 . this tells us that even if the square value is solely due to the rms of radio sources , it is unlikely that the rms make up the observed flat sfs of @xmath209 . in fact , @xcite claimed that many observed sources possess little intrinsic rm from the fact that the fractional polarization percentage is nearly constant with increasing wavelength . finally , there exists clear evidences for an extragalactic component to the rms , from three different papers taking different approaches @xcite . although @xcite and @xcite attribute the excess scatter in rm to individual intervening absorbers along the line of sight , an alternative origin which can contribute to the observed rm toward high galactic latitudes too is the igmf . as mentioned in section 1 , using simulations for the large - scale structure formation in the universe , @xcite predicted that the rm due to the igmf would have @xmath214 and @xmath215 several @xmath141 . they also predicted that the sf of the rm has a flat profile of @xmath216 amplitude at angular separations larger than @xmath217 ; the sf is expected to decrease to the order of @xmath170 at @xmath213 from @xmath218 at @xmath219 , consistent with the observation of @xcite . the work of @xcite suggests the possibility that a substantial fraction of the rm toward the ngp and sgp can be attributed to the rm due to the igmf . if so , the observed standard deviations and sfs can be explained . but reproducing the observed medians of rm still needs additional components or / and structures of the gmf . we have studied the galactic rm toward high galactic latitudes . we have considered a number of models for the global , regular components of the gmf and the electron density in the galaxy , based on observations . the turbulent , random components were modeled with three - dimensional mhd turbulence simulations . the strength of the regular magnetic field in our models is a few @xmath138 g close to the disk and smaller at high altitudes . the strength of the turbulent field is at most a few @xmath100 , and mostly @xmath101 . we obtained rm maps for @xmath26 fov toward the galactic poles , and compared the results with observations . we also considered a smaller fov of @xmath27 , designed to simulate the fov of future surveys with the ska . our models fail to simultaneously reproduce the observed medians of rms toward the ngp and sgp . the observations require vertical magnetic fields of @xmath220 toward the ngp and @xmath221 toward the sgp in the earth s vicinity , but such field geometries are not accommodated in the gmf models we considered . the pdfs of simulated rms are narrower and more sharply peaked than the observed pdfs , meaning that the standard deviations of simulated rms are smaller than the observed values . the second - order sfs of simulated rms are one to two orders of magnitude smaller than the observed sfs at small angular separations . in addition , the slopes of the sfs of simulated rms are substantially larger than the observed ones . we argue that observational errors and the intrinsic rm of background radio sources are not enough to explain the discrepancies between the statistics of our simulated rms and observed rms . we suggest that the rm due to the igmf may account for a fraction of the rm toward high galactic latitudes and could explain the discrepancies in the standard deviation and sf of the rms . as a subsequent project , the rm due to the igmf as well as observational errors and the intrinsic rm of background radio sources would be incorporated in the modeling of the rm toward high galactic latitudes , to check quantitatively whether some of the discrepancies , such as those in the standard deviation and sf , would be explained . but we expect that reproducing the observed medians will still need additional components or / and structures of the gmf . we should note , however , that current observations of rms still contain large uncertainties . new and the future observational facilities such as the jvla , mwa , lofar , askap , meerkat , and the ska will produce much better data . for instance , askap will detect radio sources with average angular separations of @xmath219 . with such a dense rm grid , the quality of the observed rm data will be dramatically improved . better quality data will hopefully enable us to better quantify the contributions due to the gmf , the igmf , the intrinsic rm of background radio sources , and observational errors . finally , we note that better measurements of the mach number of turbulence in the halo by observations of @xmath222 line profiles @xcite as well as by radio polarization gradients @xcite will help us improve the constraints on the magnitude and structures of the rm toward high galactic latitudes . the following coordinate systems are used throughout the paper : cartesian coordinates , @xmath223 , galactocentric cylindrical coordinates , @xmath224 , galactocentric polar coordinates , @xmath225 , and galactic celestial coordinates , @xmath226 , respectively , defined in figure [ figure1 ] . here , @xmath227 is the galactocentric radius , @xmath228 is the azimuth angle starting from @xmath229 and increasing in the counterclockwise direction , and the @xmath63-@xmath230 plane coincides with the galactic plane with @xmath63 pointing to @xmath229 and @xmath230 to @xmath231 . the sun is located at @xmath232 in kpc , and thus @xmath233 kpc . the ne2001 model we employ for the electron density , @xmath28 , consists of three global components ( thick disk , thin disk , and five spiral arms ) , four local components ( local hot bubble , loop i , local super bubble , and low density region in quadrant 1 ) , and more than a hundred isolated components such as clumps and voids , as displayed in figure [ figure1 ] . the electron density of the model was obtained from the average of the dm for pulsars at known distances . we take the original parameters in the ne2001 packagecordes / ne2001/. ] , except for the scale height of the thick disk , @xmath29 , and the mid - plane electron density for the thick disk , @xmath30 . @xcite obtained @xmath234 kpc from an analysis of pulsars at high galactic latitudes . we adopt @xmath235 kpc , instead of @xmath236 kpc in the package . the corresponding mid - plane electron density for the thick disk is @xmath32 . the revised scale height improves fits of the galactic rm and radio continuum emission at low and mid - galactic latitudes @xcite . the changes of the scale height and mid - plane electron density of the thick disk , however , may not be consistent with the construction of the ne2001 model , because these parameters are highly covariant with others in the model . in order to justify our modification and make sure the consistency with observations , we compare the dm and em from the ne2001 model with the original scale height and mid - plane density ( hereafter the original ne2001 model ) and the ne2001 model with the modified scale height and mid - plane density ( our modified ne2001 model ) . we also tested the exponential model fitted by @xcite ( the plane - parallel model ) for comparison . figure [ figure11 ] shows the distribution of dm as a function of height above the galactic plane . at a height of @xmath237 kpc , both the original and modified ne2001 models reproduce the distribution of @xcite well toward the sgp , but predict smaller dm by a factor of @xmath137 toward the ngp , likely due to the local hot bubble . at the height of @xmath238 kpc , the original ne2001 model overestimates dm by a factor of @xmath239 toward south . on the other hand , the modified ne2001 model reproduces the observed dm well ; the resultant dms for the modified ne2001 are @xmath240 and @xmath241 toward the ngp and sgp , respectively , which are roughly consistent with observations @xcite . next , we calculated the pdf of em . there is a well - known issue that models with smooth electron density profile do not reproduce the observed em well ; the discrepancy can be resolved by introducing volume filling factor , @xmath242 . we adopted @xmath243 , @xmath45 in kpc , for @xmath244 kpc and @xmath245 for @xmath246 kpc @xcite . the resulting em is shown in figure [ figure12 ] . with the adopted @xmath242 , the original ne2001 does not correctly reproduce the em distribution from the wham observation @xcite . on the other hand , the modified ne2001 model and the plane - parallel model produce results , which are better consistent with the observed distribution . it is interesting to see that the cases with @xmath98 @xmath247 ( or the rms mach number of turbulence is close to @xmath248 ) better reproduce the observed em . it should be pointed that @xcite claimed smaller @xmath242 ( that is , more clumpy ) at both low and high altitudes . if smaller @xmath242 is adopted , our model gives larger dispersion of em , so that even smaller @xmath46 would be preferable . then , the corresponding standard deviation of rm would be even smaller ( figure [ figure7 ] ) . finally , figure [ figure13 ] shows the second - order sf of simulated rms in a 900 @xmath249 fov . the amplitude of sf for the original ne2001 model is somewhat larger than that for the modified ne2001 and plane - parallel models . the slope of sf does not significantly depend on the electron density model , and again the sf for the three models does not match the observed one ( see section 3.4 ) . the global , regular magnetic field , @xmath35 , is conventionally described as the combination of the disk spiral field , @xmath250 , the halo toroidal field , @xmath251 , and the halo poloidal field , @xmath252 , so that @xmath253 ( e.g. , * ? ? ? there are a number of analytic models intended to reproduce the observed structure of @xmath35 . the models have been tuned by fitting the data mostly at low and mid - galactic latitudes , but can be used for studies of the galactic rm toward high galactic latitudes as well . specifically , we employ the models introduced by @xcite for the spiral field , by @xcite for the toroidal field , by @xcite for the dipole poloidal field , and by @xcite for the x - field poloidal field , unless otherwise specified . we note that our combined model does not take account of detailed theoretical consistencies , such as the continuity , divergence - free nature , and closeness of @xmath35 . we expect that those do not significantly affect our results . the large - scale disk fields of spiral galaxies including our galaxy have been classified into two types based on the spiral pattern , the axi - symmetric spiral ( ass ) field with no dependence on the azimuthal angle , or the bi - symmetric spiral ( bss ) field with a symmetry of @xmath254 . the functional form for the disk spiral field can be written as @xmath255 where @xmath256 is the pitch angle of arms which is positive for leading spirals and negative for trailing spirals such as in our galaxy , and @xmath257 for the ass model , we adopt the ass + reversals of the magnetic field directions : @xmath258 where @xmath259 means the clockwise direction as seen from the north pole . we also adopt @xmath260 , @xmath261 , @xmath262 , @xmath263 , @xmath264 , and @xmath265 . for the bss model , we adopt the spiral structure of @xmath266 we also adopt @xmath267 , @xmath261 , @xmath268 , @xmath263 , and @xmath264 . in addition , @xmath269 and @xmath270 for @xmath271 , and otherwise @xmath272 and @xmath273 . note that the trigonometric function is minus cosine if the azimuth angle , @xmath274 , starting from @xmath231 is adopted ( e.g. , * ? ? ? reversals in the sign of rm across the galactic plane and across the galactic center ( e.g. , * ? ? ? * ) suggest the existence of a halo toroidal field . global three - dimensional mhd simulations of gas disks @xcite have indicated the azimuthal magnetic field component in the halo as a result of the buoyant escape of the azimuthal magnetic flux from the disk . the halo toroidal ( azimuthal ) field can be expressed as @xmath275 where @xmath276 is the sign of @xmath45 , and @xmath277 introduces the parity of the toroidal field configuration ; @xmath278 if we consider the asymmetries in longitude and latitude relative to the galactic plane and the center , respectively ( dipole ) , or @xmath279 if we consider the axisymmetric configuration without reversals relative to the galactic plane ( quadrupole ) . we adopt @xmath280 , @xmath281 , @xmath282 , and @xmath283 for @xmath284 and @xmath285 otherwise . we note that @xcite adopted smaller @xmath286 for high @xmath45 ( and also similarly small @xmath287 in their model ) , which results in @xmath288 of order @xmath289 at @xmath290 . we adopt the parameters based on @xcite , which gives @xmath291 at high @xmath45 ( see also figure [ figure3 ] ) . such a strong halo field at high @xmath45 is also motivated by recent studies of halo magnetic fields ( see * ? ? ? * ; * ? ? ? nonthermal filaments observed near the galactic center imply the existence of a central , vertical magnetic field @xcite , which is also predicted from global three - dimensional mhd simulations of gas disks ( e.g. , * ? ? ? * ) . the vertical field can be due to the galactic - center poloidal ( dipole ) field ; a strong dipole field would be observed as the vertical magnetic field in the earth vicinity . the dipole field can be expressed as @xmath292 here , we set @xmath293 @xcite to make the vertical component of @xmath294 from the poloidal field . the vertical field could be due to the out - of - plane x - field recently studied by @xcite . this model is motivated by the x - shaped field structure seen in radio observations of external , edge - on galaxies @xcite . we define the elevation angle of magnetic field , @xmath295 , with respect to the galactic mid - plane . then , the x - field can be expressed as @xmath296 we take the field at @xmath297 to have a constant elevation angle , @xmath298 , with respect to the mid - plane , where @xmath299 is the radius at which the field line passing through @xmath300 crosses the mid - plane . this means that if @xmath301 , then @xmath302 , @xmath303 , and @xmath304 . otherwise , @xmath305 @xmath306 and @xmath307 . we adopt @xmath308 , @xmath309 , @xmath310 , and @xmath311 . we thank the referee for constructive comments . we also thank p. l. biermann for comments on the manuscript , m. machida and x. sun for discussions on the galactic magnetic field model , s. a. mao for providing the data used in figure 6 , and j. stil for providing the data used in figures 9 and 12 . t.a . acknowledges the supports of the korea research council of fundamental science and technology ( krcf ) and the japan society for the promotion of science ( jsps ) . d.r . acknowledges the support of the national research foundation ( nrf ) of korea through grant 2007 - 0093860 . j.k . acknowledges the support by the international research & development program of the national research foundation ( nrf ) funded by the ministry of education , science and technology ( mest ) of korea ( grant k20903001740 - 11b1300 - 02610 ) . b.m.g . acknowledges the support of the australian research council through grant fl100100114 . akahori , t. , & ryu , d. 2010 , , 723 , 476 akahori , t. , & ryu , d. 2011 , , 738 , 134 beck , r. , brandenburg , a. , moss , d. , shukurov , a. , & sokoloff , d. 1996 , , 34 , 155 beck , r. 2009a , ( serie de conferencias ) , 36 , 1 beck , r. 2009b , ap&ss , 320 , 77 berkhuijsen , e. m. , mitra , d. , & mller p. , 2006 , an , 327 , 82 bernet , m. l. , miniati , f. , lilly , s. j. , accepted ( arxiv:1209.2410 ) burkhart , b. , lazarian , a. , & gaensler , b. m. 2012 , , 749 , 145 carilli , c. l. & taylor , g. b. 2002 , , 40 , 319 carilli , c. l. & rawlings s. 2004 , new astronomy review , 48 , 979 cho , j. , & ryu , d. 2009 , , 705 , l90 cordes , j. m. , & lazio , t. j. w. 2002 , preprint ( astro - ph/0207156 ) cordes , j. m. , & lazio , t. j. w. 2003 , preprint ( astro - ph/0301598 ) cox , d. p. 2005 , , 43 , 337 dolag , k. , crasso , d. , springel , v. , & tkachev , i. 2005 , j. cosmology astropart . phys . , 1 , 9 faulkner , a. j. , et al . , ska memo 122 : aperture arrays for the ska - the skads white paper " , 2010 gaensler , b. m. , beck , r. , & feretti , l. 2004 , new astronomy review , 48 , 1003 gaensler , b. m. , haverkorn , m. , staveley - smith , l. , dickey , j. m. , mcclure - griffiths n. m. , dickel , j. r. , & wolleben , m. , 2005 , science , 307 , 1610 gaensler , b. m. , madsen , g. j. , chatterjee , s. , & mao , s. a. 2008 , pasa , 25 , 184 gaensler , b. , landecker , t. , & taylor , r. 2010 , bulletin of the american astronomical society , 42 , 515 gaensler , b. m. , haverkorn , m. , burkhart , b. , newton - mcgee , k. j. , ekers , r. d. , lazarian , a. , mcclure - griffiths , n. m. , robishaw , t. , dickey , j. m. , & green , a. j. 2011 , nature , 478 , 214 giacinti , g. , kachelrie , m. , demikov , d. v. , & sigl , g. 2010 , journal of cosmology and astroparticle physics , 8 , 36 haffner , l. m. , reynolds , r. j. , tufte , s. l. , madsen , g. j. , jaehnig , k. p. , & percival , j. w. 2005 , , 149 , 405 haffner , l. m. , reynolds , r. j. , madsen , g. j. , hill , a. s. , barger , k. a. , jaehnig , k. p. , mierkiewicz , e. j. , percival , j. w. & chopra , n. 2010 , in the dynamic interstellar medium : a celebration of the canadian galactic plane survey , ed . r. kothes , t. l. landecker , & a. g. willis ( san francisco : astronomical society of the pacific ) , 388 ( arxiv:1008.0612 ) hammond , a. m. , robishaw , t. , & gaensler , b. m. , submitted ( arxiv:1209.1438 ) han , j. l. 2009 , iau symp . , 259 , 455 , arxiv:0901.1165 harten , a. 1983 , j. comp . , 49 , 357 haverkorn , m. , brown , j. c. , gaensler , b. m. , & mcclure - griffiths , n. m. 2008 , , 680 , 362 hill , a. s. , benjamin , r. a. , kowel , g. , reynolds , r. j. , haffner , l. m. , & lazarian , a. 2008 , , 686 , 363 jansson , r. , & farrar , g. r. 2012 , apj , submitted . jaffe , t. r. , leahy , j. p. , banday , a. j. , leach , s. m. , lowe , s. r. , & wilkinson , a. 2010 , , 401 , 1013 kim , j. , ryu , d. , jones , t. w. , & hong , s. s. 1999 , , 514 , 506 krause , m. , alexander , p. , bolton , r. , geisbsch , j. , green , d. a. , & riley , j. 2009 , , 400 , 646 krause , m. 2009 , ( serie de conferencias ) , 36 , 25 kronberg , p. p. , bernet , m. l. , miniati , f. , lilly , s. j. , short , m. b. , & higdon , d. m. 2008 , , 676 , 70 machida , m. , matsumoto , r. , nozawa , s. , takahashi , k. , fukui , y. , kudo , n. , torii , k. , yamamoto , h. , fujishita , m. , & tomisaki , k. 2009 , , 61 , 411 machida , m. , nakamura , k. , kudo , t. , akahori , t. , sofue , y. , & matsumoto , r. , 2012 , , submitted mao , s. a. , gaensler , b. m. , haverkorn , m. , zweibel , e. g. , madsen , g. j. , mcclure - griffiths , n. m. , shukurov , a. & kronberg , p. p. 2010 , , 714 , 1170 mao , s. a. , mcclure - griffiths , n. m. , gaensler , b. m. , brown , j. c. , van eck , c. l. , haverkorn , m. , kronberg , p. p. , stil , j. m. , shukurov , a. , & taylor , a. r. 2012 , , accepted nishikori , h. , machida , m. , & matsumoto 2006 , , 641 , 862 ohno , h. , & shibata , s. 1993 , , 262 , 953 padoan , p. , jimenez , r. , nordlund , . , & boldyrev , s. 2004 , , 92 , 19 peterson , j. d. , & webber , w. r. 2002 , , 575 , 217 pshirkov , m. s. , tinyakov , p. g. , kronberg , p. p. , & newton - mcgee , k. j. 2011 , , 738 , 192 prouza , m. , & @xmath312 , r. 2003 , , 410 , 1 ryu , d. , kang , h. , & biermann , p. l. 1998a , , 335 , 19 ryu , d. , miniati , f. , jones , t. w. & frank , a. 1998b , , 509 , 244 ryu , d. , kang , h. , cho , j. , & das , s. 2008 , science , 320 , 909 schnitzeler , d. h. f. m. 2010 , , 409 , 99 simonetti , j. h. , cordes , j. m. , & spangler , s. r. , 1984 , , 284 , 126 simonetti , j. h. , & cordes , j. m. 1986 , , 310 , 160 stasyszyn , f. , nuza , s. e. , dolag , k. , beck , r. , & donnert , j. 2010 , , 408 , 684 stil , j. m. , taylor , a. r. , & sunstrum , c. 2011 , , 726 , 4 sun , x. h. , reich , w. , waelkens , a. , & enlin , t. a. 2008 , , 477 , 573 sun , x. h. , & reich , w. 2009 , , 507 , 1087 sun , x. h. , & reich , w. 2010 , research in astronomy and astrophysics , 10 , 1287 takami , h. , & sato , k. 2010 , , 724 , 1456 taylor , a. r. , stil , j. m. , & sunstrum , c. 2009 , , 702 , 1230 tinyakov , p. g. , & tkachev , i. i. 2005 , astropart . , 24 , 32 tufte , s. l. , reynolds , r. j. , & haffner , l. m. 1999 , in interstellar turbulence , ed . j. franco , & a. carraminana ( cambridge : cambridge univ . press ) , 27 van eck , c. l. , brown , j. c. , stil , j. m. , rae , k. , mao , s. a. , gaensler , b. m. , shukurov , a. , taylor , a. r. , haverkorn , m. , kronberg , p. p. , & mcclure - griffiths , n. m. 2011 , , 728 , 97 waelkens , a. , jaffe , t. , reinecke , m. , kitaura , f. s. , & enlin , t. a. , 2009 , , 495 , 697 xu , y. , kronberg , p. p. , habib , s. , & dufton , q. w. 2006 , , 637 , 19 wu , q. , kim , j. , ryu , d. , cho , j. , & alexander , p. 2009 , , 705 , l86 wu , q. , kim , j. , & ryu , d. 2012 , , submitted lcccc adon15 & @xmath90 , 1.0 , 3.0 & @xmath91 , 1.0 , 3.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + ados15 & @xmath90 , 1.0 , 1.0 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + aqos15 & @xmath90 , 1.0 , 3.0 & @xmath91 , 1.0 , 3.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + adpn15 & @xmath90 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + adps15 & @xmath90 , 1.0 , 1.0 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + aqps15 & @xmath90 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + adxn15 & @xmath90 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + adxs15 & @xmath90 , 1.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + aqxs15 & @xmath90 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdon15 & @xmath91 , 1.0 , 3.0 & @xmath91 , 1.0 , 3.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdos15 & @xmath90 , 1.0 , 1.0 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + bqos15 & @xmath91 , 1.0 , 3.0 & @xmath91 , 1.0 , 3.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdpn15 & @xmath91 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdps15 & @xmath90 , 1.0 , 1.0 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + bqps15 & @xmath91 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdxn15 & @xmath91 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + bdxs15 & @xmath90 , 1.0 , 1.0 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 & @xmath90 , 0.5 , 0.1 + bqxs15 & @xmath91 , 1.0 , 1.0 & @xmath91 , 1.0 , 1.0 & @xmath91 , 0.5 , 0.1 & @xmath91 , 0.5 , 0.1 + adon30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + ados30 & @xmath90 , 2.0 , 1.0 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + aqos30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + adpn30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + adps30 & @xmath90 , 2.0 , 1.0 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + aqps30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + adxn30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + adxs30 & @xmath90 , 2.0 , 0.1 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + aqxs30 & @xmath90 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdon30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdos30 & @xmath90 , 2.0 , 1.0 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + bqos30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdpn30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdps30 & @xmath90 , 2.0 , 1.0 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + bqps30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdxn30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 + bdxs30 & @xmath90 , 2.0 , 1.0 & @xmath90 , 2.0 , 0.1 & @xmath90 , 1.0 , 0.1 & @xmath90 , 0.5 , 0.1 + bqxs30 & @xmath91 , 2.0 , 1.0 & @xmath91 , 2.0 , 1.0 & @xmath91 , 1.0 , 0.1 & @xmath91 , 0.5 , 0.1 llrr 0.5 & 0.1 & 1.02702 & 1.00360 + & 1.0 & 1.03139 & 1.03748 + & 10 . & 1.02755 & 1.34981 + 1.0 & 0.1 & 1.09307 & 1.01275 + & 1.0 & 1.11417 & 1.14356 + & 3.0 & 1.10303 & 1.40630 + & 10 . & 1.11177 & 2.09291 + 2.0 & 0.1 & 1.29744 & 1.04641 + & 1.0 & 1.32886 & 1.45009 + & 10 . & 1.32293 & 3.04786 lrrcrr adon & 0.00 & 0.18 & & 0.06 & 0.44 + ados & 0.02 & 1.92 & & 0.03 & 4.21 + aqos & 0.02 & 0.39 & & 0.03 & 0.90 + adpn & -4.76 & 0.45 & & -4.83 & 0.87 + adps & 5.58 & 1.95 & & 5.66 & 4.22 + aqps & 5.58 & 0.36 & & 5.65 & 0.78 + adxn & -4.12 & 0.13 & & -4.12 & 0.29 + adxs & 4.69 & 1.97 & & 4.70 & 4.28 + aqxs & 4.69 & 0.51 & & 4.70 & 1.11 + bdon & 0.00 & 0.14 & & 0.05 & 0.35 + bdos & 0.02 & 1.87 & & 0.03 & 4.08 + bqos & 0.02 & 0.32 & & 0.03 & 0.75 + bdpn & -4.76 & 0.41 & & -4.83 & 0.80 + bdps & 5.58 & 1.90 & & 5.65 & 4.09 + bqps & 5.58 & 0.31 & & 5.65 & 0.66 + bdxn & -4.12 & 0.10 & & -4.12 & 0.24 + bdxs & 4.69 & 1.91 & & 4.70 & 4.14 + bqxs & 4.68 & 0.44 & & 4.70 & 0.96

manipulating of quantum systems which consist of a few quantum dots or quantum bits forms a basis of a functioning of any quantum computing device . recent experimental advances allow to assemble and manipulate nanomagnets consisting of just a few atoms and perform measurements on nanostructures with an atomic resolution . however , a theoretical understanding of microscopic quantum systems out of equilibrium ( e.g. under constant pumping or continuous measurement by a quantum probe ) is far from being complete , apart from simplest cases like a single two - level system or a quantum harmonic oscillator under external pumping or in contact with a reservoir . coupling of a quantum system with an environment ( or with a measuring apparatus ) is described under standard assumptions @xcite , @xcite in the framework of a lindblad master equation ( lme ) for a reduced density matrix , where a unitary evolving part is complemented with a lindblad dissipative action . under the lme dynamics , a system with gradients evolves towards a nonequilibrium steady state ( ness ) with currents . the heisenberg model of interacting spins is an oldest many - body quantum model of a strongly correlated systems , important both from a theoretical and an experimental point of view . energy and magnetization transport in the @xmath0 heisenberg spin chain is extensively studied by different methods , see @xcite for reviews , including bethe ansatz @xcite , bosonization @xcite , @xcite , lagrange multipliers @xcite , exact diagonalization @xcite , lanczos method @xcite , quantum monte carlo @xcite , dmrg @xcite . investigation of nonequilibrium spin chains from the point of view of markovian master equations contributes to the `` equilibrium '' studies . in quasi one - dimensional spin chain materials like srcuo@xmath5 many transport characteristics are measurable experimentally @xcite . investigation of ness in driven spin chains is hindered by enormous technical difficulties due to an exponentially growing hilbert space , although time - dependent dmrg method allows to reach somewhat larger sizes @xcite . in many studies the ness are produced by coupling a spin chain at the edges to boundary baths of polarization aligning boundary spins along the anisotropy axis @xmath6 , so that the first and the last spin tend to be antiparallel @xcite . for this particular choice of boundary baths , a substantial progress in understanding the properties of a driven @xmath0 model have been recently achieved @xcite , including an exact solution of the problem in case of strong boundary driving @xcite , see also @xcite and references therein . note that within this boundary setup , one has the transport of a @xmath7-component of magnetization @xmath8 , caused by boundary gradients along the @xmath7-axis , thus probing diagonal elements of a diffusion matrix and a conductivity tensor . in present paper we carry out a systematic study of nonequilibrium steady states in driven @xmath0 spin chain resulting from boundary gradients in the @xmath2 direction , transverse to the anisotropy axis . in the context of spin and energy transport and for appropriate limits , this would correspond to probing the nondiagonal elements of the diffusion and conductivity tensors . to this end , we bring an @xmath0 spin chain in contact with reservoirs aligning the boundary spins along the @xmath9-axis at one end and along the @xmath10-axis in the other end , thus trying to impose a twisting angle in @xmath2-plane of @xmath11 between the first and the last spin . within the lme formalism , twisting angles @xmath12 or @xmath13 result in a complete suppression of the magnetization current , see a remark after eq.([currentnonballistic ] ) , making a setup with @xmath14 the simplest nontrivial one . we derive general symmetries of the model which indicate that magnetization current should alternate its sign with system size , for any value of boundary coupling . on this basis we rule out the possibility of a ballistic spin current , which is typically observed in integrable @xmath0 model under other kinds of perturbations @xcite,@xcite . we argue , further , that the phenomenon of the spin current alternation is rather general and in particularly is not related to the integrability . to confirm further our findings we investigate the nonequilibrium steady state of the @xmath0 model numerically and analytically , for varying amplitudes of boundary couplings and boundary gradients . we find complete agreement with our predictions which in particular are confirmed analytically , in the regime of strong coupling . in this regime , full analytic solution of the lindblad equation for large lindblad amplitudes demonstrates a very nonmonotonic dependence of the magnetization current on the system size and anisotropy . the plan of the paper is the following . in sec.[sec::the model ] we introduce the model . symmetries of the lindblad master equation are obtained in sec.[sec::symmetries of the lindblad equation and restrictions on energy and magnetization currents ] . using the symmetries general ness properties for arbitrary system size and coupling are revealed , including an admissibility of spin and energy currents . parity - depending lme symmetries lead to a phenomenon of magnetization current sign alternation with the system size . in sec . [ sec::strongly driven xxz chain : numerical and analytical study ] our predictions are tested further by calculating the ness analytically and numerically for small system sizes . the cases of @xmath15 model and of @xmath0 model are treated separately in sec.[subsec::free fermion point . arbitrary n ] and sec.[subsec::arbitrary delta . finite size results ] . limits of weak coupling and of weak driving force are briefly discussed at the end of sec.[subsec::generic gamma and arbitrary driving ] . appendix contain necessary technical details . in conclusion we summarize our findings . we study an open chain of @xmath16 quantum spins in contact with boundary reservoirs . the time evolution of the reduced density matrix @xmath17 is described by a quantum master equation in the lindblad form @xcite,@xcite , @xcite ( here and below we set @xmath18 ) @xmath19 + \gamma(% % tcimacro{\tcilaplace}% % beginexpansion \mathcal{l}% % endexpansion _ { l}[\rho]+% % tcimacro{\tcilaplace}% % beginexpansion \mathcal{l}% % endexpansion _ { r}[\rho ] ) , \label{lme}%\ ] ] where @xmath20 is the hamiltonian of an extended quantum system , and @xmath21 $ ] and @xmath22 $ ] are lindblad dissipators acting on spatial edges of the system . this setting , shown schematically in fig.[fig_lindbladreservoirs ] , is a common starting point in studies of nonequilibrium transport , a particular choice of @xmath23 and @xmath24 depending on the application . often , a choice of the lindblad action is operational , i.e. is determined by a requirement to favour a relaxation of a system or its part towards a target state with given properties , e.g. to a state with given polarization or temperature @xcite . in this way , one describes an effective coupling of the system , or a part of it , with a respective bath - environment . note that within the quantum protocol of repeated interactions , @xcite , @xcite the equation ( [ lme ] ) appears to describe an exact time evolution provided that the coupling to reservoir is rescaled appropriately with the time interval between consecutive interactions of the system with the reservoirs . we specify the hamiltonian to describe an open @xmath0 spin chain with an anisotropy @xmath3 @xmath25 while @xmath23 and @xmath24 are lindblad dissipators favouring a relaxation of boundary spins @xmath26 and @xmath27 towards states described by one - site density matrices @xmath28 and @xmath29 , i.e. @xmath30=0 $ ] and @xmath31=0 $ ] . general form of the matrices @xmath28 and @xmath29 is @xmath32 , where @xmath33 are pauli matrices , and @xmath34 are real constants satisfying @xmath35 or @xmath36 . for a general choice of the lindblad dissipator @xmath37=\sum_{i , j=1}^{3}a_{ij}^{l , r}\left ( \sigma^{i}.\text { } \sigma ^{j}-\frac{1}{2}\{\text { } .\text { } , \sigma^{j}\sigma^{i}\}\right ) $ ] , each of the @xmath38 matrices @xmath28 and @xmath29 , if not constant , belongs to at most one - dimensional manifold @xcite . we choose boundary reservoirs which tend to align the spin at the left edge along the @xmath39 direction and the spin at the right edge along the @xmath40 direction . consequently , the reservoirs try to establish a twisting angle of @xmath11 between the first and the last spin in the chain . such a setting is achieved by taking the lindblad action in the form @xmath41 & = -\frac{1}{2}\sum_{m=1}^{2}\left\ { \rho , w_{m}^{\dag}% w_{m}\right\ } + \sum_{m=1}^{2}w_{m}\rho w_{m}^{\dag},\label{lindbladaction}\\% % tcimacro{\tcilaplace}% % beginexpansion \mathcal{l}% % endexpansion _ { r}[\rho ] & = -\frac{1}{2}\sum_{m=1}^{2}\left\ { \rho , v_{m}^{\dag}% v_{m}\right\ } + \sum_{m=1}^{2}v_{m}\rho v_{m}^{\dag}.\nonumber\end{aligned}\ ] ] here @xmath42 and @xmath43 are polarization targeting lindblad operators , which act on the first and on the last spin respectively , @xmath44 in absence of the unitary term in ( [ lme ] ) the boundary spins relax with a characteristic time @xmath45 @xcite to specific states described via _ one - site _ density matrices @xmath28 and @xmath29 , satisfying @xmath30=0 $ ] and @xmath31=0 $ ] , where @xmath46 from the definition of a one - site density matrix @xmath47 , we see that the lindblad superoperators @xmath23 and @xmath24 indeed try to impose a twisting angle of @xmath11 in @xmath2 plane between the first and the last spin @xcite . the twisting gradient drives the system in a steady - state with currents . in the following we restrict to a symmetric choice@xmath48 the parameter @xmath49 determines the amplitude of the gradient between the left and right boundary , and therefore plays a fundamentally different role from the coupling strength @xmath50 . the limits @xmath51 and @xmath52 will be referred to as the strong driving and weak driving case respectively . the two limits describe very different physical situations as exemplified in other ness studies . the reason for choosing the lindblad dissipators which impose a twisting gradient in @xmath2-plane perpendicular to the heisenberg chain anisotropy is two - fold . firstly , our choice is `` orthogonal '' and therefore complementary to usually considered boundary setup @xcite-@xcite,@xcite-@xcite , where alignment of spins at the boundary is parallel to the anisotropy axis @xmath6 . our system is not mappable to a quadratic fermionic system , and is therefore not amenable to a analysis via a canonical quantization in the fock space of operators @xcite,@xcite , which can be used for usual boundary setup . secondly , one realizes that imposing both @xmath9- and @xmath10- boundary gradients is necessary to create a nonvanishing stationary @xmath6-magnetization current . a boundary setup with parallel or antiparallel boundary spins , aligned in the same direction in @xmath2- plane leads to zero steady magnetization current , see remark after ( [ j_heat ] ) . consequently , by imposing a twisting angle in @xmath2-plane of @xmath11 we choose a minimal nontrivial setup . transport properties in the @xmath0 spin chain @xmath53 are governed by spin and energy current operators which are defined by lattice continuity equations @xmath54 , @xmath55 where @xmath56@xmath57 we show next that the magnetization current @xmath58 alternates its sign with the system size , while the energy transport is switched off completely , if @xmath59 , despite the boundary gradients . being locally conserved quantities , in the stationary state the currents @xmath60 @xmath61 are equal across all the bonds . for the following is essential to note , that for our choice of the lindblad dissipator ( [ lindbladaction ] ) , ( [ lindbladoperators ] ) , the system ( [ lme ] ) has a _ unique _ nonequilibrium steady state . its existence and uniqueness for any coupling @xmath50 is guaranteed by the completeness of the algebra , generated by the set of operators @xmath62 under multiplication and addition @xcite , and is verified straighforwardly as in @xcite . let us denote by @xmath63 a time - dependent solution of the lindblad equation ( [ lme ] ) , ( [ lindbladoperators ] ) for a system of @xmath16 sites , and by @xmath64 @xmath63 its steady state solution , where @xmath3 is the anisotropy of the @xmath0 hamiltonian ( [ hamiltonian ] ) . there exist transformations which map one lme solution to another . some of the transformations depend on the parity of @xmath16 . for _ even _ @xmath16 , introducing @xmath65 , we find that @xmath66 is a solution of the same lindblad equation with the anisotropy @xmath67 . we denote by @xmath68 a complex conjugated matrix in the basis where @xmath69 is diagonal . however , if a nonequilibrium steady state is unique , then a relation follows which relates the ness for positive and negative @xmath3 , @xmath70 for _ odd _ @xmath16 we obtain another relation for ness , introducing a unitary transformation @xmath71 @xmath72 another lme symmetry does not depend on parity of @xmath16 and maps a ness onto itself provided that an additional condition ( [ symmetryu - uprime ] ) is satisfied : @xmath73 where @xmath74 is a left - right reflection , and the diagonal matrix @xmath75 is a rotation in @xmath2 plane , @xmath76 @xmath77 , @xmath78 , and @xmath79 . note also that at the point @xmath80 symmetries ( [ symmetryeven ] ) and ( [ symmetryodd ] ) map a ness onto itself , and are independent on ( [ symmetryglobal ] ) . the relations ( [ symmetryeven ] ) ( [ symmetryglobal ] ) impose restrictions on the ness density matrices and respectively on all observables including the magnetic and energy currents @xmath81 and @xmath82 where @xmath83 and @xmath84 are given by ( [ jz ] ) and ( [ q ] ) . we drop the subscripts since in the steady state the currents @xmath85 are equal across all bonds . by virtue of the symmetry ( [ symmetryeven ] ) , we obtain for @xmath16 even @xmath86 in the last passage we used ( i ) @xmath87 @xmath88 in the basis where @xmath69 is diagonal ( ii ) @xmath89 ( iii ) the fact that an observable @xmath58 is a real number . analogously , for odd @xmath16 , we obtain using ( [ symmetryodd ] ) , that @xmath90 . so , we find that the magnetization current changes its parity from even to odd as a function of the anisotropy @xmath3 : @xmath91 we argue below , see sec.[subsec::free fermion point . arbitrary n ] , that at the free fermion point @xmath80 the spin current vanishes always , @xmath92 . for odd @xmath16 it follows from ( [ j_odd ] ) , while for even @xmath16 it follows from a perturbative expansion . in addition , we make another suprising observation : due to ( [ j_even ] ) , ( [ j_odd ] ) the spin current alternates its sign with the system size since by increasing a system size by one unit @xmath93 the current dependence on @xmath3 changes from even to odd or vice versa . for our case it amounts to @xmath94 because in the region @xmath4 the sign of @xmath95 can change depending on other parameters , see sec.[subsec::arbitrary delta . finite size results ] . the alternating current effect is a consequence of an existence of different lme symmetries for even and odd number of sites ( [ symmetryeven ] ) , ( [ symmetryodd ] ) and as such is quite general . indeed the derivation of ( [ j_even]),([j_odd ] ) does not depend neither on coupling @xmath50 now on gradients @xmath49 and @xmath96 as given by ( [ lindbladoperators ] ) . also , the phenomenon of alternating current does not rely on an integrability of the @xmath0 model . indeed , an addition of a staggered anisotropy to the hamiltonian ( [ hamiltonian ] ) breaks integrability but the eqs . ( [ symmetryeven ] ) , ( [ symmetryodd ] ) remain valid . moreover , even though our conclusion relied on the validity of eqs . ( [ j_even ] ) , ( [ j_odd ] ) , the alternating current was also observed in a recent study of a driven @xmath0 chain with an @xmath2 boundary setup different from ( [ lindbladoperators ] ) , where one of two symmetries ( eq([j_even ] ) ) was absent @xcite . we suggest that the magnetization current sign alternation resulting from applying transverse gradients can be a rather general phenomenon , which might be observable under appropriate experimental conditions in artificially assembled nanomagnets consisting of just a few atoms @xcite . another consequence of the current sign alteration is that the spin current _ can not be ballistic_. indeed , for large @xmath16 we should not be able to differentiate between even and odd @xmath16 , so that the stationary magnetization current @xmath58 must vanish in the thermodynamic limit:@xmath97 this conclusion is a rather unexpected one , since in the @xmath0 integrable system , in the critical region @xmath4 one usually expects a ballistic current @xcite . the magnetization current is also ballistic @xmath4 for reservoirs creating boundary gradients along the anisotropy axis @xcite . up to now we did not make use of the symmetry ( [ symmetryglobal ] ) . in fact , it does not impose any further restrictions for spin current , while the energy current changes its sign under the symmetry ( [ symmetryglobal ] ) and must therefore vanish @xmath98 consequently , the energy and spin currents decouple : the spin current can flow , while the energy current is totally suppressed ( [ j_heat ] ) . vanishing of the energy current is a consequence of a symmetric choice @xmath99 ( [ symmetryu - uprime ] ) , and will be lifted if @xmath100 . * remark*. note that if both boundary reservoirs were aligning boundary spins along the same direction in @xmath2-plane , the magnetization current will be completely suppressed for all @xmath3 , despite the existence of boundary gradients . e.g. for a choice of the set of lindblad operators @xmath101 @xmath102 @xmath103 in ( [ lindbladaction ] ) , which favours the alignment of the boundary spins along the @xmath9 axis , the current suppression follows from a symmetry @xmath104 of the ness , since @xmath105 . thus , our choice ( [ lindbladaction ] ) is the minimal nontrivial one , to observe a nonzero magnetization current . in the following we check our findings and analyze the model further by computing exact ness numerically for finite @xmath50 and analytically for @xmath106 using a perturbative expansion . we performed studies of the lme for small system sizes @xmath16 in order to support our findings ( [ j_even]-[j_heat ] ) , which should be valid for all @xmath107 and @xmath50 . for simplicity we restrict here to the strongly driven case @xmath108 . being a problem of exponential growing complexity , the steady state for a finite @xmath50 is a very complicated function of the parameters @xmath109 , for any @xmath110 , and can not reported otherwise than in a graphical form . we solve the linear system of equations which determine the full steady state , numerically , making use of the global symmetry ( [ symmetryglobal ] ) which decrease the number of uknowns by roughly the factor of @xmath1 . for @xmath111 the magnetization currents as function of the anisotropy , for various values of @xmath50 are reported in fig.[fig_j3],[fig_j4 ] . as expected from ( [ j_even]),([j_odd ] ) , the current is an odd ( even ) function of @xmath3 for @xmath112 . interestingly , as the coupling @xmath50 increases , the current @xmath113 approaches quickly a limiting curve drawn by a dashed line . for the dashed line , which corresponds to the the limit of infinitely strong couplings @xmath106 , a simple closed analytic form can be obtained . here it is worthwhile to note that lindblad master equation dynamics ( [ lme ] ) with large effective coupling to a reservoir @xmath114 is realizable within the experimental protocol involving repeated periodic interactions of a boundary spin with a infinite bath system of identical ancilla @xcite . vs @xmath3 for @xmath115 for different @xmath116 ( from bottom up ) . for finite @xmath50 , the data are obtained numerically . dashed line marks @xmath95 in the limit @xmath117 from ( [ j3]),scaledwidth=40.0% ] vs @xmath3 for @xmath118 for different @xmath119 ( from bottom up ) . the @xmath95 curves continue symmetrically in the negative @xmath3 region . for finite @xmath50 the data are obtained numerically . dashed line marks @xmath95 in the limit @xmath106 from ( [ j4 ] ) . , scaledwidth=40.0% ] to investigate the strong coupling limit @xmath114 we search for a stationary solution of the lindblad equation in the form of a perturbative expansion @xmath120 as stressed in the end of sec.[sec::the model ] , the stationary solution @xmath121 is unique . the zeroth order term of the expansion ( [ pt_largecouplings ] ) @xmath122 @xmath123 satisfies @xmath124=0 $ ] ( here and below we denote by @xmath125 the sum of the lindblad actions in ( [ lme ] ) ) . this enforces a factorized form @xmath126 where @xmath28 and @xmath29 are one - site density matrices ( [ rol ] ) , ( [ ror ] ) and @xmath127 is a matrix to be determined self - consistently later . we shall drop @xmath3- and @xmath16-dependence in @xmath17 for brevity of notations . we separate the identity matrix @xmath128 for future convenience , so that @xmath129 and @xmath127 is traceless . inserting the ( [ pt_largecouplings ] ) into ( [ lme ] ) , and comparing the orders of @xmath130 , we obtain recurrence relations@xmath131=\frac{1}{2}% % tcimacro{\tcilaplace}% % beginexpansion \mathcal{l}% % endexpansion _ { lr}\rho_{k+1}\text { , \ } k=0,1,2 ... \ ] ] a formal solution of the above is @xmath132 where @xmath133 $ ] . note however that the operator @xmath134 has a nonempty kernel subspace , and is not invertible on the elements from it . the kernel subspace @xmath135 consists of all matrices of type @xmath136 where @xmath137 is an arbitrary @xmath138 @xmath139 matrix . therefore a necessary and sufficient condition for @xmath140 to exist is to require a null overlap @xmath141\cap\ker(% % tcimacro{\tcilaplace}% % beginexpansion \mathcal{l}% % endexpansion _ { lr})=\varnothing\label{secularconditionsgeneral}%\ ] ] we shall name ( [ secularconditionsgeneral ] ) the secular conditions . for our choice of the lindblad operator the secular condition are equivalent ( see the appendix ) to the requirement of a null partial trace @xmath142)=0\text { , \ } k=0,1,2 .... \label{secularconditions}%\ ] ] the remaining missing ingredient of the perturbation theory is obtained by noting that @xmath140 is defined up to an arbitrary element from @xmath135 , so we have @xmath143)+\rho_{l}\otimes m_{k+1}\otimes\rho_{r}\text { , \ \ \ } k=0,1,2 ... \label{recurrence}%\ ] ] eqs ( [ initialchoicero0 ] ) , ( [ recurrence ] ) and ( [ secularconditions ] ) define a perturbation theory for the lindblad equation ( [ lme ] ) . at each order of the expansion , we must satisfy the secular conditions ( [ secularconditionsgeneral ] ) , before proceeding to the next order . construction of the inverse @xmath144 is illustrated in the appendix . applying the perturbation theory to our model , we find that the unknown apriori matrices @xmath145 are fully determined by secular conditions ( [ secularconditions ] ) for @xmath146 , as will be illustrated further on an example . for any @xmath147 the set of matrices @xmath148 is nontrivial . the case of zero anisotropy @xmath80 turns out to be special in that all @xmath149 and is discussed separately below . for @xmath80 we can guess a zero approximation @xmath150 for arbitrary @xmath16 : @xmath151 comparing with ( [ initialchoicero0 ] ) , we see that @xmath152 . all further @xmath153 and all secular conditions ( [ secularconditions ] ) are trivially satisfied at all orders . this fact was checked explicitly for small system sizes @xmath154 and is conjectured to be true for arbitrary @xmath16 . consequently , the general recurrence ( [ recurrence ] ) is reduced to @xmath155)$ ] for all @xmath156 . at the first order of expansion , we obtain @xmath157=q_{1}$ ] , where @xmath158 which is easily inverted @xmath159 and so on . at all orders we notice @xmath160 , which allows to conjecture that magnetization current vanishes for @xmath80 in spite of twisting boundary gradients , @xmath161 note that for odd @xmath16 , the statement ( [ zerocurrdelta0 ] ) also follows from ( [ j_odd ] ) . for @xmath115 , the perturbative expansion ( [ pt_largecouplings ] ) can be summed up for all orders due to the fact that it closes at the third step : @xmath162 . consequently , ( [ pt_largecouplings ] ) is rewritten as @xmath163 where @xmath164 and @xmath165 . summing up the series , we obtain @xmath166 it can be verified straightforwardly that ( [ rhode0_n=3 ] ) is a stationary solution of lme ( [ lme ] ) for @xmath167 . the sum ( [ n3_de0_strongcoupling ] ) converges for @xmath168 . however , since the lme solution is unique , the expression ( [ rhode0_n=3 ] ) is valid for any @xmath50 . in particular , for @xmath169 we obtain the weak coupling limit @xmath170 , which is also unique by a continuity argument , @xmath171 as expected , @xmath172 commutes with the @xmath173 hamitonian , @xmath174=0 $ ] , and corresponds to zero currents , @xmath175 . from the above example we see that perturbative series ( [ pt_largecouplings ] ) are useful even outside of their radius of convergence : if the perturbative series for @xmath176 or for a particular observable @xmath177 can be summed , the result of the summation is valid for all @xmath50 within its analyticity range , see also ( [ delta1_n3 ] ) , ( [ delta1_n4 ] ) . for any nonzero @xmath3 , the matrices @xmath179 from ( [ recurrence ] ) are nontrivial and are determined from secular conditions ( [ secularconditions ] ) . the secular conditions for two consecutive orders @xmath180 reduce to a system of linear equations which determine @xmath181 and @xmath182 . in such a way , the perturbative series ( [ pt_largecouplings ] ) is uniquely defined at all orders . below we present analytic results for @xmath127 and small @xmath16 which define the ness through eq.([initialchoicero0 ] ) . @xmath127 and @xmath183 are obtained by satisfying secular conditions ( [ secularconditions ] ) for @xmath184 and @xmath26 . the procedure is illustrated below for the simplest case @xmath111 . for @xmath115 , the most general form of the matrices @xmath185 , @xmath186 , by virtue of the symmetry ( [ symmetryglobal ] ) , is @xmath187 , @xmath188 where @xmath189 are unknown constants . the secular conditions ( [ secularconditions ] ) for @xmath184 do not give any constraints on @xmath190 , while those for @xmath26 give two nontrivial relations @xmath191 and @xmath192 from which both @xmath127 and @xmath183 are determined . for @xmath118 each of the matrices @xmath193 compatible with the symmetry ( [ symmetryglobal ] ) , contains @xmath194 unknowns , all fixed by the secular conditions ( [ secularconditions ] ) for @xmath195 , and so on . for @xmath111 we obtain @xmath196@xmath197 where @xmath198 . matrices @xmath127 obey the symmetries ( [ symmetryeven])([symmetryglobal ] ) . steady state spin currents @xmath199 in the limit @xmath200 are @xmath201@xmath202 while the energy current vanishes @xmath203 as expected from ( [ j_heat ] ) . we note that the current for @xmath118 is nonzero at @xmath80 which seems to contradict our prediction ( [ zerocurrdelta0 ] ) . the contradiction arises because the limits @xmath204 and @xmath205 do not commute . indeed , we find that even orders of the perturbative expansion @xmath206 contain terms which become singular for @xmath205 . the leading singular term at the order @xmath207 of the expansion is @xmath208 , where @xmath209 . since the singularities become stronger with @xmath210 , for any given infinitesimal @xmath211 there will be a region @xmath212 where the expansion ( [ pt_largecouplings ] ) will diverge . the origin behind the non - commuting limits is the existence of an extra symmetry ( [ j_even ] ) of the ness at the point @xmath80 . similar situation is encountered e.g. in kubo linear response theory , where a nonergodicity at zero point in the momentum space @xmath213 is inflicted by an extra conservation law at this point @xcite . reversing order of the limits @xmath214 and @xmath205 does yield @xmath215 for any @xmath50 , so that the current expression for @xmath216 has to be modified as @xmath217 while @xmath218 as given by ( [ r0_delta0 ] ) . for any finite @xmath50 the dependence @xmath95 is smooth and does not contain any singularities , see fig.[fig_j4 ] . we find the noncommutativity of the limits @xmath214 , @xmath219 for all even system sizes @xmath220 , see also ( [ j6 ] ) . since for large @xmath16 one should not be able to differentiate between even and odd system sizes , we expect that the @xmath221 for @xmath222 vanishes in the thermodynamic limit : @xmath223 . the number of equations to solve ( [ secularconditions ] ) at each order of perturbation theory grows exponentially with the system size @xmath16 , although the symmetry ( [ symmetryglobal ] ) restricts the number of independent variables . the matrices @xmath127 for @xmath224 are given by lengthy expressions and are not reproduced here . instead , we report ness currents at @xmath225 , given for @xmath226 by @xmath227@xmath228 the ness currents @xmath229 for odd / even number of sites @xmath16 are even / odd functions of @xmath3 respectively , as expected from ( [ j_even ] ) , ( [ j_odd ] ) and depend rather nontrivially on the anisotropy @xmath3 , see fig.[fig_j5 - 7 ] . in particular for @xmath224 one observes sharp peaks in the @xmath221 resembling resonances at which the current changes its sign and amplitude . one gets an insight by comparing average magnetization profiles @xmath230 and @xmath231 for values of @xmath3 inside and outside the resonance region . the average magnetization at boundaries @xmath232 is fixed by the lindblad reservoirs @xmath233 , @xmath234 and @xmath235 , @xmath236 so that the lowest interpolating fourier mode has the wavelength @xmath237 where @xmath238 is the lattice constant . in the resonance region we observe that @xmath239 @xmath240 becomes the dominant fourier mode , see fig.[fig_xyprofiles](b ) . one can interpret this frequency tripling as a triple increase of a twisting angle along the chain . note , see fig.[fig_xyprofiles](a , b ) that there is a @xmath11 phase shift between the dominant harmonics describing @xmath230 and @xmath231 profiles . in particular , in the resonance region , fig.[fig_xyprofiles](b ) , @xmath230 @xmath241 @xmath242 , where @xmath243 . in the meanfield approximation the respective magnetization current is @xmath244 , which is amazingly close to the exact current value at the resonance @xmath245 . with the increase of system size , one might expect an appearance of further resonances at wavelengths @xmath246 corresponding to larger twisting angles @xmath247 etc .. why higher fourier harmonics become stable under a strong lindblad action for narrow lacunae of @xmath3 , and how to predict a location of the peaks remains an intriguing open question . as coupling @xmath50 decreases , the resonances become smoother and then ultimately disappear , but their precursors can still be seen for @xmath50 of the order of the heisenberg exchange interaction and larger ( data not shown ) . examining analytic expressions for the current for small @xmath16 , we can extrapolate its behaviour for @xmath248 and @xmath249 and finite @xmath16 as @xmath250@xmath251 the eq.([limitjfordeltalarge ] ) implies an exponential decay of the current with the system size @xmath16 for fixed large @xmath3 , @xmath252 with @xmath253 , and is generically expected from an insulating nature of the @xmath0 chain in the @xmath248 ising limit . similar behaviour of the ness current at large @xmath3 was found for driven chain with @xmath6-polarized boundary baths @xcite . on the other hand , eq.([limitjfordeltasmall ] ) implies a complete flattening of the current near @xmath80 in the thermodynamic limit @xmath254 , which is a new unusual feature . the flattening is already well seen in fig.[fig_j5 - 7 ] for @xmath255 . note that the results ( [ limitjfordeltalarge ] ) , ( [ limitjfordeltasmall ] ) would be very difficult to obtain by studying perturbative series in @xmath3 or in @xmath256 : in both cases @xmath222 and @xmath257 the first nonzero term will appear only in the order @xmath258 of the expansion ! vs @xmath50 for @xmath111 ( thick and thin line , respectively ) for isotropic heisenberg model @xmath259 , given by ( [ delta1_n3]),([delta1_n4 ] ) . dashed lines indicate limiting values of the current for @xmath106.,scaledwidth=40.0% ] the analytic results derived in previous subsection concerned the limits of @xmath106 and strong driving @xmath108 . what can we say about generic @xmath50 and @xmath49 ? examining perturbative series ( [ pt_largecouplings ] ) for various @xmath16 and @xmath3 we notice that nonzero contributions to ness current appear only at even orders of the expansion @xmath260 . if the ness current @xmath113 is an analytic function of @xmath50 , then we can expect the current to be an even function of @xmath261 . in particular , for small @xmath262 one expects the the current to behave as @xmath263 , since the current must vanish at zero coupling @xmath264 . our expectations are confirmed by exact expressions for the current as function of @xmath50 , obtained for special case of isotropic @xmath265 model , for which one can sum up the series ( [ pt_largecouplings ] ) and obtain @xmath266 indeed the above expressions contain only even @xmath50 orders and @xmath263 , see also fig.[fig_jversusga ] . numerically , we observe @xmath263 for other @xmath3 values as well . generically , we find the current amplitude to be a smooth function of @xmath50 for a fixed value of the anisotropy @xmath3 . for large couplings , it asymptotically approaches its limiting value @xmath221 at @xmath200 . interestingly , the magnetization current is not suppressed for large couplings @xmath50 , as it happens for @xmath6-polarized lindblad baths @xcite,@xcite due to a quantum zeno effect . indeed for our choice of the bath ( [ lindbladoperators ] ) and large coupling , the quantum zeno effect `` freezes '' @xmath267 and @xmath268 spin components at the boundaries , while hoppings in the @xmath69 component , contributing to the magnetization current , may still occur . finally , we investigated the current dependence on the boundary gradient @xmath49 , which can be manipulated by a choice of the amplitudes of the lindblad operators in ( [ lindbladoperators ] ) . by fourier law , we may expect the current to be proportional to @xmath49 , for small gradients @xmath49 , and introduce a conductivity tensor @xmath269 as a proportionality coefficient between the current @xmath270 and the infinitesimal boundary gradient @xmath271 in the spin component @xmath272 , @xmath273 . with our perturbative method we construct iteratively an expansion ( [ pt_largecouplings ] ) for a ness for an arbitrary amplitude of the boundary gradient @xmath49 from ( [ lindbladoperators ] ) . we observe that the current is a rather nontrivial function of @xmath49 , and in particular that it vanishes as @xmath274 or faster for small @xmath49 . consequently , @xmath275 so that the off - diagonal elements of the conductivity tensor @xmath276 vanish . note that diagonal elements of the conductivity tensor are nonzero @xcite . numerically , we find that the leading order of expansion of @xmath277 for small @xmath49 depends on @xmath16 and @xmath3 in a nontrivial way . more details will be given elsewhere . finally , note that our model shows a behaviour fundamentally different from that with usually considered boundary driving along the @xmath6-axis , referred to below as a `` scalar '' setup . just to mention two crucial differences : the scalar setup leads to a ballistic current in the critical gapless region of the @xmath0 model @xmath278 @xcite , while in the @xmath2-driven chain the alternating signs phenomenon ( [ symmetryeven]-[currentsignalternation ] ) rules out the possibility of a ballistic current . in the `` scalar '' setup the current at strong coupling is suppressed due to a zeno effect , @xmath279 , while in the @xmath2-driven chain of finite size the spin current in the limit @xmath106 remains finite . we considered an open @xmath0 spin 1/2 chain with a twisting in @xmath2- plane , imposed by boundary gradients . symmetries of the lindblad equation were found which impose drastic restrictions on the steady state reduced density matrix , and various observables . for our boundary setup the ness energy current vanishes , and the magnetization current alternates its sign with system size , which rules out the possibility of a ballistic current . we argued that the current sign alternation under trasverse gradients is a robust phenomenon in quantum transport which does not rely on integrability . the qualitative difference in quantum transport in even- and odd - sized systems adds to a list of similar phenomena observed in quantum diffusion in periodic chains @xcite . to support our findings , we constructed a nonequilibrium steady state solution of the lindblad master equation in the form of a perturbative expansion in orders of @xmath45 , and calculated the zeroth - order term analytically for small system sizes , by solving a set of linear equations which guarantee the self - consistency of the expansion , the secular conditions ( [ secularconditions ] ) . we find that , for @xmath147 , the current remains finite even in the limit of infinitely strong effective coupling . further on , for large couplings we find a nontrivial dependence of the magnetization current on the anysotropy @xmath95 , and observed sharp peaks in @xmath95 inside the critical gapless region of the @xmath0 model @xmath278 , the origin of which was attributed to an appearance of higher fourier harmonics in magnetization profiles . we observe an anomalous flattening of the current near @xmath80 point for odd @xmath16 and noncommutativity of limits @xmath205 and @xmath204 for even @xmath16 . for weak driving @xmath52 and small system sizes , we find @xmath275 for all values of @xmath3 , signalizing subdiffusive current in the direction trasverse to @xmath2 boundary gradients . * acknowledgements * the author thanks n. plakida , r. vaja , m. g. pini , m. salerno and r. moessner for stimulating discussions . support from the italian miur through prin 20083c8xfz initiative is acknowledged . @xmath23 and @xmath24 are linear super - operators acting on a matrix @xmath17 as defined by ( [ lindbladaction ] ) . in our case , each super - operator act locally on a single qubit only . we find the eigen - basis @xmath280 of @xmath281 in the form @xmath282 where @xmath283 are pauli matrices and @xmath284 . the respective eigenvalues are @xmath285 . analogously , the eigen - basis and eigenvalues of the eigenproblem @xmath286 are @xmath287 and @xmath288 , where @xmath289 . since the basises @xmath290 and @xmath291 are complete , any matrix @xmath292 acting in the appropriate hilbert space can be expanded as @xmath293 in the unique way . indeed , let us introduce complementary basises @xmath294 and @xmath295 , trace - orthonormal to the @xmath296 respectively , @xmath297 , @xmath298 . then , the coefficients of the expansion ( [ ro_expansion ] ) are given by @xmath299 . on the other hand , in terms of the expansion ( [ ro_expansion ] ) the superoperator inverse @xmath300 is simply@xmath301 note however that the above sum contains a singular term with @xmath302 , because @xmath303 to eliminate the singularity , one must require @xmath304 , which produces the secular conditions ( [ secularconditions ] ) . within the repeated interaction scheme @xcite , @xcite , the lme ( [ lme ] ) describes a system evolution where coupling to reservoir is rescaled as @xmath305 where @xmath306 is a time interval between the two successive interactions with the ancilla bath . the relaxation times depend on spin component , the targeted spin component relaxing two times faster than other spin components . e.g. at the left boundary , @xmath39- and @xmath7-spin components of @xmath17 relax as @xmath307 while the @xmath40-spin component relax as @xmath308 , see @xcite .

dynamic coupling of ferromagnets has been of interest in the field of magnetism . the dipole interaction has been the basic interaction to excite the coupled motion of the magnetizations , and is applied to magnetic recording @xcite . another method to realize the coupling is to use the spin - transfer effect @xcite , where the application of an electric current to ferromagnetic / nonmagnetic multilayers results in the magnetization switching and self - oscillation @xcite . the coupled dynamics through pure spin current generated in spin pumping @xcite and nonlocal @xcite geometries have also been observed . it should be emphasized that these couplings are strongly restricted by the characteristic length scales . for example , the dipole coupling decays according to the inverse cube detection law , whereas the spin transfer effect by a spin - polarized electric or pure spin current occurs in a system smaller than the spin diffusion length . recently , physical phenomena , such as the spin torque @xcite and magnetoresistance effects @xcite , due to the spin hall effect @xcite in bilayers consisting of an insulating or metallic ferromagnet and a nonmagnetic heavy metal have attracted much attention . the latter , known as the spin hall magnetoresistance , originates from the charge - spin conversion of an external electric current by the direct and inverse spin hall effects , and has been observed by measuring the longitudinal and transverse electric currents , which are given by @xmath0 @xmath1 respectively , where @xmath2 is the electric current density generated by the external electric field . the definitions of the dimensionless coefficients , @xmath3 , @xmath4 , and @xmath5 , are given below . it should be emphasized that the currents given by eqs . ( [ eq : longitudinal_current ] ) and ( [ eq : transverse_current ] ) depend on the magnetization direction @xmath6 . when another ferromagnet is placed onto the same nonmagnet , these currents will be converted to spin current by the spin hall effect again , and excite spin torque on this additional ferromagnet . then , the magnetization dynamics of two ferromagnets will be coupled through the angular dependencies of the electric current given by eqs . ( [ eq : longitudinal_current ] ) and ( [ eq : transverse_current ] ) . this coupling is unavoidable whenever several ferromagnets are placed onto the same nonmagnet , and is not restricted by the distance between the ferromagnets because it is carried by the electric current . since the structure consisting of several ferromagnets on the nonmagnetic heavy metal will be important from the viewpoints of both fundamental physics and practical applications based on the spin hall effect , such as magnetic random access memory , spin torque oscillators , and bio - inspired computing @xcite , it is of interest to clarify the role of this coupling . in this paper , we investigate the coupled dynamics of magnetizations in ferromagnets in the presence of the spin hall effect by solving the landau - lifshitz - gilbert ( llg ) equation numerically for both longitudinal and transverse geometries . in addition to the external electric current , the current contributing to the spin hall magnetoresistance also excites the spin torque . the strength of this additional torque is estimated from the theory of the spin hall magnetoresistance extended to the system consisting of several ferromagnets . the conventional spin torque is proportional to the spin hall angle @xmath7 , whereas the new torque is on the order of @xmath8 , and therefore , its value is two orders of magnitude smaller than the conventional spin torque . nevertheless , it is found that this additional new torque affects the phase difference of the magnetization in the self - oscillation state . the numerical simulation reveals that the in - phase or antiphase synchronization is observed in the transverse geometry . on the other hand , in addition to them , the phase difference becomes a quarter of a period in the longitudinal geometry . it is found that these phase differences depend on the values of the gilbert damping constant and the dimensionless field - like torque strength . an analytical theory clarifying the relation among the current , frequency , and phase difference is also developed . the paper is organized as follows . in sec . [ sec : system description and llg equation ] , we describe the system under consideration , and discuss the theoretical formula of the spin torque excited by the spin hall effect in the presence of the several ferromagnets and the spin hall magnetoresistance effect . in sec . [ sec : numerical analysis of synchronization ] , we study the phase differences in the synchronized state of the magnetizations for both the longitudinal and transverse geometries by solving the llg equation numerically . in sec . [ sec : theoretical analysis of current - frequency relation ] , the theory clarifying the relation among the current , frequency , and phase difference is developed based on the llg equation averaged over constant energy curves . the summary of the paper is given in sec . [ sec : conclusion ] . ( @xmath9 ) are placed onto the same nonmagnet n. the external electric field is applied to the @xmath10 direction . ( b ) schematic view of the generations of electric currents by the direct and inverse spin hall effect ( she and ishe ) in the longitudinal geometry . the total electric current is the sum of the conventional electric current @xmath11 and the current generated near the f@xmath12/n and f@xmath13/n interfaces , @xmath14 and @xmath15 . ] in this section , we describe the system adopted in this study , and show the spin torque formulas including the coupling torques between the ferromagnets . the system we consider is schematically shown in fig . [ fig : fig1](a ) , where three ferromagnets f@xmath16 ( @xmath9 ) are placed onto the same nonmagnet n. we assume that the material parameters in the ferromagnets are identical , for simplicity . the external electric field , @xmath17 , is applied to the @xmath10 direction , inducing the electric current density @xmath11 , where @xmath18 is the conductivity of the nonmagnet . the direct and inverse spin hall effects produce electric currents in the longitudinal ( @xmath10 ) and transverse ( @xmath19 ) directions . these electric currents are converted to the spin current and injected into the f@xmath13 and f@xmath20 layers due to the spin hall effect , resulting in the excitation of the spin torque . then , the magnetization dynamics in the f@xmath16 ( @xmath21 ) layer is affected by that in the f@xmath12 layer , and vice versa . we call the coupling between the f@xmath12 and f@xmath13 layers the longitudinal coupling , whereas that between the f@xmath12 and f@xmath20 layers the transverse coupling . we assume that both the ferromagnet and nonmagnet are metallic because metallic bilayers are generally used to measure the magnetization switching and oscillation by the spin hall effect @xcite . although the spin hall magnetoresistance was originally studied for insulating ferromagnets @xcite , a large spin hall magnetoresistance in metallic system has also been reported recently @xcite . the dimensionless coefficients , @xmath3 , @xmath4 , and @xmath5 , in eqs . ( [ eq : longitudinal_current ] ) and ( [ eq : transverse_current ] ) for single ferromagnets have been derived for an insulating @xcite or metallic @xcite ferromagnet , which are given by @xmath22 \tanh^{2 } \left ( \frac{d_{\rm n}}{2 \lambda_{\rm n } } \right ) , \label{eq : chi } \end{split}\ ] ] @xmath23 @xmath24 where @xmath25 and @xmath26 are thickness and spin diffusion length of the nonmagnet , respectively , whereas @xmath27 with the cross section area of the ferromagnetic / nonmagnetic interface @xmath28 . the dimensionless mixing conductance @xmath29 consists of its real and imaginary parts @xcite , and @xmath30 is defined as @xmath31 here , @xmath32 is the sum of the conductances of the spin - up and spin - down electrons , whereas @xmath33 is its spin polarization . the ferromagnetic / nonmagnetic interface resistance @xmath34 is related to @xmath35 via @xmath36 . we also introduce @xmath37 , where @xmath38 is the conductivity of the ferromagnet and @xmath39 is its spin polarization . the thickness and spin diffusion length of the ferromagnet are denoted as @xmath40 and @xmath41 , respectively . the term related to @xmath30 is neglected when the ferromagnet is an insulator @xcite , i.e. , @xmath42 . the following quantities correspond to the effective spin polarizations of the damping - like ( or slonczewski @xcite ) torque and the field - like torque , respectively , @xmath43 for the later discussion , we introduce @xmath44 which corresponds to the ratio of the field - like torque to the damping - like torque . the values of the parameters used in the following calculations are derived from recent experiments on the w / cofeb heterostructure @xcite , where @xmath45 k@xmath46 nm , @xmath47 , @xmath48 nm , @xmath49 k@xmath46 nm , @xmath50 nm , and @xmath51 , whereas the thicknesses are assumed to be @xmath52 nm and @xmath53 the interface conductances were not evaluated in ref . @xcite by assuming a transparent interface . instead , we use typical values of the interface conductances obtained from the first - principles calculations @xcite , @xmath54 k@xmath46nm@xmath55 , @xmath56 , and @xmath57 nm@xmath58 . we note that the imaginary part of the mixing conductance , @xmath59 , is either positive or negative , depending on the material and thickness @xcite . the sign of @xmath59 determines those of @xmath4 and @xmath60 , or equivalently , @xmath61 . for example , when @xmath62 nm@xmath58 , @xmath63 , @xmath64 , @xmath65 , @xmath66 , and @xmath67 ( @xmath68 ) ; see appendix [ sec : values of parameters in numerical simulations ] . in the following calculations , we study the magnetization dynamics for several values of @xmath61 . equation ( [ eq : longitudinal_current ] ) was derived for a system having single ferromagnet . in this case , @xmath2 is the external electric current density , whereas @xmath69 is the current density generated as a result of the charge - spin conversion by the direct and inverse spin hall effects . in the longitudinal geometry in the present study , on the other hand , two ferromagnetic / nonmagnetic interfaces , i.e. , f@xmath12/n and f@xmath13/n interfaces , contribute to the generation of the longitudinal current through the direct and inverse spin hall effects , as schematically shown in fig . [ fig : fig1](b ) . let us denote the electric current density generated by these effects near the f@xmath16/n interface as @xmath70 ( @xmath71 ) . this current density is determined by the conservation law of the electric current , as follows . considering in a similar manner to the case of the single ferromagnet , the electric current @xmath72 is converted to the spin current by the spin hall effect near the f@xmath13/n interface , and this spin current produces an additional electric current @xmath73 by the inverse spin hall effect . therefore , the total longitudinal electric current density near the f@xmath13/n interface is @xmath74 . similarly , the electric current density near the f@xmath12/n interface can be expressed as @xmath75 . these currents should be equal to the total electric current density , @xmath76 , according to the conservation law of the electric current . then , we find that @xmath70 is given by @xmath77 where we neglect the higher order terms of @xmath7 . then , the spin torque excited on the f@xmath16 layer by the spin hall effect in the longitudinal geometry is obtained by replacing the external electric current density @xmath11 in the previous work @xcite with @xmath78 , and is given by @xmath79 where @xmath80 or @xmath81 . the unit vector pointing in the magnetization direction of the f@xmath16 layer is @xmath82 , and @xmath83 , @xmath84 , and @xmath85 are the gyromagnetic ratio , the saturation magnetization , and the thickness of the ferromagnet , respectively . note that the terms , @xmath86 in eq . ( [ eq : torque_l ] ) are the conventional spin torques generated by the external electric current @xmath2 and the spin hall effect , and are often referred to as the damping - like torque and the field - like torque , respectively . on the other hand , the terms @xmath87 in eq . ( [ eq : torque_l ] ) originate from the current @xmath88 , and are newly introduced in this study . it should be emphasized that eq . ( [ eq : torque_l_coupling ] ) depends on the magnetization direction of the other ferromagnet f@xmath89 ( @xmath90 or @xmath91 ) , @xmath92 , resulting in the coupling between the f@xmath12 and f@xmath13 layers . let us next show the spin torque formulas in the transverse geometry . we denote the electric current density flowing in the @xmath19 direction generated near the f@xmath16/n ( @xmath93 ) interface by the inverse spin hall effect as @xmath94 . the conservation law of the total electric current density , @xmath95 , gives ( see also appendix [ sec : derivation of coupling torque in transverse geometry ] ) @xmath96 j_{0 } \\ & \simeq -\left ( \chi m_{\ell x } m_{\ell y } + \chi^{\prime } m_{\ell z } \right ) j_{0}. \label{eq : electric_current_y } \end{split}\ ] ] in addition to the conventional spin torque , eq . ( [ eq : torque_conventional ] ) , this transverse electric current also excites the spin torque on the other ferromagnet . in total , the spin torque acting on the f@xmath16 layer is given by [ @xmath97 or @xmath98 @xmath99 the last two terms represent the coupling torque between the f@xmath12 and f@xmath20 layers . note that the direction of this coupling torque is different from that of the conventional torque because the currents @xmath2 and @xmath94 flow in different directions . in the following , the torques related to @xmath3 , @xmath4 , and @xmath5 in eqs . ( [ eq : torque_l ] ) and ( [ eq : torque_t ] ) are referred to as coupling torque . the ratio of these new torques to the conventional spin torque is on the order of @xmath100 . since the conventional spin torque due is proportional to the spin hall angle @xmath101 , the coupling torque is proportional to @xmath8 . although the spin hall angle is usually a small quantity , it will be shown that the coupling torques play a non - negligible role on the magnetization dynamics , as shown below . in the following sections , we study the magnetization dynamics excited by the spin torque given by eq . ( [ eq : torque_l ] ) or ( [ eq : torque_t ] ) both numerically and analytically . we neglect the transverse coupling when the role of the longitudinal coupling is studied , and vice versa , for simplicity , which corresponds to considering a system consisting of the f@xmath12 and f@xmath13 layers , or f@xmath12 and f@xmath20 layers . the magnetization dynamics in the f@xmath16 ( @xmath9 ) is described by the llg equation , @xmath102 where the gilbert damping constant is denoted as @xmath103 . in the following calculations , we use the values of the material parameters , @xmath104 rad/(oe s ) and @xmath105 , derived from the experiments @xcite . for the later discussion , it is useful to show the explicit forms of the llg equation in the longitudinal and transverse geometries . equation ( [ eq : llg ] ) for the longitudinal geometry is @xmath106 whereas that for the transverse geometry is @xmath107 in this section , we study the magnetization dynamics in the ferromagnets in the presence of the coupling torques by solving eq . ( [ eq : llg ] ) numerically . the self - oscillation of the magnetization provides an interesting example to understand the role of the coupling torque . note that the coupling torques are proportional to the parameters @xmath3 , @xmath4 , and @xmath5 , and their products to other parameters such as @xmath108 , the values of which are relatively small , as can be seen in eqs . ( [ eq : llg_l_exact ] ) and ( [ eq : llg_t_exact ] ) . nevertheless , the coupling torques lead to the phase synchronization , as shown below . the self - oscillation of the magnetization in single ferromagnets by the spin hall effect has been observed for in - plane magnetized ferromagnets @xcite , and is induced by the conventional spin torque given by eq . ( [ eq : torque_conventional ] ) . therefore , in the following , we assume that the magnetic field , @xmath109 consists of an in - plane anisotropy field @xmath110 along the @xmath19 direction and a demagnetization field @xmath111 in the @xmath112 direction , which we assume as @xmath113 oe and @xmath114 emu / c.c . @xcite in the following calculations . it is known for the case of the single ferromagnet @xcite that the in - plane self - oscillation can be excited when the electric current density @xmath2 is in the range of @xmath115 , where @xmath116 @xmath117 which in this study are @xmath118 and @xmath119 ma / cm@xmath55 . ) numerically for the transverse geometry . the dotted lines indicate the oscillation direction . ( b ) the time evolutions of @xmath121 and @xmath122 near the initial state . the time variations of the @xmath10 and @xmath19 components of the magnetizations in the steady state are shown in ( c ) and ( d ) for @xmath123 and ( e ) and ( f ) for @xmath124 . the solid red lines correspond to time evolution in f@xmath12 , and dotted blue lines to those in f@xmath20 for ( b ) through ( f ) . ] let us first investigate the magnetization dynamics in the transverse geometry because this geometry provides a simple example of the coupled motion . we start with solving the llg equation given by eq . ( [ eq : llg_t_exact ] ) for the f@xmath12 and f@xmath20 layers . figure [ fig : fig2](a ) shows an example of the trajectory of the magnetization dynamics obtained by solving eq . ( [ eq : llg_t_exact ] ) numerically , where @xmath125 ma / cm@xmath55 . as shown , the in - plane oscillation is observed in the steady state . the initial conditions set for @xmath126 and @xmath127 are different as @xmath128 and @xmath129 . therefore , the dynamics of @xmath126 and @xmath127 near the initial time are different , as shown in fig . [ fig : fig2](b ) , where the time evolutions of @xmath121 and @xmath130 in @xmath131 ns are shown . nevertheless , the dynamics of @xmath126 and @xmath127 synchronize gradually , and finally , synchronization of @xmath132 and @xmath133 is realized , as shown in figs . [ fig : fig2](c ) and [ fig : fig2](d ) . we emphasize here that the dynamics shown in these figures are obtained for @xmath123 . the mutual , as well as self , synchronization of the spin torque induced magnetization oscillation by using spin waves , electric current , microwave field , or dipole coupling has been an exciting topic from the viewpoints of both nonlinear science and practical applications @xcite . the key quantity of the synchronization is the phase difference @xmath134 between each magnetization to enhance the emission power from the spin torque oscillators and to investigate new practical applications such as neuromorphic architectures @xcite . the synchronization found here , i.e. , @xmath135 , is called the in - phase synchronization . we should emphasize here that , although the results shown in figs . [ fig : fig2](b ) and [ fig : fig2](c ) are shown for one certain initial condition , the in - phase synchronizations are confirmed for the present set of the parameters even when the initial conditions are changed . , for several values of @xmath103 on the field - like torque strength @xmath61 in the transverse geometry . @xmath136 and @xmath137 correspond to the in - phase and antiphase , respectively . the values of the current density , @xmath2 , is increased linearly , where @xmath125 ma / cm@xmath55 for @xmath105 . ] on the other hand , it was shown for the case of the current - injection locking of the spin torque oscillator that the phase difference between the magnetization oscillation and the alternative current depends on the strength of the field - like torque @xcite . the field - like torque in the present system can be either positive or negative , as mentioned above . these facts motivate us to study the magnetization dynamics for different values of @xmath61 . when @xmath138 , synchronized dynamics is observed in a similar manner , but in this case , the phase difference is antiphase , i.e. , @xmath139 , as shown in figs . [ fig : fig2](e ) and [ fig : fig2](f ) . figure [ fig : fig3 ] summarizes the dependences of the phase difference , @xmath134 , between the magnetizations on the field - like torque strength @xmath61 for the several values of the damping constant @xmath103 . the vertical axis in this figure represents the phase difference in the unit of the oscillation period ; i.e. , @xmath136 corresponds to the in - phase synchronization . on the other hand , @xmath140 means that the phase difference is half of a period , and thus , is antiphase . the algorithm evaluating @xmath134 in the numerical simulation is summarized in appendix [ sec : values of parameters in numerical simulations ] . the results indicate that the phase difference is antiphase for positive @xmath61 , whereas it becomes in - phase when @xmath61 becomes smaller than a certain value , except for the narrow intermediate region near @xmath141 , where the phase difference is in between in - phase and antiphase . and @xmath19 components of the magnetizations in the f@xmath12 ( solid , red ) and f@xmath13 ( dotted , blue ) in the longitudinal geometry are shown in ( a ) , ( c ) and ( b ) , ( d ) respectively , where @xmath123 . the initial conditions of @xmath142 are different between ( a ) , ( b ) and ( c ) , ( d ) . the magnetization dynamics for @xmath124 are shown in ( e ) and ( f ) ] next , we study the magnetization dynamics in the longitudinal geometry between f@xmath12 and f@xmath13 layers by solving eq . ( [ eq : llg_l_exact ] ) numerically . figure [ fig : fig4](a ) and [ fig : fig4](b ) show @xmath143 and @xmath144 in a steady state , where @xmath123 . the initial conditions in these figures are @xmath128 and @xmath145 . an antiphase synchronization is observed in this case . we notice , however , that the in - phase synchronization can also be realized when the initial conditions are changed . figure [ fig : fig4](c ) and [ fig : fig4](d ) show such an example , where @xmath128 and @xmath146 are assumed . these numerical results indicate that both the in - phase and antiphase synchronizations are stable in this case , and whether the phase difference finally becomes in - phase or antiphase depends on the initial conditions , material parameters , and current magnitude . we also notice that the phase difference is changed when the value of @xmath61 is changed . figures [ fig : fig4](e ) and [ fig : fig4](f ) show @xmath147 and @xmath148 for @xmath124 . in this case , the phase difference of the magnetizations is a quarter of a precession period . , for several values of @xmath103 on the field - like torque strength @xmath61 in the longitudinal geometry , where @xmath136 and @xmath137 correspond to the in - phase and antiphase , respectively . @xmath149 means that the phase difference is a quarter of a period . the values of the current density , @xmath2 , is increased linearly , where @xmath125 ma / cm@xmath55 for @xmath105 . ] the dependences of the phase difference on the field - like torque strength @xmath61 for several values of the gilbert damping constant @xmath103 are summarized in fig . [ fig : fig5 ] , where @xmath149 in this figure means that the phase difference is a quarter of a period . the phase difference is found to become a quarter of the period for positive @xmath61 , whereas it becomes in - phase or antiphase for negative @xmath61 , depending on the initial states of the magnetizations . let us summarize the results of the numerical simulations here . in the transverse geometry , the coupling torque induces the synchronized oscillation of the magnetizations and finally stabilizes the configuration in the in - phase or antiphase state , depending on the values of the field - like torque strength @xmath61 and the damping constant @xmath103 . the phase difference in the stable synchronized state in the longitudinal geometry also depends on the values of @xmath61 and @xmath103 , as well as the initial conditions . in this case , however , in addition to the in - phase or antiphase state , a phase difference with a quarter of a period is generated . ) for in - phase synchronization ( solid line ) and antiphase synchronization ( dotted line ) . ( b ) the longitudinal electric current densities given by eq . ( [ eq : electric_current_x_total ] ) for in - phase or antiphase synchronization ( solid line ) and phase difference of a quarter of a period ( dotted line ) . ] the phase difference can be measured from the spin hall magnetoresistance effect . according to eq . ( [ eq : electric_current_y ] ) , the total electric current density in the transverse direction is @xmath150 then , an oscillating current appears in the transverse direction for in - phase synchronization , whereas the transverse current becomes zero for antiphase synchronization , as shown in fig . [ fig : fig6](a ) . the fourier transformation of eq . ( [ eq : electric_current_y_total ] ) for in - phase synchronization has peaks at the frequencies of @xmath151 ( @xmath152 ) , where @xmath153 is the lowest frequency ; see appendix [ sec : analytical solution of magnetization on a constant energy curve ] . similarly , the total electric current density in the longitudinal direction is @xmath154 the oscillation frequency of this current is different for the synchronizations having the phase difference of in - phase or antiphase and that of a quarter of a period , as shown in fig . [ fig : fig6](b ) . the fourier transformation of eq.([eq : electric_current_x_total ] ) has the peaks at @xmath155 for in - phase or antiphase and @xmath156 when the phase difference is a quarter of a period . an interesting question regarding these numerical results is to clarify the reason why the phase difference finally becomes a certain value for a given set of the parameters , i.e. , which phase difference is an attractor of the limit cycle . it is , however , difficult to answer this question directly due to the following reason . we note that the present model includes several small - valued parameters , @xmath103 , @xmath61 , @xmath3 , @xmath4 , and @xmath5 , as shown in eqs . ( [ eq : llg_l_exact ] ) and ( [ eq : llg_t_exact ] ) , and is complicated . the torque related to the lowest order of @xmath61 in these equations is the conventional field - like torque given by @xmath157 \mathbf{m}_{\ell } \times \mathbf{e}_{y}$ ] . it should be noted that the phase difference is not determined solely by this term because this lowest order term of the field - like torque does not include the coupling between the magnetizations . similarly , the attractor is not determined solely by the lowest order term of the coupling torque , which is @xmath158 \chi m_{\ell^{\prime } y}^{2 } \mathbf{m}_{\ell } \times \left ( \mathbf{e}_{y } \times \mathbf{m}_{\ell } \right)$ ] in eq . ( [ eq : llg_l_exact ] ) and @xmath158 \chi m_{\ell^{\prime } x } m_{\ell^{\prime } y } \mathbf{m}_{\ell } \times \left ( \mathbf{e}_{x } \times \mathbf{m}_{\ell } \right)$ ] in eq . ( [ eq : llg_t_exact ] ) , because this torque does not include the field - like torque . their combinations or the higher order terms including both @xmath61 and the coupling torques related to @xmath3 , @xmath4 , and @xmath5 should be taken into account to answer the question , which is difficult due to the nonlinearity and complexity of the llg equation . nevertheless , it is possible to reveal the relation between the current and frequency in the synchronized oscillation state by assuming a certain value of the phase difference between the magnetizations . the current - frequency relation has been often measured in the experiments , and therefore , it will be useful to develop a theory clarifying the role of the coupling on the current - frequency relation . in the next section , we will investigate this subject . the purpose of this section is to develop an analytical theory of the synchronization revealing the relation among the current , frequency , and the phase difference of the magnetizations in the synchronized oscillation state . here , let us mention the basis of our theoretical analysis . it is difficult to solve the llg equation exactly because of its nonlinearity . instead , we employ the averaging technique of the llg equation on constant energy curves @xcite . this approach has been used to study the spin torque switching in thermally activated regions @xcite and spin torque oscillators @xcite , as well as the microwave assisted magnetization reversal @xcite , but has not been applied to the coupled system . this approach is valid when the magnetic energy is changed slowly compared with the oscillation period . we note that only the lowest order terms in the llg equation is necessary to derive the current - frequency relation , as far as several parameters such as @xmath61 are small . thus , we use the simplified llg equation maintaining only the dominant terms . the llg equation used in this section for the longitudinal geometry is @xmath159 whereas that for the transverse geometry is @xmath160 in the self - oscillation state , the damping torque during the precession is balanced with the spin torque , and the torque due to the magnetic field , corresponding to the first term on the right hand side of eq . ( [ eq : llg ] ) , becomes the dominant term determining the magnetization dynamics . the dynamic trajectory given by this field torque corresponds to constant energy curves of the energy density , @xmath161 , where its explicit form is @xmath162 the minimum and saddle points of eq . ( [ eq : energy ] ) are @xmath163 and @xmath164 , where the corresponding energy densities are @xmath165 and @xmath166 . the solution of @xmath82 precessing on a constant energy curve is described by the jacobi elliptic function @xcite as ( see also appendix [ sec : analytical solution of magnetization on a constant energy curve ] summarizing the derivations ) @xmath167 , \label{eq : mx_sol}\ ] ] @xmath168 , \label{eq : my_sol}\ ] ] @xmath169 , \label{eq : mz_sol}\ ] ] where @xmath170 is the initial phase . the period of @xmath171 and @xmath172 is @xmath173 , whereas that of @xmath174 is @xmath175 , where the precession frequency @xmath176 is given by @xmath177 where @xmath178 is the first kind of complete elliptic integral with the modulus @xmath179 note that eq . ( [ eq : frequency ] ) reproduces the ferromagnetic resonance ( fmr ) frequency , @xmath180 , in the limit of @xmath181 . identifying @xmath182 and @xmath170 corresponds to the determination of the initial condition . the averaged technique investigates the energy change during a precession on a constant energy curve , which is obtained from the llg equation as @xmath183 where @xmath184 is the energy change by the conventional spin torque due to the spin hall effect , whereas @xmath185 is the dissipation due to the damping torque . the integral range is over the precession period . the explicit forms of @xmath184 and @xmath185 are given by @xcite @xmath186 \\ & = \frac{\pi \hbar \theta_{\rm r } j_{0 } \left ( h_{\rm k } + 2e / m \right)}{ed_{\rm f } \sqrt{h_{\rm k } \left ( h_{\rm k } + 4 \pi m \right ) } } , \label{eq : melnikov_s } \end{split}\ ] ] @xmath187 \\ & = -4 \alpha m \sqrt{\frac{4\pi m-2e / m}{h_{\rm k } } } \left [ \frac{2e}{m } \mathsf{k}(k ) + h_{\rm k } \mathsf{e}(k ) \right ] , \label{eq : melnikov_alpha } \end{split}\ ] ] where @xmath188 is the second kind of complete elliptic integral . on the other hand , @xmath189 represents the work done by the coupling torque in the longitudinal or transverse geometry , corresponding to the last term in eq . ( [ eq : llg_l_approx ] ) or ( [ eq : llg_t_approx ] ) . the explicit forms of @xmath190 and @xmath191 are shown in the following sections . for both cases , the relation between the current and frequency is clarified as follows . in the self - oscillation state , since the spin torque balances the damping torque , the following condition should be satisfied , @xmath192 the current density @xmath2 satisfying eq . ( [ eq : oscillation_condition ] ) is the current necessary to excite the self - oscillation on the constant energy curve of @xmath182 , and is denoted as @xmath193 . the relation between the current and frequency in the self - oscillation state is given by this @xmath193 and @xmath194 given by eq . ( [ eq : frequency ] ) . it should be emphasized that the current density @xmath193 depends on the phase difference between the magnetizations through @xmath189 . we will therefore study the relation between the phase difference and the current - frequency relation in line with this deduction . in this section , we investigate the current - frequency relation in the transverse geometry . the work done by the coupling torque is defined as @xmath195 . \label{eq : melnikov_t_def}\ ] ] before advancing the calculation , let us briefly mention the definition of the phase difference in the present approach . if the oscillation trajectory is a circle , the phase difference is easily defined , i.e. , the antiphase corresponds to @xmath196 , whereas @xmath136 is the in - phase . in the present case , on the other hand , the oscillation trajectory is described by the elliptic function , as shown in eqs . ( [ eq : mx_sol ] ) , ( [ eq : my_sol ] ) , and ( [ eq : mz_sol ] ) . in this case , the phase difference is defined using the elliptic integral @xmath178 , where @xmath136 for the in - phase synchronization , and the antiphase synchronization corresponds to @xmath197 . it is useful to note that @xmath197 becomes @xmath198 in the limit of @xmath199 , corresponding to the case that the oscillation trajectory becomes a circle . similarly , @xmath200 means that the phase difference is a quarter of a period . equation ( [ eq : melnikov_t_def ] ) for an arbitrary phase difference is evaluated by numerically calculating the integral with the solutions of @xmath82 and @xmath92 shown in appendix [ sec : details of calculations ] . it is , however , useful to derive the analytical solutions of eq . ( [ eq : melnikov_t_def ] ) for specific values of the phase difference . equation ( [ eq : melnikov_t_def ] ) for both the in - phase ( @xmath136 ) and antiphase [ @xmath197 ] becomes @xmath202 where the double sign means the upper for the in - phase ( @xmath136 ) synchronization and the lower for the antiphase [ @xmath197 ] synchronization . equation ( [ eq : melnikov_t_sol ] ) is zero at @xmath203 and @xmath204 , and is negative ( positive ) for the energy density @xmath182 in the rage of @xmath205 when @xmath136 [ @xmath206 . this means that the coupling torque acts as a damping ( an antidamping ) torque when the phase difference is in - phase ( antiphase ) . we also note that @xmath207 when @xmath208 ; i.e. , the phase difference is a quarter of a period . the calculations necessary to obtain these specific values of @xmath191 are also summarized in appendix [ sec : details of calculations ] . we note that the sign change of @xmath191 with respect to the phase difference is related to the fact that the coupling torque in the transverse geometry , eq . ( [ eq : torque_t ] ) , has the angular dependence of @xmath209 . because of this angular dependence , the coupling torque acts as an anti - damping ( a damping ) torque when @xmath147 and @xmath210 have the opposite ( same ) signs , resulting in the increase ( decrease ) of the energy supplied to the ferromagnets by the coupling torque . in summary , the work done by the coupling torque , @xmath191 , is negative and minimized at the in - phase ( @xmath136 ) , zero for @xmath200 , and positive and maximized at the antiphase [ @xmath211 . , necessary to excite the self - oscillation in the transverse geometry as a function of the oscillation frequency @xmath194 and the phase difference @xmath134 of the magnetizations . the phase difference is in the unit of @xmath178 . ( b ) dependence of @xmath193 for the transverse geometry , on @xmath134 at @xmath212 ghz . the dotted line represents @xmath193 in the absence of the coupling . ( c ) the relation among @xmath193 , @xmath194 , and @xmath134 in the longitudinal geometry . ( d ) the current density @xmath193 for @xmath212 ghz in the longitudinal geometry . ] the current @xmath2 in the transverse geometry is defined as @xmath213 where @xmath214 and @xmath215 are defined as @xmath216,\ ] ] @xmath217 \\ & + \gamma \chi \oint dt m_{\ell^{\prime } x } m_{\ell^{\prime } y } \left [ \mathbf{e}_{x } \cdot \mathbf{h}_{\ell } - \left ( \mathbf{m}_{\ell } \cdot \mathbf{e}_{x } \right ) \left ( \mathbf{m}_{\ell } \cdot \mathbf{h}_{\ell } \right ) \right ] . \end{split}\ ] ] the explicit form of @xmath214 is obtained from eq . ( [ eq : melnikov_alpha ] ) as @xmath218.\ ] ] on the other hand , @xmath215 for the in - phase or antiphase is obtained from eqs . ( [ eq : melnikov_s ] ) and ( [ eq : melnikov_t_sol ] ) as @xmath219}{\sqrt{h_{\rm k } \left ( h_{\rm k } + 4 \pi m \right ) } } , \end{split}\ ] ] where the double sign means the upper for the in - phase synchronization and the lower for the antiphase synchronization . figure [ fig : fig7](a ) shows @xmath193 as functions of the oscillation frequency @xmath194 and the phase difference @xmath134 . the current - frequency relation in the transverse geometry can be obtained from this figure . to reveal the role of the phase difference more clearly , we show @xmath193 for a certain value of @xmath220 in fig . [ fig : fig7](b ) . note that @xmath193 is smaller than that in the absence of the coupling , which is shown by the dotted line , and minimized when @xmath197 , i.e. , the antiphase . this is because the work done by the coupling torque is positive and maximized at the antiphase . on the other hand , @xmath193 is maximized at the in - phase , and is larger than that in the absence of the coupling because the work done by the coupling torque is negative and minimized at the in - phase . we notice that the phase differences observed in the numerical simulation in sec . [ sec : numerical analysis of synchronization ] , i.e. , the in - phase and antiphase , correspond to @xmath134 satisfying @xmath221 or equivalently , @xmath222 in other words , the phase differences observed in the numerical simulations correspond to those giving the extrema of @xmath2 ( @xmath191 ) . let us investigate the theoretical relation between the current and frequency in the longitudinal geometry . in this case , the averaged llg equation is given by @xmath223 where @xmath184 and @xmath185 are given by eqs . ( [ eq : melnikov_s ] ) and ( [ eq : melnikov_alpha ] ) . on the other hand , @xmath190 representing the energy change due to the longitudinal coupling is defined as @xmath224 . \label{eq : melnikov_l_def}\ ] ] for both the in - phase and antiphase , eq . ( [ eq : melnikov_l_def ] ) becomes ( see appendix [ sec : details of calculations ] ) @xmath225}{h_{\rm k } \left ( h_{\rm k } + 4 \pi m \right)}. \label{eq : melnikov_l_sol } \end{split}\ ] ] on the other hand , when the phase difference is a quarter of a period [ @xmath200 ] , eq . ( [ eq : melnikov_l_def ] ) becomes @xmath227 it should be emphasized that @xmath190 is always positive for an arbitrary phase difference . this is because the coupling torque in eq . ( [ eq : torque_l ] ) always acts as an anti - damping torque . we can determine the current density @xmath193 satisfying eq . ( [ eq : oscillation_condition ] ) in the longitudinal geometry , as in the case of the transverse geometry , by replacing @xmath215 in eq . ( [ eq : je_def ] ) with @xmath228 \\ & + \gamma \chi \oint dt m_{\ell^{\prime } y}^{2 } \left [ \mathbf{e}_{y } \cdot \mathbf{h}_{\ell } - \left ( \mathbf{m}_{\ell } \cdot \mathbf{e}_{y } \right ) \left ( \mathbf{m}_{\ell } \cdot \mathbf{h}_{\ell } \right ) \right ] . \end{split}\ ] ] the explicit form of @xmath229 for both the in - phase and antiphase is obtained from eqs . ( [ eq : melnikov_s ] ) and ( [ eq : melnikov_l_sol ] ) as @xmath230 } { \left [ h_{\rm k } \left ( h_{\rm k } + 4\pi m \right ) \right]^{3/2 } } , \end{split}\ ] ] whereas that when the phase difference is a quarter of a period is obtained from eqs . ( [ eq : melnikov_s ] ) and ( [ eq : melnikov_l_sol_quarter ] ) as @xmath231 figure [ fig : fig7](c ) shows @xmath193 as functions of @xmath194 and @xmath134 . the current - frequency relation in the longitudinal geometry can be obtained from this figure . it is noted that @xmath193 is always smaller than that in the absence of the coupling because the coupling torque in the longitudinal geometry always points to the anti - damping direction , and therefore , the work done by the coupling torque is always positive . figure [ fig : fig7](b ) shows @xmath193 as a function of @xmath134 at a certain value of @xmath194 . as shown , @xmath193 has minima at both the in - phase ( @xmath136 ) and the antiphase [ @xmath211 , whereas it is maximized when the phase difference is a quarter of a period [ @xmath200 ] . we again notice that these phase differences found in the numerical simulations in sec . [ sec : numerical analysis of synchronization ] correspond to @xmath134 satisfying @xmath232 , or equivalently , @xmath233 equation ( [ eq : llg_averaged ] ) describes the slow change of the magnetic energy in the oscillation state . the magnetization dynamics is regarded as a motion of a point particle in an effective potential given by its right - hand side . equations ( [ eq : stable_phase_condition_t ] ) and ( [ eq : stable_phase_condition_l ] ) correspond to the stability conditions of the point particle in this effective potential . therefore , the phase difference found in the numerical simulation finally converges to one of these values satisfying eq . ( [ eq : stable_phase_condition_t ] ) or eq . ( [ eq : stable_phase_condition_l ] ) . whether the in - phase , antiphase , or the phase difference with a quarter of a period becomes the attractor depends on the higher order terms of the small parameters , as well as the initial states of the magnetizations , as mentioned at the end of sec . [ sec : numerical analysis of synchronization ] . this discussion is beyond the scope of this paper . at mentioned at the beginning of sec . [ sec : numerical analysis of synchronization ] , the in - plane self - oscillation for a single ferromagnet is stabilized when the current density is in the range of @xmath115 , where @xmath234 and @xmath235 are given by eqs . ( [ eq : j_c_single ] ) and ( [ eq : j_switch_single ] ) , respectively . at the end of this section , let us briefly discuss the effect of the coupling on these scaling currents . let us remind the readers that @xmath234 is the current density necessary to destabilize the magnetization in equilibrium , whereas @xmath235 is the current necessary to overcome the energy barrier , @xmath236 . these current densities are theoretically defined as @xcite @xmath237 @xmath238 it is confirmed that eqs . ( [ eq : j_c_single ] ) and ( [ eq : j_switch_single ] ) are reproduced by substituting eqs . ( [ eq : melnikov_s ] ) and ( [ eq : melnikov_alpha ] ) in the definition of @xmath193 in the absence of the coupling . on the other hand , in the presence of the transverse coupling , it is confirmed from eq . ( [ eq : melnikov_t_sol ] ) that a factor @xmath239 $ ] should be multiplied to the denominator of eq . ( [ eq : j_c_single ] ) when the phase difference between the magnetizations is in - phase , whereas this factor is replaced by @xmath240 $ ] when the phase difference is antiphase . the other scaling current , @xmath235 , is unchanged for these phase differences . in the longitudinal geometry , we see from eqs . ( [ eq : melnikov_l_sol ] ) and ( [ eq : melnikov_l_sol_quarter ] ) that the factor @xmath241 should be multiplied to the denominator of eq . ( [ eq : j_c_single ] ) when the phase difference is in - phase , antiphase , or a quarter of a period , whereas , for @xmath235 , the factor becomes @xmath242 $ ] for in - phase and antiphase , and @xmath243 when the phase difference is a quarter of a period . in conclusion , the coupled magnetization dynamics in the ferromagnets through the spin hall magnetoresistance effect was investigated . the coupling appears in both the longitudinal and transverse directions of the alignment of the ferromagnets . the in - phase or antiphase synchronization of the magnetization oscillation was found in the transverse geometry by solving the llg equation numerically . on the other hand , in addition to them , the synchronization having the phase difference of a quarter of a period is also found in the longitudinal geometry . it was shown that these phase differences depend on the values of the damping constant and the field - like torque strength . the analytical theory revealing the relation among the current , frequency , and phase difference was also developed . it was shown that the phase differences observed in the numerical simulations correspond to that giving the fixed points of the energy supplied by the coupling torque . the author is grateful to takehiko yorozu for his contributions to the analytical calculations , and to hitoshi kubota , sumito tsunegi , yoichi shiota , shingo tamaru , tazumi nagasawa , kiwamu kudo , and yoji kawamura for valuable discussions . the author is also thankful to satoshi iba , aurelie spiesser , atsushi sugihara , takahide kubota , hiroki maehara , and ai emura for their support and encouragement . this work was supported by jsps kakenhi grant - in - aid for young scientists ( b ) 16k17486 . the exact values of the parameters used in the simulations , evaluated from the parameters found in the experiment @xcite , are @xmath244 , @xmath245 , @xmath246 , @xmath247 , and @xmath248 for @xmath249 nm@xmath58 . in the main text , @xmath250 and @xmath138 correspond to @xmath245 and @xmath251 , respectively . strictly speaking , the change of the value of @xmath59 affects not only @xmath61 but also other quantities such as @xmath252 , @xmath3 , and @xmath4 . we , however , change the value of @xmath61 only in the numerical simulation , for simplicity , because the results do not change significantly unless @xmath253 . the llg equation with these parameters is solved by using the fourth - order runge - kutta method from @xmath254 to @xmath255 @xmath256s with the time step of @xmath257 fs ; i.e. , the number of the time mesh is @xmath258 . the present system has two stable states at @xmath259 . for convention , we assume that the magnetizations initially stay near one equilibrium , @xmath260 . for in - phase synchronizations , such as those shown in figs . [ fig : fig2](c ) and [ fig : fig2](d ) , the @xmath112 components are also synchronized with in - phase , i.e. , @xmath261 . on the other hand , for antiphase synchronization shown in , for example in figs . [ fig : fig2](e ) and [ fig : fig2](f ) , the @xmath112 components are also synchronized with antiphase , @xmath262 . the algorithm evaluating the phase differences shown in figs . [ fig : fig3 ] and [ fig : fig5 ] from the discrete numerical data is as follows . we gathered @xmath263 data of @xmath264 ( @xmath9 ) from @xmath265 to @xmath266 @xmath256s . then , the averaged periods @xmath267 of the oscillation of each magnetization were evaluated from the peaks of @xmath264 in this time range as @xmath268/\left ( n_{\ell}-1 \right)= \left ( t_{\ell , a}-t_{\ell,1 } \right)/\left ( n_{\ell}-1 \right)$ ] , where @xmath269 is the number of the peaks in @xmath264 , whereas @xmath270 is the time corresponding to the @xmath271-th peak . then , the phase difference is evaluated as @xmath272 , where @xmath273 $ ] and @xmath274 with @xmath97 for the transverse geometry , whereas that is @xmath275 for the longitudinal geometry . for the in - phase synchronization , this @xmath134 is zero because @xmath276 . when the phase difference is antiphase , @xmath140 because @xmath277 in this case . similarly , @xmath134 is @xmath278 when the phase difference is a quarter of a period . note that the critical current density to excite the self - oscillation , given by eq . ( [ eq : j_c_single ] ) , is proportional to the damping constant @xmath103 . therefore , the value of the current density should be increased to observe the self - oscillation when @xmath103 is varied , as in the case of figs . [ fig : fig3 ] and [ fig : fig5 ] . in these figures , @xmath2 is assumed as @xmath279 ma / cm@xmath55 for @xmath280 ( @xmath281 ) . the numerical simulation in fig . [ fig : fig2](e ) indicates that the antiphase synchronization is an attractor for @xmath138 . an exception is that if the initial conditions are set to be identical , the final state becomes in - phase synchronization due to the symmetry of the llg equation with respect to the change of @xmath282 . since eq . ( [ eq : stable_phase_condition_t ] ) is satisfied , the phase difference is fixed to in - phase even if it is unstable . similar situations occur in other cases for such specific initial conditions . in a ferromagnetic / nonmagnetic bilayer , the spin current density flowing in the @xmath283 direction ( @xmath284 ) with the spin polarization in the @xmath285 direction is related to the electrochemical potential @xmath286 and the spin accumulation @xmath287 via @xmath288 where @xmath289 is the electric field in the @xmath290 direction , and therefore , @xmath291 is the electric current density . we assume that this equation is extended to @xmath292 in the transverse geometry , where @xmath293 is the spin current density flowing near the f@xmath16/n interface in the @xmath112 direction with the spin polarization in the @xmath285 direction . the spin accumulation obeys the diffusion equation , and the boundary conditions of the diffusion equation are given by the spin current density at the boundaries . using eq . ( [ eq : spin_current_t ] ) , the solution of the spin accumulation is given by @xcite @xmath294 \right\ } , \label{eq : spin_accumulation_sol_t } \end{split}\ ] ] where we assume that the nonmagnet is in the region of @xmath295 . the spin current density , @xmath296 , at the f@xmath16/n interface is given by @xcite @xmath297 \\ & - \frac{\hbar}{2e } j_{\ell^{\prime } y } \left [ \vartheta_{\rm r } \mathbf{m}_{\ell } \times \left ( \mathbf{e}_{x } \times \mathbf{m}_{\ell } \right ) + \vartheta_{\rm i } \mathbf{e}_{x } \times \mathbf{m}_{\ell } \right ] , \label{eq : spin_current_fn_t } \end{split}\ ] ] where the vector notation in boldface represents the direction of the spin polarization , whereas the spatial direction of eq . ( [ eq : spin_current_fn_t ] ) is defined as the positive direction , i.e. , from the nonmagnet to the ferromagnet . on the other hand , the electric current density in the nonmagnet flowing in the @xmath283 direction is given by @xmath298 in the present case , the electric current density near the f@xmath16/n interface flowing in the @xmath19 direction becomes @xmath299 substituting eqs . ( [ eq : spin_accumulation_sol_t ] ) and ( [ eq : spin_current_fn_t ] ) into eq . ( [ eq : electric_current_t ] ) , and averaging along the @xmath112 direction as @xmath300 , we find that @xmath301 the conservation law of the electric current along the @xmath19 direction requires that @xmath302 . solving this equation for @xmath303 and @xmath304 , we obtain eq . ( [ eq : electric_current_y ] ) . the spin torque is defined from eq . ( [ eq : spin_current_fn_t ] ) as @xmath305 in this appendix , we shows the derivation of the analytical solution of the magnetization on a constant energy curve . for simplicity , we remove the subscript @xmath306 distinguishing the ferromagnets . the magnetization dynamics on a constant energy curve is described by the landau - lifshitz ( ll ) equation @xmath307 the magnetic field , eq . ( [ eq : field ] ) , is related to the magnetic energy density @xmath182 via @xmath308 , as mentioned in the main text . using the relation @xmath309 , eq . ( [ eq : energy ] ) is rewritten as @xmath310 this equation indicates that @xmath172 and @xmath171 can be expressed as @xmath311 and @xmath312 , respectively , where @xmath313 and @xmath314 are defined as @xmath315 and @xmath316 . then , @xmath317=\left ( v / m_{z } \right ) \left ( d m_{x}/dt \right)$ ] , which becomes , from eq . ( [ eq : ll ] ) , @xmath318 introducing new variable @xmath319 , this equation gives @xmath320 where the modulus @xmath321 is given by eq . ( [ eq : modulus ] ) . the modulus monotonically varies from @xmath322 to @xmath323 by changing the energy density @xmath182 from its minimum @xmath324 to saddle @xmath204 . we also notice that @xmath325 . equation ( [ eq : elliptic_integral_def ] ) indicates that @xmath326 is given by @xmath327,\ ] ] where @xmath328 is the jacobi elliptic function , and @xmath170 is the initial phase determined by the initial condition . using the relations @xmath329 and @xmath330 , we find that the solution of @xmath331 on the constant energy curve is given by eqs . ( [ eq : mx_sol ] ) , ( [ eq : my_sol ] ) , and ( [ eq : mz_sol ] ) . the peak frequencies of the fourier transformation of eq . ( [ eq : electric_current_y_total ] ) in the transverse geometry are discussed as follows . we note that eq . ( [ eq : electric_current_y_total ] ) for in - phase synchronization is proportional to @xmath332 and @xmath333 . substituting the following formulas @xcite , @xmath334,\ ] ] @xmath335,\ ] ] @xmath336 , \end{split}\ ] ] to eqs . ( [ eq : mx_sol ] ) , ( [ eq : my_sol ] ) , ( [ eq : mz_sol ] ) , where @xmath337 $ ] , it is found that the peak frequencies of eq . ( [ eq : electric_current_y_total ] ) appear at @xmath151 ( @xmath152 ) , where the lowest frequency @xmath153 is given by eq . ( [ eq : frequency ] ) . in the longitudinal geometry , eq . ( [ eq : electric_current_x_total ] ) is proportional to @xmath338 . when the phase difference of the magnetizations is in - phase or antiphase , it becomes @xmath339 . in this case , using the formula @xcite @xmath340,\ ] ] it is found that the fourier transformation of eq . ( [ eq : electric_current_x_total ] ) has the peaks at @xmath155 . on the other hand , when the phase difference is a quarter of a period , eq . ( [ eq : electric_current_x_total ] ) is proportional to @xmath341 = { \rm dn}^{2}(u , k)+\left [ \left ( 1-k^{2 } \right)/{\rm dn}^{2}(u , k ) \right]$ ] . we notice that @xmath342=g(u)$ ] , indicating that the fourier transformation of eq . ( [ eq : electric_current_x_total ] ) in this case has the peaks at @xmath156 . equations ( [ eq : melnikov_t_def ] ) and ( [ eq : melnikov_l_def ] ) can be calculated by substituting the solution of @xmath82 on a constant energy curve to the integrals . as emphasized in the main text , the phase difference @xmath134 between the magnetizations is an important quantity . according to eqs . ( [ eq : mx_sol ] ) , ( [ eq : my_sol ] ) , and ( [ eq : mz_sol ] ) , we set @xmath82 and @xmath92 as @xmath343 , \label{eq : mx_sol_1}\ ] ] @xmath344 , \label{eq : my_sol_1}\ ] ] @xmath345 , \label{eq : mz_sol_1}\ ] ] and @xmath346 , \label{eq : mx_sol_2}\ ] ] @xmath347 , \label{eq : my_sol_2}\ ] ] @xmath348 . \label{eq : mz_sol_2}\ ] ] the value of @xmath134 varies in the rage of @xmath349 . @xmath136 corresponds to the in - phase synchronization , whereas @xmath134 is @xmath350 for the antiphase synchronization . the analytical formulas of eqs . ( [ eq : melnikov_t_def ] ) and ( [ eq : melnikov_l_def ] ) for the in - phase and antiphase synchronizations can be obtained as follows . first , since the elliptic functions satisfy @xmath351=-{\rm sn}(u , k)$ ] , @xmath352=-{\rm cn}(u , k)$ ] , and @xmath353={\rm dn}(u , k)$ ] , @xmath191 has the same magnitude but different sign for @xmath136 and @xmath354 , whereas @xmath190 is the same for the in - phase and antiphase . therefore , it is sufficient to calculate @xmath191 and @xmath190 for the in - phase case . in this case , it is unnecessary to distinguish @xmath82 and @xmath92 . next , it should be noted that eq . ( [ eq : melnikov_t_def ] ) includes the following two integrals , @xmath355 @xmath356 by replacing the integral variable from @xmath357 with @xmath358 , and noting that @xmath359 , these integrals are calculated as @xmath360 + \left ( 4 - 3 k^{2 } \right)\sin^{-1}x}{8 } , \end{split}\ ] ] @xmath361 using these integrals , eq . ( [ eq : melnikov_t_sol ] ) is obtained . on the other hand , eq . ( [ eq : melnikov_l_def ] ) includes the following three integrals , @xmath362 @xmath363+\left ( 8 - 8k^{2}+3 k^{4 } \right)\sin^{-1}x}{8 } , \end{split}\ ] ] @xmath364 + \left ( 4 - k^{2 } \right ) \sin^{-1}x}{8}. \end{split}\ ] ] using these integrals , eq . ( [ eq : melnikov_l_sol ] ) is obtained . when the phase difference is a quarter of a period ( @xmath200 ) , the relations , @xmath365={\rm cn}(u , k)/{\rm dn}(u , k)$ ] , @xmath366=-\sqrt{1-k^{2}}{\rm cn}(u , k)/{\rm dn}(u , k)$ ] , @xmath367=\sqrt{1-k^{2}}/{\rm dn}(u , k)$ ] , are used to evaluate eqs . ( [ eq : melnikov_t_def ] ) and ( [ eq : melnikov_l_def ] ) . in this case , we notice that all the integrands in eq . ( [ eq : melnikov_t_def ] ) become odd functions of @xmath368 , and the integrals over @xmath369 are zero . therefore , @xmath207 for @xmath200 . on the other hand , the following integrals are necessary to obtain eq . ( [ eq : melnikov_l_sol_quarter ] ) , @xmath370}{\sqrt{1-k^{2 } } } , \end{split}\ ] ] @xmath371 @xmath372}{k^{2}}. \end{split}\ ] ]

in search for unusual magnetic ground states and spin dynamics , frustrated systems have been the main focus of both theoretical and experimental investigations in recent years.@xcite in frustrated magnetic systems,@xcite the energies of the various spin interactions compete and therefore they can not be simultaneously minimized . this competition can be induced by the geometry of the lattice on which the spins are arranged , in which case the phenomenon is called _ geometrical frustration _ and it often leads to quite unusual low temperature spin structures and dynamics . the nearest neighbor ( nn ) heisenberg antiferromagnet ( af ) on the corner - sharing kagom lattice is one of the most studied systems since it has all the ingredients such as low dimensionality , strong frustration , and low coordination number , required for a disordered `` spin liquid '' ground state.@xcite ( color online ) three possible nel states of the kagom lattice nn af : ( a ) * q*=0 with positive chirality , ( b ) * q*=0 with negative chirality , and ( c ) @xmath3 structure . the plus / minus signs indicate the chirality of spins on the elemental triangles . the shaded ( yellow ) areas indicate the magnetic unit cells . note that the nel sates shown above have continuous degeneracy as the spins on a line ( a - b ) and on a hexagon ( c ) can be rotated out of the plane as shown by ellipses without changing the classical ground state energy . , width=302 ] figure [ kagome ] shows three possible nel states of the kagom lattice nn antiferromagnet . in the classical limit , all ground states satisfy the 120@xmath4 structure " , in which the angle between each nn pair of spins is 120@xmath4 . [ kagome]a - b shows * q*=0 type ordering with respectively positive and negative chirality . for positive ( negative ) chirality the direction in which the spins rotate as one traverses clockwise the triangle of sites is clockwise ( counterclockwise ) . [ kagome]c shows the @xmath5 spin structure which has triangles with both positive and negative chirality . from the spin configurations shown in fig . [ kagome ] , it is clear the the classical ground state has a continuous degeneracy due to the weathervane " rotation of the spins@xcite and therefore no long range magnetic order is expected even at zero temperature . however , in real systems , there are small perturbations such as next nearest neighbor ( nnn ) interactions,@xcite anisotropies or defects . it has been also shown that thermal or quantum fluctuations could also lift some of the continuous degeneracy known as order by disorder , " yielding coplanar spin structures.@xcite despite the extensive theoretical studies that suggest many possible fascinating frustrated ground states for the kagom nn af , few experimental realizations exist for the system . the initial experimental studies were focused on the layered garnet , srcr@xmath6ga@xmath7o@xmath8.@xcite however the interpretation of the magnetic properties of this system were complicated by the presence of an additional triangular lattice interposed between kagom layers and by the inherent configurational randomness associated with the random alloying . in recent years , it has been shown that the jarosite family of minerals @xmath9,@xcite ( where a is monovalent ion such as k@xmath10 , and m is trivalent cation such as fe@xmath11 , cr@xmath11 , or v@xmath11 ) form much better realization of the two dimensional ( 2d ) kagom lattice . as shown in fig . [ alunite ] for m = fe jarosite , the magnetic ion @xmath12 is centered in a slightly distorted and tilted oxygen octahedra , and it forms a kagom lattice in the @xmath13-plane . the kagom planes are widely separated by nonmagnetic @xmath14 and @xmath15 ions . with an exception of the hydronium jarosite @xmath16 , all members of jarosite family are found to exhibit long range magnetic order ( lro ) at finite temperatures with varying ground state spin configurations and exchange interaction strengths depending on the magnetic ion m. the strength of exchange interaction is the greatest for the parent m = fe@xmath11 ( @xmath17 ) jarosite with curie - weiss temperature @xmath18 k and @xmath19 k. replacement of the fe@xmath11 centers by cr@xmath11 ( @xmath20 also affords an antiferromanetically ordered material but with a significant reduced @xmath21 k and @xmath22 k. interestingly , when the magnetic ion is replaced by v@xmath11 , the ground state is changed to ferromagnetic kagom layers which are coupled antiferromanetically.@xcite as we will discuss detail below , such a ferromagnetic ordering on the kagom lattice is allowed by the representation theory in contrast to a previous analysis.@xcite among the members of the jarosite family , @xmath23 ( fej ) is probably the most studied one . the magnetic ground state of fej was first investigated by townsend _ _ using neutron diffraction.@xcite however , later it became clear that the proposed spin configuration in ref . was not quite correct.@xcite inami _ et al_@xcite presented a detailed neutron scattering study and determined that the fej has 2d kagom planes with the * q*=0 spin - structure with positive chirality as shown in fig . [ kagome]a . they observed that the unit cell along the @xmath24-axis is doubled in the magnetic phase , suggesting the the kagom planes are antiferromagnetically coupled . field - dependent magnetization measurements@xcite have suggested that each kagom plane has a small ferromagnetic component due to the canting of the spins in the * q*=0 structure with positive chirality ( this is the so called `` umbrella '' configuration ) . however to the best of our knowledge , there is no experimental data which indicates the direction of this spin - canting whose determination would allow one to deduce the sign of some of the anisotropic terms in the hamiltonian , as we will discuss in detail later . ( color online ) conventional unit cell of fej . for clarity only fe , big spheres ( blue ) and the octahedral oxygens , small spheres ( red ) are shown . the tilting angle , @xmath25 of the feo@xmath26 octahedra is also shown.,width=188 ] the fact that fej exhibits lro at finite temperature indicates that there are interactions other than nn isotropic af superexchange , such as nnn interactions , anisotropies , etc . et al._@xcite have pointed out that in fej , the oxygen - octahedra are significantly tilted ( see fig . [ alunite ] ) , thereby inducing a strong single - ion crystal field ( cf ) anisotropy . they used this cf term in the hamiltonian to explain the observed spin configuration and to calculate the spin - wave spectrum . however in their spin - wave calculations , they neglected the canting of the spins out of the kagom plane . as we will discuss in detail below , this neglect leads to qualitatively wrong results for the spin - wave energy gaps at the high symmetry points in the brillouin zone . elhajal _ et al._@xcite have pointed out that cf terms in fej should be fairly small because they are second order in the spin - orbit coupling and fe@xmath11 has a spherical charge distribution ( _ i . @xmath27 ) . hence as an alternative source of anisotropy , they suggested that the dzyaloshinskii - moriya ( dm ) interaction of the form:@xcite @xmath28 could stabilize the experimentally observed spin structure . indeed , the @xmath29-component of the dm vector ( the definition of its components is given below ) forces the spins to lie in the @xmath30plane and therefore effectively acts like an easy - plane anisotropy . the sign of @xmath31 can distinguish between the @xmath32=0 states with negative and positive chirality . since the @xmath33 state has triangles with both positive and negative chirality , the @xmath31 term alone will not affect the energy of this state . furthermore if we introduce a @xmath34 component , we break the rotational symmetry around the @xmath24-axis and create a small anisotropy with respect to in - plane orientations . the effect of @xmath34 is also to ca nt the spins ( so that they have a small out - of - plane component ) into the observed umbrella " spin configuration . finally , since the dm interaction occurs at first - order in the spin - orbit coupling , it is expected to be larger than the cf terms of inami _ et al._@xcite in order to identify the origin of the magnetic interactions that are responsible for the lro in fej compounds , clearly we need more experimental data such as the observed spin - wave spectrum . fortunately , thanks to the very recent progress in the synthesis of high - quality single crystals of fej compounds,@xcite such spin - wave data has been recently become available.@xcite here we present a detailed theory for the spin - wave spectrum in fej compound for generic nn and nnn superexchange interactions including the cf and dm terms discussed above . as we shall see below , the dm term along with the nn and nnn isotropic interactions can explain quite well not only the observed spin configuration but also the observed spin - wave spectrum . briefly this paper is organized as follows . in the next section we first discuss the symmetry of the fej structure in the paramagnetic phase and then present a brief representation analysis of the allowed magnetic structures . in doing so , we discovered that in wills s analysis,@xcite some of the ferromagnetic wavefunctions were missing . indeed such a structure consisting of ferromagnetic easy - plane kagom planes has been observed for a nav - jarosite.@xcite in this section , we also argue that characterization by irreducible representations is more fundamental than by the chirality of the * q*=0 state . in sec . iii we discuss the generic magnetic hamiltonian that we use in our calculations . in this section , we first discuss the symmetry of the hamiltonian matrix and then its representation in a coordinate system in which the local @xmath29-axis coincides with the local direction of the spin moments . here we also treat the canting of the spin orientations . in sec . iv , we derive analytic results for the spin - wave energy gaps at the @xmath35 , @xmath36 and @xmath37-points of the brillouin zone . in sec . v , we present spin - wave spectra from numerical calculations and compare the results with the recent spin - wave data of matan _ et al_.@xcite here we consider several models with increasing complexity . we find that the dm terms combined with nn and nnn isotropic exchange interactions give an excellent fit to the data . our conclusions are summarized in sec . the details of the diagonalization of the spin hamiltonian were given in appendices . in appendix a , we describe the transformation to the boson representations of the spins . in appendix b , we discuss how to obtain the normal mode energies and eigenvectors . appendix c describes the actual calculations of the matrix elements and appendix d gives the results for the dynamical structure factor in terms of the normal mode energies and eigenvectors . we first discuss the symmetry of the paramagnetic phase . in fig . [ alunite ] we show the conventional unit cell which contains three formula units of fej . the primitive unit cell ( which contains one formula unit of fej ) is rhombohedral , with basis lattice vectors @xmath38 where @xmath24 denotes the height of the conventional unit cell of fig . [ alunite ] . the space group of fej is r@xmath39 m , which is # 166 in the international tables of crystallography.@xcite in order to understand the magnetic properties of fej we show in fig . [ kplane ] only the fe @xmath40 spins . their locations within the primitive unit cell , denoted @xmath41 , are @xmath42 where the components are given with respect to the orthogonal axes of fig . [ kplane ] . notice that each plane ( e. g. the filled circles ) forms a kagom plane lattice . as one moves from one plane to the adjacent plane at more positive @xmath29 , the kagom lattice is translated by @xmath43 , or , equivalently by either @xmath44 or @xmath45 . ( color online . ) kagom planes . the filled circles are @xmath46-@xmath47 planes at @xmath48 , the filled squares are @xmath46-@xmath47 planes at @xmath49 , and the filled triangles are @xmath46-@xmath47 planes at @xmath50 . the positive @xmath29-axis is out of the plane . representative sublattice numbers @xmath51 are given.,height=188 ] apart from translations the generators of the space group may be taken to be @xmath52 , spatial inversion about the center of a hexagon , @xmath53 , a two - fold rotation about an axis parallel to the @xmath46-axis and passing through the center of the hexagon , and @xmath54 , a three - fold rotation about an axis passing through the center of the hexagon perpendicular to the kagom plane . since the center of a hexagon in one plane lies just above a triangle in an adjacent plane , the three - fold axis can be taken at the center of a triangle . the above operations imply the existence of mirror planes perpendicular to the kagom plane and bisecting the sides of the hexagons . the reciprocal lattice basis vectors @xmath55 , which satisfy @xmath56 , where @xmath57 is the kronecker delta , are @xmath58 since interactions between adjacent kagom planes are very small , we may approximately describe the structure in terms of a two - dimensional hexagonal lattice with basis vectors , @xmath59 , given by @xmath60 the associated reciprocal lattice of this hexagonal lattice has basis vectors @xmath61 in contrast the three - dimensional reciprocal lattice vectors in the @xmath48 plane for fej are @xmath62 as shown in fig . [ fejfig ] , the vectors of eq . ( [ hexeq ] ) define the hexagonal brillouin zones and those of eq . ( [ fejeq ] ) define the boundaries of the fej brillouin zone in the @xmath48 plane . ( color online . ) reciprocal lattices for the fej system ( blue squares ) and for the two - dimensional hexagonal lattice ( red circles ) . the first brillouin zone for the two - dimensional hexagonal system is indicated by heavy solid lines in red and that ( in the @xmath48 plane ) for the fej system by heavy dashed lines in blue . the light solid lines in red mark the boundaries of neighboring hexagonal zones . the fej brillouin zone is three times as large as the hexagonal zone.,height=188 ] the first brillouin zone of the fej system in the hexagonal plane has an area three times that of the two - dimensional hexagonal system . if there were absolutely no interactions between kagom planes , the spin - wave dispersion relation in the region in the first brillouin zone of the fej system but outside that of thetwo - dimensional hexagonal system could be mapped into the dispersion relation inside the first brillouin zone of the two - dimensional system and in this paper we adopt this picture . for the two dimensional system , spin - wave energies at points just outside the hexagonal brillouin zone near point y are exactly the same as at the corresponding point just inside the zone near y. for the three dimensional fej system this is only approximately true . the spectra of these two points will differ due to interplanar interactions . similarly , for the two - dimensional system the spin - wave energies at point z are identical to those at @xmath35 , whereas for the fej system they will differ due to interplanar interactions . we will show the spin - wave spectrum observed by matan et al.@xcite along the line @xmath63 , where @xmath64 is equivalent to @xmath37 if interplanar interactions are neglected . a continuous phase transition may be described by a landau expansion in powers of @xmath65 , the fourier components of the spin order parameters which are defined as @xmath66 where @xmath67 is the spin at site @xmath51 in the unit cell at @xmath68 , @xmath69 denotes a thermal average , and @xmath32 is the wavevector of the ordering . at quadratic order the landau expansion for the free energy , @xmath70 , assumes the form @xmath71 where @xmath72 labels cartesian components and @xmath73 . the ordering that actually occurs corresponds to the eigenvector associated with that eigenvalue of the quadratic form of eq . ( [ feq2 ] ) which first becomes negative ( unstable ) as the temperature is lowered starting from the paramagnetic phase . in view of the invariance of @xmath70 with respect to the symmetry operations of the crystal , one can say that the eigenvectors give rise to an irreducible representation ( irrep ) of the space group of the crystal.@xcite this analysis is straightforward and has been given previously,@xcite although the results we present below seem to contradict that reference . because spin is a pseudovector , spatial inversion takes a spin into its spatially inverted location but does not reorient the spin . since all spins are located at centers of inversion symmetry , we only need to consider irreps which are invariant under spatial inversion , and we denote these by @xmath74 , where the label @xmath75 assumes the values @xmath76 and the superscript indicates invariance under spatial inversion . for the one dimensional irreps ( @xmath77 and @xmath78 ) the situation is quite simple : each eigenvector of @xmath70 is also an eigenvector of each operation of the space group . thus we show in fig . [ g1g2 ] the possible spin configurations which correspond to one dimensional irreps . the actual spin configuration associated with irrep @xmath78 is a linearcombination of the two configurations shown . spin configurations which transform according to one dimensional irreps . solid circles indicate spins pointing out of the page . top : a configuration of @xmath77 for which the eigenvalue of @xmath53 is @xmath79 and @xmath54 is @xmath79 . bottom : two configurations for @xmath78 for which the eigenvalue of @xmath53 is @xmath80 and @xmath54 is @xmath79.,height=226 ] for the two dimensional irrep @xmath81 the situation is more complicated . here one has @xmath82 wavefunctions @xmath83 , for @xmath84 which transform as the first row of the matrix representation of @xmath81 and their @xmath82 partners @xmath85 which transform as the second row of the matrix representation of @xmath81 , where for fej @xmath86.@xcite we show these wavefunctions in fig . [ g3 ] . to be specific , if @xmath87 is a symmetry operation and @xmath88 is a matrix representation of it according to @xmath81 , then we have spin configurations which transform according to two dimensional irrep @xmath81 . in the left ( right ) column we give the wavefunctions which transform according to the first ( second ) row of the matrices of the irrep . arrows represent spins of unit length . solid ( open ) circles indicate spins pointing perpendicularly into ( out of ) the kagom plane . for @xmath89 the magnitudes of the inward spins are both equal to half that of the outward spin which is set to be @xmath90 . for @xmath91 the nonzero magnitudes of the spins are both @xmath92.,height=321 ] @xmath93 these wavefunctions of @xmath81 correspond to the choice @xmath94 \ , \ \ \ \ { \cal m}(r ) = \left [ \begin{array } { c c } -1/2 & - \sqrt 3 /2 \\ \sqrt 3 /2 & -1/2 \\ \end{array } \right ] \ .\end{aligned}\ ] ] one can check that the wavefunctions shown in fig . [ g3 ] actually do satisfy eq . ( [ trans ] ) . of course , all the wavefunctions shown here are orthogonal to one another and span the original subspace of spin components . ( that appears not to be the case with the results of ref . . in addition , his states do not seem to allow the net moment for a single kagom layer to lie in the kagom plane . since this does happens for a na jarosite,@xcite the statement that the structure of the chromate jarosite must be the umbrella structure is not justified . ) we now discuss the implications of the above results . the above results hold for two separate cases : in the ferromagnetic case , all kagom planes have the same spin structure ( apart from a translation ) , whereas in the antiferromagnetic case successive kagom plane have their spins inverted . the ferromagnetic case corresponds to wavevector @xmath95 , whereas the antiferromagnetic case corresponds to @xmath96.@xcite note that irrep @xmath78 has two wavefunctions . this means that any such structure generically consists of a linear combination of a ferromagnetic state ( in a layer ) with spins perpendicular to the plane and some in - plane ordering as well . this is the so - called `` umbrella '' state . the implication is that if one has an easy axis ferromagnet with spins perpendicular to the kagom plane , one unavoidably will have some in - plane order . conversely , in the case of fej one has order of irrep @xmath81 in each plane which must be accompanied by a net out - of plane moment . note that the irreps @xmath77 and @xmath78 are ising - like in that within these representations one can not uniformly rotate the spins . thus the excitation spectrum must have a significant energy gap . in contrast irrep @xmath81 is @xmath46-@xmath47-like . one spin structure is some linear combination of the wavefunctions of the left column of fig . alternatively , the free energy is unchanged if one instead takes the _ same _ linear combination of the wavefunctions of the right column of fig . the actual wavefunction is therefore characterized by four constants , @xmath72 , @xmath97 , @xmath98 and @xmath99 : @xmath100 \nonumber \\ & & \ + \sin \theta \left [ \alpha \psi^{(3,2)}_a + \beta \psi^{(3,2)}_b + \gamma \psi^{(3,2)}_c \right ] \ , \end{aligned}\ ] ] which can not be fixed by symmetry . at quadratic level in the landau expansion , the free energy does not depend on @xmath99 and one would expect a goldstone ( gapless ) mode . however , due to the discrete symmetry of the kagom plane , the anisotropy energy , @xmath101 , which at lowest order is @xmath102 will lead to a gap . if @xmath103 , the system will favor @xmath104 ( or equivalent angles ) whereas if @xmath105 the system will favor angles equivalent to @xmath106 . quadratic energies , such as the dipolar energy , will not lead to a gap . accordingly , it is not surprising that we found that for @xmath107 the dipolar energy of the system was independent of @xmath99 . it is interesting to note that if one assumes isotropic nearest neighbor ( nn ) and next - nearest neighbor ( nnn ) interactions , then spin configurations @xmath108 and @xmath109 have the same energy . however , in general irrep @xmath110 will have lower energy because it can accommodate a distortion into the configuration @xmath111 . in principle , local single - ion anisotropy@xcite and/or dm interactions@xcite can select a ground state from among these functions . in the literature , there are frequent references to `` chirality . '' configurations have positive chirality , if when one traverses a triangle of spins clockwise , the spins rotate clockwise by @xmath112 , whereas for negative chirality @xmath113 . in our opinion , the characterization by irreps is more fundamental . one sees that the wavefunction of irrep @xmath77 and one of those of @xmath114 correspond to positive chirality and two of the wavefunctions for @xmath81 correspond to negative chirality . however , the sign of the chirality , of itself , does not indicate the presence or not of a gap in the excitation spectrum . nor does the sign of the chirality correlate in an obvious way with the nature of other wavefunctions which may be admixed . spin structure of fej as determined in refs . , , and .,height=188 ] the actual magnetic structure@xcite of fej is that shown in fig . [ spin ] associated with the wavevector @xmath115 . this means that successive kagom planes are translated ( as they would be in the paramagnetic phase ) , but then in successive planes the spin orientations are reversed . this causes a doubling of the unit cell because after three translations , the sites are back to their original positions , but the spin orientations are reversed . thus the magnetic unit cell contains twice as many layers as in the paramagnetic phase . although each layer individually has a small ferromagnetic moment perpendicular to the kagom plane , the sign of this moment alternates in sign from one kagom net to the next , so that the sample exhibits no overall net moment . however , as in la@xmath116cuo@xmath117,@xcite one has a spin - flop transition@xcite when a magnetic field perpendicular to the kagom plane rearranges the planes so that their ferromagnetic moments are parallel . ( color online ) spin structure of fej . the squares ( blue ) are sites in the kagom plane at @xmath49 , the triangles ( yellow ) are sites in the plane at @xmath118 and the circles ( red ) are sites in the plane at @xmath48 . the two symmetry related choices for spin orientations in the ordered phase are shown in the two panels.,height=132 ] here we discuss some aspects of broken symmetry . one could ask whether the direction of the net moment of a kagom plane is fixed by interactions within the plane or is accidentally selected . first , we should note that if we had a single isolated kagom plane , then the two orientations or the net moment perpendicular to the plane would have identical free energy . as we will see , the direction of the net moment is fixed by the sign of the dm interaction . if we had a single kagom plane , such an interaction would not be allowed . so , the uniqueness of the direction of the moment perpendicular to the plane , must depend on the details of the three - dimensional structure and , because @xmath119 and @xmath120 are equivalent in the paramagnetic crystal , on the in - plane magnetic ordering . to see how this is possible consider fig . the actual spin structure is shown in the two panels of fig . the selection of one of these states is arbitrary and represents an example of broken symmetry . let us see how the direction of the net moment perpendicular to the plane is fixed relative to the in - plane ordering . suppose we have the broken symmetry selection of the structure shown in the left panel . there one sees that the in - plane spin moments point away from sites in the @xmath118 plane and towards sites in the @xmath49 plane . this indicates that positive and negative @xmath29 are not equivalent _ due to the presence of long - range magnetic order in the three - dimensional crystal structure_. this means that when the spin order is as in the left panel , the free energy will select the direction of the perpendicular moment . when the spin order is as in the right panel , the perpendicular moment will be reversed relative to what it was in the left panel . as we will see , when the in - plane order is selected , the direction of the out - of plane moment is determined by the sign of the @xmath47-component of the dm vector . it is interesting to note that the direction of the perpendicular moment can not be related to chirality . the structures shown in fig . [ lro ] have positive chirality whether seen by a viewer from in front of the page or behind the page . so associating a direction with chirality can not distinguish between directions into or out of the page . we neglect interactions between adjacent kagom planes . accordingly , we now discuss the hamiltonian of a single kagom plane . we first parameterize the nn interaction between spins # 1 and # 2 in fig . [ bonds ] , which we write as @xmath121 where the greek superscripts label cartesian components . the effect of the mirror plane perpendicularly bisecting the 1 - 2 bond is to change the sign of the @xmath47- and @xmath29-components of spin and simultaneously interchange spin labels . thus taking the transpose of the matrix and changing the signs of the @xmath47- and @xmath29-components must leave the matrix invariant . so the interaction matrix must be of the form @xmath122 \ , \label{nn } \end{aligned}\ ] ] where the superscript indicates an nn interaction and we introduce symmetric anisotropic exchange tensor @xmath123 and antisymmetric dm interactions,@xcite characterized by a dm vector @xmath124 . note that the definition of eq . ( [ defeq ] ) implies that @xmath125 independent of what the local symmetry may be , where tilde indicates a transposed matrix . numbering of sites used to define nn and nnn interactions . the axes fixed in the crystal are shown.,height=188 ] we next characterize the nnn interaction between sites # 5 and # 2 in fig . [ bonds ] . this interaction has to be invariant under the two - fold rotation about the @xmath46-axis passing through the center of the bond . so the interaction matrix in cartesian coordinates must be of the form @xmath126 \ , \label{nnn } \end{aligned}\ ] ] where @xmath127 is the nnn dm interaction and @xmath128 is the symmetric nnn interaction . we also include a single - ion anisotropy energy @xmath129 for site @xmath130 of the form @xmath131_i^{\alpha \beta } [ s_i^\alpha s_i^\beta + s_i^\beta s_i^\alpha ] \ .\end{aligned}\ ] ] the mirror @xmath46-plane through site # 3 implies that the single - ion anisotropy matrix for this site is @xmath132 \ .\end{aligned}\ ] ] a direct , but inefficient , procedure would be to use the symmetry of the crystal to express all the other nn and nnn interactions in terms of the matrices of eqs . ( [ nn ] ) and ( [ nnn ] ) , respectively . instead , we proceed as follows . we express the nn and nnn interactions in terms of local axes which facilitate a spin - wave expansion . these local axes are defined so that the local positive @xmath29 axis coincides with the projection of the local spin moment onto the kagom plane . these axes are illustrated in fig . [ local ] . the rotation matrices to transform to these local axes are @xmath133 \ , \nonumber \\ { \cal r}(\tau_2 ) & = & \left [ \begin{array } { c c c } { 1 \over 2 } & 0 & { \sqrt 3 \over 2 } \\ { \sqrt 3 \over 2 } & 0 & - { 1\over 2 } \\ 0 & 1 & 0 \\ \end{array } \right ] \ , \nonumber \\ { \cal r}(\tau_3 ) & = & \left [ \begin{array } { c c c } -1 & 0 & 0 \\ 0 & 0 & 1 \\ 0 & 1 & 0 \\ \end{array } \right ] \ .\end{aligned}\ ] ] the transformed interaction matrices , denoted @xmath52 , are such that @xmath134 and @xmath135 where now the spin components are taken in the local axes of fig . [ local ] so that @xmath136 here @xmath137 \ , \label{effsw } \end{aligned}\ ] ] where @xmath138 where @xmath139 so that @xmath140 is the isotropic part of the @xmath75th neighbor interaction and we will only keep the anisotropic nn interactions , so that henceforth @xmath141 and @xmath142 . we define @xmath143 with a sign so that the interactions appear ferromagnetic in the local frame . the matrix elements @xmath144 lead to a canted spin structure exhibiting weak ferromagnetism , as we will see in the next subsection . the single - ion anisotropy in the local frame can be written as @xmath145 \ . \label{cparam } \end{aligned}\ ] ] a big advantage of expressing the interactions in terms of local axes is that all single - ion anisotropies and all nn interaction matrices are identical ( apart from transposing if site indices are interchanged ) . using the three - fold axis and inversion symmetry we find that @xmath146 where the sites are numbered as in fig . [ local ] . local axes of various sites . the local @xmath29-axis ( indicated by the full line with arrow ) is along the projection of the direction of the local magnetization onto the kagom plane . the local @xmath47-axis is perpendicularly out of the kagom plane . the local @xmath46-axis is in the kagom plane and is indicated by the dashed line with arrow.,height=188 ] just as for nearest neighbors , all the next nearest neighbor interaction matrices are the same , providing we take the indices in the correct order . up to now we have taken the local @xmath29-axes to lie in the kagom plane . however , the matrix elements @xmath144 for @xmath147 or @xmath148 gives rise to a field which induces a uniform @xmath47 component of spin . for the spin - wave calculation it is convenient to define `` canted local axes '' such that the canted @xmath29-axes lie along the direction of the canted spins . this direction is found by minimizing the energy . we write the ground state energy per site , @xmath149 , for @xmath150 and @xmath151 as @xmath152 \cos^2 \theta \nonumber \\ & & \ + [ c_{yy } - 2 e_y^{(1)}-2e_y^{(2 ) } ] \sin^2 \theta \nonumber \\ & & \ + 2 [ c_{yz } + 2 e_{yz}^{(1)}+2e_{yz}^{(2 ) } ] \sin \theta \cos \theta \nonumber \\ & \equiv & - a_{cc } \cos^2 \theta - a_{ss } \sin^2 \theta + 2 a_{cs } \sin \theta \cos \theta \nonumber \\ & = & - { 1 \over 2 } [ a_{cc}+a_{ss } ] - h_z \cos ( 2 \theta ) \nonumber \\ & & \ - h_y \sin ( 2 \theta ) \ , \end{aligned}\ ] ] where @xmath153/2 \ , \ \ \ \ h_y = - a_{cs } \ . \end{aligned}\ ] ] the energy is minimized by setting @xmath154 where @xmath155 . if we neglect the effect of anisotropic nnn interactions on the canting angle , we have @xmath156 = -c_{yz } - \sqrt 3 d_y + j_{yz } \nonumber \\ h_z & = & e_z^{(1 ) } - e_y^{(1 ) } - { 1 \over 2 } c_{zz } + { 1 \over 2 } c_{yy } \nonumber \\ & & \ + e_z^{(2 ) } - e_y^{(2 ) } \nonumber \\ & = & { 3 \over 2 } j_1 - { \sqrt 3 \over 2 } d_z + { c_{y - z } \over 2 } + { 3 \over 2 } j_2 + { 3 \delta_j - \eta_j \over 8 } \ , \label{hyzeq } \end{aligned}\ ] ] where @xmath157 . here and below the superscript `` 1 '' for nn s is implied in the anisotropic interactions . the ` initial ' and ` final ' local @xmath47 and @xmath29 axes . initially the @xmath29 axis is in the kagom plane . the sign of @xmath99 is the same that of @xmath158 , which is the negative of @xmath159 , i .e . the negative of @xmath34 . ] this means that we want to transform the initial local coordinates with moments in the kagom plane into the canted local coordinates via @xmath160 as illustrated in fig . [ axes ] , where @xmath161 and @xmath162 . then , if we include this transformation we finally arrive at the local anisotropy matrix as @xmath163 \ , \label{cdef } \end{aligned}\ ] ] and the transformed local interaction matrices are @xmath164 \ , \end{aligned}\ ] ] and similarly for @xmath165 , where @xmath166 \sin ( 2 \theta ) \ , \nonumber \\ \overline e_{yz}^{(n ) } & = & e_{yz}^{(n ) } \cos ( 2 \theta ) + { 1 \over 2 } [ e_z^{(n ) } - e_y^{(n ) } ] \sin(2 \theta ) \nonumber \\ \overline d_z^{(n ) } & = & c d_z^{(n ) } + s d_y^{(n ) } \nonumber \\ \overline d_y^{(n ) } & = & c d_y^{(n ) } - s d_z^{(n ) } \ . \label{tilde } \end{aligned}\ ] ] an equivalent condition for equilibrium is that @xmath167 when the @xmath46-@xmath29 elements are summed over all neighbors they give a vanishing effective field . indeed , group theory guarantees that this is the case for general interactions . in appendix [ boson ] we write the hamiltonian in terms of boson operators and in appendix [ modes ] we obtain the wavevector dependent spin - wave dynamical matrix ( swdm ) @xmath168 whose eigenvalues are the spin - wave energies at wavevector @xmath32 and which is written as @xmath169 \ .\end{aligned}\ ] ] the matrices @xmath170 and @xmath171 are given in eqs . ( [ adef ] ) and ( [ bdef ] ) , respectively . we write them as @xmath172 and @xmath173 where @xmath174 , @xmath175 is the unit matrix , and for @xmath147 or 2 @xmath176 is @xmath177 \gamma_{1,2}^{(n)}({\bf q } ) & + 2 i \overline d_z^{(n ) } ] \gamma_{1,3}^{(n)}({\bf q } ) \\ \hline [ -e_x^{(n ) } - \overline e_y^{(n ) } & 4\overline e_z^{(n ) } & [ -e_x^{(n ) } - \overline e_y^{(n ) } \\ + 2 i \overline d_z^{(n ) } ] \gamma_{1,2}^{(n)}({\bf q } ) & & -2 i \overline d_z^{(n ) } ] \gamma_{2,3}^{(n)}({\bf q } ) \\ \hline [ -e_x^{(n ) } - \overline e_y^{(n ) } & [ -e_x^{(n ) } - \overline e_y^{(n ) } & 4\overline e_z^{(n ) } \\ -2 i \overline d_z^{(n ) } ] \gamma_{1,3}^{(n)}({\bf q } ) & + 2 i \overline d_z^{(n ) } ] \gamma_{2,3}^{(n)}({\bf q } ) & \\ \hline \end{array } \right ] \ , \label{mataeq } \end{aligned}\ ] ] and @xmath178 is @xmath179 & [ - e_x^{(n ) } + \overline e_y^{(n ) } ] \\ & \times \gamma_{1,2}^{(n)}({\bf q } ) & \times \gamma_{1,3}^{(n)}({\bf q } ) \\ \hline [ - e_x^{(n ) } + \overline e_y^{(n ) } ] & 0 & [ - e_x^{(n ) } + \overline e_y^{(n ) } ] \\ \times \gamma_{1,2}^{(n)}({\bf q } ) & & \times \gamma_{2,3}^{(n)}({\bf q } ) \\ \hline [ - e_x^{(n ) } + \overline e_y^{(n ) } ] & [ - e_x^{(n ) } + \overline e_y^{(n ) } ] & 0 \\ \times \gamma_{1,3}^{(n)}({\bf q } ) & \times \gamma_{2,3}^{(n)}({\bf q } ) & \\ \hline \end{array } \right ] \ , \label{matbeq } \end{aligned}\ ] ] where @xmath180 is the normalized form factor for the @xmath75th shell of neighbors . here we we will only treat @xmath181 . for in - plane interactions we have @xmath182 \ , \nonumber \\ \gamma_{23}^{(1)}({\bf q } ) & = & \cos [ a ( q_x - \sqrt 3 q_y)/4 ] \ , \nonumber \\ \gamma_{12}^{(2)}({\bf q } ) & = & \cos ( a q_y \sqrt 3 /2 ) \ , \nonumber \\ \gamma_{13}^{(2)}({\bf q } ) & = & \cos [ a ( 3 q_x - \sqrt 3 q_y)/4 ] \ , \nonumber \\ \gamma_{23}^{(2)}({\bf q } ) & = & \cos [ a ( 3 q_x + \sqrt 3 q_y)/4 ] \ .\end{aligned}\ ] ] in eqs . ( [ mataeq ] ) and ( [ matbeq ] ) one one uses eq . ( [ tilde ] ) to relate the overlined coefficients to the bare coefficients , which are defined in eqs . ( [ cdef ] ) and ( [ param ] ) . to simplify the expressions we will now specialize to the case when the nnn interactions are isotropic and all interactions further than nnn are ignored . then we write the matrix @xmath170 in the form @xmath183 \ , \label{nmataeq } \end{aligned}\ ] ] where @xmath184 denotes @xmath185 . the matrix @xmath171 is of the form @xmath186 \ . \label{nmatbeq } \end{aligned}\ ] ] in appendix [ evaluate ] we evaluate of the constants in the matrices for the general hamiltonian introduced above . in the interest of simplicity we now limit consideration to the following generic hamiltonian : @xmath187 \nonumber \\ & & \ + \sum_{k < l \in nnn } j_2 s_k \cdot s_l \nonumber \\ & + & d \sum_{i } ( s_i^{y'})^2 - e \sum_{i } [ ( s_i^{z'})^2 -(s_i^{x'})^2 ] \label{model_h}\end{aligned}\ ] ] where @xmath188 ( @xmath189 ) indicates that the sum is over nn s ( nnn s ) , @xmath190 is the dzyaloskinskii - moriya vector for bond @xmath191 as shown in fig . [ bonds ] and the single - ion anisotropy terms are those used by nishiyama _ et al._@xcite in their treatment of the spin - wave spectrum in jarosites . here the prime - spin components refer to the local axis associated with the rotated oxygen octahedra shown in fig . 1 ( see ref . for details ) . the @xmath192 and @xmath193 are all expressible in terms of the parameters @xmath34 and @xmath31 as discussed in sec . the single - ion anisotropy constants @xmath194 and @xmath195 used in ref . are related to our generic single - ion matrix @xmath196 defined in the uncanted - local frame given in eq . ( [ cparam ] ) as @xmath197 where @xmath198 is the rotation angle of the octahedra around fe - ion ( see fig . [ alunite ] ) and is @xmath199.@xcite in what follows we will often have recourse to two simple models . in the first of these , which we call the `` dm '' model , we neglect the single ion anisotropy and the exchange anisotropy , so that the only nonzero parameters are @xmath1 , @xmath2 , @xmath34 , and @xmath31 . in the second model , which we call the `` cf '' model , all the anisotropy is incorporated by the single - ion crystal field , so that the only nonzero parameters are @xmath1 , @xmath2 , and @xmath200 . in either case , we assume @xmath1 to be the dominant interaction . accordingly we give here the expressions for the matrix elements of the swdm which pertain to these two cases : @xmath201 \ , \nonumber \\ a_1 & = & { 1 \over 2 } j_1 + { \sqrt 3 \over 2 } d_z + { 1 \over 6j_1 } \biggl [ 3d_y^2-c_{yz}^2 \biggr ] \ , \nonumber \\ b_0 & = & c_{xx}-c_{yy } - { 2c_{yz}^2 \over 3j_1 } \ , \nonumber \\ b_1 & = & - { 3 \over 2 } j_1 + { \sqrt 3 \over 2 } d_z + { c_{yz}^2 - 3 d_y^2 \over 6j_1 } \ , \nonumber \\ \alpha_1 & = & { 1 \over \sqrt 3 } c_{yz } + { 2 \sqrt 3 d_y d_z \over 3j_1 } - { d_y j_2 \over j_1 } \ , \nonumber \\ \alpha_2 & = & { d_y j_2 \over j_1 } \ . \label{eq47}\end{aligned}\ ] ] where @xmath202 . at zero wavevector the swdm , @xmath203 , whose eigenvalues give @xmath204 , is | c | c | c || + & dm model & cf model + @xmath205 & @xmath206 & @xmath207^{1/2 } + { \cal o}(c^{5/2}/j^{3/2})$ ] + @xmath208 & @xmath209 & @xmath210^{1/2 } + { \cal o}(c^{5/2}/j^{3/2})$ ] + @xmath211 & @xmath212 & @xmath213 + + & dm model & cf model + @xmath205 & @xmath214 & @xmath215^{1/2}$ ] + @xmath216 & @xmath217 & @xmath218^{1/2}$ ] + @xmath219 & @xmath220 & @xmath214 + @xmath221 \ , \nonumber \\ & & \label{gmeq}\end{aligned}\ ] ] where @xmath222 and we include the effects of @xmath2 by replacing @xmath1 in eq . ( [ eq47 ] ) everywhere by @xmath223 . take @xmath224 to have components @xmath225 and @xmath226 to have components @xmath227 . then @xmath228 from this we find the spin - wave energy @xmath205 to be @xmath229 now take @xmath230 to have components @xmath231 , @xmath232 to have components @xmath233 , @xmath234 to have components @xmath235 , and @xmath236 to have components @xmath237 . in this subspace @xmath238 \ , \end{aligned}\ ] ] where @xmath239 , @xmath240 , and @xmath222 . thereby we find that the other two spin - wave energies , @xmath241 , are given by @xmath242 thus for zero wavevector we find the frequencies @xmath243 for the dm and cf models to be those given in table [ omega ] ( under the heading `` with canting '' ) . when @xmath72 is nonzero , we split the degeneracy between the two heretofore degenerate frequencies . this splitting is linear in @xmath72 . in contrast , when @xmath244 , the splitting between this two - fold degenerate mode and the other mode will be found to be of order @xmath245 , where @xmath246 involves anisotropic exchange or dm interactions . note that for @xmath247 , we have stability for @xmath248 , where @xmath249\ , \label{stable } \end{aligned}\ ] ] which is similar to elhajal et al@xcite who give @xmath250 . for @xmath251 , we have stability ( in the large @xmath252 limit ) if both @xmath253 and @xmath254 are positive , so that the @xmath29-axis is really the easiest axis . the x point is at @xmath255 \ , \ \ \ \ q_y = q_z=0 \\end{aligned}\ ] ] and the form factors are @xmath256 and @xmath257 | c | c | c || + & dm model & cf model + @xmath205 & @xmath258 & @xmath259 + & @xmath260^{1/2 } + { \cal o } ( d^{5/2}/j^{3/2})$ ] & @xmath261^{1/2 } + { \cal o}(c^{5/2}/j^{3/2})$ ] + @xmath208 & @xmath262- { 5 \sqrt 6 \over 4}d_z + { \cal o}(d^2/j)$ ] & @xmath262+{\sqrt 2 \over 2 } ( 2c_{x - z}+c_{y - z } ) + { \cal o}(c^2/j)$ ] + @xmath211 & @xmath263 & @xmath264 + + & dm model & cf model + @xmath205 & @xmath265^{1/2}$ ] & @xmath266^{1/2}$ ] + @xmath216 & @xmath262- { 5 \sqrt 6 \over 4}d_z + { \cal o}(d^2/j)$ ] & @xmath262 + { \sqrt 2 \over 2 } ( 2c_{x - z}+c_{y - z } ) + { \cal o}(c^2/j)$ ] + @xmath219 & @xmath267 & @xmath214 + thus the swdm is of the form @xmath268 \ , \end{aligned}\ ] ] where , to the same accuracy as before @xmath269 @xmath270/2 + j_2/2 - i \alpha_2 \ , \nonumber \\ & = & -{1 \over 4 } j_1 + { 1 \over 2 } j_2 - { \sqrt 3 \over 4 } d_z - { d_y^2 \over 4j_1 } \nonumber \\ & & \ - c_{yz}^2 /(12j_1 ) - { 3i \over 2 } { d_y j_2 \over j_1 } \nonumber \\ & & \ + i \sqrt 3 [ c_{yz } + 2(d_yd_z / j_1)]/6 \nonumber \\ & \equiv & a_{12}^\prime + i a_{12}^{\prime \prime}\end{aligned}\ ] ] @xmath271 and @xmath272 where we have neglected symmetric anisotropic exchange and only kept terms that seem to be most relevant . now we diagonalize the submatrices @xmath273 , @xmath274 , and @xmath275 . the eigenvectors of these 3 @xmath276 3 matrices are @xmath277 \ , \label{eigvec } \end{aligned}\ ] ] where @xmath278 , so that @xmath279 , @xmath280 , and @xmath281 . one can show that the vector given in eq . ( [ eigvec ] ) is an eigenvector of @xmath273 with the associated eigenvalue @xmath282 and is simultaneously an eigenvector of @xmath275 with eigenvalue @xmath283 the three spin - wave energies are given by @xmath284^{1/2 } \nonumber \\ & & \ + \left ( { a(\lambda ) - a(\lambda^ * ) \over 2 } \right ) \ .\end{aligned}\ ] ] so that @xmath285^{1/2 } \ , \tilde \omega_{\pm } & = & [ ( a_{11 } - a_{12}^\prime ) ^2 - ( b_{11 } - b_{12})^2]^{1/2 } \pm \sqrt 3 a_{12}^{\prime \prime } \ .\end{aligned}\ ] ] we thus find the spin - wave energies to be those given in table [ omegax ] . the y point is at @xmath286 , @xmath287 , so that @xmath288 , @xmath289 , @xmath290 , and @xmath291 . therefore the swdm at the y point is of the form @xmath292 \ , \end{aligned}\ ] ] | c | c | c || + & dm model & cf model + @xmath293 & @xmath294 & @xmath295 + & @xmath296^{1/2 } + { \cal o } ( d^{5/2}/j^{3/2})$ ] & @xmath297^{1/2 } + { \cal o}(c^{5/2}/j^{3/2})$ ] + @xmath298 & @xmath299 & @xmath300 + @xmath301 & @xmath302 & @xmath303 + + & dm model & cf model + @xmath293 & @xmath304^{1/2}$ ] & @xmath305 + & & @xmath306^{1/2}$ ] + @xmath307 & @xmath299 & @xmath308 + @xmath301 & @xmath302 & @xmath303 + where @xmath309 @xmath310 - j_2/2 + i \alpha_2 \ , \nonumber \\ & = & ( j_1 - j_2)/2 + { \sqrt 3 \over 2 } d_z + { d_y^2 \over 2j_1 } - { c_{yz}^2 \over 6j_1 } \nonumber \\ & & \ - i \sqrt 3 [ c_{yz } + 2(d_yd_z / j_1 ) - 2d_yj_2/j_1 ] /3 \nonumber \\ & \equiv & a_{12}^\prime + i a_{12}^{\prime \prime}\end{aligned}\ ] ] @xmath311 and @xmath312 one root for @xmath313 ( coming from rows and columns # 3 and # 6 ) is @xmath314 and the remaining eigenvalues come from the projection of the swdm , @xmath203 , into the remaining subspace : @xmath315 \ . \label{remain } \end{aligned}\ ] ] as usual , in our analytic work , we work to second order in the anisotropies , in the absence of which @xmath316 is real . in that case we have the unperturbed mode energies @xmath317 the perturbed ( i. e. for @xmath318 ) secular equation is @xmath319 [ \tilde \omega^2 - ( \tilde \omega_+^{(0)})^2 ] + ( a_{12}^{\prime \prime})^4 \nonumber \\ & & \ -2(a_{12}^{\prime \prime})^2 [ \tilde \omega^2 + a_{11}^2 - ( a_{12}^\prime)^2 -b_{11}^2 + b_{12}^2 ] \ .\end{aligned}\ ] ] thereby we find the approximate solutions @xmath320 } \nonumber \\ & & \ \times \biggl [ ( \tilde \omega_\pm^{(0)})^2 + a_{11}^2 - ( a_{12}^\prime)^2 - b_{11}^2 + b_{12}^2 \biggr ] \ .\end{aligned}\ ] ] in table [ omegay ] we give analytical results correct only to leading order in the anisotropies , in which case @xmath321 . recently matan _ et al._@xcite have measured spin - wave spectrum in fej using inelastic neutron scattering and here we discuss the comparison of our calculations with their results . the experimental spectrum@xcite shown in fig . [ spinwave_exp ] clearly indicates three spin - bands . one of these bands is quite dispersionless and it is therefore identified to be the zero - energy " mode , which is lifted to about 8 mev due to anisotropic magnetic interactions which we wish to identify . in addition to this lifted zero - energy mode " , the spectrum shows two additional gaps at 1.7 mev ( an in - plane mode ) and around 7 mev ( an out - of - plane mode ) . the spin - wave spectrum extends up to 20 mev at the x - point . it is of interest to develop a model hamiltonian which reproduces the observed spectrum . in particular , we will see that the observed energy splittings at the various high symmetry points provide sensitive constraints on the interaction parameters . in order to identify magnetic interactions in fej , in this section , we will present numerical calculations of the spin - wave spectrum for a large number of models and compare the results with the experimental data shown in fig . [ spinwave_exp ] . for a quantitative comparison we define an error - factor @xmath54 ( i.e. the goodness - of - fit ) as follows : @xmath322 where @xmath323 is the number of experimental data points , and @xmath324 and @xmath325 are respectively the experimental and calculated spin wave energies at wave - vector @xmath326 . one should recognize that @xmath54 describes the percentage error averaged over the entire spectrum . it is important to reproduce the splittings at the high symmetry points even though this only weakly affects the value of @xmath54 . model . in these and succeeding figures the values of the parameters of the fit are given in mev.,width=264 ] we start by comparing the experimental spectrum to a simple model where we consider only the isotropic nearest neighbor interaction @xmath1 and a non - zero @xmath31 in dm interaction vector ( i.e. @xmath327 ) . the best fit from this simple two - parameter model is shown in fig . the dm interaction with only the @xmath31 term lifts the zero - energy mode to the dispersionless energy @xmath328 , as observed experimentally . however this term alone can not give the experimentally observed small dispersion of this mode . without @xmath34 , we do not get any canting of spins either . the effect of the @xmath31 term is almost identical to an easy plane anisotropy . the lifted zero - energy mode " is found to be degenerate with the other spin gap ( _ i.e. _ the out - of - plane spin - gap ) at the @xmath35 point . however , we do not get the in - plane gap as experimentally observed at 1.7 mev . ( analytic results are given in table [ omega ] . ) the agreement between calculations and experiment can be improved significantly by turning on @xmath34 in the dm - vector as shown in fig . [ j1dydz ] . the main effect of @xmath34 term is to create a small in - plane anisotropy and therefore it can be adjusted to give an in - plane gap of 1.7 mev , as experimentally observed . this model gives three gaps at @xmath35 point , for which the analytic results are given in table [ omega ] . from these analytic results , one sees that the in - plane gap is proportional to @xmath329 and the small splitting between the flat - mode and the out - of - plane gap is equal to @xmath330 , which is quite small . although the experiment does not give clear results for the splittings of the high - energy modes at the x and y points , we see from table [ omegax ] that the splitting , @xmath246 , at the x point is expected to be quite small ( it is second order in the perturbations ) . table [ omegay ] indicates that the splitting of the high frequency modes at the y point is expected to be larger than at the x point . the @xmath34 term also causes the spins to be canted with a canting angle of about 2.5@xmath4 . however despite a significant improvement with the addition of @xmath34 term , we still do not get any dispersion to the flat mode in contradiction to the experimental finding . we also studied the single ion term @xmath194 and @xmath195 as given in eq . ( [ de_relation ] ) with only @xmath1 and obtained similar results . with only dm or single ion anisotropy without further neighbor interactions , it was not possible to obtain the small dispersion of the flat mode as experimentally observed . model . , width=264 ] in order to get the experimentally observed dispersion along with the correct average energy of the nearly flat mode , one needs to consider next - nearest neighbor interactions . in fig . [ j1j2 ] , we show the best fit from a simple isotropic nn ( @xmath1 ) and nnn ( @xmath2 ) interactions , which is in good agreement with the previous work.@xcite we note that the nnn @xmath2 interaction gives the zero - energy mode a significant dispersion . however in order to correctly reproduce the flat - mode energy requires a @xmath2 which gives too strong a dispersion . since the hamiltonian with isotropic @xmath1 and @xmath2 is rotationally invariant , we do not get any gap at the @xmath35 point , in contrast to experimentally observed three gaps . and @xmath2.,width=264 ] from above discussion , it is clear that one needs both a nnn @xmath2 interaction and either a dm or a single ion anisotropy . in fig . [ j1j2dydz ] , we show the best fit from a model which has the isotropic nn @xmath1 , nnn @xmath2 interactions and the dm vector @xmath331 . the fit to the data is very good . we reproduce not only the three gaps but also the small dispersion of the flat - mode . the spin gaps at the @xmath35 , x , and y points obtained numerically are in good agreement with the analytic results given in tables [ omega ] , [ omegax ] , and [ omegay ] . the spin - canting angle is about 2.1@xmath4 , in reasonable agreement with the experimentally estimated value of 1.5@xmath4.@xcite \(a ) and @xmath2 and ( a ) dm - vector @xmath331 and ( b ) single ion anisotropy terms @xmath194 and @xmath195 . , title="fig:",width=264 ] + ( b ) and @xmath2 and ( a ) dm - vector @xmath331 and ( b ) single ion anisotropy terms @xmath194 and @xmath195 . , title="fig:",width=264 ] as an alternative to dm interactions , one can also try to explain the experimental data based on a model which considers only the single ion anisotropies @xmath194 and @xmath195 as given in eq . ( [ de_relation]).@xcite in fig . [ j1j2dydz]b , we show the best fit from such a model which has an isotropic nn @xmath1 and nnn @xmath2 and the @xmath332 single ion anisotropy instead of the dm interactions . the r - factor for this fit is slightly larger than the r - factor obtained from dm interaction . the main differences are the larger splittings of the optical modes at the @xmath35 and x points and the smaller canting angle for the cf model in comparison to the dm model . the canting angle from the cf model is only 0.8@xmath4 , which is smaller than the experimental value@xcite of 1.5@xmath4 . the biggest problem with the cf model is the large gaps at the @xmath35 and x - points as shown in fig . [ j1j2dydz]b . the experimental splitting of the modes at these points are much smaller . numerical results give a splitting of about 0.75 mev at the @xmath35 point and 0.4 mev at the x - point . the splitting at the y - point is quite small . these results are in good agreement with the analytical results given in tables [ omegax ] and [ omegay ] . in conclusion , the observed spin - wave spectrum can be explained by a simple model which has only nearest and next - nearest isotropic interactions . the dm interaction gives the best fit to explain the observed gaps and the spin canting . even though the single ion terms also fit the data well , the quality of the fit is not as good as the one from dm term at the @xmath35 and x points . for the fe ion , the single ion anisotropy is expected to be smaller than the dm term ( because the single - ion term is second order in the spin - orbit coupling while the dm term appears in first order ) . for these reasons , we think that dm term is a natural source of anisotropy in fej compound , which not only stabilizes the observed experimental spin structure but also gives the right spin - wave spectrum with the proper gaps and dispersion . here we give results when the transformation to the canted structure is omitted , i.e. @xmath99 is forced to be zero . one purpose of giving these results is to show the importance of making this transformation when calculating the gaps in the spectrum which depend on small perturbations . also , we could check our calculations by comparing these results to those of nishiyama et al@xcite who omitted this transformation . when one distorts a collinear spin structure by the application of a small transverse field , the effects on the spectrum are second order in the distortion angle . here , however , the situation is different because the spins before distortion are noncollinear . as a result of this noncollinearity the matrix element @xmath333 in eq . ( [ effsw ] ) ( which would be zero for a collinear system ) is of order @xmath252 . thus the matrix element @xmath334 in eq.([tilde ] ) has a contribution , @xmath335 , induced by canting of the same order as that , @xmath336 , which existed before canting . this fact indicates that canting should not be ignored in the calculation of the spin - wave spectrum . to obtain results when canting is ignored we replace quantities with overbars in eq . ( [ tilde ] ) by the corresponding quantities without overbars . here ( where @xmath252 denotes @xmath223 ) @xmath337 @xmath338 @xmath339 @xmath340 @xmath341 then , when the anisotropy in @xmath140 is neglected , we find the frequencies listed in table [ omega ] under the heading `` neglecting canting . '' the treatment including canting is now modified so that @xmath342/2 + j_2/2 - i \alpha_2 \nonumber \\ & = & -j_1/4 + j_2/2 - \sqrt 3 d_z/4 -i d_y/2 \ , \nonumber \\ b_{11 } & = & c_{xx}-c_{yy } \ , \nonumber \\ b_{12}&= & -b_1/2 - 3j_2/2 \nonumber \\ & = & 3j_1/4 -3j_2/2 - \sqrt 3 d_z/4 \ .\end{aligned}\ ] ] then we find the spin - wave frequencies as listed in table [ omegax ] . now we have @xmath343 so that the frequencies are those given in table [ omegay ] . unless the only anisotropy is due to @xmath34 , these results are the same as those given in table [ omegay ] with canting , so that , except for this case , the effect of canting at the y point is very small . the effect of canting can be deduced from the epxressions for the matrix elemets given in appendix c. one sees that the rotation to the canted spin structure induces terms of the form @xmath344 into the swdm matrix elements . also the quantity @xmath345 is transformed ( by the rotation to the canted system of coordinates ) from the value @xmath34 ( neglecting the anisotropy in @xmath140 ) to @xmath346 . suprprisingly perhaps , this rotation through an angle of order @xmath347 also introduces a contribution to @xmath348 of order @xmath349 . a qualitative understanding of what is going on here is obtained by noting that the coefficient @xmath350 is associated with an interaction proportional to @xmath351 , where the components are taken in the `` final '' _ i. e. _ canted coordinate system . this term directly affects the spin - wave energies . this term arises from rotating an interaction in the local uncanted coordinate system of the form @xmath352 , a term which for collinear systems is zero , but for this noncollinear system is of order @xmath1 , as given in eq . ( [ param ] ) . since these quantities control several of the splittings , it is clear that neglecting canting can have a major impact on these splittings , as shown in fig . [ no_canting ] and the tables . from fig . [ j1j2dydz](a ) , it is clear that the quality of the fit is significantly degraded for the dm term , in particular at the @xmath35 and @xmath36-points where big gaps are opened . \(a ) + ( b ) the effect of neglecting canting is opposite for the cf term . the quality of the fit is improved because the large splitting at @xmath35 ( see fig . [ j1j2dydz](b ) ) is now zero ( fig . [ no_canting](b ) ) . for instance , for zero wavevector the splitting @xmath353 ( given by @xmath354 ) is linear in @xmath34 when canting is neglected , whereas when the canting was taken into account the splitting was found to be proportional to @xmath355 . to calculate this splitting it is therefore not qualitatively correct to ignore the canting . in addition , when canting is ignored and single - ion anisotropy is not present , neglecting canting leads to @xmath356 , even when @xmath124 is nonzero . at the x point the effects of canting are similar . for the low energy mode the transformation to the canted structure introduces a term proportional to @xmath357 which is only important if all the other anisotropies are very small or vanishing . however , as table [ omegax ] indicates , the splitting between @xmath358 is proportional to @xmath34 when canting is neglected , but is proportional to @xmath359 when canting is taken into account . at the y point the effects of canting are less important . indeed @xmath360 and @xmath361 are insensitive to canting and only in @xmath362 does one see that canting introduces a term of order @xmath357 . in conclusion , it is important that one should not ignore the spin - canting when spin - wave spectrum is calculated . even though the cf single ion term gives much better fit when the canting is ignored ( see figs . [ j1j2dydz ] and [ no_canting ] ) , it is not an appropriate approximation , as we have shown . when the cf single ion term is properly treated , it actually gives a worse fit to the data and therefore it is more logical to conclude that the dominant source of the anisotropy in fej compound is the dm term . we have presented a detailed study of the magnetic structure and spin - wave spectrum in fej for the most generic nn and nnn exchange interactions including dzyaloshinskii - moriya and the single ion anisotropy terms of inami _ et al._@xcite our conclusions can be summarized as follows . * we first discuss the allowed magnetic structures from a representations analysis . the experimentally observed umbrella " structure of * q*=0 with positive chirality arises from a one dimensional representation which exhibits three - fold symmetry . our analysis and discussion of broken symmetry suggests that the characterization by irreducible representations is more fundamental than that by chirality , even though chirality has been widely invoked in the literature . * representation analysis reveals the existence of an @xmath46-@xmath47-like two dimensional representation which has sixth order in - plane anisotropy . in this representation one can have kagom layers with a net in - plane moment which can be stacked either ferromagnetically or antiferromagnetically . the analysis of ref . did not find wavefunctions of this type . however , the observed magnetic ground state of a nav - jarosite@xcite consists of such ferromagnetic kagom planes which are coupled antiferromagnetically . * we presented detailed spin - wave calculations for the most generic spin hamiltonian allowed by symmetry . we derived analytic expressions for the spin - wave energies at the @xmath35 , @xmath36 , and @xmath37 points in the brillouin zone . the agreement between our analytic and numerical results tends to support the correctness of both approaches . since , the spectrum is invariant under a change of sign of @xmath34 , the sign of this parameter remains undetermined by experiment . * we obtained the spin - wave spectrum numerically throughout the brillouin zone for a large number of models with increasing complexity and compared the results with the recent experimental data.@xcite we obtained a nn exchange constant of @xmath363 mev , which is in good agreement with previous measurements.@xcite we found that the observed small dispersion can be explained by a nnn isotropic interaction of @xmath364 . the observed gaps at the @xmath35 , @xmath36 , and @xmath37 points are best explained by considering only the dm interaction with @xmath365 mev and @xmath366 mev together with nn and nnn interactions . the d@xmath367 term is adjusted to lift the zero - energy mode to about 8 mev with a small dispersion due to @xmath2 interaction . a fit of similar quality can also be obtained using the crystal field ( cf ) model of ref . but the resulting anisotropy is about 10% of @xmath1 , too large a value for the s state fe@xmath11 ion which is supposed to be quite isotropic . the calculated canting angle is 2.07@xmath4 and 0.77@xmath4 for dm and cf models , respectively , compared to the experimentally deduced value@xcite of @xmath368 . * finally we discuss the effect of the canting on the spin - wave quantization axis which was neglected by inami _ et al._@xcite we found that neglecting canting is not a good approximation for calculating the spin - wave energy gaps at the high symmetry points in the brillouin zone . when the cf is treated properly , it gives larger gaps at the @xmath35 point than those from dm interaction . the experiment does not indicate significant splitting at the @xmath35 point and therefore it suggests that the dm term works better to explain the data . * acknowledgement : * we thank k. matan and y. s. lee for many fruitful discussions and for sharing their spin - wave data prior to publication . to construct the boson hamiltonian it is convenient to introduce spherical components ( @xmath369 , @xmath370 ) for interactions between site @xmath51 in unit cell @xmath130 located at @xmath371 and site @xmath372 in unit cell @xmath373 located at @xmath374 . if these sites are @xmath75th nearest neighbors , then we write @xmath375 \nonumber \\ \overline{\cal i}^{\pm \mp}_{\tau \tau'}(i , j ) & = & { 1 \over 4 } [ \overline { \cal i}_{xx}^{(n ) } \pm i \overline { \cal i}_{xy}^{(n ) } \mp \overline { \cal i}_{yx}^{(n ) } + \overline { \cal i}_{yy}^{(n ) } ] \end{aligned}\ ] ] and similarly for the single - ion anisotropy matrix @xmath376 . now we invoke the holstein - primakoff transformation in the form @xmath377 then apart from a constant the hamiltonian in terms of bose operators is @xmath378 \nonumber \\ & & \ + s { \sum_{i \tau , j \tau'}}^\prime \biggl [ - { 1 \over 2 } \overline { \cal i}_{\tau \tau'}^{zz}(i , j ) ( a_{i \tau}^\dagger a_{i \tau } + a_{j \tau'}^\dagger a_{j \tau ' } ) \nonumber \\ & & \ + \overline{\cal i}_{\tau \tau'}^{++ } ( i , j ) a_{i \tau } a_{j\tau ' } + \overline{\cal i}_{\tau \tau'}^{-- } ( i , j ) a_{i \tau}^\dagger a_{j\tau'}^\dagger \nonumber \\ & & \ + \overline{\cal i}_{\tau \tau'}^{+- } ( i , j ) a_{i \tau } a_{j\tau'}^\dagger + \overline{\cal i}_{\tau \tau'}^{-+ } ( i , j ) a_{i \tau}^\dagger a_{j\tau ' } \biggr ] \ , \end{aligned}\ ] ] where the prime on the summation excludes the term with @xmath379 . now write @xmath380 so that @xmath381 where @xmath382 is the ground state energy and @xmath383 where @xmath384 \delta_{\tau , \tau ' } + \overline { \cal i}^{+-}_{\tau ' \tau}({\bf q } ) + \overline { \cal i}^{-+}_{\tau \tau'}(-{\bf q } ) \biggr ) \nonumber \\ & = & \biggl ( [ c_{xx } + \overline c_{yy } - 2\overline c_{zz } - \sum_{\tau '' } \overline { \cal i}_{\tau,\tau''}^{zz } ( q=0 ) ] \delta_{\tau , \tau ' } \nonumber \\ & & \ + \overline { \cal i}^{+-}_{\tau ' \tau}({\bf q } ) + \overline { \cal i}^{-+}_{\tau \tau'}(-{\bf q } ) \biggr ) \nonumber \\ \label{adef } \end{aligned}\ ] ] @xmath385 \delta_{\tau,\tau ' } + 2 \overline { \cal i}_{\tau \tau'}^{--}(-{\bf q } ) \ , \label{bdef } \end{aligned}\ ] ] where @xmath386 where @xmath387 and the superscripts assume the values @xmath388 , @xmath389 , and @xmath29 . since each site is a center of inversion symmetry , the exponential factor in the sum over @xmath130 in eq . ( [ gammaeq ] ) can be replaced by a cosine : @xmath390 \label{coseq } \end{aligned}\ ] ] and @xmath391 is an even function of @xmath32 . if @xmath392 is a vector of the reciprocal lattice ( such that @xmath393 is a multiple of @xmath394 ) , then @xmath395 \nonumber \\ b_{\tau \tau'}({\bf q } + { \bf g } ) & = & b_{\tau \tau'}({\bf q } ) e^{i { \bf g } \cdot ( \tauv'-\tauv)}\nonumber \\ & = & b_{\tau \tau'}({\bf q } ) \cos [ { \bf g } \cdot ( \tauv'-\tau ) ] \ . \label{gvec } \end{aligned}\ ] ] here we used the fact that for the kagom lattice , the @xmath396 s are half a lattice vector , so that @xmath397= \pm 1 $ ] . we write the normal mode operators as @xmath398 \ , \end{aligned}\ ] ] and these are determined by @xmath399 = \omega^{(\mu ) } ( { \bf q } ) x^\dagger_\mu({\bf q } ) \ . \label{modeeq } \end{aligned}\ ] ] if there are @xmath82 spins per unit cell , then we expect that @xmath82 of the @xmath400 will be nonnegative and @xmath82 will be nonpositive . equation ( [ modeeq ] ) gives @xmath401 \nonumber \\ & = & \sum_{\alpha , \beta } \biggl [ c_\alpha^\mu ( { \bf q } ) a_{\beta , \alpha}({\bf q } ) a_\beta^\dagger ( { \bf q } ) + c_\alpha^\mu ( { \bf q } ) b_{\alpha , \beta}^*({\bf q } ) a_\beta(-{\bf q } ) \biggr ] \nonumber \\ & & \ \ - \sum_{\alpha , \beta } \biggl [ d_\alpha^\mu ( { \bf q } ) a_{\alpha , \beta}(-{\bf q } ) a_\beta(-{\bf q } ) \nonumber \\ & & \ + d_\alpha^\mu ( { \bf q } ) b_{\alpha , \beta}(-{\bf q } ) a_\beta^\dagger({\bf q } ) \biggr ] \ , \end{aligned}\ ] ] where again @xmath402 . this leads to the eigenvalue problem @xmath403 \nonumber \\ \tilde \omega^{(\mu ) } ( { \bf q } ) d_\alpha^\mu ( { \bf q } ) & = & \sum_\beta [ b_{\beta , \alpha}({\bf q})^ * c_\beta^\mu ( { \bf q } ) - a_{\beta , \alpha } ( -{\bf q } ) d_\beta^\mu ( { \bf q } ) ] \ , \end{aligned}\ ] ] which can be written as @xmath404 \left [ \begin{array } { c } { \bf c}^\mu ( { \bf q } ) \\ { \bf d}^\mu ( { \bf q } ) \\ \end{array } \right ] & = & \tilde \omega^{(\mu ) } ( { \bf q } ) \left [ \begin{array } { c } { \bf c}^\mu({\bf q } ) \\ { \bf d}^\mu ( { \bf q } ) \\ \end{array } \right ] \ , \end{aligned}\ ] ] where @xmath405 is a column vector with components @xmath406 , @xmath407 , @xmath408 , @xmath409 , where @xmath86 is the number of spins in the unit cell . we now note that because the hamiltonian is hermitian and taking account of inversion symmetry , we have @xmath410 similarly we write @xmath411 thus the eigenvalue problem is @xmath412 \left [ \begin{array } { c } { \bf c}^\mu ( { \bf q } ) \\ { \bf d}^\mu ( { \bf q } ) \\ \end{array } \right ] & = & \tilde \omega^{(\mu ) } ( { \bf q } ) \left [ \begin{array } { c } { \bf c}^\mu({\bf q } ) \\ { \bf d}^\mu ( { \bf q } ) \\ \end{array } \right ] \ . \label{meq } \end{aligned}\ ] ] we now show that the roots come in pairs with opposite signs . equation ( [ meq ] ) is @xmath413 take the complex conjugate of these equations , change the signs of both sides of each equations , and reorder the equations to get @xmath414^ * - { \bf b } [ { \bf c}^\mu]^ * & = & - \tilde \omega^{(\mu ) } [ { \bf d}^\mu]^ * \nonumber \\ { \bf b}^ * [ { \bf d}^\mu]^ * - { \bf a}^ * [ { \bf c}^\mu]^ * & = & - \tilde \omega^{(\mu ) } [ { \bf c}^\mu]^ * \ , \end{aligned}\ ] ] where we used the fact that the @xmath243 s are real . this shows that from an eigenvector with eigenvalue @xmath415 we can construct a related eigenvector with eigenvalue @xmath416 . one can show that the eigenvalues at wavevector @xmath32 are identical to those at wavevector @xmath417 , as expected in view of the inversion symmetry of the lattice . thus we may write @xmath418 \left [ \begin{array } { c c } { \bf c}({\bf q } ) & { \bf d}({\bf q})^ * \\ { \bf d}({\bf q } ) & { \bf c}({\bf q})^ * \\ \end{array } \right ] \nonumber \\ & = & \left [ \begin{array } { c c } { \bf c}({\bf q } ) & { \bf d}({\bf q})^ * \\ { \bf d}({\bf q } ) & { \bf c}({\bf q})^ * \\ \end{array } \right ] \left [ \begin{array } { c c } \tilde \omegav ( { \bf q } ) & 0 \\ 0 & - \tilde \omegav ( { \bf q } ) \\ \end{array } \right ] \ , \label{54eq } \end{aligned}\ ] ] where @xmath419 and @xmath420 are matrices whose columns are the column vectors @xmath421 and @xmath422 , respectively , and @xmath423 is a diagonal matrix with entries @xmath424 . recall the dependence of the matrices on the reciprocal lattice vector as recorded in eq . ( [ gvec ] ) . then the eigenvalue problem at wavevector @xmath425 may be written in the form @xmath426 e^{i{\bf g } \cdot ( \tauv_\beta - \tauv_\alpha ) } \nonumber \\ & & \tilde \omega^{(\mu ) } ( { \bf q}+{\bf g } ) d_\alpha^\mu ( { \bf q}+{\bf g } ) = \sum_\beta [ b_{\beta , \alpha}({\bf q})^ * c_\beta^\mu ( { \bf q}+{\bf g } ) \nonumber \\ & & \ - a_{\beta , \alpha } ( -{\bf q } ) d_\beta^\mu ( { \bf q}+{\bf g } ) ] e^{i{\bf g } \cdot ( \tauv_\beta - \tauv_\alpha ) } \ .\end{aligned}\ ] ] set @xmath427 and @xmath428 then we recover the previous equations . so we conclude that under addition of @xmath392 , the eigenvalues remain invariant , but the wavefunctions change according to eq . ( [ gtrans ] ) . taking the hermitian conjugate of eigenvalue equations , we get @xmath429 so that @xmath430 \left [ \begin{array } { c c } { \bf a}({\bf q } ) & - { \bf b}({\bf q } ) \\ { \bf b}({\bf q})^ * & -{\bf a } ( { \bf q})^ * \\ \end{array } \right ] \nonumber \\ & & \ = \tilde \omega^{(\mu ) } ( { \bf q } ) \left [ \begin{array } { c } { \bf c}^\mu({\bf q})^\dagger \ , - { \bf d}^\mu ( { \bf q})^\dagger \end{array } \right ] \ .\end{aligned}\ ] ] thus from any right solution we have constructed the associated left solution . now consider @xmath431 \left [ \begin{array } { c c } { \bf a}({\bf q } ) & - { \bf b}({\bf q } ) \\ { \bf b}({\bf q})^ * & -{\bf a } ( { \bf q})^ * \\ \end{array } \right ] \nonumber \\ & & \ \times \left [ \begin{array } { c } c^\nu ( { \bf q } ) \\ d^\nu ( { \bf q } ) \\ \end{array } \right ] \ .\end{aligned}\ ] ] using the fact that the left and right vectors are eigenvectors , we have @xmath432 \left [ \begin{array } { c } c^\nu ( { \bf q } ) \\ d^\nu ( { \bf q } ) \\ \end{array } \right ] \nonumber \\ & = & \tilde \omega^{(\nu ) } ( { \bf q } ) \left [ \begin{array } { c } { \bf c}^\mu({\bf q})^\dagger \ , - { \bf d}^\mu({\bf q})^\dagger \end{array } \right ] \left [ \begin{array } { c } c^\nu ( { \bf q } ) \\ d^\nu ( { \bf q } ) \\ \end{array } \right ] \ .\end{aligned}\ ] ] if @xmath433 this shows that @xmath434 is real . also @xmath435 \left [ \begin{array } { c } c^\nu ( { \bf q } ) \\ d^\nu ( { \bf q } ) \\ \end{array } \right ] = 0 , \ \ \mu \not= \nu \ .\end{aligned}\ ] ] ( if we have degenerate eigenvalues , we can choose the wavefunctions to preserve this orthogonality . ) a more general orthogonality is @xmath436 \left [ \begin{array } { c c } { \bf c}({\bf q } ) & { \bf d}({\bf q})^ * \\ { \bf d}({\bf q } ) & { \bf c}({\bf q})^ * \\ \end{array } \right ] & = & { \cal i } \ , \end{aligned}\ ] ] where @xmath437 is the @xmath438 unit matrix . we have the transformation to normal modes as @xmath398 \ , \label{nego } \end{aligned}\ ] ] and also @xmath439 \ , \end{aligned}\ ] ] where @xmath440 is restricted so that @xmath441 . we require that @xmath442 obey bose commutation relations : @xmath443 = \delta_{\mu,\nu } \ .\end{aligned}\ ] ] this gives @xmath444 the inverse transformation is @xmath445 \ , \nonumber \\ a_\alpha ( { \bf q } ) & = & \sum_\mu [ c_\alpha^\mu({\bf q } ) x_\mu({\bf q } ) - d_\alpha^\mu ( { \bf q})^ * x_\mu ( -{\bf q})^\dagger ] \ .\end{aligned}\ ] ] we record here the dependence on reciprocal lattice vector : @xmath446 e^{i{\bf g } \cdot \tauv_\alpha } \ , \nonumber \\ a_\alpha ( { \bf q}+{\bf g } ) & = & \sum_\mu [ c_\alpha^\mu({\bf q } ) x_\mu({\bf q } ) - d_\alpha^\mu ( { \bf q})^ * x_\mu ( -{\bf q})^\dagger ] e^{-i{\bf g } \cdot \tauv_\alpha } \ .\end{aligned}\ ] ] to use this we would assume that @xmath32 is in the first brillouin zone and @xmath392 is the vector needed to bring the actual wavevector back into the first brillouin zone . we now develop explicit expressions for the constants appearing in these matrices . we use eq . ( [ tilde ] ) to relate the overlined coefficients to the bare coefficients , which are defined in eqs . ( [ cdef ] ) and ( [ param ] ) . we will work to second order in the perturbations from isotropic exchange . to this order @xmath447 and @xmath448 where @xmath158 and @xmath449 are given in eq . ( [ hyzeq ] ) , so that @xmath450 } \ , \label{zeq } \end{aligned}\ ] ] where @xmath451 then @xmath452 equation ( [ tilde ] ) gives @xmath453 then , keeping only relevant anisotropy corrections , we find that the constants in eq . ( [ nmataeq ] ) and ( [ nmatbeq ] ) are @xmath454 \cos ( 2 \theta ) \nonumber \\ & & \ - [ 4e_{yz}^{(1)}+3c_{yz } ] \sin ( 2 \theta ) \nonumber \\ & = & \delta + { 3 \over 2 } j_{xx } - { 1 \over 2 } j_{yy } - 2 j_{zz } - \sqrt 3 d_z \nonumber \\ & & \ + \biggl [ { 3 \over 2 } j_{xx } - { 1 \over 2 } j_{yy } + 2 j_{zz } - \sqrt 3 d_z \biggr]\nonumber \\ & & \ \times \biggl [ 1 - { 2[j_{yz } - c_{yz } - \sqrt 3 d_y]^2 \over 9j_1 ^ 2 } \biggr ] \nonumber \\ & & - { 1 \over 3j_1 } \biggl [ 4\sqrt 3 d_y - 4j_{yz } + 6 c_{yz } \biggr ] \biggl [ j_{yz } - c_{yz } - \sqrt 3 d_y \biggr ] \nonumber \\ & = & 2j + \delta + { 5 \over 6 } \delta_j + { 3 \over 2 } \eta_j - 2 \sqrt 3 d_z \nonumber \\ & & \ + { 2 \over 3j_1 } [ j_{yz } - c_{yz } - \sqrt 3 d_y][j_{yz } - \sqrt 3 d_y - 2 c_{yz } ] \ , \nonumber \\ % % a_1 & = & - e_x^{(1 ) } - \overline e_y^{(1)}\nonumber \\ & = & - e_x^{(1 ) } - [ e_z^{(1 ) } + e_y^{(1)}]/2\nonumber \\ & & \ - { 1 \over 2 } [ e_y^{(1 ) } - e_z^{(1 ) } ] \cos ( 2 \theta ) - e_{yz}^{(1 ) } \sin ( 2 \theta ) \nonumber \\ & = & { 1 \over 4 } j_{xx } - { 3 \over 4 } j_{yy } + { \sqrt 3 \over 2 } d_z - { 3 \over 8 } j_{xx}\nonumber \\ & & \ + { 1 \over 8 } j_{yy } + { 1 \over 2 } j_{zz } + { \sqrt 3 \over 4 } d_z \nonumber \\ & & \ + \biggl [ { 1 \over 2 } j_{zz } + { 3 \over 8 } j_{xx } - { 1 \over 8 } j_{yy } - { \sqrt 3 \over 4 } d_z \biggr ] \nonumber \\ & & \ \times \biggl [ 1 - { 2[j_{yz } - c_{yz } - \sqrt 3 d_y]^2 \over 9j_1 ^ 2 } \biggr ] \nonumber \\ & & \ - { 1 \over 3j_1 } [ \sqrt 3 d_y - j_{yz } ] [ j_{yz } - c_{yz } - \sqrt 3 d_y ] \nonumber \\ & = & { 1 \over 2 } j_1 - { 3 \over 8 } \eta_j + { 11 \over 24 } \delta_j + { \sqrt 3 \over 2 } d_z \nonumber \\ & & \ + { 1 \over 6j_1 } \biggl [ j_{yz } - c_{yz } - \sqrt 3 d_y \biggr ] \biggl [ - \sqrt 3 d_y + j_{yz } + c_{yz } \biggr ] \ , \nonumber \\ % % b_0 & = & c_{xx}- \overline c_{yy } \ , \nonumber \\ & = & c_{xx } - { 1 \over 2 } \biggl [ c_{yy } + c_{zz } \biggr ] + { 1 \over 2 } \biggl [ c_{zz}-c_{yy } \biggr ] \cos ( 2 \theta ) \nonumber \\ & & \ + c_{yz } \sin ( 2\theta ) \nonumber \\ & = & c_{xx}-c_{yy } + { 2c_{yz } \over 3j_1 } [ j_{yz } - c_{yz } - \sqrt 3 d_y ] \ , % % \nonumber \\ b_1 & = & - e_x^{(1 ) } + \overline e_y^{(1 ) } \nonumber \\ & = & -e_x^{(1 ) } + { 1 \over 2 } \biggl [ e_y^{(1 ) } + e_z^{(1 ) } \biggr ] + e_{yz}^{(1 ) } \sin ( 2 \theta ) \nonumber \\ & & \ + { 1 \over 2 } \biggl [ e_y^{(1)}-e_z^{(1 ) } \biggr ] \cos ( 2 \theta ) \nonumber \\ & = & { 5 \over 8 } j_{xx } - { 7 \over 8 } j_{yy } - { 1 \over 2 } j_{zz } + { \sqrt 3 \over 4 } d_z \nonumber \\ & & \ - { 1 \over 2 } \biggl [ j_{zz } + { 3 \over 4 } j_{xx } - { 1 \over 4 } j_{yy } - { \sqrt 3 \over 2 } d_z \biggr ] \nonumber \\ & & \ \times \biggl [ 1 - { 2[j_{yz } - c_{yz } - \sqrt 3 d_y]^2 \over 9j_1 ^ 2 } \biggr ] \nonumber \\ & & \ + { 1 \over 3j_1 } [ \sqrt 3 d_y - j_{yz } ] [ j_{yz } - c_{yz } - \sqrt 3 d_y ] \nonumber \\ & = & - { 3 \over 2 } j_1 + { 5 \over 8 } \eta_j + { 1 \over 8 } \delta_j + { \sqrt 3 \over 2 } d_z\nonumber \\ & & \ + { [ j_{yz}-c_{yz } - \sqrt 3 d_y ] \over 6j } [ \sqrt 3 d_y - j_{yz } - c_{yz } ] \ , \nonumber \\ % % \alpha^{(1 ) } & = & 2 \overline d_z^{(1 ) } = 2d_z^{(1 ) } \cos \theta + 2d_y^{(1 ) } \sin \theta \nonumber \\ & = & 2d_z^{(1 ) } + d_y^{(1 ) } \sin ( 2 \theta ) \nonumber \\ & = & -(d_y + \sqrt 3 j_{yz } ) \cos \theta \nonumber \\ & + & \left(- \sqrt 3 j_1 - d_z + { \sqrt 3 \delta_j \over 12 } - { \sqrt 3 \eta_j \over 4 } \right ) \sin \theta \ .\end{aligned}\ ] ] but using eq . ( [ zeq ] ) this gives @xmath455 \left [ - \sqrt 3 j_1 - d_z + { \sqrt 3 \delta_j \over 12 } - { \sqrt 3 \eta_j \over 4 } \right ] \nonumber \\ & = & - { 4 \sqrt 3 \over 3 } j_{yz } + { c_{yz } \over \sqrt 3 } + { j_{yz } - c_{yz } - \sqrt 3 d_y \over 3 \sqrt 3 j_1 } \nonumber \\ & & \ \times \left [ \delta_j - \eta_j - 2 \sqrt 3 d_z + c_{y - x } + 3j_2 \right ] \ . \label{alphaeq } \end{aligned}\ ] ] also @xmath456 \ . \label{alphaeq2 } \end{aligned}\ ] ] we first define the green s function @xmath457 \ .\end{aligned}\ ] ] then we will need the frequency and wave vector dependent susceptibility given by @xmath458 in terms of which the dynamical structure factor is @xmath459 \ , \end{aligned}\ ] ] where @xmath460^{-1}$ ] . we have that @xmath461 n(\omega ) \nonumber \\ & = & \sum_{m , n } \biggl\ { p_n \delta(\omega - e_m + e_n ) \langle n|{\cal s}^{\alpha } ( { \bf q } ) |m \rangle \langle m| { \cal s}^{\beta } ( -{\bf q})|n \rangle \nonumber \\ & & \ \ - p_m \delta(\omega + e_n - e_m ) \langle m|{\cal s}^{\beta } ( -{\bf q } ) |n \rangle \langle n| { \cal s}^{\alpha } ( { \bf q})|m \rangle \biggr\ } n(\omega ) \nonumber \\ & = & \sum_{m , n } p_n \delta(\omega - e_m + e_n ) [ 1 - e^{-\beta(e_m - e_n ) } ] n(\omega ) \nonumber \\ & & \ \ \times \langle n|{\cal s}^{\alpha } ( { \bf q } ) |m \rangle \langle m| { \cal s}^{\beta } ( -{\bf q})|n \rangle \nonumber \\ & = & \sum_{m , n } p_n \delta(\omega - e_m + e_n ) { 1 - e^{-\beta(e_m - e_n ) } ] \over e^{\beta(e_m - e_n ) } -1 } \nonumber \\ & & \ \ \times \langle n|{\cal s}^{\alpha } ( { \bf q } ) |m \rangle \langle m| { \cal s}^{\beta } ( -{\bf q})|n \rangle \nonumber \\ & = & \sum_{m , n } p_m \delta(\omega - e_m + e_n ) \langle n|{\cal s}^{\alpha } ( { \bf q } ) |m \rangle \langle m| { \cal s}^{\beta } ( -{\bf q})|n \rangle \ .\end{aligned}\ ] ] we write @xmath462 where the subscript @xmath463 indicates a component with respect to the local canted axes . then , to linear order in the bose operators we have @xmath464 \nonumber \\ & = & \sqrt { s \over 2 n } \sum_{\tauv , { \bf r } } e^{i { \bf q } \cdot ( \tauv + { \bf r } ) } \nonumber \\ & & \ \times \biggl ( { \cal r}_{\alpha x } ( \tau ) [ a({\bf r } , \tauv ) + a^\dagger({\bf r } , \tauv ) ] \nonumber \\ & & \ \ -i { \cal r}_{\alpha y } ( \tau ) [ a({\bf r } , \tauv ) - a^\dagger({\bf r } , \tauv ) ] \biggr ) \ .\end{aligned}\ ] ] now set @xmath465 then @xmath466 \nonumber \\ & = & \sqrt{s \over 2 } \sum_\tauv \biggl [ \gamma_\alpha ( \tau ) a^\dagger_\tau ( { \bf q } ) + \gamma_\alpha ( \tau)^ * a_\tau ( -{\bf q } ) \biggr ] \nonumber \\ & = & \sqrt{s \over 2 } \sum_{\tauv \mu } \biggl [ \gamma_\alpha ( \tau ) \biggl ( c^{(\mu)}_\tau ( { \bf q})^ * x_\mu^\dagger ( { \bf q } ) - d_\tau^{(\mu ) } ( { \bf q } ) x_\mu(-{\bf q})\biggr ) \nonumber \\ & & \ \ + \gamma_\alpha ( \tau)^ * \biggl ( c_\tau^{(\mu)}(-{\bf q } ) x_\mu(-{\bf q } ) - d_\tau^{(\mu ) } ( -{\bf q})^ * x_\mu^\dagger ( { \bf q})\biggr ) \biggr ] \nonumber \\ & = & \sqrt{s \over 2 } \sum_{\tauv \mu } \biggl [ \biggl ( \gamma_\alpha ( \tau ) c_\tau^{(\mu)}({\bf q})^ * - \gamma_\alpha(\tau)^ * d_\tau^{(\mu ) } ( { \bf q})^ * \biggr ) x_\mu^\dagger ( { \bf q } ) \nonumber \\ & & \ \ + \biggl ( \gamma_\alpha ( \tau)^ * c_\tau^{(\mu)}({\bf q } ) - \gamma_\alpha(\tau ) d_\tau^{(\mu ) } ( { \bf q } ) \biggr ) x_\mu(-{\bf q } ) \biggr ] \ .\end{aligned}\ ] ]

helioseismology is one of the possible ways to study solar physics ( see e.g. @xcite ) . in 1916 plaskett observed fluctuations in the velocity of superficial elements exploiting doppler effect . in 1962 leighton detected an oscillation with period @xmath1 that was damped after a few minutes . the first detections were characterized by limited spatial and temporal extensions and , hence , they were typically interpreted as local phenomena in the solar atmosphere . oscillations in the solar diameter were detected in 1975 by hill and , for the first time , the global nature of these oscillations was suggested . today helioseismology is able to provide quantitative information about many solar properties including the speed of sound , chemical composition , the thickness of the convective zone and internal rotation ( see e.g. @xcite ) . one of the issues that will be investigated in this article is the possibility to exploit solar physics to study the so - called _ chameleon particles_. these chameleons are quantum fields ( typically scalar ) with a mass which is an increasing function of the matter density of the environment and they have been originally introduced in @xcite . for a general introduction to chameleon fields see @xcite and references therein . the careful reader may wonder whether the coupling matter - chameleon leads to detectable deviations from standard general relativity ( gr ) . after all , in the solar system the matter density is small and , hence , the chameleon can be very light ( this issue is model dependent ) . it is common knowledge that it is not possible to introduce in a model a very light scalar degree of freedom with a generic coupling to matter because the light scalar particle would mediate a long - range force and phenomenological constraints must be faced . naturally , when the matter density is large , the chameleon is heavy and , consequently , the chameleon mediated fifth - force is screened . however , in the solar system the field can be light and another mechanism is required to hide the chameleon in gravity tests : it is called thin - shell mechanism . when this mechanism is operative in the sun , where @xmath2 is the mass of the chameleon inside the object ( e.g. the sun ) and @xmath3 is the radius of the object . ] , the chameleon field outside the sun is determined , to a good approximation , by a thin - shell just below the solar surface . the thin - shell mechanism suppresses the chameleon force and , in this way , phenomenological constraints on deviations from gr can be faced in the solar system . remarkably , the chameleon effect plays also a very important role for constructing viable modified gravities , see for example @xcite . the purposes of this paper are : * we analyze the effects of a chameleon - induced deviation from standard gravity in the solar thin - shell . we make explicit numerical simulations exploiting solar models that take into account the presence of the chameleon field . in particular , we discuss , on the one hand , a standard chameleonic scenario where the mass scale of the potential is fine - tuned to the mev dark energy scale and , on the other hand , a second chameleonic model where this fine - tuning is removed . we parametrize the presence of the chameleon through a modification of the newton constant inside the shell . interestingly , our solar models are _ inconsistent _ with the standard observables of helioseismic observations , namely , a ) the thickness of the convective zone ; b ) the helium abundance on the solar surface ; c ) the speed of sound as a function of the distance from the center of the sun . in these models the thickness of the shell is @xmath4 of the solar radius . * the thickness of the thin - shell is bounded by lunar laser ranging ( see for example @xcite ) to be less than approximately @xmath5 . for this reason we developed also solar models with shells much thinner than @xmath6 . in particular , we considered also models where the thickness of the shell is @xmath7 and @xmath8 . in the case we mentioned last , the shell is the thinnest that can be analyzed by the computer and the results are basically indistinguishable from a standard solar model ( ssm ) . * in order to explore further the connection between chameleon fields and solar physics , we consider a modified fujii s model ( mfm ) recently analyzed in @xcite and we show that , in this model , a conceivable interpretation of the solar oscillations is given by quantum vacuum fluctuations . the equations of helioseismology are _ classical _ hydrodynamical equations where the attractive contribution of gravity is balanced by pressure ( i.e. electromagnetic ) force . in this article , however , we suggest to exploit quantum physics to study the oscillations in the sun . in particular , the solar oscillations are interpreted as quantum vacuum fluctuations of a scalar field ( the dilaton ) while the electromagnetic pressure force of the classical picture is reinterpreted as a repulsive contribution due to ( the gradient of the ) quantum vacuum energy . as we will see , the mfm is telling us that helioseismology could be quantum physics . the connection between solar physics and chameleon fields has also been discussed in @xcite . as far as the organization of this article is concerned , in section [ approx ] we discuss some useful formulas for the standard chameleon field ; in section [ smodels ] we build solar models which are properly modified to take into account the presence of a standard chameleon field . in section [ model ] we present the lagrangian of the mfm . some aspects of solar physics are investigated from the standpoint of the mfm in paragraph [ helio ] , in particular , we discuss quantum helioseismology . in the final part of the paper , we briefly touch upon some concluding remarks . the standard chameleon mechanism is obtained through a competition between two different contributions to the effective potential : on the one hand , a bare potential ( which is not necessarily run - away ) and , on the other hand , a matter branch . when the matter density is large , the matter branch is uplifted and a large mass is obtained ( i.e. the field is stabilized ) . on the contrary , on cosmological distances , the matter density is very small , the effective potential is basically indistinguishable from the bare one and the chameleon can be very light . in this section we consider one example of chameleonic potential which has been discussed in the literature . in particular , we have the following contributions to the effective potential . ] : 1 ) a run - away bare potential of the form @xmath9 , where @xmath10 is the chameleon , @xmath11 is an integer , @xmath12 is the mass scale of the potential and @xmath13 is a real number ; 2 ) a matter branch contribution given by an exponential function . in other words , the effective potential is @xmath14 where @xmath15 is a constant parameter ( that can be also much larger than one ) , @xmath16 is the equation of state of the i - th particle specie and @xmath17 is the matter density of the i - th particle specie . the acceleration induced by the chameleon is given by @xmath18 . the variation of the chameleon field inside the shell can be estimated with @xmath19 , \label{formuletta}\ ] ] where @xmath20 is the matter density inside the object , @xmath3 is the solar radius , @xmath21 is the value of the chameleon field inside the object and the thin - shell is in the region @xmath22 . we can estimate the absolute value of the chameleonic acceleration [ chameleonforce ] near the surface of the object as a , where @xmath23 is the thickness of the shell and @xmath24 is the variation of the chameleon field inside the shell . let us gather some useful formulas for the chameleon inside the sun . as we will show explicitly , if we choose @xmath25 ( where @xmath3 is the solar radius ) , @xmath26 everywhere constant and @xmath27 , we obtain a chameleonic acceleration near the surface of the sun that turns out to be roughly similar to the metric - induced acceleration near the surface of the sun . now to the calculations . we call @xmath10 the chameleon field . as already pointed out in @xcite , there are models where the chameleon field is basically constant inside the shell . this result can be obtained removing the fine - tuning on the scale of the potential . hence , let us choose @xmath28 gev . moreover , we choose @xmath29 . if we define @xmath30 and @xmath31 , we can write [ formuletta ] as c= c - c_c _ c r_c^2 [ ] _ c r_c^2 [ ^2/2 ] , where we made the approximation @xmath32 in the shell . the best thing would be to evaluate @xmath33 numerically , but we will simply assume @xmath34 . in this way we have a = 3 3 = 3 , where in the last step we introduced @xmath35 defined by @xmath36 . now we can write ( in international units ) : a^2 3 ^2 3 10 ^ -19 km / m^2 ^2 3 m / s^2 g _ , therefore , if we choose @xmath37 we have ( near the surface of the sun ) a chameleon - induced acceleration comparable to the metric - induced acceleration . the first possibility we explore is a model inspired to a standard chameleonic scenario but without fine - tuning on the scale of the potential . we choose a large and constant planck mass and we fix the matter density to a reasonable value by - hand . in particular , in the model of section [ approx ] , we choose @xmath25 , where @xmath3 is the solar radius . @xmath26 everywhere constant . @xmath27 . with our choice of parameters the chameleon is basically constant everywhere inside the sun . one approximation in the chameleonic literature is to consider a constant value of the chameleon for @xmath38 ( see for example @xcite ) . this means that we are writing the scalar field inside the sun as the sum of ( see @xcite for further details ) ( 1 ) a constant background value and ( 2 ) a fluctuating component . in other words we exploit a background field gauge ( see for example section 17.4 of reference @xcite ) with a constant background field . we infer that the chameleon - mediated force is vanishing when averaged over the sphere @xmath38 because the constant background field has vanishing gradient . the fluctuating component is still present , it mediates a force on short distances because in the uv the chameleon has non - vanishing gradient . hence , an interesting solar model can be constructed requiring a vanishing modification of gravity in the region @xmath38 . as far as the thin - shell is concerned , we exploit the calculations of section [ approx ] and , consequently , we modify gravity in the shell by substituting @xmath39 . we stress once again that inside the sphere @xmath38 , gravity is modified at short distances : the fluctuating component of the chameleon is the relevant one in the uv , it is not constant , it has a non - vanishing gradient and , hence , it mediates a force . the force is vanishing only when averaged in the sphere @xmath38 , because the background field is constant . our second example is the standard chameleon , where we fine - tune the scale of the potential to the dark energy scale ( @xmath40 ) and we choose a large planck mass @xmath41 . moreover we choose @xmath27 . interestingly , the formulas shown for model 1 do not depend on the scale of the potential @xmath12 and , consequently , if we assume that the thin - shell is roughly 1/10 of the solar radius , we obtain the same modification of gravity of model 1 , namely , @xmath39 inside the shell . needless to say , the scale of the potential is related to the mass of the chameleon . in this section we will exploit the possibility of using the sun as a sort of laboratory for testing the existence of a chameleon field . we will see how the introduction of a modification of gravity in a outer shell of the sun ( as in model 1 and model 2 discussed above ) alters significantly some solar properties . note that elsewhere in the literature , the sun , or other stellar structures , have been used for testing the possible existence of a new physics , see e.g.@xcite . let us first summarize some useful concepts in solar physics . a standard solar model ( ssm ) can be defined as a description of the solar structure providing ( within uncertainties ) the observed properties of the sun and exploiting initial chemical - physical characteristics chosen in a proper range @xcite . usually , in these models magnetic fields and rotation are neglected and , consequently , the structure is spherically symmetric . under these conditions , the physical / chemical structure of the sun and its evolution in time is obtained solving a set of fundamental equations : mass conservation law , hydrostatic equilibrium , energy production , energy transport , equation of state and time variation of elemental abundances due to nuclear burning and microscopic diffusion . in addition , for solving this set of equations one has to know other important functions : the solar opacity @xmath42 ( the product @xmath43 represents the inverse of the photon mean free path in the stellar interior ) and the nuclear cross sections , see e.g @xcite for a detailed description of stellar equilibrium equations . in particular in our calculations , the numerical solution of the equations mentioned above is obtained by using the stellar evolutionary code franec @xcite , by adopting livermore 2006 equation of state ( eos ) @xcite , livermore radiative opacity tables @xcite , molecular opacities from ref . @xcite , conductive opacities from ref . @xcite and nuclear reaction rates from nacre compilation @xcite , taking into account the astrophysical factors @xmath44 and @xmath45@xcite and the @xmath46 electron capture rate from @xcite . so in order to build a ssm one studies the evolution of an initially homogeneous solar mass model up to the sun age ( @xmath47 @xcite ) so as to reproduce the main solar observables : luminosity @xmath48erg / s @xcite , radius @xmath49 cm @xcite and photospheric chemical composition , usually indicated with the ratio @xmath50 where @xmath51 indicates the ( relative ) abundance in mass of all the elements heavier than helium , usually called metals , and @xmath52 indicates the ( relative ) abundance in mass of hydrogen . we adopted the most recent determination of photospheric abundances @xcite , corresponding to @xmath53 . to achieve as outputs the three solar observable mentioned above , one can play with three input parameters : 1 ) the initial abundance of helium ( @xmath54 ) , i.e. the helium abundance at the zero age of the sun . the solar luminosity basically depends on @xmath54 : the smaller @xmath54 , the smaller the temperature , the smaller the luminosity for fixed solar age . 2 ) the initial abundance of heavy elements ( @xmath55 ) . this quantity essentially determines the present metal content in the photosphere , the relationship being due to the efficiency of diffusion mechanism , since metals are not burnt in the sun . furthermore @xmath55 affects the solar luminosity since the opacity of a star is strongly influenced by the presence of heavy elements , see e.g. @xcite . 3 ) the so called mixing - length parameter ( @xmath56 ) . the radius is related to convection which dominates the energy transfer in the outer layers of the sun . the smaller @xmath56 , the larger the local gradient of temperature , see e.g. @xcite for a detailed description of the mixing length theory . we infer that a smaller temperature on the surface is obtained and , since @xmath57 is fixed , a bigger radius is necessary granted that we approximate the solar spectrum with a black - body one . now it is crucial to observe that helioseismology has added other solar observables which have to be reproduced : the photospheric helium abundance @xmath58 , the depth of the solar convective zone @xmath59 and the sound speed profile in the solar interior . these quantities have been accurately determined in the last years by helioseismic measurements ( see e.g. @xcite and references therein ) and are crucial in testing solar model calculations ( see e.g. @xcite ) . indeed the three input parameters can not be considered free anymore . in table [ tab1 ] and fig . [ figu ] we report the main results of our ssm , obtained with all the ingredients mentioned above in this section . we also report the helioseismic determinations with their more conservative uncertainties as derived in @xcite . in particular in fig . [ figu ] ( left panel ) we report the relative difference of the squared isothermal sound speed ( @xmath60 ) with respect to the helioseismic determination . one can easily see that our ssm does not recover the helioseimic data . this is a well known and still open problem in solar physics : standard solar models including most recent data on phothospheric composition @xcite do not recover helioseismic observations , whereas adopting older composition @xcite some agreement is achieved ( see e.g. @xcite ) . , for ssm ( continuous blu line ) and non - ssm with a thin - shell of thickness @xmath6(dashed red line ) . the yellow shaded area represents the conservative @xmath61 helioseismic uncertainties on squared isothermal sound speed as derived in @xcite . lower panel . we show the behaviour of opacity along the solar profile for our solar models , same notations as in upper panel.,title="fig:",width=8,height=6 ] , for ssm ( continuous blu line ) and non - ssm with a thin - shell of thickness @xmath6(dashed red line ) . the yellow shaded area represents the conservative @xmath61 helioseismic uncertainties on squared isothermal sound speed as derived in @xcite . lower panel . we show the behaviour of opacity along the solar profile for our solar models , same notations as in upper panel.,title="fig:",width=8,height=6 ] now let us see what occurs in the solar interior when the chameleon effect is taken into account . we have built a non - ssm in which we have modified artificially the gravity in the external solar shell , i.e. @xmath39 for @xmath62 . the results of such model are reported again in table [ tab1 ] and fig . one can immediately see that the existence of a chameleon field in the external solar shell does not improve the agreement with helioseismology at all . let us discuss in some details our non - ssm model , taking into account that the aim of this section is not a detailed astrophysical discussion about solar modeling , but only the suggestion of how the chameleon physics can be phenomenologically studied in the solar interior . in order to reproduce , at the solar age , the main solar observables ( luminosity , radius , photospheric abundances ) we have varied the input parameters in our non - ssm , see table 1 . in particular , the effect of extra gravity in the outer part of the sun determined a more efficient gravitational settling at the solar surface , so that in order to recover the observed photospheric @xmath63 one has to start with a higher @xmath55 . let us observe that the initial metal content clearly influences the present inner values , they are substantially coincident apart for a 10% variation due to elemental diffusion . consequently with the only increasing of @xmath55 we would have an increasing of the solar opacity in the interior and consequently ( at fixed age ) a less bright sun . to compensate this effect we have to increase the initial helium abundance @xmath54 . but , even if we have increased the initial helium abundance , one sees from table [ tab1 ] that the present photospheric helium abundance results lower than the values obtained in the ssm . the gravity modification occurring in a external region alters in such a way the equilibrium condition of the stellar structure , resulting in significant variations of physical and chemical properties of the sun well below 0.9@xmath64 . in particular we observe that the diffusion efficiency of the elements in the region around 0.7 - 0.8 @xmath64 is increased : in fig . [ figelements ] we report the helium and metals profiles along the solar interior and we can see that the gap in abundance between the radiative and the convective is almost doubled when we pass from ssm to non - ssm model . concerning the depth the convective zone @xmath59 , in the solar model with modified gravity the transition between radiative regime and convective one occurs in more external region . the substantial modification of the elemental abundances in this part of the sun determines a decreasing of the solar opacity in this region . in fig . [ figu ] we show the profile of the solar opacity along the solar structure , both for ssm and non - ssm model . since in non - ssm the opacity is decreased for solar radius greater than @xmath65 , the energy transport by radiation remains efficient for a larger part of the sun . finally we observe that the squared sound speed profile is completely different with respect to the helioseismic observations , when we are above 0.7@xmath66 . around @xmath67 the agreement is clearly better , also with respect to the ssm case , this is due to the fact that generally a less opaque sun is preferred by helioseismic measurements of the sound speed , see e.g.@xcite . in conlusion we can say that the non - standard solar model with modified gravity in an external shell , is in disagreement with helioseismic observations . one can observe that we start from a standard solar model already in disagreement , but on the other hand we can observe that if we were able to build a standard solar model perfectly consistent with the central value of helioseismic data , the introduction of a chameleon effect would induce a strong variation on @xmath58 , @xmath59 and sound speed profile well above the existing conservative uncertainties on such helioseismic observables . as we have seen , if we consider a standard chameleonic scenario , the modification of gravity inside the sun induced by the presence of a chameleon field is _ not _ compatible with helioseismology . in particular , we considered two different models , namely model 2 ( a standard chameleonic scenario where the planck mass is constant and the scale of the potential is fine - tuned to the mev scale ) and model 1 ( a chameleonic scenario where the planck mass is constant but the fine - tuning on the scale of the potential is absent ) . these two models are both inconsistent with helioseismology . as we already mentioned above , we considered a thin - shell with @xmath68 . the careful reader may wonder what are the theoretical grounds supporting this particular thickness and what are the phenomenological constraints on it . the evaluation of the thickness of the shell in the general case requires a numerical calculation ( see @xcite ) , however , an estimate can be obtained from the formula ( see for example @xcite ) : = where @xmath69 is the newtonian potential at the surface of the object and @xmath70 is the value of the chameleon field in the background outside the sun . remarkably , lunar laser ranging constrains the difference in the free - fall acceleration of the moon and the earth towards the sun to be less than approximately one part in @xmath71 ( see for example @xcite ) and , consequently , the bound < 10 ^ -7 is obtained . this translates into a bound for the solar thin - shell given by 10 ^ -3 < 10 ^ -10 . for this reason , we developed also solar models where the thin - shell is much smaller than @xmath72 . the results are shown in fig . [ modellinuovi ] and table 1 . the dashed red line corresponds to @xmath68 . the dashed green line corresponds to @xmath73 . the pink line corresponds to @xmath74 and it is the thinnest shell that can be analyzed by our numerical simulations . one can see that solar models with a shell thinner than @xmath8 are , at the moment , basically indistinguishable from ssm . at this stage , llr provides a constraint stronger than the one coming from solar models . , for ssm ( continuous blu line ) and non - ssms . the dashed red line corresponds to @xmath68 . the dashed green line corresponds to @xmath73 . the pink line corresponds to @xmath74 . the yellow shaded area represents the conservative @xmath61 helioseismic uncertainties on squared isothermal sound speed as derived in @xcite . , height=6 ] .we report the values for : initial helium abundance @xmath54 , initial metals abundance @xmath55 , photospheric helium ( metal ) abundance , @xmath58 ( @xmath75 ) , depth of the convective zone , @xmath59 , both for standard solar model ( ssm ) , and for non standard solar models with shell thickness in the range from @xmath76 to @xmath8 , see text . we also report the helioseismic values for @xmath58 and @xmath59 with their conservative @xmath77 uncertainties as derived in @xcite [ cols="^,^,^,^,^,^",options="header " , ] + . in the remaining part of this article , we will further explore this connection between chameleons and solar physics . however , we will leave the standard chameleonic scenario and we will discuss a different set - up that we call modified fujii s model ( mfm ) . indeed the mfm is the result of a lagrangian discussed by fujii in @xcite and recently modified by one of us to discuss the cosmological constant problem in @xcite . one remark is necessary . the lagrangian of the mfm is written in the string frame and , once a conformal transformation to the e - frame is performed , we obtain a chameleonic competition between a run - away potential and the @xmath78 term of gluons @xcite . in other words , in [ veff ] the minimum in the effective potential is generated by @xmath79 , but in the mfm , once we move to the e - frame , the minimum is formed through the presence of the @xmath80 term of gluons with reasonable assumptions about the couplings @xcite ( the interactions must be a decreasing function of energy and this is natural because the ir region corresponds to the weak coupling of the string ) . here is the model . we have two different conformal frames . on the one hand , we have the string frame ( s - frame ) where the cosmological constant ( cc ) is large ( for example planckian ) and the fields are stabilized ( including the dilaton @xmath81 ) . on the other hand , there is the einstein frame ( e - frame ) , where the dilaton @xmath82 is a chameleon field . remarkably , in the e - frame the amount of scale invariance is parametrized by @xmath82 : scale invariance is abundantly broken locally ( i.e. in this room ) and it is almost restored globally ( i.e. on cosmological distances ) . consequently , the cc is around the mev - scale in this frame but the vacuum energy is large locally . we write the string frame lagrangian as ( the gauge part is not written explicitly but it is present in the theory ) = l_si + l_sb , [ ltotale]where the scale - invariant part of the lagrangian is given by : @xmath83 @xmath3 is the curvature . @xmath84 is a scalar field representative of matter fields , @xmath85 , @xmath86 , @xmath87 and @xmath88 . we can also write terms of the form @xmath89 , @xmath90 and @xmath91 which are multiplied by dimensionless couplings . however we will not include these terms . indeed , the first two terms can be removed exploiting symmetries of strong interaction @xcite and the @xmath91 term does not clash with the solution to the cc problem , because the renormalized planck mass in the ir region is an exponentially decreasing function of @xmath82 ( see also @xcite ) . the symmetry breaking lagrangian @xmath92 is supposed to contain scale - non - invariant terms , in particular , a stabilizing ( stringy ) potential for @xmath81 in the s - frame . for this reason we write : _ sb=- ( a ^2 + b + c ) . [ sb ] @xmath92 is a stabilizing contribution . the reader can find a possible choice of parameters in @xcite . for a discussion of the e - frame lagrangian and of the different dynamical behaviour of the dilaton in the two frames see @xcite . interestingly , the masses of the matter particles in the e - frame are determined by the planck mass @xcite . as pointed out in @xcite , this model satisfies a chameleonic equivalence postulate ( cep ) . here is the cep : _ for each pair of vacua v1 and v2 allowed by the theory there is a conformal transformation that connects them and such that the mass of matter fields @xmath93 ( i.e. @xmath94 evaluated in v1 ) is mapped to @xmath95 ( i.e. @xmath94 evaluated in v2 ) . when a conformal transformation connects two vacua with a different amount of conformal symmetry , an additional term ( in the form of a conformal anomaly ) must be included in the field equations and this additional term is equivalent to the gravitational field . _ in reference @xcite this has been presented as an equivalence postulate for quantum gravity . one aspect of the cep will be particularly useful in this paper : _ all _ the ground states of the model ( including the e - frame vacua ) can be mapped into each other by a conformal transformation . we will come back to this issue in the next section . in this section we will start summarizing the hydrodynamical equations of the classical theory of solar oscillations and then we will proceed with a reinterpretation of the solar oscillations through quantum vacuum fluctuations ( i.e. a quantum portrait ) . here is the classical theory . a hydrodynamical system is specified by some functions of position * r * and time @xmath96 : the speed @xmath97 , the density @xmath98 and the pressure @xmath99 . mass conservation is given by the continuity equation : + ( ) = 0 . [ cont ] if we neglect the viscosity of the gas , the relevant interactions are due to pressure and gravity and , hence , we can write the equation of motion as : ( + * v*)=- p + , [ moto ] where @xmath100 is the gravitational acceleration , connected to the potential @xmath84 by the definition @xmath101 . @xmath84 satisfies the poisson equation = 4 g . [ poisson ] our last equation will be the first principle of thermodynamics @xmath102 . now we suggest a different interpretation of the solar oscillations from the standpoint of the e - frame of the mfm : a quantum picture . let us start with some preliminary remarks . in this paper we assume that the cep is satisfied by the mfm s lagrangian and we discuss some of its consequences . it is not our intention to explore in this paper the theoretical grounds supporting this assumption ( this has been already done in @xcite and related references ) . we are well aware of the fact that the ideas we are going to present in this section can not be considered as a rigorous proved statement , however , we think it s useful to mention this quantum picture of helioseismology as a guideline towards future research developments . as already mentioned above , the cep is telling us that , in the mfm , _ all _ the ground states ( including the e - frame vacua ) can be mapped into each other by a conformal transformation . we follow @xcite and we suppose to live in the e - frame . let us map the ground state of a composite hadron of the e - frame into a macroscopic ground state in the e - frame . when we perform this conformal transformation the size of the hadron is stretched to macroscopic length scales , but also the de broglie wavelength is rescaled to macroscopic size by the same conformal factor . we infer that quantum physics is stretched to macroscopic length scales . let us exploit these considerations to map the ground state of a composite hadron into the ground state of the sun . this point must be further analyzed . as far as the hadronic radius is concerned , we can exploit the standard qcd formula c 200 mev fm connecting the hadronic radius to @xmath103 . the cep is telling us that the same physics can be stretched to macroscopic distances . let us explore the consequences of this statement in the case of the sun . solar matter is globally a gauge singlet under @xmath104 and , hence , we choose a sterile spinor field to represent solar matter . needless to say , we can not neglect gravity and the masses of particles , therefore , the spinor field is massive . if we exploit the cep , the standard qcd condensation formula is mapped into c _ r _ , where @xmath64 is the solar radius and @xmath105 is the condensation scale . now , what is the value of the condensation scale @xmath105 ? the mass of the spinor field is much larger than the planckian value of the particles masses in the sun , because , in harmony with the cep , we assume the spinor to be a _ composite _ field representative of _ many _ solar particles ( analogously to a nucleon inside the hadron ) and consequently @xmath106 . moreover , @xmath107 . hence the solar @xmath108 is extremely large with respect to the hadronic value . we infer that @xmath109 is an increasing function of @xmath82 correspond to the ir region ( see @xcite ) . ] and this is in harmony with the fact that quantum physics is stretched to solar length scales . happily this peculiar @xmath109 is fully compatible with a chameleonic behaviour of the de broglie wavelength already discussed in this model in @xcite . this is a good point to discuss more carefully the theoretical grounds supporting quantum helioseismology . the reader may wonder whether the cep is the only motivation for quantum helioseismology . we point out that the terms of the form @xmath110 and @xmath111 in the lagrangian [ bsl1 - 96 ] are present in the ginzburg - landau approach to superconductivity and they can be obtained with a mean field approximation starting from the microscopic theory ( see for example @xcite ) . in other words , the structure of the lagrangian of the mfm is supporting fermion condensation in a natural way and , therefore , the description of the sun as a macroscopic fermionic condensate , namely the quantum helioseismology portrait , is compatible with [ bsl1 - 96 ] ( after a proper conformal transformation to the e - frame ) . to proceed further , let us analyze helioseismology from the standpoint of our quantum portrait of the sun . the number density of solar particles is a decreasing function of @xmath82 because also the solar spinor field is rescaled after the conformal transformation to the e - frame . hence , large @xmath82 corresponds to small number density of solar particles and hence to a large solar radius . we infer that solar oscillations are related to @xmath82-fluctuations . the variation of the solar radius is very small ( with respect to the radius ) and we suggest to interpret these oscillations as small quantum fluctuations of the dilaton around its minimum . interestingly , the model of [ ltotale ] is motivated , at least partially , by string theory @xcite and the string length is chameleonic in this model @xcite . we infer that the strings near the solar surface might be macroscopic and quantum in this model : the oscillations of the solar radius , in this model , might be simply macroscopic ( and quantum ! ) string vibrations . an interesting line of development will try to explore potential phenomenological consequences of this comment on higher - spin fields @xcite ( remarkably when the matter density is small the string mass is small and the massive spectrum of the string is lowered to smaller energies ) . we are led to a reinterpretation of the forces which govern the dynamics of the sun . classical helioseismology is telling us that the solar radius is stabilized by the competition between gravity and pressure ( which is the result of electromagnetic interaction ) . interestingly , in quantum helioseismology , the stabilization of the chameleonic dilaton ( i.e. of the solar radius ) is the result of a competition between gravity , which is attractive , and the gradient of the vacuum energy , which gives a repulsive contribution to matter . since pressure creates the correct competition with gravity in the classical picture , the two contributions ( pressure and gravity ) must have a similar energy scale already at the classical level . this result is automatic in the quantum picture , because pressure is related to the ( gradient of the ) planckian vacuum energy and its value is fixed by the conformal anomaly which is equivalent to gravitation @xcite . in classical helioseismology the oscillations near the surface of the sun are maintained by pressure ( the so - called p - modes ) and , hence , in the mfm , the solar oscillations near the surface are a direct result of a variation of the chameleonic vacuum energy . in other words , the role of the _ electromagnetic _ interaction in classical helioseismology is played , in quantum helioseismology , by the variation of the quantum vacuum energy and this vacuum energy is determined , in our model , by the _ gravitational _ interaction . the length scale of quantum fluctuations is related to the mass of the field and , therefore , in quantum helioseismology the length scale of solar fluctuations is related to the mass of the chameleon . one interesting lesson from quantum helioseismology is that the qcd dynamics is mapped , through a conformal transformation , to a ( quantum ) gravitational dynamics governing solar physics . in other words , it seems at this stage that , in the mfm , ( quantum ) gravitation is the ir limit of qcd . in this sense , there might be no difference in this model between strong interaction and gravity : they might be just two aspects of the same interaction . in a modern language , it seems that the cep might support a unification of all the interactions ( a natural result in heterotic string theory ) and the strength of this interaction is simply parametrized by the value of @xmath82 . further research efforts are necessary to support properly this idea . some words of caution are necessary . first of all , the stringy origin of the lagrangian [ ltotale ] is still under investigation , the connection to string theory is only partial at the moment . moreover , more detailed and quantitative motivations for introducing quantum helioseismology would be welcome and many aspects of this proposal are still waiting for a careful investigation . for these reasons , quantum helioseismology can be considered a conjecture at this stage , even if we assume to describe nature through our chameleonic model ( where the cep is satisfied ) . in this article we discussed some aspects of solar physics from the standpoint of the so - called chameleon fields ( i.e. quantum fields , typically scalar , where the mass is an increasing function of the matter density of the environment ) . here is a summary of our results . + 1 ) we analyzed the effects of a chameleon - induced deviation from standard gravity just below the surface of the sun . in particular , we developed solar models which take into account the presence of the chameleon and we showed that they are inconsistent with helioseismic observations . this inconsistency presents itself not only with the typical chameleon set - up discussed in the literature ( where the mass scale of the potential is fine - tuned to the mev ) , but also if we remove the fine - tuning on the scale of the potential . however , if we modify standard gravity only in a shell of thickness @xmath0 just below the solar surface , the model is basically indistinguishable from a standard solar model . at this stage lunar laser ranging provides a stronger constraint than the one coming from solar models . + 2 ) we pointed out that , in a model recently considered in @xcite , a conceivable interpretation of the solar oscillations is given by quantum vacuum fluctuations . in this model , called mfm , helioseismology is quantum physics . in particular , we assumed that the cep is satisfied by the mfm s lagrangian and we discussed some of its consequences . it was not our intention to explore in this paper the theoretical grounds supporting this assumption ( this has been already done in @xcite and related references ) . we are well aware of the fact that these ideas can not be considered as a rigorous proved statement , however , we think it s useful to illustrate this quantum picture of helioseismology as a guideline towards future research developments . interestingly , the lagrangian of the mfm can be obtained , at least partially , from string theory . if the connection with the string was complete , we could consider quantum helioseismology as a step forward in string phenomenology . needless to say , we live in the lhc era , but we think that it is important to search for alternative ways of creating links between particle physics and experiments . special thanks are due to giovanni fiorentini and franco strocchi for useful comments and suggestions . is extremely grateful to scilla deglinnocenti and to pisa astroparticle group for useful comments and for providing the updated version of opacity tables . we also acknowledge the referees of the journal for pointing out useful comments about a first edition of the manuscript . s. turck - chieze and s. couvidat , `` solar neutrinos , helioseismology and the solar internal dynamics , '' http://dx.doi.org/10.1088/0034-4885/74/8/086901[_rept.prog.phys._ * 74 * ( 2011 ) 086901 ] , http://arxiv.org/abs/1009.0852[arxiv:1009.0852 [ astro-ph.sr ] ] . s. basu and h. antia , `` helioseismology and solar abundances , '' http://dx.doi.org/10.1016/j.physrep.2007.12.002[_phys.rept._ * 457 * ( 2008 ) 217283 ] , http://arxiv.org/abs/0711.4590[arxiv:0711.4590 [ astro - ph ] ] . j. khoury and a. weltman , `` chameleon cosmology , '' http://dx.doi.org/10.1103/physrevd.69.044026[_phys . _ * d69 * ( 2004 ) 044026 ] , http://arxiv.org/abs/astro-ph/0309411[arxiv:astro-ph/0309411 ] . j. khoury and a. weltman , `` chameleon fields : awaiting surprises for tests of gravity in space , '' http://dx.doi.org/10.1103/physrevlett.93.171104[_phys . _ * 93 * ( 2004 ) 171104 ] , http://arxiv.org/abs/astro-ph/0309300[arxiv:astro-ph/0309300 ] . d. f. mota and d. j. shaw , `` evading equivalence principle violations , astrophysical and cosmological constraints in scalar field theories with a strong coupling to matter , '' http://dx.doi.org/10.1103/physrevd.75.063501[_phys . _ * d75 * ( 2007 ) 063501 ] , http://arxiv.org/abs/hep-ph/0608078[arxiv:hep-ph/0608078 ] . j. khoury , `` chameleon field theories , '' http://arxiv.org/abs/1306.4326[arxiv:1306.4326 [ astro-ph.co ] ] . s. nojiri and s. d. odintsov , `` unified cosmic history in modified gravity : from f(r ) theory to lorentz non - invariant models , '' phys . rept . * 505 * , 59 ( 2011 ) [ arxiv:1011.0544 [ gr - qc ] ] . a. zanzi , `` chameleonic dilaton , nonequivalent frames , and the cosmological constant problem in quantum string theory , '' http://dx.doi.org/10.1103/physrevd.82.044006[_phys . rev . _ * d82 * ( 2010 ) 044006 ] , http://arxiv.org/abs/1008.0103[arxiv:1008.0103 [ hep - th ] ] . brax and k. zioutas , `` solar chameleons , '' http://arxiv.org/abs/1004.1846[arxiv:1004.1846 [ astro-ph.sr ] ] . a. c. vincent , p. scott , and r. trampedach , `` light bosons and photospheric solutions to the solar abundance problem , '' http://dx.doi.org/10.1093/mnras/stt683[_mon.not.roy.astron.soc._ * 432 * ( 2013 ) 33323339 ] , http://arxiv.org/abs/1206.4315[arxiv:1206.4315 [ astro-ph.sr ] ] . p. brax , a. lindner , and k. zioutas , `` detection prospects for solar and terrestrial chameleons , '' http://dx.doi.org/10.1103/physrevd.85.043014[_phys.rev._ * d85 * ( 2012 ) 043014 ] , http://arxiv.org/abs/1110.2583[arxiv:1110.2583 [ hep - ph ] ] . o. baker , a. lindner , y. semertzidis , a. upadhye , and k. zioutas , `` detection of radiation pressure from solar chameleons , '' http://arxiv.org/abs/1201.6508[arxiv:1201.6508 [ astro-ph.im ] ] . a. zanzi , `` chameleonic theories : a short review , '' http://arxiv.org/abs/submitted to acta physica polonica[submitted to acta physica polonica ] . a. zanzi , `` species , chameleonic strings and the concept of particle , '' http://arxiv.org/abs/1206.4794[arxiv:1206.4794 [ hep - th ] ] . s. weinberg , `` the quantum theory of fields . 2 : modern applications , '' . cambridge university press - 1996 . a. bottino , g. fiorentini , n. fornengo , b. ricci , s. scopel , _ et al . _ , `` does solar physics provide constraints to weakly interacting massive particles ? , '' http://dx.doi.org/10.1103/physrevd.66.053005 [ _ phys.rev . _ * d66 * ( 2002 ) 053005 ] , http://arxiv.org/abs/hep-ph/0206211[arxiv:hep-ph/0206211 [ hep - ph ] ] . s. cassisi , v. castellani , s. deglinnocenti , g. fiorentini , and b. ricci , `` stellar evolution and large extra dimensions , '' http://dx.doi.org/10.1016/s0370-2693(00)00442-1[_phys.lett._ * b481 * ( 2000 ) 323332 ] , http://arxiv.org/abs/astro-ph/0002182[arxiv:astro-ph/0002182 [ astro - ph ] ] . s. deglinnocenti , w. dziembowski , g. fiorentini , and b. ricci , `` helioseismology and standard solar models , '' http://dx.doi.org/10.1016/s0927-6505(97)00004-2[_astropart.phys._ * 7 * ( 1997 ) 7795 ] , http://arxiv.org/abs/astro-ph/9612053[arxiv:astro-ph/9612053 [ astro - ph ] ] . m. stix , `` the sun , '' http://arxiv.org/abs/springer-verlag ( 1991 ) - berlin [ springer - verlag ( 1991 ) - berlin ] . a. chieffo and o. straniero _ apj suppl . _ * 17 * ( 1989 ) 47 . f. ciacio , s. deglinnocenti , and b. ricci , `` updating standard solar models , '' http://dx.doi.org/10.1051/aas:1997168 [ _ astron.astrophys.suppl.ser . _ * 123 * ( 1997 ) 449454 ] , http://arxiv.org/abs/astro-ph/9605157[arxiv:astro-ph/9605157 [ astro - ph ] ] . f. j. rogers , j. swenson , fritz , and a. iglesias , carlos , `` opal equation of state tables for astrophysical applications , '' http://dx.doi.org/10.1086/176705[_astrophys.j._ * 456 * ( 1996 ) 902 ] . c. a. iglesias and f. j. rogers , `` updated opal opacities , '' http://dx.doi.org/10.1086/177381[_astrophys.j._ * 464 * ( 1996 ) 943 ] . j. w. ferguson , d. r. alexander , f. allard , t. barman , j. g. bodnarik , _ et al . _ , `` low temperature opacities , '' http://dx.doi.org/10.1086/428642[_astrophys.j._ * 623 * ( 2005 ) 585596 ] , http://arxiv.org/abs/astro-ph/0502045[arxiv:astro-ph/0502045 [ astro - ph ] ] . s. cassisi , a. potekhin , a. pietrinferni , m. catelan , and m. salaris , `` updated electron - conduction opacities : the impact on low - mass stellar models , '' http://dx.doi.org/10.1086/516819[_astrophys.j._ * 661 * ( 2007 ) 10941104 ] , http://arxiv.org/abs/astro-ph/0703011[arxiv:astro-ph/0703011 [ astro - ph ] ] . c. angulo , m. arnould , m. rayet , p. descouvemont , d. baye , _ et al . _ , `` a compilation of charged - particle induced thermonuclear reaction rates , '' http://dx.doi.org/10.1016/s0375-9474(99)00030-5[_nucl.phys._ * a656 * ( 1999 ) 3183 ] . m. marta , a. formicola , g. gyurky , d. bemmerer , c. broggini , _ et al . _ , `` precision study of ground state capture in the n-14(p , gamma ) o-15 reaction , '' http://dx.doi.org/10.1103/physrevc.78.022802 [ _ phys.rev . _ * c78 * ( 2008 ) 022802 ] , http://arxiv.org/abs/0807.4919[arxiv:0807.4919 [ nucl - ex ] ] . h. costantini , d. bemmerer , f. confortola , a. formicola , g. gyurky , _ et al . _ , `` the he-3(alpha , gamma)be-7 s - factor at solar energies : the prompt gamma experiment at luna , '' http://dx.doi.org/10.1016/j.nuclphysa.2008.09.014[_nucl.phys._ * a814 * ( 2008 ) 144158 ] , http://arxiv.org/abs/0809.5269[arxiv:0809.5269 [ nucl - ex ] ] . a. junghans , e. mohrmann , k. snover , t. steiger , e. adelberger , _ et al . _ , `` precise measurement of the be-7(p , gamma ) b-8 s factor , '' http://dx.doi.org/10.1103/physrevc.68.065803[_phys.rev._ * c68 * ( 2003 ) 065803 ] , http://arxiv.org/abs/nucl-ex/0308003[arxiv:nucl-ex/0308003 [ nucl - ex ] ] . e. g. adelberger , s. m. austin , j. n. bahcall , a. balantekin , g. bogaert , _ et al . _ , `` solar fusion cross - sections , '' http://dx.doi.org/10.1103/revmodphys.70.1265[_rev.mod.phys._ * 70 * ( 1998 ) 12651292 ] , http://arxiv.org/abs/astro-ph/9805121[arxiv:astro-ph/9805121 [ astro - ph ] ] . j. n. bahcall and m. pinsonneault , `` solar models with helium and heavy element diffusion , '' http://dx.doi.org/10.1103/revmodphys.67.781 [ _ rev.mod.phys . _ * 67 * ( 1995 ) 781808 ] , http://arxiv.org/abs/hep-ph/9505425[arxiv:hep-ph/9505425 [ hep - ph ] ] . j. n. bahcall , a. m. serenelli , and m. pinsonneault , `` how accurately can we calculate the depth of the solar convective zone ? , '' http://dx.doi.org/10.1086/423027[_astrophys.j._ * 614 * ( 2004 ) 464 ] , http://arxiv.org/abs/astro-ph/0403604[arxiv:astro-ph/0403604 [ astro - ph ] ] . c. w. allen , `` astrophysical quantities , '' _ the athlone press - university of london . _ m. asplund , n. grevesse , a. j. sauval , and p. scott , `` the chemical composition of the sun , '' http://dx.doi.org/10.1146/annurev.astro.46.060407.145222 [ _ ann.rev.astron.astrophys . _ * 47 * ( 2009 ) 481522 ] , http://arxiv.org/abs/0909.0948[arxiv:0909.0948 [ astro-ph.sr ] ] . s. basu and h. antia , `` constraining solar abundances using helioseismology , '' http://dx.doi.org/10.1086/421110 [ _ astrophys.j . _ * 606 * ( 2004 ) l85 ] , http://arxiv.org/abs/astro-ph/0403485[arxiv:astro-ph/0403485 [ astro - ph ] ] . w. dziembowski , g. fiorentini , b. ricci , and r. sienkiewicz , `` helioseismology and the solar age , '' _ astron.astrophys . _ * 343 * ( 1999 ) 990 , http://arxiv.org/abs/astro-ph/9809361[arxiv:astro-ph/9809361 [ astro - ph ] ] . s. deglinnocenti and b. ricci , `` a mixed solar core , solar neutrinos and helioseismology , '' http://dx.doi.org/10.1016/s0927-6505(97)00057-1[_astropart.phys._ * 8 * ( 1998 ) 293296 ] , http://arxiv.org/abs/astro-ph/9710292[arxiv:astro-ph/9710292 [ astro - ph ] ] . n. grevesse and a. j. sauval , `` standard solar composition , '' http://dx.doi.org/10.1023/a:1005161325181[_space sci.rev._ * 85 * ( 1998 ) 161174 ] . f. villante , a. serenelli , f. delahaye , and m. pinsonneault , `` the chemical composition of the sun from helioseismic and solar neutrino data , '' http://arxiv.org/abs/1312.3885[arxiv:1312.3885 [ astro-ph.sr ] ] . t. p. waterhouse , `` an introduction to chameleon gravity , '' http://arxiv.org/abs/astro-ph/0611816[arxiv:astro-ph/0611816 ] . y. fujii , `` mass of the dilaton and the cosmological constant , '' http://dx.doi.org/10.1143/ptp.110.433[_prog . * 110 * ( 2003 ) 433439 ] , http://arxiv.org/abs/gr-qc/0212030[arxiv:gr-qc/0212030 ] . y. fujii and k. maeda , `` the scalar - tensor theory of gravitation , '' . cambridge univ . press ( 2003 ) 240 p. a. zanzi , `` chameleonic equivalence postulate and wave function collapse , '' http://arxiv.org/abs/arxiv:1404.1942 [ quant - ph ] . submitted to electronic journal of theoretical physics[arxiv:1404.1942 [ quant - ph ] . submitted to electronic journal of theoretical physics ] . a. zanzi , `` chameleonic dilaton and conformal transformations , '' http://arxiv.org/abs/1206.4463[arxiv:1206.4463 [ hep - th ] ] . a. zanzi , `` dilaton stabilization and composite dark matter in the string frame of heterotic - m - theory , '' http://arxiv.org/abs/1210.4615[arxiv:1210.4615 [ hep - th ] ] . a. zanzi , `` neutrino dark energy and moduli stabilization in a bps braneworld scenario , '' http://dx.doi.org/10.1103/physrevd.73.124010[_phys . rev . _ * d73 * ( 2006 ) 124010 ] , http://arxiv.org/abs/hep-ph/0603026[arxiv:hep-ph/0603026 ] . a. l. fetter and j. d. walecka , `` quantum theory of many - particle systems , '' _ dover publications _ ( 1971 ) . a. sagnotti and m. tsulaia , `` on higher spins and the tensionless limit of string theory , '' http://dx.doi.org/10.1016/j.nuclphysb.2004.01.024[_nucl . _ * b682 * ( 2004 ) 83116 ] , http://arxiv.org/abs/hep-th/0311257[arxiv:hep-th/0311257 ] .

the precision test of the standard model ( sm ) at the lep gave a possibility not only to determine all the parameters and particle masses at the level of radiative corrections but also afforded an opportunity for searching for signals of new heavy particles beyond the energy scale of it . on the base of the lep2 experiments the low bounds on parameters of various models extending the sm have been estimated and the scale of new physics was obtained @xcite . although no new particles were discovered , a general believe is that the energy scale of new physics to be of order 1 tev , that may serve as a guide for experiments at the tevatron and lhc . in this situation , any information about new heavy particles obtained on the base of the present day data is desirable and important . numerous extended models include the @xmath0 gauge boson massive neutral vector particle associated with the extra @xmath1 subgroup of an underlying group . searching for this particle as a virtual state is widely discussed in the literature ( see for references @xcite ) . in the content of searching for @xmath0 at the lhc and the ilc an essential information and prospects for future investigations are given in lectures @xcite . such aspects as the mass of @xmath0 , couplings to the sm particles , @xmath2 mixing and its influence in various processes and particles parameters , distinctions between different models are discussed in details . we shall turn to these papers in what follows . as concerned a searching for @xmath0 in the lep experiments and the experiments at tevatron @xcite , it was carried out mainly in a model - dependent way . a wide class of popular models has been investigated and low bounds on the mass @xmath3 were estimated ( see , @xcite ) . as it is occurred , the low masses are varying in a wide energy interval 400 - 1800 gev dependently on a specific model . these bounds are a little bit different in the lep and tevatron experiments . in this situation a model - independent analysis is of interest . in the papers @xcite of the present authors a new approach for the model - independent search for @xmath0-boson was proposed which , in contrast to other model - independent searches , gives a possibility to pick out uniquely this virtual state and determine its characteristics . the corresponding observables have also been introduced and applied to analyze the lep2 experiment data . our consideration is based on two constituents : 1 ) the relations between the effective low - energy couplings derived from the renormalization group ( rg ) equation for fermion scattering amplitudes . we called them the rg relations . due to these relations , a number of unknown @xmath0 parameters entering the amplitudes of different scattering processes considerably decreases . 2 ) when these relations are accounted for , some kinematics properties of the amplitude become uniquely correlated with this virtual state and the @xmath0 signals exhibit themselves . the rg relations allow to introduce observables correlated uniquely with the @xmath0-boson . comparing the mean values of the observables with the necessary specific values , one could arrive at a conclusion about the @xmath0 existence . the confidence level ( cl ) of these values has been estimated and adduced in addition . without taking into consideration the rg relations the determination of @xmath0-boson requires a supplementary specification due to a larger number of different couplings contributing to the observables . similar situation takes place in the `` helicity model fits '' of lep collaborations @xcite when different virtual states contribute to each of the specific models ( aa , vv , and so on ) . therefore these fits had the goal to discover any signals of new physics independently of the particular states which may cause deviations from the sm . note that the lep collaborations saw no indications of new contact four fermion interactions in these fits . in refs . @xcite the one - parametric observables were introduced and the signals ( hints in fact ) of the @xmath0 have been determined at the 1@xmath4 cl in the @xmath5 process , and at the 2@xmath4 cl in the bhabha process . the @xmath0 mass was estimated to be 11.2 tev . an increase in statistics could make these signals more pronounced and there is a good chance to discover this particle at the lhc . in ref . @xcite the updated results of the one - parameter fit and the complete many - parametric fit of the lep2 data were performed with the goal to estimate a possible signal of the @xmath0-boson with accounting for the final data of the lep collaborations delphi and opal @xcite . usually , in a many - parametric fit the uncertainty of the result increases drastically because of extra parameters . on the contrary , in our approach due to the rg relations between the low - energy couplings there are only 2 - 3 independent parameters for the investigated leptonic scattering processes . as it was showed in ref . @xcite , an inevitable increase of confidence areas in the many - parametric space was compensated due to accounting for all accessible experimental information . therefore , the uncertainty of the many - parametric fit was estimated as the comparable with previous one - parametric fits in refs . @xcite . in this approach the combined data fit for all lepton processes is also possible . from the results obtained on the searching for abelian @xmath0 within the lep experiment data set we conclude that it is insufficient for convincing discovery of this particle as the virtual state . in this situation it is reasonable to analyze the data by using the neural network approach which is able to make a realistic prognoses for the parameters of interest . this investigation was done within the two parametric global fit of the lep2 data on the bhabha scattering process . as the result of all these considerations we derive at the 2@xmath4 cl the characteristics of the @xmath0 ( the vector @xmath6 and axial - vector @xmath7 couplings of the @xmath0 with sm leptons and the @xmath8 mixing ) . the @xmath0 mass is also estimated . due to the universality of the @xmath7 we also derived the model independent estimate of the @xmath0 axial - vector couplings to quarks , @xmath9 . note that the hints for the @xmath0 have been determined in all the processes considered that increases the reliability of the signal . these results may serve as a good input into the future lhc and ilc experiments and used in various aspects . to underline the importance of them we mention that there are many tools at the lhc for the identification of @xmath0 . but many of them are only applicable if @xmath0 is relatively light . the knowledge of the @xmath0 couplings to sm fermions also have important consequences . the paper is organized as follows . in sect . 2 we give a necessary information about the description of @xmath0 at low energies . in sects . 3 - 5 we discuss the origin of the rg relations , their explicit forms for the case of the heavy @xmath0 and consequences of the relations for scattering processes investigated . in sect . 6 the cross sections and the observables to pick out uniquely the virtual @xmath0 in the @xmath10 processes are given . the fits of data are described and discussed . then in sect . 7 the same is present for the bhabha process @xmath11 . the one parametric and two parametric fits are discussed . in sect . 8 the analysis of this process is carried out by using the neuron network approach . the criteria for training the network are introduced which guarantee the 2@xmath4 cl deviations of the data from the model containing the sm with extra @xmath12 . the obtained parameters of the @xmath0 practically coincide with that of derived in the one parameter analysis . in this way we determine the characteristics of the @xmath0 coming from the lep experiments . in sect . 9 we discuss the role of the present model - independent analysis for the lhc experiments . the discussion and comparison with results of other approaches are given in sect . 10 . in the appendix we describe the two - mass - scale yukawa model and analyze in detail how the decoupling of the loop contributions due to heavy virtual states is realized when the mixing of fields is taken into consideration . this point is an essential element of the approach developed . let us adduce a necessary information about the abelian @xmath0-boson . this particle is predicted by a number of grand unification models . among them the @xmath13 and @xmath14 based models @xcite ( for instance , lr , @xmath15 and so on ) are often discussed in the literature . in all the models , the abelian @xmath0-boson is described by a low - energy @xmath16 gauge subgroup originated in some symmetry breaking pattern . at low energies , the @xmath0-boson can manifest itself by means of the couplings to the sm fermions and scalars as a virtual intermediate state . moreover , the @xmath17-boson couplings are also modified due to a @xmath17@xmath0 mixing . in principle , arbitrary effective @xmath0 interactions to the sm fields could be considered at low energies . however , the couplings of non - renormalizable types have to be suppressed by heavy mass scales because of decoupling . therefore , significant signals beyond the sm can be inspired by the couplings of renormalizable types . such couplings can be derived by adding new @xmath16-terms to the electroweak covariant derivatives @xmath18 in the lagrangian @xcite ( review , @xcite ) @xmath19 where summation over all the sm left - handed fermion doublets , leptons and quarks , @xmath20 , and the right - handed singlets , @xmath21 , is understood . @xmath22 denotes the charge of @xmath23 in positron charge units , @xmath24 , and @xmath25 for leptons and 1/3 for quarks . for general purposes we derive the rg relations for the @xmath0 beyond the sm with two light higgs doublets ( thdm ) @xcite . @xmath0 interactions with the scalar doublets can be parametrized in a model - independent way as follows , @xmath26 in these formulas , @xmath27 are the charges associated with the @xmath28 and the @xmath0 gauge groups , respectively , @xmath29 are the pauli matrices , @xmath30 is the generator corresponding to the gauge group of the @xmath0 boson , and @xmath31 is the @xmath32 hypercharge . the yukawa lagrangian can be written in the form @xmath33 \right.\nonumber\\&&\left . + g_{f_{u , i } } [ \bar{f}_l \phi_i^c ( f_u)_r + ( \bar{f}_u)_r \phi^{c+}_i f_l]\right),\end{aligned}\ ] ] where @xmath34 is the charge conjugated scalar doublet . the lagrangian ( [ lscal ] ) leads to the @xmath17@xmath0 mixing . the @xmath17@xmath0 mixing angle @xmath35 is determined by the coupling @xmath36 as follows @xmath37 where @xmath38 is the sm weinberg angle , and @xmath39 is the electromagnetic fine structure constant . although the mixing angle is a small quantity of order @xmath40 , it contributes to the @xmath17-boson exchange amplitude and can not be neglected at the lep energies . in what follows we will also use the @xmath0 couplings to the vector and axial - vector fermion currents defined as @xmath41 the lagrangian ( [ lf ] ) leads to the following interactions between the fermions and the @xmath17 and @xmath0 mass eigenstates : @xmath42f , \nonumber\\ { \cal l}_{z'\bar{f}f}&=&\frac{1}{2 } z'_\mu\bar{f}\gamma^\mu\left [ ( v_f+\gamma^5 a_f)\cos\theta_0 -\right.\nonumber\\&&\quad\left . -(v^\mathrm{sm}_{fz}+\gamma^5 a^\mathrm{sm}_{fz})\sin\theta_0\right]f,\end{aligned}\ ] ] where @xmath23 is an arbitrary sm fermion state ; @xmath43 , @xmath44 are the sm couplings of the @xmath17-boson . since the @xmath0 couplings enter the cross - section together with the inverse @xmath0 mass , it is convenient to introduce the dimensionless couplings @xmath45 which can be constrained by experiments . low energy parameters @xmath46 , @xmath47 , @xmath48 , @xmath49 must be fitted in experiments . in most investigations they were considered as independent ones . in a particular model , the couplings @xmath50 , @xmath51 take some specific values . in case when the model is unknown , these parameters remain potentially arbitrary numbers . however , this is not the case if one assumes that the underlying extended model is a renormalizable one . in the papers @xcite it was shown that these parameters are correlated due to renormalizability . we called them the rg relations . since this notion is a key - point of our consideration , we discuss it in detail . what is rg relation ? generally speaking , this is a correlation between low energy parameters of interactions of a heavy new particle with known light particles of the sm following from the requirement that full unknown yet theory extending sm is to be renormalizable . strictly speaking , rg relations are the consequence of two constituencies : 1 . rg equation for a scattering amplitude ; 2 . decoupling theorem . the latter one describes the modification of both the rg operator @xmath52 and an amplitude at the energy threshold @xmath53 of new physics . here , @xmath54- and @xmath55-functions correspond to all the charges @xmath56 and fields and masses @xmath57 of the underlying theory . the rg equation for a scattering amplitude @xmath23 reads , @xmath58 where @xmath23 accounts for as intermediate states either the light or heavy virtual particles of the full theory . the standard usage of the rg equation is to improve the amplitude by solving this equation for the operator @xmath59 calculated in a given order of perturbation theory . however , to search for heavy virtual particles , we will use eq . ( [ rge ] ) in another way . first we note that for any renormalizable theory , the rg equation is just identity , if @xmath23 and @xmath59 are calculated in a given order of loop expansion . in this case ( [ rge ] ) expresses the well known fact that the structure of the divergent term coincides with the structure of the corresponding term in a tree - level lagrangian . for example , in massless qed , the tree - level plus one - loop one - particle - irreducible vertex function describing scattering of electron in an external electromagnetic field @xmath60 , @xmath61 , is if we calculate the rg operator in one - loop order @xmath62 where @xmath63 , @xmath64 are the beta - function and the anomalous dimensions of electromagnetic and electron fields , respectively , and apply it to @xmath65 , we obtain @xmath66 then , accounting for the values of @xmath67 and the factor @xmath68 in @xmath69 , we observe that the first and the last terms in the r.h.s . cancel . since @xmath70-dependent term in @xmath71 is @xmath72 , we see that eq.([rge1 ] ) is identity in the order @xmath73 . next important point is that in a theory with different mass scales the decoupling of heavy - loop contributions at the threshold of heavy masses , @xmath53 , results in the following property : the running of all functions is regulated by the loops of light particles . therefore , the @xmath74 and @xmath75 functions at low energies are determined by the sm particles , only . this fact is the consequence of the decoupling theorem @xcite . the decoupling results in the redefinition of parameters at the scale @xmath53 and removing heavy - particle loop contributions from rg equation @xcite : @xmath76 where @xmath77 and @xmath78 denote the parameters of the sm . they are calculated assuming that no heavy particles are excited inside loops . the matching between both sets of parameters @xmath77 , @xmath78 and @xmath56 , @xmath57 is chosen at the normalization point @xmath79 , @xmath80 the differential operator @xmath81 in the rg equation is in fact unique ; the apparently different @xmath81 in both theories are the same ! note that if a theory with different mass scales is specified one can freely replace the parameters @xmath77 , @xmath78 and @xmath56 , @xmath57 by each other @xcite . an example of the derivation and the main features of the rg relations are shown in the appendix for a simple model with different mass scales . if underlying theory is not specified , the set of @xmath56 , @xmath57 is unknown . the low energy theory consists of the sm plus the effective lagrangian generated by the interactions of light particles with virtual heavy particle states . the low energy parameters @xmath82 of these interactions are arbitrary numbers which must be constrained by experiments . by calculating the rg operator @xmath81 and the scattering amplitudes of light particles in this ` external field ' in a chosen order of loop expansion , it is possible to obtain the model - independent correlations between @xmath82 . these are just the re relations . let us derive the correlations between @xmath50 , @xmath83 appearing due to renormalizability of the underlying theory containing @xmath0 . in our case , the rg invariance of the vertex leads to the equation @xmath84 where @xmath85 and @xmath86 are computed with taking into account the loops of light particles . now we derive the rg relations following from the one - loop consideration . in accordance with the previous sections , the one - loop rg equation for the vertex function reads @xmath87 where @xmath88 and @xmath89 are the tree - level and one - loop contributions to the fermion-@xmath0 vertex . @xmath90 is the one - loop level part of the rg operator . vertex at the one - loop level . ] to calculate these functions , only the divergent parts of the one - loop vertices are to be calculated . the corresponding diagrams are shown in fig . [ f-2 ] . the fermion anomalous dimensions @xmath91 can be calculated by using the diagrams in fig . [ f-3 ] . then , eq.([rgez ] ) leads to algebraic equations for the parameters @xmath50 , @xmath92 and @xmath49 which have two sets of solutions @xcite : @xmath93 and @xmath94 here @xmath23 and @xmath95 are the partners of the @xmath96 fermion doublet ( @xmath97 and @xmath98 ) , @xmath99 is the third component of weak isospin . the first of these relations describes the @xmath0 boson analogous to the third component of the @xmath96 gauge field . the couplings to the right - handed singlet are absent . the second relation corresponds to the abelian @xmath0 . in this case the sm lagrangian appears to be invariant with respect to the @xmath16 group associated with the @xmath0 . the last relation in eq.([rgr2 ] ) ensures the @xmath100 eq.([lyukawa ] ) is to be invariant with respect to the @xmath16 transformations . introducing the @xmath0 couplings to the vector and axial - vector fermion currents ( [ av ] ) , the last line in eq . ( [ rgr2 ] ) yields @xmath101 the couplings of the abelian @xmath0 to the axial - vector fermion current have a universal absolute value proportional to the @xmath0 coupling to the scalar doublet . these relations are model independent . in particular , they hold in all the known models containing the abelian @xmath0 . the most discussed models are derived from the @xmath102 group ( the so called lr , @xmath103-@xmath104 models ) . the tree - level @xmath0 couplings to the sm fermions in the models are shown in table 1 . [ zpcoup ] .the @xmath105 couplings to the sm fermions in the most discussed @xmath102-based models . [ cols="<,^,^,^,^ " , ] the standard @xmath106-fit gives the following constraints for the coupling @xmath107 at the 68% cl : @xmath108 hence it is seen that the most precise data of delphi and opal collaborations give no signal of the chiral @xmath0 at the 1@xmath4 cl . the combined value also shows no signal at the 1@xmath4 cl . from the combined fit the 95% cl bound on the value of @xmath107 can be derived , @xmath109 . supposing the @xmath0 coupling constant @xmath110 to be of the order of the electroweak one , @xmath111 , the corresponding @xmath0 mass has to be larger than @xmath112 tev . a complete two parametric fit of experimental data based directly on the differential cross - sections has been carried out in ref . @xcite . due to only two independent couplings this fit is efficient . in the fitting the available final data for the differential cross - sections of the bhabha process were used . the data set consists of 299 bins including the data of aleph at 130 - 183 gev , delphi at 189 - 207 gev , l3 at 183 - 189 gev , and opal at 130 - 207 gev @xcite . the fitting procedure is similar to that of discussed above for the abelian @xmath0 . the results can be summarized as follows . the parameter space of the chiral @xmath0 is the plane ( @xmath113 , @xmath114 ) . the minimum of the @xmath106-function , @xmath115 , is reached at zero value of @xmath113 ( @xmath116 ) and almost independent of the value of @xmath114 ( the maximal - likelihood values of the couplings ) . the 95% cl area ( @xmath117 ) is shown in figure 3 . [ fig - ch - many ] as one can see , the zero point , @xmath118 ( the absence of the chiral @xmath0 boson ) is inside the confidence area . the value of @xmath106 in this point ( 238.62 ) is indistinguishable from the @xmath119 . in other words , the set of experimental data can not determine the signal of the chiral @xmath0-boson . as also is seen , the value of @xmath113 is constrained as @xmath120 at the 95% cl . this upper bound is in an agreement with the corresponding result of the one - parameter fit ( @xmath121 ) . thus , the @xmath0 mass has to be larger than @xmath122 tev , if the @xmath0 coupling constant @xmath110 is again supposed to be of the order of the electroweak one , @xmath111 . the fit of the differential cross - sections leads to a better accuracy for @xmath113 than the fit of the integrated cross - sections based on the same data . without accounting for the model - independent relations between the @xmath0 couplings it is impossible to obtain such results in this section we discuss all the assumptions giving a possibility to pick out the @xmath0 signal and determine its characteristics in a model independent way . we also note the role of the present results for the future lhc and ilc experiments . as it was already stressed , in searching for this particle at the lep and tevatron a model dependent analysis was mainly used . the motivation for this was the different number of chiral fermions involved in different models ( see , for example , @xcite ) . in this approach , the low bounds on @xmath3 have been estimated and a smallness of the @xmath2 mixing was also observed . on the contrary , in our model independent approach the rg relations between the parameters of the effective low energy lagrangians have been accounted for that gave a possibility to determine not only the bounds but also the mass and other parameters of the @xmath0 . first we note all the assumptions used in our considerations . we analyzed the four - fermion scattering amplitudes of order @xmath123 generated by the @xmath0 virtual states . the vertices linear in @xmath0 were included into the effective low - energy lagrangian . we also impose a number of natural conditions . the interactions of a renormalizable type are dominant at low energies @xmath124 . the non - renormalizable interactions generated at high energies due to radiation corrections are suppressed by the inverse heavy mass @xmath125 and neglected . we also assumed that the @xmath126 gauge group of the sm is a subgroup of the gut group . as a consequence , all structure constants connecting two sm gauge bosons with @xmath0 are to be zero . hence , the interactions of gauge fields of the types @xmath127 , and other are absent at the tree level . our effective lagrangian is also consistent with the absence of the tree - level flavor - changing neutral currents ( fcncs ) in the fermion sector . the renormalizable interactions of fermions and scalars are described by the yukawa lagrangian ( [ lyukawa ] ) that accounts for different possibilities of the yukawa sector without the tree - level fcncs . these assumptions are quite general and satisfied in a wide class of @xmath13 inspired models . within these constraints for the low energy effective lagrangian the rg correlations have been derived . correspondingly , the model independent estimates of the mass @xmath3 and other parameters are regulated by the noted requirements . therefore , the extended underlying model has also to accept them . in this regard , let us discuss the role of the obtained estimates for the lhc . as it is well known ( see , for example , @xcite ) , there are many tools at the lhc for @xmath0 identification . but many of them are only applicable if this particle is relatively light . our results are in favor to this case . next important point is the determination of @xmath0 couplings to the various sm fermions . as we have shown , the axial - vector couplings of the @xmath0 to the sm fermions are universal and proportional to its coupling to the higgs field . hence we have obtained an estimate of the @xmath128 couplings for both leptons and quarks . this is an essential input because experimental analyses for the lhc have mainly concentrated on being able to distinguish models and not on actual couplings . the vector coupling @xmath129 was also estimated that , in particular , may help to distinguish the decay of the @xmath0 resonance state to @xmath130 pairs . since the couplings @xmath131 and @xmath129 were estimated there is a possibility to distinguish this process from the decay of the @xmath132 system . in the literature @xcite on searching for the @xmath0 it is also mentioned that the determination of the @xmath0 couplings to fermions could be fulfilled channel by channel , @xmath133 , . in almost all these considerations the relations between the parameters have not been taken into account . but this is very essential for treating of experimental data and introducing relevant observables to measure . other parameter is the @xmath2 mixing , which is responsible for different decay processes and the effective interaction vertices generated at the lhc @xcite . it is also determined by the axial - vector coupling ( see eq . ( [ grgav ] ) ) and estimated in a model independent way . remind that in our analysis ( in contrast to the approaches of the lep collaborations ) the mixing was systematically taken into account . its value is of the same order of magnitude as the parameters that were fitted in experiments . note also that the existence of other heavy particles with masses @xmath134 does not influence the rg relations which are the consequences of the necessary condition for renormalizability . in fact , this condition ( the structure of a divergence generated by radiation corrections coincides with that of the tree - level vertex ) holds for each renormalizable type interaction . an important role of the model independent results for searching for @xmath0 at the lhc and ilc consists , in particular , in possibility to determine the particle as a virtual state due to a large amount of relevant events . we mentioned already that , in principle , lep experiments were able to determine it if the statistics was sufficiently large . experiments at the ilc will increase numerously the data set of interest . in fact , the observables , introduced in sects . 6 and 7 for picking out uniquely @xmath135 and @xmath136 couplings in the leptonic scattering process , are also effective at energies @xmath137 gev and could be applied in future experiments at ilc . other model independent methods of searching for the @xmath0 as a resonance state are proposed in the literature ( see refs . we do not discuss them here because they take no relations between the parameters into consideration . besides , the main goal of the present paper is to adduce model independent information about the @xmath0 followed from experiments at low energies . different aspects of @xmath0 physics at the lhc are out of the scope of it . in this section we collect in a convenient form all the results obtained and make a comparison with other investigations on searching for @xmath0 at low energies . in fact , this is a large area to discuss . references to numerous results obtained in either model dependent or model independent approaches can be found in the surveys @xcite . further subdivision can be done into the considerations accounting for any type correlations between the parameters of the low energy effective interactions and that of assuming complete independence of them . because of a large amount of fitting parameters the latter are less predictable . now , for a convenience of readers let us present the results of fits of the @xmath0 parameters in terms of the popular notations @xcite . the lagrangian reads @xmath138f , \nonumber\\ { \cal l}_{z'\bar{f}f}&=&\frac{1}{2 } z'_\mu\bar{f}\gamma^\mu\left [ ( v'_f-\gamma^5 a'_f)\right]f,\end{aligned}\ ] ] with the sm values of the @xmath17 couplings @xmath139 where @xmath68 is the positron charge , @xmath22 is the fermion charge in the units of @xmath68 , @xmath140 for the neutrinos and @xmath141-type quarks , and @xmath142 for the charged leptons and @xmath143-type quarks . @xmath144 & @xmath145 , @xmath146 & @xmath147 , @xmath146 & @xmath148 , @xmath144 + + @xmath149 & @xmath150 & - & @xmath151 & 0.437 + + @xmath149 & - & @xmath152 & - & - + @xmath153 & @xmath154 & - & @xmath155 & 1.278 + @xmath156 & @xmath157 & - & @xmath158 & 0.464 + + @xmath149 , @xmath159 & - & @xmath160 & - & - + @xmath144 & @xmath145 , @xmath146 & @xmath147 , @xmath146 & @xmath148 , @xmath144 + + @xmath149 & 68% & @xmath161 & - & @xmath162 & @xmath163 + + @xmath149 & 95% & - & @xmath164 & - & - + @xmath153 & 95% & @xmath165 & - & @xmath166 & @xmath167 + @xmath153 , & & & & & + @xmath168 & 95% & @xmath169 & - & @xmath170 & @xmath171 + + @xmath149 , & & & & & + @xmath153 , & 95% & @xmath172 & @xmath173 & @xmath174 & @xmath175 + @xmath168 & & & & & + @xmath149 , & & & & & + @xmath159 & 68% & @xmath176 & @xmath177 & @xmath178 & @xmath179 + the results of the fits of the @xmath0 couplings to the sm leptons obtained from the analysis of the lep experiments are adduced in the tables [ fitmlv]-[fitcli ] and fig . [ fig : fitfigs ] . remind that due to the universality of the axial - vector coupling @xmath180 the same estimates take also place for quarks . first of all , one parameter fits of lep experiments as well as the many - parameter fit for the @xmath181 backward bins show the hints of the @xmath0 boson at the 1 - 2@xmath4 cl . due to this fact , the fits allow to determine the maximum likelihood values of @xmath0 parameters . in spite of uncertainties , these values can be used as a guiding line for the estimation of possible @xmath0 effects in the lhc experiments . the maximum likelihood values are given in table [ fitmlv ] . as is seen , different fits and processes lead to the comparable values of the @xmath0 parameters . in table [ fitcli ] we present the confidence intervals for the fitted parameters . with this table one is able to estimate the uncertainty of the @xmath0 couplings as well as the lower bounds on the parameters . + + in fig . [ fig : fitfigs ] the maximum likelihood values and the cl intervals are drawn for the different values of the @xmath0 mass . all the plots exploit the same color scheme . the values excluded at 95% cl by the many - parameter fit of all the lep2 leptonic processes @xmath182 are shown in gray . the 95% confidence intervals from the one - parameter fit of lep2 @xmath183 are drawn in pink with the maximum likelihood values as the dashed red line . the corresponding results with taking into account the @xmath184 process only are shown in yellow with solid red line for the maximum likelihood values . the maximum likelihood values from the lep1 experiments are represented as dotted blue line . the 95% confidence interval from the one - parameter fit of the lep2 bhabha scattering is shown as the blue crosshatched region with the maximum likelihood values as the dashed blue line . finally , the 68% confidence interval and the maximum likelihood values from the many - parameter fit of the backward bins of lep2 bhabha scattering are shown in green . now we compare the above results with the ones of other fits accounting for the @xmath0 presence . as it was mentioned in introduction , lep collaborations have determined the model dependent low bounds on the @xmath0 mass which vary in a wide energy interval dependently on a model . the same has also been done for tevatron experiments . the modern low bound is @xmath185 gev . it is also well known that though almost all the present day data are described by the sm @xcite , the overall fit to the standard model is not very good . in ref . @xcite it was showed that the large difference in @xmath186 from the forward - backward asymmetry @xmath187 of the bottom quarks and the measurements from the slac sld experiment can be explained for physically reasonable higgs boson mass if one allows for one or more extra @xmath1 fields , that is @xmath0 . a specific model to describe @xmath0 physics of interest was proposed which introduces two type couplings to the hyper charge @xmath188 and to the baryon - minus - lepton number @xmath189 . within this model by using a number of precision measurements from lep1 , lep2 , sld and tevatron experiments the parameters @xmath190 and @xmath191 of the model were fitted . the presence of @xmath0 was not excluded at 68% cl . the value of @xmath190 was estimated to be of the same order of magnitude as in our analysis and is comparable with values of other parameters detected in the lep experiments . the erroneous claim that @xmath190 is two order less then the value derived from our table 4 is , probably , a consequence of some missed factors . the upper limit on the mass was also obtained @xmath192 tev . these two analyzes are different but complementary . a common feature of them is an accounting for the @xmath0 gauge boson as a necessary element of the data fits . the results are in accordance at 68 - 95% cl with the existence of the not heavy @xmath0 which has a good chance to be discovered at tevatron and/or lhc . in this appendix we are going to investigate the yukawa model with a heavy scalar field @xmath103 and a light scalar field @xmath193 @xcite . the goals of our investigation are two fold : 1 ) to derive the one - loop rg relation for the four - fermion scattering amplitude in the decoupling region and 2 ) to find out the possibility of reducing this relation in the equation for vertex describing the scattering of light particles on the external field when the mixing between heavy and light virtual states takes place . consider the four - fermion scattering @xmath196 . the @xmath197-matrix element at the one - loop level is given by @xmath198,\nonumber \\ s_\mathrm{1pr}&=&\sum\limits_{{\phi}_{1},{\phi}_{2}=\varphi , \chi}g_{{\phi}_{1 } } g_{{\phi}_{2}}\left ( \frac{{\delta}_{{\phi}_{1}{\phi}_{2}}}{s - m^2_{{\phi}_{1 } } } + \frac{\pi_{\phi_1\phi_2}(s)}{(s- m^2_{\phi_1})(s- m^2_{\phi_2})}\right ) \nonumber\\ & & \times { \bar\psi}\left(p_{4}\right)\left[1+\gamma\left ( p_{3 } , -p_{4}- p_{3}\right ) \right ] \psi\left(p_{3}\right)\nonumber \\ & & \times { \bar\psi}\left ( p_{1}\right ) \left[1+\gamma\left ( p_{2 } , -p_{1}- p_{2}\right ) \right ] \psi\left ( p_{2}\right),\end{aligned}\ ] ] where @xmath199 , @xmath200 is the contribution from the one - particle reducible diagrams shown in figs . [ fig : tree]-[fig : loop ] , and @xmath201 is the contribution from box diagrams . the one - loop polarization operator of scalar fields @xmath202 and the one - loop vertex function @xmath65 are related to the green functions as @xmath203 where @xmath204 is the spinor propagator in the momentum representation . using the dimensional regularization ( the dimension of the momentum space is @xmath206 ) and the @xmath207 renormalization scheme one can compute the renormalization constants @xmath208 , \nonumber\\ \delta\lambda^2&=&\frac{1}{4\pi^2\varepsilon } \left[\left(g_\chi^2 + 6\xi\right)\lambda^2 - 6g_\chi^2m^2-\rho m^2\right ] , \nonumber\\ z_g^{-1}&=&\left[1-\frac{3}{16\pi^2\varepsilon}\left(g_\varphi^2+g_\chi^2\right)\right ] { \left(z_\phi^{1/2}\right)}^\mathrm{t}.\end{aligned}\ ] ] from eq . ( [ app5 ] ) we obtain the appropriate @xmath74 and @xmath75 functions at the one - loop level : @xmath209 then , the @xmath197-matrix element can be expressed in terms of the renormalized quantities . the contribution from the one - particle reducible diagrams becomes @xmath210\psi(p_3 ) \nonumber\\&&\times \bar\psi(p_1)\left[1+\gamma^\mathrm{(fin)}\left(p_2,-p_1-p_2\right)\right]\psi(p_2),\end{aligned}\ ] ] where the functions @xmath211 and @xmath212 are the expressions @xmath213 and @xmath65 without the terms proportional to @xmath214 . since the quantity @xmath201 is finite , the renormalization leaves it without changes . introducing the rg operator at the one - loop level @xcite @xmath215 we determine that the following relation holds for the @xmath197-matrix element @xmath216 where the @xmath217 and the @xmath218 are the contributions to the @xmath200 at the tree level and at the one - loop level , respectively : @xmath219 @xmath220.\end{aligned}\ ] ] the first term in eq . ( [ app11 ] ) is originated from the one - loop correction to the fermion - scalar vertex . the rest terms are connected with the polarization operator of scalars . the third term describes the one - loop mixing between the scalar fields . it is canceled in the rg relation ( [ app9 ] ) by the mass - dependent terms in the @xmath74 functions produced by the non - diagonal elements in @xmath221 . ( [ app9 ] ) is the consequence of the renormalizability of the model . it insures the leading logarithm terms of the one - loop @xmath197-matrix element to reproduce the appropriate tree - level structure . in contrast to the familiar treatment we are not going to improve scattering amplitudes by solving eq . ( [ app9 ] ) . we will use it as an algebraic identity implemented in the renormalizable theory . naturally if one knows the explicit couplings expressed in terms of the basic set of parameters of the model , this rg relation is trivially fulfilled . but the situation changes when the couplings are represented by unknown arbitrary parameters . in this case the rg relations are the algebraic equations dependent on these parameters and appropriate @xmath74 and @xmath75 functions . in the presence of a symmetry the number of @xmath74 and @xmath75 functions is less than the number of rg relations . so , one has non trivial system of equations relating the unknown couplings . for example , such a scenario is realized for the gauge coupling . although the considered simple model has no gauge couplings , we are able to demonstrate the general procedure of deriving the rg relations . at energies @xmath222 the heavy scalar field @xmath103 is decoupled . this means , that the four - fermion scattering amplitude is described by the model with no heavy field @xmath103 plus terms of the order @xmath223 . at the tree level , this is the obvious consequence of the expansion of the heavy scalar propagator @xmath224,\ ] ] which is resulted in the effective contact four - fermion interaction in eq . ( [ app10 ] ) @xmath225 so , the tree level contribution to the scattering amplitude becomes @xmath226 \bar\psi\psi\times\bar\psi\psi,\ ] ] and the lowest order effects of the heavy scalar in the decoupling region are described by the parameter @xmath227 , only . decoupling of heavy particles is present also at the level of radiative corrections . the radiative corrections are generally described by various loop integrals in the momentum space ( the passarino veltman functions ) . considering a passarino veltman function with at least one heavy mass @xmath228 inside loop in the low - energy limit , one can see the following asymptotic behavior : the function splits into 1 ) possible energy - independent divergent part ( including also @xmath229 ) and 2 ) energy - dependent finite part which can be expanded by inverse powers of @xmath53 and vanishes at the small energies . the important property is that the @xmath229-term in the divergent part reproduces the logarithm of the cut - off scale . so , such a potentially large term has to be automatically absorbed by the renormalization at low energies and leads to no observable effects . however , if the renormalization is actually performed at high energies ( as in the @xmath207 renormalization scheme ) the potentially large @xmath229-terms should be re - summed manually by the redefinition of the physical couplings and masses at the scale @xmath53 . what is the form of the rg relations in the limit of large @xmath53 ? the method of constructing the rg equation in the decoupling region was proposed in @xcite . it introduces the redefinition of the parameters of the model allowing to remove all the heavy particle loop contributions to eq . ( [ app11 ] ) . let us define a new set of fields , charges and masses @xmath230 , @xmath231 , @xmath232 , @xmath233 , @xmath234 , @xmath235 @xmath236 where dots stand for the higher powers of @xmath229 responsible for the decoupling at higher loop orders . the differential operator ( [ app8 ] ) ban be rewritten in terms of these new low - energy parameters : @xmath237 where @xmath238 and @xmath239 functions are obtained from the one - loop relations ( [ app6 ] ) and ( [ app15 ] ) @xmath240 hence , one immediately notices that @xmath238 and @xmath239 functions contain only the light particle loop contributions , and all the heavy particle loop terms are completely removed from them . the @xmath197-matrix element expressed in terms of new parameters satisfies the following rg relation @xmath241 @xmath242 @xmath243 where @xmath244 is the redefined effective four - fermion coupling . as one can see , eq . ( [ app20 ] ) includes all the terms of eq . ( [ app11 ] ) except for the heavy particle loop contributions . it depends on the low energy quantities @xmath230 , @xmath231 , @xmath245 , @xmath246 , @xmath234 , @xmath235 . the first and the second terms in eq . ( [ app20 ] ) are just the one - loop amplitude calculated within the model with no heavy particles . the third and the fourth terms describe the light particle loop correction to the effective four - fermion coupling and the mixing of heavy and light virtual fields . due to the mixing term it is impossible to split the rg relation ( [ app18 ] ) for the @xmath197-matrix element into the one for vertices . hence , we are not able to consider eq . ( [ app18 ] ) in the framework of the scattering of light particles on an external field induced by the heavy virtual scalar . but this is an important step in deriving the rg relation for el parameters . fortunately , there is a simple procedure allowing to avoid the mixing in eq . ( [ app20 ] ) . the way is to incorporate the diagonalization of the leading logarithm terms of the scalar polarization operator into the redefinition of the @xmath247 , @xmath248 , @xmath231 , @xmath232 : @xmath249 \left(\zeta^{-1/2}\right)^\mathrm{t } \left(\begin{array}{c}\tilde{g}_\varphi \\ \tilde{g}_\chi \end{array}\right ) , \\ \nonumber\zeta^{1/2}&= & 1-\frac{\tilde{g}_\varphi\tilde{g}_\chi}{8\pi^2\left(\tilde\lambda^2-\tilde{m}^2\right ) } \log\frac{\kappa^2}{\tilde\lambda^2 } % \nonumber\\&&\times \left(\begin{array}{cc}0 & \tilde\lambda^2 - 6\tilde{m}^2\\-\tilde{m}^2 - 6\tilde{m}^2 & 0\end{array}\right).\end{aligned}\ ] ] the appropriate @xmath238 functions @xmath250 contain no terms connected with mixing between light and heavy scalars . so , the fourth term in eq . ( [ app20 ] ) is removed , and the rg relation for the @xmath197-matrix element becomes at @xmath254 eq . ( [ app23 ] ) is just the rg identity for the scattering amplitude calculated in the absence of the heavy particles . the terms of order @xmath245 describe the rg relation for the effective low - energy four - fermion interaction in the decoupling region . the last one can be reduced in the rg relation for the vertex describing the scattering of the light particle ( fermion ) on the external field @xmath255 substituting the virtual heavy scalar : @xmath256 where @xmath257 eqs . ( [ app23])-([app27 ] ) is the main result of our investigation . one can derive them with only the knowledge about the low - energy couplings of heavy particle ( [ app13 ] ) and the lagrangian of the model with no heavy particles . one also has to ignore all the heavy particle loop contributions to the rg relation and the one - loop mixing between the heavy and the light fields . ( [ app23])-([app27 ] ) depend on the effective low - energy parameters , only . but as the difference between the original set of parameters and the low - energy one is of one - loop order , one may freely substitute them in eqs . ( [ app23])-([app26 ] ) . it is also possible to reduce the rg relation for scattering amplitudes to the one for vertex describing the scattering of light particles on the ` external ' field induced by the heavy virtual particle . in fact , this result is independent on the specific features of the considered model . 99 the lep collaborations aleph , delphi , l3 , opal , and the lep electroweak working group , hep - ex/0612034 . g. abbiendi _ et al . _ [ opal collaboration ] , eur . j. * c33 * ( 2004 ) 173 [ hep - ex/0309053 ] ; eur . j. * c6 * ( 1999 ) 1 ; k. ackerstaff _ et al . _ [ opal collaboration ] , eur . phys . j. * c2 * ( 1998 ) 441 . j. abdallah _ et al . _ [ delphi collaboration ] , eur . phys . j. * c45 * ( 2006 ) 589 [ hep - ex/0512012 ] . a. leike , phys . rep . * 317 * ( 1999 ) 143 . p. langacker , arxiv:0801.1345 [ hep - ph ] . t. rizzo , hep - ph/0610104 . a. ferroglia , a. lorca , and j.j . van der bij , annalen phys * 16 * ( 2007 ) 563 - 578 [ hep - ph/0611174 ] . a. gulov and v. skalozub , eur . j. * c17 * ( 2000 ) 685 . a. gulov and v. skalozub , phys . rev . * d61 * ( 2000 ) 055007 . v. demchik , a. gulov , v. skalozub , and a. tishchenko , phys . * 67 * ( 2004 ) 1312 [ yad . fiz . * 67 * ( 2004 ) 1335 ] . a. gulov and v. skalozub , phys . rev . * d70 * ( 2004 ) 115010 . a. gulov and v. skalozub , phys . * d76 * ( 2007 ) 075008 . j.hewett and t.rizzo , phys . rep . * 183 * ( 1989 ) 193 . m. cvetic and b.w . lynn , phys . rev . * d35 * ( 1987 ) 51 . g. degrassi and a. sirlin , phys . rev . * d40 * ( 1989 ) 3066 . t. appelquist and j. carazzone , phys . rev . * d11 * ( 1975 ) 2856 . j. collins , f. wilczek , and a. zee , phys . rev . * d18 * ( 1978 ) 242 . m. bando , t. kugo , n. maekawa , and h. nakano , progress of theor . * 90 * ( 1993 ) 405 . t. rizzo , phys . * d55 * ( 1997 ) 5483 . r. barate _ [ aleph collaboration ] , eur . j. * c12 * ( 2000 ) 183 [ hep - ex/9904011 ] . m. acciarri _ et al . _ [ l3 collaboration ] , phys . * b479 * ( 2000 ) 101 [ hep - ex/0002034 ] . a. babich , g. della ricca , j. holt , p. osland , a. pankov , and n. paver , eur . j. * c29 * ( 2003 ) 103 . w. eadie , d. dryard , f. james , m. roos , and b. sadoulet , _ statistical methods in experimental physics _ , amsterdam , north - holland , 1971 . aleph collaboration , delphi collaboration , l3 collaboration , opal collaboration , sld collaboration , lep electroweak working group , sld electroweak group , and sld heavy flavour group , phys . * 427 * ( 2006 ) 257 [ hep - ex/0509008 ] . v. abazov _ [ d0 collaboration ] , phys . * b517 * ( 2001 ) 282 . a. buryk and v. skalozub , arxiv:0802.1486 [ hep - ph ] . schwind / mlpfit.html a. gulov and v. skalozub , phys . * 70 * ( 2007 ) 1100 - 1106 [ hep - ph/0510354 ] . m. dittmar , a .- s . nicollerat , and a. djouadi , phys . b583 * ( 2004 ) 111 - 120 [ hep - ph/0307020 ] . c. coriano , a. faraggi , and m. guzzi , phys . * d78 * ( 2008 ) 015012 [ arxiv:0802.1792 ] . f. petriello and s. quackenbush , phys . rev . * d77 * ( 2008 ) 115004 [ arxiv:0801.4389 ] . a. gulov and v. skalozub , phys . atom . nucl . * 63 * ( 2000 ) 139 - 143 [ yad . fiz . * 63 * ( 2000 ) 152 - 157 ] .

access to positive , negative and unlabeled examples is a standard assumption for most semi - supervised binary classification techniques . in many domains , however , a sample from one of the classes ( say , negatives ) may not be available , leading to the setting of learning from positive and unlabeled data @xcite . positive - unlabeled learning often emerges in sciences and commerce where an observation of a positive example ( say , that a protein catalyzes reactions or that a social network user likes a particular product ) is usually reliable . here , however , the absence of a positive observation can not be interpreted as a negative example . in molecular biology , for example , an attempt to label a data point as positive ( say , that a protein is an enzyme ) may be unsuccessful for a variety of experimental and biological reasons , whereas in social networks an explicit dislike of a product may not be possible . both scenarios lead to a situation where negative examples can not be actively collected . fortunately , the absence of negatively labeled examples can be tackled by incorporating unlabeled examples as negatives , leading to the development of non - traditional classifiers . here we follow the terminology by @xcite that a traditional classifier predicts whether an example is positive or negative , whereas a non - traditional classifier predicts whether the example is positive or unlabeled . positive vs. unlabeled ( non - traditional ) training is reasonable because the class posterior and also the optimum scoring function for composite losses @xcite in the traditional setting is monotonically related to the posterior in the non - traditional setting . however , the true posterior can be fully recovered from the non - traditional posterior only if we know the class prior ; i.e. , the proportion of positives in unlabeled data . the knowledge of the class prior is also necessary for estimation of the performance criteria such as the error rate , balanced error rate or f - measure , and also for finding the right threshold for the non - traditional scoring function that leads to an optimal classifier with respect to some criteria @xcite . class prior estimation in a nonparametric setting has been actively researched in the past decade offering an extensive theory of identifiability @xcite and a few practical solutions @xcite . application of these algorithms to real data , however , is limited in that none of the proposed algorithms simultaneously deals with noise in the labels and practical estimation for high - dimensional data . much of the theory on learning class priors relies on the assumption that either the distribution of positives is known or that the positive sample is clean . in practice , however , labeled data sets contain class - label noise , where an unspecified amount of negative examples contaminates the positive sample . this is a realistic scenario in experimental sciences where technological advances enabled generation of high - throughput data at a cost of occasional errors . one example for this comes from the studies of proteins using analytical chemistry technology ; i.e. , mass spectrometry . for example , in the process of peptide identification @xcite , bioinformatics methods are usually set to report results with specified false discovery rate thresholds ( e.g. , 1% ) . unfortunately , statistical assumptions in these experiments are sometimes violated thereby leading to substantial noise in reported results , as in the case of identifying protein post - translational modifications . similar amounts of noise might appear in social networks such as facebook , where some users select ` like ' , even when they do not actually like a particular post . further , the only approach that does consider similar such noise @xcite requires density estimation , which is known to be problematic for high - dimensional data . in this work , we propose the first classification algorithm , with class prior estimation , designed particularly for high - dimensional data with noise in the labeling of positive data . we first formalize the problem of class prior estimation from noisy positive and unlabeled data . we extend the existing identifiability theory for class prior estimation from positive - unlabeled data to this noise setting . we then show that we can practically estimate class priors and the posterior distributions by first transforming the input space to a univariate space , where density estimation is reliable . we prove that these transformations preserve class priors and show that they correspond to training a non - traditional classifier . we derive a parametric algorithm and a nonparametric algorithm to learn the class priors . finally , we carry out experiments on synthetic and real - life data and provide evidence that the new approaches are sound and effective . consider a binary classification problem of mapping an input space @xmath0 to an output space @xmath1 . let @xmath2 be the true distribution of inputs . it can be represented as the following mixture @xmath3 where @xmath4 , @xmath5 , @xmath6 are distributions over @xmath0 for the positive ( @xmath7 ) and negative ( @xmath8 ) class , respectively ; and @xmath9 is the class prior or the proportion of the positive examples in @xmath2 . we will refer to a sample from @xmath2 as unlabeled data . let @xmath10 be the distribution of inputs for the labeled data . because the labeled sample contains some mislabeled examples , the corresponding distribution is also a mixture of @xmath11 and a small proportion , say @xmath12 , of @xmath13 . that is , @xmath14 where @xmath15 $ ] . observe that both mixtures have the same components but different mixing proportions . the simplest scenario is that the mixing components @xmath13 and @xmath11 correspond to the class - conditional distributions @xmath16 and @xmath17 , respectively . however , our approach also permits transformations of the input space @xmath0 , thus resulting in a more general setup . the objective of this work is to study the estimation of the class prior @xmath18 and propose practical algorithms for estimating @xmath19 . the efficacy of this estimation is clearly tied to @xmath20 , where as @xmath20 gets smaller , the noise in the positive labels becomes larger . we will discuss identifiability of @xmath19 and @xmath20 and give a practical algorithm for estimating @xmath19 ( and @xmath20 ) . we will then use these results to estimate the posterior distribution of the class variable , @xmath21 , despite the fact that the labeled set does not contain any negative examples . the class prior is identifiable if there is a unique class prior for a given pair @xmath22 . much of the identifiability characterization in this section has already been considered as the case of asymmetric noise @xcite ; see section [ sec_related ] on related work . we recreate these results here , with the aim to introduce required notation , to highlight several important results for later algorithm development and to include a few missing results needed for our approach . though the proof techniques are themselves quite different and could be of interest , we include them in the appendix due to space . there are typically two aspects to address with identifiability . first , one needs to determine if a problem is identifiable , and , second , if it is not , propose a canonical form that is identifiable . in this section we will see that class prior is not identifiable in general because @xmath23 can be a mixture containing @xmath24 and vice versa . to ensure identifiability , it is necessary to choose a canonical form that prefers a class prior that makes the two components as different as possible ; this canonical form was introduced as the mutual irreducibility condition @xcite and is related to the proper novelty distribution @xcite and the max - canonical form @xcite . we discuss identifiability in terms of measures . let @xmath25 , @xmath26 , @xmath27 and @xmath28 be probability measures defined on some @xmath29-algebra @xmath30 on @xmath0 , corresponding to @xmath2 , @xmath10 , @xmath13 and @xmath11 , respectively . it follows that @xmath31 consider a family of pairs of mixtures having the same components @xmath32 where @xmath33 is some set of pairs of probability measures defined on @xmath30 . the family is parametrized by the quadruple @xmath34 . the condition @xmath35 means that @xmath26 has a greater proportion of @xmath28 compared to @xmath25 . this is consistent with our assumption that the labeled sample mainly contains positives . the most general choice for @xmath33 is @xmath36 where @xmath37 is the set of all probability measures defined on @xmath30 and @xmath38 is the set of pairs with equal distributions . removing equal pairs prevents @xmath25 and @xmath26 from being identical . we now define the maximum proportion of a component @xmath39 in a mixture @xmath40 , which is used in the results below and to specify the criterion that enables identifiability ; more specifically , @xmath41:\lambda=\alpha\lambda_1+(1-\alpha)\lambda_0 , \lambda_0 \in { { \ensuremath \mathcal{p}}^{\text{all}}}}.\ ] ] of particular interest is the case when @xmath42 , which should be read as @xmath40 is not a mixture containing @xmath39 " . we finally define the set all possible @xmath43 that generate @xmath25 and @xmath26 when @xmath44 varies in @xmath33 : @xmath45 if @xmath46 is a singleton set for all @xmath47 , then @xmath48 is identifiable in @xmath43 . first , we show that the most general choice for @xmath33 , @xmath49 , leads to unidentifiability ( ) . fortunately , however , by choosing a restricted set @xmath50 as @xmath33 , we do obtain identifiability ( ) . in words , @xmath51 contains pairs of distributions , where each distribution in a pair can not be expressed as a mixture containing the other . the proofs of the results below are in the appendix . lemmaunidentifiability [ lem : fpiall ] given a pair of mixtures @xmath52 , let parameters @xmath34 generate @xmath53 and @xmath54 . it follows that 1 . [ st : oneone ] there is a one - to - one relation between @xmath44 and @xmath43 and @xmath55 2 . [ st : validmeasure ] both expressions on the right - hand side of equation [ eq : mu01 ] are well defined probability measures if and only if @xmath56 and @xmath57 . [ st : atwoall ] @xmath58 . [ st : uniden1 ] @xmath59 is unidentifiable in @xmath43 ; i.e. , @xmath43 is not uniquely determined from @xmath53 . [ st : uniden2 ] @xmath59 is unidentifiable in @xmath19 and @xmath20 , individually ; i.e. , neither @xmath19 nor @xmath20 is uniquely determined from @xmath53 . observe that the definition of @xmath60 and @xmath61 imply @xmath62 and , consequently , any @xmath63 satisfies @xmath64 , as expected . theoremidentifiability [ thm : fpires ] given @xmath52 , let @xmath65 and @xmath66 . let @xmath67 , @xmath68 and @xmath69 it follows that 1 . [ st : generate ] @xmath70 generate @xmath53 2 . [ st : frompires ] @xmath71 and , consequently , @xmath72 , @xmath73 . [ st : samesets ] @xmath74 contains all pairs of mixtures in @xmath59 . [ st : atwores ] @xmath75 . [ st : iden]@xmath74 is identifiable in @xmath43 ; i.e. , @xmath43 is uniquely determined from @xmath53 . we refer to the expressions of @xmath25 and @xmath26 as mixtures of components @xmath27 and @xmath28 as a max - canonical form when @xmath44 is picked from @xmath51 . this form enforces that @xmath28 is not a mixture containing @xmath27 and vice versa , which leads to @xmath27 and @xmath28 having maximum separation , while still generating @xmath25 and @xmath26 . each pair of distributions in @xmath74 is represented in this form . identifiability of @xmath74 in @xmath43 occurs precisely when @xmath75 , i.e. , @xmath76 is the only pair of mixing proportions that can appear in a max - canonical form of @xmath25 and @xmath26 . moreover , statement [ st : generate ] in and statement [ st : oneone ] in imply that the max - canonical form is unique and completely specified by @xmath70 , with @xmath77 following from equation [ eq : correction ] . thus , using @xmath74 to model the unlabeled and labeled data distributions makes estimation of not only @xmath19 , the class prior , but also @xmath78 a well - posed problem . moreover , due to statement [ st : samesets ] in , there is no loss in the modeling capability by using @xmath74 instead of @xmath59 . overall , identifiability , absence of loss of modeling capability and maximum separation between @xmath27 and @xmath28 combine to justify estimating @xmath79 as the class prior . taking values from @xmath80 . in the first step , the sample is randomly selected to attempt labeling , with some probability independent of @xmath81 or @xmath82 . if it is not selected , it is added to the unlabeled " set . if it is selected , then labeling is attempted . if the true label is @xmath83 , then with probability @xmath84 , the labeling will succeed and it is added to noisy positives " . otherwise , it is added to the dropped " set . if the true label is @xmath85 , then the attempted labeling is much more likely to fail , but because of noise , could succeed . the attempted label of @xmath86 succeeds with probability @xmath87 , and is added to noisy positives " , even though it is actually a negative instance . @xmath88 leads to the no noise case and the noise increases as @xmath87 increases . @xmath89 , gives the proportion of positives in the noisy positives " . ] the theory and algorithms for class prior estimation are agnostic to the dimensionality of the data ; in practice , however , this dimensionality can have important consequences . parametric gaussian mixture models trained via expectation - maximization ( em ) are known to strongly suffer from co - linearity in high - dimensional data . nonparametric ( kernel ) density estimation is also known to have curse - of - dimensionality issues , both in theory @xcite and in practice @xcite . we address the curse of dimensionality by transforming the data to a single dimension . the transformation @xmath90 , surprisingly , is simply an output of a non - traditional classifier trained to separate labeled sample , @xmath91 , from unlabeled sample , @xmath92 . the transform is similar to that in @xcite , except that it is not required to be calibrated like a posterior distribution ; as shown below , a good ranking function is sufficient . first , however , we introduce notation and formalize the data generation steps ( figure [ fig_labeling ] ) . let @xmath81 be a random variable taking values in @xmath0 , capturing the true distribution of inputs , @xmath25 , and @xmath82 be an unobserved random variable taking values in @xmath93 , giving the true class of the inputs . it follows that @xmath94 and @xmath95 are distributed according to @xmath27 and @xmath28 , respectively . let @xmath96 be a selection random variable , whose value in @xmath80 determines the sample to which an input @xmath97 is added ( figure [ fig_labeling ] ) . when @xmath98 , @xmath97 is added to the noisy labeled sample ; when @xmath99 , @xmath97 is added to the unlabeled sample ; and when @xmath100 , @xmath97 is not added to either of the samples . it follows that @xmath101 and @xmath102 are distributed according to @xmath25 and @xmath26 , respectively . we make the following assumptions which are consistent with the statements above : @xmath103 assumptions [ eq : sy0 ] and [ eq : sy1 ] states that the proportion of positives in the unlabeled sample and the labeled sample matches the true proportion in @xmath25 and @xmath26 , respectively . assumption [ eq : sxy ] states that the distribution of the positive inputs ( and the negative inputs ) in both the unlabeled and the labeled samples is equal and unbiased . gives the implications of these assumptions . statement [ st : seq2 ] in is particularly interesting and perhaps counter - intuitive as it states that with non - zero probability some inputs need to be dropped . lemmaassumptions [ lem : assumptions ] let @xmath81 , @xmath82 and @xmath96 be random variables taking values in @xmath0 , @xmath93 and @xmath104 , respectively , and @xmath101 and @xmath102 . for measures @xmath105 , satisfying equations [ eq : muu ] and [ eq : mul ] and @xmath106 , let @xmath107 give the distribution of @xmath81 , @xmath94 and @xmath95 , respectively . if @xmath108 and @xmath96 satisfy assumptions [ eq : sy0 ] , [ eq : sy1 ] and [ eq : sxy ] , then 1 . [ st : sx0 ] @xmath81 is independent of @xmath109 ; i.e. , @xmath110 2 . [ st : xuxl ] @xmath111 and @xmath112 are distributed according to @xmath25 and @xmath26 , respectively . st : seq2 ] @xmath113 . the proof is in the appendix . next , we highlight the conditions under which the score function @xmath114 preserves @xmath79 . observing that @xmath96 serves as the pseudo class label for labeled vs. unlabeled classification as well , we first give an expression for the posterior : @xmath115 theoremtransform [ thm : transform ] let random variables @xmath116 and measures @xmath105 be as defined in . let @xmath117 be the posterior as defined in and @xmath118 , where @xmath119 is a 1-to-1 function on @xmath120 $ ] and @xmath121 is the composition operator . assume 1 . @xmath122 , 2 . @xmath111 and @xmath112 are continuous with densities @xmath2 and @xmath10 , respectively , 3 . @xmath123 are the measures corresponding to @xmath124 , respectively , 4 . @xmath125 and @xmath126 . then @xmath127 and so @xmath114 is an @xmath79-preserving transformation . moreover , @xmath117 can also be used to compute the true posterior probability : @xmath128 the proof is in the appendix . shows that the @xmath79 is the same for the original data and the transformed data , if the transformation function @xmath114 can be expressed as a composition of @xmath117 and a one - to - one function , @xmath119 , defined on @xmath120 $ ] . trivially , @xmath117 itself is one such function . we emphasize , however , that @xmath79-preservation is not limited by the efficacy of the calibration algorithm ; uncalibrated scoring that ranks inputs as @xmath129 also preserves @xmath79 . further demonstrates how the true posterior , @xmath130 , can be recovered from @xmath117 by plugging in estimates of @xmath117 , @xmath131 , @xmath79 and @xmath132 in . the posterior probability @xmath117 can be estimated directly by using a probabilistic classifier or by calibrating a classifier s score @xcite ; @xmath133 serves as an estimate of @xmath131 ; gives parametric and nonparametric approaches for estimation of @xmath79 and @xmath132 . in this section , we derive a parametric and a nonparametric algorithm to estimate @xmath79 and @xmath132 from the unlabeled sample , @xmath134 , and the noisy positive sample , @xmath135 . in theory , both approaches can handle multivariate samples ; in practice , however , to circumvent the curse of dimensionality , we exploit the theory of @xmath79-preserving univariate transforms to transform the samples . * parametric approach . * the parametric approach is derived by modeling each sample as a two component gaussian mixture , sharing the same components but having different mixing proportions : @xmath136 where @xmath137 and @xmath138 , the set of all @xmath139 positive definite matrices . the algorithm is an extension to the em approach for gaussian mixture models ( gmms ) where , instead of estimating the parameters of a single mixture , the parameters of both mixtures @xmath140 are estimated simultaneously by maximizing the combined likelihood over both @xmath141 and @xmath142 . this approach , which we refer to as a multi - sample gmm ( msgmm ) , exploits the constraint that the two mixtures share the same components . the update rules and their derivation are given in the appendix . * nonparametric approach . * our nonparametric strategy directly exploits the results of and , which give a direct connection between @xmath143 ) and @xmath144 ) . therefore , for a two - component mixture sample , @xmath145 , and a sample from one of the components , @xmath146 , it only requires an algorithm to estimate the maximum proportion of @xmath147 in @xmath148 . for this purpose , we use the alphamax algorithm @xcite , briefly summarized in the appendix . specifically , our two - step approach for estimating @xmath79 and @xmath132 is as follows : ( @xmath149 ) estimate @xmath150 and @xmath151 as outputs of alphamax@xmath152 and alphamax@xmath153 , respectively ; ( @xmath154 ) estimate @xmath76 from the estimates of @xmath155 by applying . we refer to our nonparametric algorithm as alphamax - n . in this section we systematically evaluate the new algorithms in a controlled , synthetic setting as well as on a variety of data sets from the uci machine learning repository @xcite . * experiments on synthetic data : * we start by evaluating all algorithms in a univariate setting where both mixing proportions , @xmath156 and @xmath157 , are known . we generate unit - variance gaussian and unit - scale laplace - distributed i.i.d . samples and explore the impact of mixing proportions , the size of the component sample , and the separation and overlap between the mixing components on the accuracy of estimation . the class prior @xmath156 was varied from @xmath158 and the noise component @xmath157 from @xmath159 . the size of the labeled sample @xmath91 was varied from @xmath160 , whereas the size of the unlabeled sample @xmath92 was fixed at @xmath161 . * experiments on real - life data : * we considered twelve real - life data sets from the uci machine learning repository . to adjust these data to our problems , categorical features were transformed into numerical using sparse binary representation , the regression data sets were transformed into classification based on mean of the target variable , and the multi - class classification problems were converted into binary problems by combining classes . in each data set , a subset of positive and negative examples was randomly selected to provide a labeled sample while the remaining data ( without class labels ) were used as unlabeled data . the size of the labeled sample was kept at @xmath162 ( or @xmath163 for small data sets ) and the maximum size of unlabeled data was set @xmath161 . * algorithms : * we compare the alphamax - n and msgmm algorithms to the elkan - noto algorithm @xcite as well as the noiseless version of alphamax @xcite . there are several versions of the elkan - noto estimator and each can use any underlying classifier . we used the @xmath164 alternative estimator combined with the ensembles of 100 two - layer feed - forward neural networks , each with five hidden units . the out - of - bag scores of the same classifier were used as a class - prior preserving transformation that created an input to the alphamax algorithms . it is important to mention that neither elkan - noto nor alphamax algorithm was developed to handle noisy labeled data . in addition , the theory behind the elkan - noto estimator restricts its use to class - conditional distributions with non - overlapping supports . the algorithm by @xcite minimizes the same objective as the @xmath164 elkan - noto estimator and , thus , was not implemented . * evaluation : * all experiments were repeated 50 times to be able to draw conclusions with statistical significance . in real - life data , the labeled sample was created randomly by choosing an appropriate number of positive and negative examples to satisfy the condition for @xmath157 and the size of the labeled sample , while the remaining data was used as the unlabeled sample . therefore , the class prior in the unlabeled data varies with the selection of the noise parameter @xmath157 . the mean absolute difference between the true and estimated class priors was used as a performance measure . the best performing algorithm on each data set was determined by multiple hypothesis testing using the p - value of @xmath165 and bonferroni correction . * results : * the comprehensive results for synthetic data drawn from univariate gaussian and laplace distributions are shown in appendix ( table 2 ) . in these experiments no transformation was applied prior to running any of the algorithms . as expected , the results show excellent performance of the msgmm model on the gaussian data . these results significantly degrade on laplace - distributed data , suggesting sensitivity to the underlying assumptions . on the other hand , alphamax - n was accurate over all data sets and also robust to noise . these results suggest that new parametric and nonparametric algorithms perform well in these controlled settings . table 1 shows the results on twelve real data sets . here , alphamax and alphamax - n algorithms demonstrate significant robustness to noise , although the parametric version msgmm was competitive in some cases . on the other hand , the elkan - noto algorithm expectedly degrades with noise . finally , we investigated the practical usefulness of the @xmath166-preserving transform . ( appendix ) shows the results of alphamax - n and msgmm on the real data sets , with and without using the transform . because of computational and numerical issues , we reduced the dimensionality by using principal component analysis ( the original data caused matrix singularity issues for msgmm and density estimation issues for alphamax - n ) . msgmm deteriorates significantly without the transform , whereas alphamax - n preserves some signal for the class prior . alphamax - n with the transform , however , shows superior performance on most data sets . 1.01 |>x| > x| > x| > x| > x| > x| > x| > x| > x| > x| > x| & @xmath19 & @xmath20 & auc & @xmath167 & @xmath168 & @xmath169 & & & & + & 0.095 & 1.00 & 0.842 & 13 & 5188 & 45000 & 0.241 & 0.070 & * 0.037 * * & 0.163 + & 0.096 & 0.95 & 0.819 & 13 & 5188 & 45000 & 0.284 & 0.079 & * 0.036 * * & 0.155 + & 0.101 & 0.75 & 0.744 & 13 & 5188 & 45000 & 0.443 & 0.124 & * 0.040 * * & 0.127 + & 0.419 & 1.00 & 0.685 & 8 & 490 & 1030 & 0.329 & 0.141 & 0.181 & * 0.077 * * + & 0.425 & 0.95 & 0.662 & 8 & 490 & 1030 & 0.363 & 0.174 & 0.231 & * 0.095 * * + & 0.446 & 0.75 & 0.567 & 8 & 490 & 1030 & 0.531 & * 0.212 * & 0.272 & 0.233 + & 0.342 & 1.00 & 0.825 & 127 & 2565 & 5574 & 0.017 & 0.011 & 0.017 & * 0.008 * * + & 0.353 & 0.95 & 0.795 & 127 & 2565 & 5574 & 0.078 & 0.016 & * 0.006 * & 0.006 + & 0.397 & 0.75 & 0.672 & 127 & 2565 & 5574 & 0.396 & 0.137 & 0.009 & * 0.006 * * + & 0.268 & 1.00 & 0.810 & 13 & 209 & 506 & 0.159 & * 0.087 * & 0.094 & 0.209 + & 0.281 & 0.95 & 0.777 & 13 & 209 & 506 & 0.226 & * 0.094 * & 0.110 & 0.204 + & 0.330 & 0.75 & 0.651 & 13 & 209 & 506 & 0.501 & * 0.125 * & 0.134 & 0.172 + & 0.093 & 1.00 & 0.933 & 36 & 1508 & 6435 & 0.074 & 0.009 & * 0.007 * * & 0.157 + & 0.103 & 0.95 & 0.904 & 36 & 1508 & 6435 & 0.110 & 0.015 & * 0.008 * * & 0.152 + & 0.139 & 0.75 & 0.788 & 36 & 1508 & 6435 & 0.302 & 0.063 & * 0.012 * * & 0.143 + & 0.409 & 1.00 & 0.792 & 126 & 3916 & 8124 & 0.029 & * 0.015 * * & 0.022 & 0.037 + & 0.416 & 0.95 & 0.766 & 126 & 3916 & 8124 & 0.087 & 0.015 & * 0.008 * * & 0.037 + & 0.444 & 0.75 & 0.648 & 126 & 3916 & 8124 & 0.370 & 0.140 & * 0.020 * & 0.024 + & 0.086 & 1.00 & 0.885 & 10 & 560 & 5473 & 0.116 & * 0.026 * * & 0.044 & 0.129 + & 0.087 & 0.95 & 0.858 & 10 & 560 & 5473 & 0.137 & * 0.031 * * & 0.052 & 0.125 + & 0.090 & 0.75 & 0.768 & 10 & 560 & 5473 & 0.256 & * 0.041 * * & 0.064 & 0.111 + & 0.243 & 1.00 & 0.875 & 16 & 3430 & 10992 & 0.030 & * 0.006 * * & 0.009 & 0.081 + & 0.248 & 0.95 & 0.847 & 16 & 3430 & 10992 & 0.071 & 0.011 & * 0.005 * * & 0.074 + & 0.268 & 0.75 & 0.738 & 16 & 3430 & 10992 & 0.281 & 0.093 & * 0.007 * * & 0.062 + & 0.251 & 1.00 & 0.735 & 8 & 268 & 768 & 0.351 & 0.120 & * 0.111 * & 0.171 + & 0.259 & 0.95 & 0.710 & 8 & 268 & 768 & 0.408 & 0.118 & * 0.110 * & 0.168 + & 0.289 & 0.75 & 0.623 & 8 & 268 & 768 & 0.586 & * 0.144 * & 0.156 & 0.175 + & 0.139 & 1.00 & 0.929 & 9 & 8903 & 58000 & * 0.024 * * & 0.027 & 0.029 & 0.157 + & 0.140 & 0.95 & 0.903 & 9 & 8903 & 58000 & 0.052 & * 0.004 * * & 0.007 & 0.157 + & 0.143 & 0.75 & 0.802 & 9 & 8903 & 58000 & 0.199 & 0.047 & * 0.004 * * & 0.148 + & 0.226 & 1.00 & 0.842 & 57 & 1813 & 4601 & 0.184 & 0.046 & * 0.041 * & 0.059 + & 0.240 & 0.95 & 0.812 & 57 & 1813 & 4601 & 0.246 & 0.059 & * 0.042 * * & 0.063 + & 0.295 & 0.75 & 0.695 & 57 & 1813 & 4601 & 0.515 & 0.155 & * 0.044 * * & 0.059 + & 0.566 & 1.00 & 0.626 & 11 & 4113 & 6497 & 0.290 & 0.083 & * 0.060 * & 0.070 + & 0.575 & 0.95 & 0.610 & 11 & 4113 & 6497 & 0.322 & 0.113 & * 0.063 * & 0.076 + & 0.612 & 0.75 & 0.531 & 11 & 4113 & 6497 & 0.420 & 0.322 & 0.353 & * 0.293 * + class prior estimation in a semi - supervised setting , including positive - unlabeled learning , has been extensively discussed previously ; see @xcite and references therein . recently , a general setting for label noise has also been introduced , called the mutual contamination model . the aim under this model is to estimate multiple unknown base distributions , using multiple random samples that are composed of different convex combinations of those base distributions @xcite . the setting of asymmetric label noise is a subset of this more general setting , treated under general conditions by @xcite , and previously investigated under a more restrictive setting as co - training @xcite . a natural approach is to use robust estimation to learn in the presence of class noise ; this strategy , however , has been shown to be ineffective , both theoretically @xcite and empirically @xcite , indicating the need to explicitly model the noise . generative mixture model approaches have also been developed , which explicitly model the noise @xcite ; these algorithms , however , assume labeled data for each class . as the most related work , though @xcite did not explicitly treat the positive - unlabeled learning with noisy positives , their formulation can incorporate this setting by using @xmath170 and @xmath171 . the theoretical and algorithmic treatment , however , is very different . their focus is on identifiability and analyzing convergence rates and statistical properties , assuming access to some @xmath172 function which can obtain proportions between samples . they do not explicitly address issues with high - dimensional data nor focus on algorithms to obtain @xmath172 . in contrast , we focus primarily on the univariate transformation to handle high - dimensional data and practical algorithms for estimating @xmath173 . supervised learning used for class prior - preserving transformation provides a rich set of techniques to address high - dimensional data . in this paper , we developed a practical algorithm for classification of positive - unlabeled data with noise in the labeled data set . in particular , we focused on a strategy for high - dimensional data , providing a univariate transform that reduces the dimension of the data , preserves the class prior so that estimation in this reduced space remains valid and is then further useful for classification . this approach provides a simple algorithm that simultaneously improves estimation of the class prior and provides a resulting classifier . we derived a parametric and a nonparametric version of the algorithm and then evaluated its performance on a wide variety of learning scenarios and data sets . to the best of our knowledge , this algorithm represents one of the first practical and easy - to - use approaches to learning with high - dimensional positive - unlabeled data with noise in the labels . we thank prof . michael w. trosset for helpful comments . grant support : nsf dbi-1458477 , nih r01mh105524 , nih r01gm103725 , and the indiana university precision health initiative . * * we will need the following lemma for the proofs . [ thm : altchart ] let @xmath25 , @xmath28 and @xmath27 be three measures defined on @xmath30 such that @xmath174 for some @xmath175 $ ] . define @xmath176 it follows that 1 . [ st : altchart ] @xmath177 2 . [ st : resmu0 ] if @xmath178 , then @xmath179 . the proof follows from lemma 4 and theorem 3 of @xcite . as a consequence of statement 1 in , we use an alternate characterization of @xmath60 @xmath180 where @xmath181 . [ [ statement - stoneone ] ] statement [ st : oneone ] : + + + + + + + + + + + + + + + + + + + + + + because @xmath34 generate @xmath25 and @xmath26 , @xmath182 and equations [ eq : muu ] and [ eq : mul ] hold . @xmath27 can be expressed in terms of @xmath183 by eliminating @xmath28 ( @xmath184 ( equation [ eq : muu ] ) @xmath185 ( equation [ eq : mul ] ) ) . similarly @xmath28 can be expressed in terms of @xmath183 and equation [ eq : mu01 ] follows . thus , given @xmath25 and @xmath26 , @xmath186 can be uniquely determined from @xmath187 . equation [ eq : muu ] implies that for all @xmath188 such that @xmath189 , @xmath190 . existence of @xmath191 is guaranteed because @xmath106 from the definition of @xmath49 . thus , given @xmath25 , @xmath19 is uniquely determined from @xmath192 . similarly , equation [ eq : mul ] implies that , given @xmath26 , @xmath20 is uniquely determined from @xmath192 . * statement [ st : validmeasure ] : * it is easy to observe that @xmath193 is a valid probability measure , if and only if @xmath194 for all @xmath188 . because the inequality is trivially true when @xmath195 , the necessary and sufficient condition can be reduced to @xmath194 for all @xmath188 with @xmath196 . this can be rewritten as @xmath197 for all @xmath188 such that @xmath198 . equivalently , @xmath199 is a lower bound to @xmath200 ; in other words , @xmath56 due to the alternate characterization given by equation [ eq : altchar ] . similarly , @xmath201 is a valid probability measure provided @xmath57 . * statement [ st : atwoall ] : * because system of equations in [ eq : mu01 ] is essentially equivalent to equations [ eq : muu ] and [ eq : mul ] , @xmath63 if and only if , @xmath27 and @xmath28 as defined in equations [ eq : mu01 ] come from @xmath49 . when the conditions of statement [ st : validmeasure ] are not satisfied , @xmath27 , @xmath28 are not well - defined probability measures and consequently @xmath202 . on the other hand , when they are satisfied , @xmath27 , @xmath28 are well - defined probability measures . moreover , @xmath61 implies @xmath203 and consequently @xmath204 . * statement [ st : uniden2 ] : * we construct mixtures @xmath205 and @xmath206 such that there are multiple values for @xmath19 that generate the mixtures . for @xmath207 and probability measures @xmath208 on @xmath30 , let @xmath209 generate mixtures @xmath205 and @xmath206 . thus , @xmath210 . it follows that @xmath211 and @xmath212 are both strictly greater than @xmath213 because @xmath214 implies @xmath215 and @xmath216 implies @xmath217 ( using statement [ st : atwoall ] ) , which contradicts our assumption . now , @xmath218 follows trivially from statement [ st : atwoall ] . it is easy to observe that @xmath219 for any @xmath220 $ ] . thus , there are multiple values for @xmath19 that generate @xmath205 and @xmath206 . similarly , @xmath221 for any @xmath222 $ ] and there are multiple values for @xmath20 that generate @xmath205 and @xmath206 . * statement [ st : uniden1 ] : * statement [ st : uniden1 ] follows trivially from statement [ st : uniden2 ] and also by observing that @xmath223 can not be a singleton set because @xmath224 . [ [ statement - stgenerate ] ] statement [ st : generate ] : + + + + + + + + + + + + + + + + + + + + + + + + first , we show that @xmath225 are well defined probability measures . + @xmath226 : suppose @xmath227 . it follows that @xmath228 and , consequently , from , @xmath229 . however , @xmath203 because they are picked from @xmath49 . thus , @xmath230 by contradiction . similarly @xmath231 . thus , the denominator in the r.h.s . of @xmath232 and @xmath233 is not @xmath213 . + by definition @xmath234 . thus , @xmath235 when @xmath198 . consequently , @xmath236 . the inequality is trivially true when @xmath195 . thus , @xmath236 for all @xmath188 . hence , @xmath232 is a probability measure . similarly , @xmath233 is also a probability measure . + second , we show that @xmath70 generate @xmath237 . + @xmath238 : observe that @xmath225 can also be expressed as @xmath239 and @xmath240 . moreover , after some algebraic manipulation of equations labeled [ eq : correction ] , @xmath241 and @xmath242 can be derived . thus , from ( statements [ st : oneone ] and [ st : atwoall ] ) , @xmath70 generate @xmath237 . * statement [ st : frompires ] : * we show that @xmath243 and @xmath244 , giving @xmath71 . + we start by showing that @xmath245 . suppose , for some @xmath246 and @xmath247 thus , @xmath248 is a lower bound to @xmath249 and , consequently , @xmath250 . however , because @xmath251 , @xmath252 . this is a contradiction . thus , for all @xmath246 there exists some @xmath253 such that @xmath254 . we now divide the inequality on both sides by @xmath255 , observing that the divisor is strictly greater than @xmath213 because @xmath256 as @xmath233 is a probability measure . thus , @xmath257 because the choice of @xmath258 is arbitrary and @xmath231 , any lower bound of @xmath259 can not be greater than @xmath213 . thus , @xmath260 . a similar argument shows that @xmath261 therefore , @xmath243 and @xmath244 ; consequently , @xmath71 . because @xmath232 is not a mixture containing @xmath233 and because @xmath262 , from statement [ st : resmu0 ] in we get that @xmath72 . similarly , @xmath73 . * statement [ st : samesets ] : * because statements [ st : generate ] and [ st : frompires ] are true for any @xmath52 , statement [ st : samesets ] is true . * statement [ st : atwores ] : * it follows from statements [ st : generate ] and [ st : frompires ] , that @xmath263 . to show that @xmath264 contains no other element , we give a proof by contradiction . suppose @xmath265 and @xmath266 . let @xmath267 generate @xmath53 , for some @xmath268 . first , we show that @xmath269 and @xmath270 : + @xmath271 : suppose for some @xmath272 and all @xmath273 where @xmath198 , @xmath274 however , this is a contradiction because @xmath122 . thus , for every @xmath275 there exists some @xmath276 with @xmath277 such that @xmath278 . thus , @xmath279 , because @xmath258 can be made arbitrarily small , @xmath280 . however , because @xmath281 , @xmath43 also belongs to @xmath282 and consequently @xmath283 from ( statement [ st : atwoall ] ) . thus , @xmath284 . + @xmath285 : the proof is similar to @xmath284 supposing @xmath286 for some @xmath287 and all @xmath273 with @xmath288 , reaching a contradiction and following the subsequent steps . + @xmath269 , @xmath270 and implies @xmath289 and @xmath290 , which contradicts our assumption . hence , @xmath76 is the only element in @xmath264 , which proves statement [ st : atwores ] . * statement [ st : iden ] : * this statement follows by observing that @xmath264 is a singleton set . we first prove , which we can then use to construct a suitable univariate transform . [ [ statement - stsx0 ] ] statement [ st : sx0 ] : + + + + + + + + + + + + + + + + + + + observe that @xmath291 thus , @xmath81 is independent of @xmath109 . [ [ statement - stxuxl ] ] statement [ st : xuxl ] : + + + + + + + + + + + + + + + + + + + + from statement [ st : sx0 ] , @xmath111 has the same distribution as @xmath81 , which is @xmath25 . now , @xmath292 thus , the distribution of @xmath293 is @xmath294 , which is @xmath26 . [ [ statement-3 ] ] statement 3 : + + + + + + + + + + + + now , @xmath295 the probability @xmath296 is independent of @xmath97 only if @xmath297 is a constant with respect to @xmath97 . let @xmath298 , where @xmath299 is some constant . integrating over @xmath97 on both sides gives @xmath300 . since both integrals are @xmath301 , it follows that @xmath302 . thus , @xmath303 , which implies @xmath228 ; i.e. , the labeled and unlabeled samples have the same distribution . however , this implies @xmath304 , which contradicts the assumption . therefore , @xmath305 is not independent of @xmath81 . to prove the main result about the preservation properties of the univariate transform , we will make use of the following theorem . [ thm : transformpdf ] let @xmath81 and @xmath306 be random variables with densities @xmath2 and @xmath11 and measures @xmath25 and @xmath28 respectively . for @xmath307 and an abstract space @xmath308 , given any one - to - one function @xmath309 , define function @xmath310 @xmath311 where @xmath312 let @xmath313 and @xmath314 be the measures for the random variables @xmath315 , @xmath316 respectively for @xmath29-algebra @xmath317 on @xmath308 . then @xmath318 . first we prove that @xmath319 . to this end , we expand @xmath129 as follows @xmath320 the last step is justified because @xmath321 and @xmath322 . consider one - to - one functions @xmath323 and @xmath324 defined on @xmath325 . we can apply to @xmath326 to get @xmath327 and to @xmath328 to get @xmath329 . to satisfy the condition of for @xmath326 , let @xmath111 , @xmath112 and @xmath330 play the role of @xmath81 , @xmath306 and @xmath331 , respectively . now , @xmath332 because @xmath119 and @xmath326 are both one - to - one functions , @xmath330 is one - to - one as well . thus , all the conditions of are satisfied and , consequently , @xmath327 . similarly , we can use with @xmath111 , @xmath112 and @xmath333 playing the role of @xmath306 , @xmath81 and @xmath331 , respectively , now giving that @xmath329 . from statement [ st : frompires ] in and @xmath334 and , thus , @xmath335 . next we prove . let @xmath336 . rearranging the terms of , @xmath337 rearranging the terms , @xmath338 let @xmath134 and @xmath135 be the unlabeled sample and the noisy positive sample , respectively . the parametric approach is derived by modeling each sample as a two component gaussian mixture , sharing the same components but having different mixing proportions : @xmath136 where @xmath137 and @xmath138 , the set of all @xmath139 positive definite matrices . the algorithm is an extension to the em approach for gaussian mixture models ( gmms ) where , instead of estimating the parameters of a single mixture , the parameters of both mixtures @xmath140 are estimated simultaneously by maximizing the combined likelihood over both @xmath141 and @xmath142 . this approach , that we refer to as a multi - sample gmm ( msgmm ) , exploits the constraint that the two mixtures share the same components . + to derive the update equations , we introduce missing variables @xmath339 that give the true class of the @xmath340 and @xmath341 example in @xmath92 and @xmath91 , respectively . the variables @xmath339 are bernoulli distributed ; i.e. , @xmath342 and @xmath343 . for @xmath344 the quartet @xmath345 forms the observed and unobserved variables in the em framework . the complete data log - likelihood , @xmath346 is given by , @xmath347 where @xmath348 is the density of @xmath349 . our goal is to maximize @xmath350 $ ] . to do so we take the conditional expectation of @xmath346 with respect to @xmath351 and @xmath352 given @xmath92 and @xmath91 . for @xmath353=\frac{{\alpha}{\phi_1}({x^{u}}_i)}{{\alpha}{\phi_1}({x^{u}}_i)+(1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\alpha}){\phi_0}({x^{u}}_i)},\\ { \bar{w}^l}_i = e[{w^{l}}_i|{{x^{l}}_i={x^{l}}_i}]=\frac{{\beta}{\phi_1}({x^{l}}_i)}{{\beta}{\phi_1}({x^{l}}_i)+(1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\beta}){\phi_0}({x^{l}}_i)},\end{aligned}\ ] ] we obtain @xmath354 & = \sum_{i=1}^{|{u}| } { \bar{w}^u}_i \log\brac*{{\alpha}{\phi_1}({x^{u}}_i ) } + ( 1-{\bar{w}^u}_i ) \log\brac*{(1-{\alpha}){\phi_0}({x^{u}}_i ) } \\ & \hspace{1em } + \sum_{i=1}^{|{l}| } { \bar{w}^l}_i \log\brac*{{\beta}{\phi_1}({x^{l}}_i ) } + ( 1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\bar{w}^l}_i ) \log\brac*{(1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\beta}){\phi_0}({x^{l}}_i)}\end{aligned}\ ] ] which up to constants , that are ignored in the optimization , can be explicitly written as @xmath355 = & \sum_{i=1}^{|{u}| } { \bar{w}^u}_i \brac * { \log{\alpha}{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}\frac{1}{2}\log |{\sigma_1}| { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}({x^{u}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_1})^{t}{\sigma_1}^{-1}({x^{u}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_1 } ) } \\ & + \sum_{i=1}^{|{u}| } ( 1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\bar{w}^u}_i ) \brac * { \log(1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\alpha } ) { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}\frac{1}{2}\log |{\sigma_0}| { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}({x^{u}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_0})^{t}{\sigma_0}^{-1}({x^{u}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_0 } ) } \\ & + \sum_{i=1}^{|{l}| } { \bar{w}^l}_i \brac * { \log{\beta}{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}\frac{1}{2}\log |{\sigma_1}| { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}({x^{l}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_1})^{t}{\sigma_1}^{-1}({x^{l}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_1 } ) } \\ & + \sum_{i=1}^{|{l}| } ( 1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\bar{w}^l}_i ) \brac * { \log(1{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{\beta } ) { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}\frac{1}{2}\log |{\sigma_0}| { \raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}({x^{l}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_0})^{t}{\sigma_0}^{-1}({x^{l}}_i{\raisebox{0.05em}{$\scalebox{0.8}{$-$}$}}{u_0})}.\end{aligned}\ ] ] finally , we obtain the parameter update equations by maximizing @xmath350 $ ] with respect to @xmath356 : @xmath357 the update rules reduce to the standard gmm when the labeled sample is not provided . further generalization to more than two samples and/or mixing components is straightforward . for a mixture sample @xmath148 and a component sample @xmath147 , alphamax@xmath358 estimates the maximum proportion of @xmath147 in @xmath148 @xcite . alphamax is based on the constrained maximization of the log likelihood of samples @xmath145 and @xmath146 , derived using nonparametric estimates of their densities @xmath359 and @xmath360 , respectively . we list the main steps of alphamax below . 1 . estimate @xmath360 nonparameterically as @xmath361 using sample @xmath146 . obtain the weights , @xmath362 , and components @xmath363 from nonparametric density estimation of @xmath359 as a @xmath364-component mixture , @xmath365 , using @xmath145 . 2 . construct two density functions @xmath366 and @xmath367 from @xmath368 and @xmath361 parameterized by a @xmath364-dimensional weight vector @xmath369 $ ] , @xmath370 , which re - weights components @xmath363 : @xmath371 3 . maximize the log likelihood of @xmath145 and @xmath146 constructed with @xmath372 and @xmath373 under the constraint @xmath374 for many values of @xmath375 equispaced in @xmath120 $ ] . @xmath376 4 . estimate the maximum proportion of @xmath360 in @xmath359 , @xmath377 ( minor abuse of notation is a minor abuse of notation ; replacing the densities with the underlying measures makes the notation correct . ] ) , as the @xmath97-coordinate of the elbow in the @xmath378 versus @xmath375 graph . the densities @xmath367 and @xmath366 are constructed to approximate @xmath359 and @xmath360 . the efficacy of the approximation depends on the value of @xmath379 ; there exists @xmath379 such that @xmath367 and @xmath366 are good approximations provided @xmath380 , however , the approximation deteriorates progressively , even with the optimum @xmath379 , as @xmath381 moves beyond @xmath377 . this suggests that the graph of @xmath378 versus @xmath375 should be approximately a flat line from @xmath213 to @xmath377 and decrease progressively beyond @xmath377 exposing an elbow at @xmath377 , which is detected in the last step . the pseudo code for elbow detection is provided in @xcite . the practical implementation , backed by the @xmath166-preservation theory , reduces the dimension of the data to a single dimension by using the scoring function of a non - traditional classifier and employs histograms as the nonparametric method to obtain @xmath382 and @xmath361 . the bin - width is chosen to cover the component sample s ( after the transformation ) range and reveal the shape of its distribution , using the default option in matlab s ` histogram ` function . more bins with the same bin - width are subsequently added to cover the mixture sample s range . the total computation includes the time to ( a ) train a classifier ( b ) perform density estimation in 1d and ( c ) perform optimization in alphamax . the complexity for ( a ) varies ; for neural networks it is @xmath383 per epoch . for ( b ) we used @xmath384 bin histograms , where @xmath384 can be @xmath385 to @xmath386 depending on the bandwidth selection rule , giving @xmath387 complexity . for ( c ) , ( size @xmath384 optimization ) the computation of the objective and gradient is @xmath387 per step ; e.g. , lbfgs is @xmath387 per step . an execution of alphamax takes about 10 minutes on a laptop computer for the shuttle data set ( for results reported in table 1 ; i.e. , 1000 labeled and 10000 unlabeled examples ) ; 12 hours on the entire data ( 8903 labeled and 49097 unlabeled examples ) . [ [ results - for - univariate - synthetic - data . ] ] results for univariate synthetic data . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + 1.01 |>x| > x| > x| > x| > x| > x| > x| > x| > x| > x| > x| & & & & & & + & @xmath156 & @xmath157 & 100 & 1000 & 100 & 1000 & 100 & 1000 & 100 & 1000 + & 0.05 & 1.00 & 0.473 & 0.460 & 0.079 & 0.102 & 0.064 & 0.085 & * 0.034 * * & * 0.020 * * + & 0.25 & 1.00 & 0.484 & 0.435 & 0.132 & 0.160 & 0.123 & 0.122 & * 0.063 * * & * 0.044 * * + & 0.50 & 1.00 & 0.395 & 0.347 & 0.155 & 0.125 & 0.173 & 0.096 & * 0.073 * * & * 0.040 * * + & 0.05 & 0.95 & 0.484 & 0.496 & 0.099 & 0.124 & 0.076 & 0.099 & * 0.039 * * & * 0.022 * * + & 0.25 & 0.95 & 0.510 & 0.469 & 0.127 & 0.167 & 0.115 & 0.111 & * 0.074 * * & * 0.037 * * + & 0.50 & 0.95 & 0.433 & 0.378 & 0.180 & 0.152 & 0.186 & 0.087 & * 0.068 * * & * 0.048 * * + & 0.05 & 0.75 & 0.663 & 0.630 & 0.124 & 0.135 & 0.089 & 0.076 & * 0.056 * * & * 0.024 * * + & 0.25 & 0.75 & 0.641 & 0.608 & 0.152 & 0.209 & 0.141 & 0.120 & * 0.141 * & * 0.060 * * + & 0.50 & 0.75 & 0.548 & 0.485 & 0.219 & 0.218 & 0.244 & 0.137 & * 0.040 * * & * 0.068 * * + & 0.05 & 1.00 & 0.112 & 0.136 & 0.018 & 0.016 & 0.017 & 0.015 & * 0.006 * * & * 0.004 * * + & 0.25 & 1.00 & 0.177 & 0.168 & 0.050 & 0.049 & 0.049 & 0.042 & * 0.018 * * & * 0.010 * * + & 0.50 & 1.00 & 0.219 & 0.153 & 0.104 & 0.053 & 0.109 & 0.043 & * 0.024 * * & * 0.015 * * + & 0.05 & 0.95 & 0.125 & 0.162 & 0.015 & 0.019 & 0.015 & 0.015 & * 0.009 * * & * 0.004 * * + & 0.25 & 0.95 & 0.212 & 0.208 & 0.063 & 0.061 & 0.060 & 0.043 & * 0.025 * * & * 0.011 * * + & 0.50 & 0.95 & 0.271 & 0.211 & 0.099 & 0.077 & 0.106 & 0.046 & * 0.023 * * & * 0.018 * * + & 0.05 & 0.75 & 0.215 & 0.285 & 0.030 & 0.036 & * 0.022 * & 0.012 & 0.034 & * 0.005 * * + & 0.25 & 0.75 & 0.405 & 0.403 & 0.100 & 0.139 & 0.084 & 0.036 & * 0.025 * * & * 0.014 * * + & 0.50 & 0.75 & 0.457 & 0.415 & 0.156 & 0.184 & 0.159 & 0.048 & * 0.030 * * & * 0.020 * * + & 0.05 & 1.00 & 0.010 & 0.013 & 0.022 & 0.019 & 0.022 & 0.019 & * 0.001 * * & * 0.001 * * + & 0.25 & 1.00 & 0.032 & 0.018 & 0.022 & 0.004 & 0.027 & 0.008 & * 0.002 * * & * 0.002 * * + & 0.50 & 1.00 & 0.070 & 0.020 & 0.032 & 0.005 & 0.039 & 0.012 & * 0.002 * * & * 0.002 * * + & 0.05 & 0.95 & 0.018 & 0.029 & 0.018 & 0.014 & 0.019 & 0.015 & * 0.001 * * & * 0.001 * * + & 0.25 & 0.95 & 0.063 & 0.056 & 0.010 & 0.017 & 0.013 & 0.009 & * 0.002 * * & * 0.001 * * + & 0.50 & 0.95 & 0.115 & 0.083 & 0.040 & 0.032 & 0.045 & 0.010 & * 0.002 * * & * 0.001 * * + & 0.05 & 0.75 & 0.062 & 0.114 & 0.011 & 0.006 & 0.018 & 0.015 & * 0.001 * * & * 0.001 * * + & 0.25 & 0.75 & 0.236 & 0.249 & 0.063 & 0.090 & 0.026 & 0.005 & * 0.002 * * & * 0.002 * * + & 0.50 & 0.75 & 0.380 & 0.353 & 0.130 & 0.168 & 0.106 & 0.021 & * 0.002 * * & * 0.002 * * + & 0.05 & 1.00 & 0.410 & 0.389 & 0.195 & 0.256 & * 0.147 * * & * 0.190 * * & 0.418 & 0.390 + & 0.25 & 1.00 & 0.410 & 0.356 & 0.151 & 0.209 & * 0.103 * * & * 0.117 * * & 0.148 & 0.183 + & 0.50 & 1.00 & 0.367 & 0.299 & 0.195 & 0.154 & * 0.190 * & * 0.038 * * & 0.243 & 0.236 + & 0.05 & 0.95 & 0.455 & 0.430 & 0.204 & 0.271 & * 0.140 * * & * 0.192 * * & 0.424 & 0.406 + & 0.25 & 0.95 & 0.455 & 0.401 & 0.180 & 0.230 & * 0.115 * * & * 0.112 * * & 0.157 & 0.187 + & 0.50 & 0.95 & 0.412 & 0.331 & 0.222 & 0.179 & * 0.214 * & * 0.047 * * & 0.249 & 0.241 + & 0.05 & 0.75 & 0.593 & 0.578 & 0.225 & 0.325 & * 0.133 * * & * 0.181 * * & 0.415 & 0.440 + & 0.25 & 0.75 & 0.602 & 0.562 & 0.191 & 0.327 & * 0.119 * * & * 0.123 * * & 0.210 & 0.230 + & 0.50 & 0.75 & 0.520 & 0.470 & 0.264 & 0.278 & 0.272 & * 0.053 * * & * 0.254 * & 0.247 + & 0.05 & 1.00 & 0.116 & 0.123 & 0.052 & 0.061 & * 0.045 * & 0.056 & 0.448 & * 0.014 * * + & 0.25 & 1.00 & 0.158 & 0.132 & 0.054 & 0.067 & 0.041 & 0.051 & * 0.037 * & * 0.027 * * + & 0.50 & 1.00 & 0.186 & 0.125 & 0.077 & 0.049 & * 0.074 * & * 0.021 * * & 0.130 & 0.068 + & 0.05 & 0.95 & 0.131 & 0.146 & 0.053 & 0.066 & * 0.042 * * & 0.053 & 0.455 & * 0.016 * * + & 0.25 & 0.95 & 0.186 & 0.175 & 0.071 & 0.082 & * 0.049 * & 0.049 & 0.055 & * 0.021 * * + & 0.50 & 0.95 & 0.239 & 0.183 & 0.082 & 0.082 & * 0.073 * & * 0.024 * * & 0.134 & 0.075 + & 0.05 & 0.75 & 0.230 & 0.268 & 0.079 & 0.097 & * 0.046 * * & * 0.052 * * & 0.461 & 0.261 + & 0.25 & 0.75 & 0.375 & 0.377 & 0.123 & 0.159 & * 0.090 * * & * 0.039 * & 0.191 & 0.050 + & 0.50 & 0.75 & 0.450 & 0.406 & 0.201 & 0.212 & 0.216 & * 0.045 * & * 0.101 * * & 0.067 + & 0.05 & 1.00 & 0.015 & 0.020 & 0.024 & 0.020 & 0.025 & 0.021 & * 0.015 * & * 0.011 * * + & 0.25 & 1.00 & 0.040 & 0.025 & 0.018 & * 0.005 * & 0.022 & 0.005 & * 0.009 * * & 0.009 + & 0.50 & 1.00 & 0.094 & 0.027 & 0.022 & 0.005 & 0.028 & 0.006 & * 0.002 * * & * 0.002 * * + & 0.05 & 0.95 & 0.024 & 0.037 & * 0.019 * & 0.017 & 0.020 & 0.018 & 0.022 & * 0.012 * * + & 0.25 & 0.95 & 0.074 & 0.063 & 0.013 & 0.021 & 0.011 & 0.011 & * 0.009 * & * 0.009 * * + & 0.50 & 0.95 & 0.141 & 0.089 & 0.028 & 0.034 & 0.026 & 0.008 & * 0.002 * * & * 0.002 * * + & 0.05 & 0.75 & 0.072 & 0.124 & * 0.008 * * & * 0.005 * * & 0.015 & 0.010 & 0.034 & 0.012 + & 0.25 & 0.75 & 0.244 & 0.258 & 0.076 & 0.095 & 0.033 & 0.009 & * 0.009 * * & * 0.008 * + & 0.50 & 0.75 & 0.389 & 0.355 & 0.128 & 0.173 & 0.106 & 0.014 & * 0.002 * * & * 0.002 * * + to demonstrate the efficacy of the class prior preserving transform , we implemented the multivariate versions of alphamax - n and msgmm and evaluated them on the twelve real data sets without applying the transform . there were significant stability and computational issues related to the high - dimensional nature of the data sets . msgmm was numerically unstable because of singular / nearly - singular covariance matrix , whereas alphamax - n became computationally demanding because the number of bins ( for histogram based density estimation ) grow exponentially with the dimension , resulting in a large parameter vector @xmath379 and , consequently , a large optimization problem , even after removing the zero - count bins . this is expected , as density estimation for multivariate data is known to be problematic , which is one of the main reasons for introducing our transform . to make estimation feasible under these stability and computational issues , we used dimensionality reduction . though not all data sets posed the same level of difficulty , to have a standard approach and permit effective density estimation , we used the top three principal components , obtained via principal component analysis on the z - score normalized data ( mixture and component samples combined ) , as input to the two algorithms . we also attempted using top @xmath384 principal components that preserve 75 percent of the total variance , however , for some of the data sets , the dimension was still too high . in the same manner as in the univariate case , we used histograms in the multivariate implementation of alphamax - n . the bin - width for a dimension was selected to minimize the asymptotic mean integrated squared error ( amise ) with a normal reference rule , using the component sample , @xmath147 . the formula for the bin - width of dimension @xmath384 is given by : @xmath388 where @xmath167 is the total number of dimensions , @xmath389 is the standard deviation of the @xmath390 dimension and @xmath391 is the size of the component sample . bins were added to cover the range of the entire data , mixture and component combined , and empty bins were removed to reduce the size of the optimization problem . contains the results of alphamax - n and msgmm on the real - life data sets , using the top three principal components under column headings alphamax - nm ( m for multi - dimensional ) and msgmm , respectively . the results of alphamax - n and msgmm - t with the class - prior preserving transform are also provided for comparison . notice that , though alphamax - nm ( without transform ) performs well , alphamax - n ( with transform ) is significantly better in terms of estimation error , despite having a lower computational cost . also notice the deterioration in the performance of msgmm ( without transform ) compared to msgmm - t ( with transform ) . 1.01 |>x| > x| > x| > x| > x| > x| > x| > x| & @xmath19 & @xmath20 & % variance & & & & + & 0.095 & 1.00 & & 0.037 & * 0.028 * * & 0.163 & 0.528 + & 0.096 & 0.95 & 40 & 0.036 & * 0.032 * & 0.155 & 0.574 + & 0.101 & 0.75 & & * 0.040 * & 0.047 & 0.127 & 0.580 + & 0.419 & 1.00 & & 0.181 & 0.276 & 0.077 & * 0.020 * * + & 0.425 & 0.95 & 60 & 0.231 & 0.269 & 0.095 & * 0.028 * * + & 0.446 & 0.75 & & 0.272 & 0.320 & 0.233 & * 0.063 * * + & 0.342 & 1.00 & & 0.017 & 0.030 & * 0.008 * * & 0.585 + & 0.353 & 0.95 & 81 & * 0.006 * & 0.021 & 0.006 & 0.575 + & 0.397 & 0.75 & & 0.009 & 0.064 & * 0.006 * * & 0.533 + & 0.268 & 1.00 & & * 0.094 * * & 0.132 & 0.209 & 0.316 + & 0.281 & 0.95 & 69 & 0.110 & * 0.110 * & 0.204 & 0.308 + & 0.330 & 0.75 & & * 0.134 * & 0.205 & 0.172 & 0.283 + & 0.093 & 1.00 & & * 0.007 * & 0.008 & 0.157 & 0.443 + & 0.103 & 0.95 & 89 & * 0.008 * * & 0.028 & 0.152 & 0.298 + & 0.139 & 0.75 & & * 0.012 * * & 0.053 & 0.143 & 0.270 + & 0.409 & 1.00 & & * 0.022 * * & 0.075 & 0.037 & 0.432 + & 0.416 & 0.95 & 24 & * 0.008 * * & 0.021 & 0.037 & 0.398 + & 0.444 & 0.75 & & * 0.020 * & 0.050 & 0.024 & 0.375 + & 0.086 & 1.00 & & * 0.044 * & 0.046 & 0.129 & 0.178 + & 0.087 & 0.95 & 72 & 0.052 & * 0.040 * * & 0.125 & 0.178 + & 0.090 & 0.75 & & 0.064 & * 0.031 * * & 0.111 & 0.188 + & 0.243 & 1.00 & & * 0.009 * * & 0.071 & 0.081 & 0.289 + & 0.248 & 0.95 & 65 & * 0.005 * * & 0.070 & 0.074 & 0.286 + & 0.268 & 0.75 & & * 0.007 * * & 0.092 & 0.062 & 0.260 + & 0.251 & 1.00 & & * 0.111 * & 0.123 & 0.171 & 0.299 + & 0.259 & 0.95 & 60 & * 0.110 * * & 0.156 & 0.168 & 0.292 + & 0.289 & 0.75 & & * 0.156 * & 0.178 & 0.175 & 0.286 + & 0.139 & 1.00 & & * 0.029 * * & 0.064 & 0.157 & 0.232 + & 0.140 & 0.95 & 67 & * 0.007 * * & 0.055 & 0.157 & 0.227 + & 0.143 & 0.75 & & * 0.004 * * & 0.015 & 0.148 & 0.356 + & 0.226 & 1.00 & & 0.041 & * 0.034 * & 0.059 & 0.487 + & 0.240 & 0.95 & 22 & 0.042 & * 0.041 * & 0.063 & 0.485 + & 0.295 & 0.75 & & * 0.044 * * & 0.072 & 0.059 & 0.434 + & 0.566 & 1.00 & & * 0.060 * & 0.258 & 0.070 & 0.134 + & 0.575 & 0.95 & 65 & * 0.063 * & 0.256 & 0.076 & 0.126 + & 0.612 & 0.75 & & 0.353 & 0.302 & 0.293 & * 0.096 * * +

the merging of two galaxies will produce a binary black hole at the center of the newly formed galaxy . numerical relativity simulations predict that black holes can be kicked out of the galaxy in binary black hole mergers with velocities up to several thousand @xmath2 ( e.g. * ? ? ? detection of a rapidly recoiling black hole would be a confirmation of the gravitational wave and shed new light on the cosmic evolution of super - massive black holes . @xcite reported the first candidate , sdss j092712.65 + 294344.0 , which have two sets of emission lines with velocity difference of 2650 @xmath2 and the red set has only narrow emission lines ( nels ) . and since then several other recoiling black hole candidates have been reported ( e.g. * ? ? ? . however , all the candidates have interpretations other than recoiling black hole , such as peculiar double - peak emitters , close black hole binaries or chance superpositions ( e.g. * ? ? ? * ; * ? ? ? . there have been many efforts to search for recoiling black hole candidates in large area survey . most of them used shifted broad emission lines ( bels ) relative to nels as an indicator of recoiling black holes ( see * ? ? ? * for a review ) . in addition , @xcite suggested an `` off - nuclear agn '' character , which means that a recoiling active black hole moves away from the center of its host galaxy , can also be used as the indicator of recoiling black holes . we have run a systematic search for low - redshift rapidly recoiling black hole candidates based on sdss data release 7 . the quasar catalog of @xcite was used as our parent sample . we first selected those with @xmath4 so that complete h@xmath5 profiles are available using sdss data . we then obtained the profiles of h@xmath5 bels by subtracting continuum and nels . after that , we detected the peaks from the smoothed profiles and measured their velocity shifts relative to nels . quasars which show peaks on both blue and red sides are excluded because these are obvious double - peak emitters and difficult to be explained by recoiling black holes . we found 22 sources with velocity shifts greater than 2000 @xmath2 . among this sample , sbs 1421 + 511 , which is a system consisting of a radio quasar at z = 0.276 and a nearby extended source ( figure 1(b ) ) , is the only one showing an off - nuclear agn character . the position offset between the quasar and the extended source suggests that the quasar might be kicked out of its original host galaxy . it can be seen from the sdss spectrum ( figure 1(a ) ) that the profile of balmer bels is blue - skewed and the peak velocity of the profile is @xmath13300 @xmath2 relative to nels . both the blue - skewed profile and the off - nuclear agn character suggest a super - massive recoiling black hole . if confirmed , sbs 1421 + 511 may shed new light on the study of recoiling black holes due to its off - nuclear agn character which is rare among the candidates . in this paper , we report the results of the new keck - ii echellette spectrograph and imager ( esi ) long - slit spectrographic observation of the system sbs 1421 + 511 , which took the spectra of the quasar and the extended source at the same time . we identified a host galaxy in the quasar position , and therefore the off - nuclear agn character is not due to a recoiling black hole . the quasar bels are well modeled as emissions from a circular disk with a blueshift velocity of @xmath11400 @xmath2 relative to nels , suggesting a super - massive black hole which is recoiling at a velocity of @xmath11400 @xmath2 , and meanwhile other possible interpretations exist . the physical justifications of these interpretations are discussed in section 4.2 . the spectrum of the extended source shows that it is an galaxy with a velocity difference relative to the quasar host galaxy of only 120 @xmath2 and a projected distance of only 12.5 kpc at the redshift . therefore , the extended source is likely to be an companion galaxy and we refer to the extended source as `` the companion '' hereafter . the companion shows seyfert - like line ratios and a high [ oiii ] luminosity , suggesting that sbs 1421 + 511 is a dual agn system . dual agns have valuable potential as observational tools for studies of galaxy evolution , including measurements of galaxy merger rates , mass growth rates and merger rates of super - massive black holes @xcite . [ oiii ] emission is often used as a indicator of agn in surveys of dual agns . for example , double - peak [ oiii ] nels are used as signatures of dual agns in large area survey ( e.g. * ? ? ? * ) and separated [ oiii ] nuclei are important to confirming the dual agn candidates ( e.g. * ? ? ? * ) . intriguingly , we found an extended emission line region ( eelr ) southern to the quasar . the existence of the eelr suggests that a similar being may coincide with the companion and the companion is not necessarily active . we considered other possible interpretations such as a pair of eelrs with a bipolar structure and gas in the companion galaxy illuminated by the quasar , which are further discussed in section 4.3 . the eelr itself is interesting . in spite of multiple observations , the nature and origin of eelrs ( e.g. * ? ? ? * ) are still not understood completely and there is no generally interpretation for all the eelrs ( see * ? ? ? * for a review ) . they are initially thought to be tidal features , but later observations found that only a few of them show structures like tidal tails . only a small fraction of eelrs show accompanied large scale radio features , indicating that they are likely driven by radio jets @xcite . @xcite presented that the luminosity of the eelr is strongly correlated with nuclear [ oiii ] luminosity , suggesting that most of the eelrs originate in agn - driven galactic wind . sbs 1421 + 511 is likely to be tidally interacting with a companion , and shows radio structures and large scale outflows at the same time , thus provide an ideal laboratory to study the nature of the eelr . the possible origins of the eelr in sbs 1421 + 511 are discussed in section 4.4 . we obtained the keck esi optical long slit spectrum of sbs 1421 + 511 on 2013 may 8 using the echelle mode . the width of the slit is 0.75@xmath0 and the instrumental dispersion is @xmath6 22 @xmath2 . the slit was aligned in a north - south direction ( figure 1(b ) ) , so the quasar and the companion can fall into the slit at the same time . the exposure time was 20 minutes . after the exposure , we immediately obtained a spectrum of the standard star bd+33 2642 with the same setting and 2 minutes exposure to perform flux calibration and generate point spread function ( psf ) . the seeing was @xmath7 measured from the full width of half maximum ( fwhm ) of the psf . we reduced the original data through a routine based on idl program package * * xidl * * to obtain two - dimensional ( hereafter 2-d ) spectral images . using the spectral images , we can easily obtain one - dimensional ( hereafter 1-d ) spectra by collapsing them along the spatial direction and obtain spatial brightness profiles along the slit by collapsing them along the spectral direction . bias and flat correction , wavelength calibration , sky subtraction were performed first . cosmic - rays were removed using * lacosmic * @xcite . we resampled the clean spectral data around the tracks of the quasar for 10 echelle orders so that one pixel corresponds to 10 @xmath2 in wavelength direction and 0.15@xmath0 in spatial direction , using 2-d wavelength calibration from arc - lamp image and pixel scales from archival maps . we then jointed the spectral images of the 10 orders . the joint 2-d spectral image covers the wavelength range of 400010100 after removing low response regions . we then ran the same routine to obtain the spectral image of the standard star and then used the extracted 1-d spectrum of the standard star to make flux calibration of the source . we also obtained the spatial brightness profiles in different wavelength ranges for the standard star to generate 1-d psf . there is a chance that the psf may not be adequate for analysing the spectral image of the source due to seeing variance . however , the variance is minor because the seeing was stable at that time according to the observation log and the time interval between the two exposures was only @xmath112 minutes . sbs 1421 + 511 shows an off - nuclear agn character . it could be explained by a single - galaxy model that the quasar is naked and the companion is the original host galaxy of which the quasar might be kicked out . there also exists an alternative double - galaxy model that the quasar has a host galaxy and the companion is indeed a companion galaxy . eracleous & halpern ( 2003 , hereafter eh03 ) presented the analysis of optical spectra of sbs 1421 + 511 obtained with kpno 2.1 m and mdm 2.4 m telescopes during 1997 and 2000 . according to the results of eh03 , there is starlight emission underneath the quasar and the flux fraction of the starlight is 0.28@xmath80.2 . the starlight can be interpreted as the host galaxy of the quasar according to the double - galaxy model , and as the outer part of the single galaxy according to the single - galaxy model . we could not judge the two based on eh03 s results because we had no spectrum of the companion and the starlight fraction has large uncertainty . we extracted the spectra of the quasar and the companion from the new esi 2-d spectral image with 3@xmath0 apertures . as can be seen from figure 2 , the quasar spectrum clearly shows absorption features of old stellar populations such as caii k band around 3934 , mg ib feature around 5177 , na i d @xmath9 5890 , 5896 doublet lines , proving that there is starlight underneath the quasar . and the companion spectrum shows strong these features , and thus is dominated by old stellar populations . the absorption line redshifts are close for the two spectra . according to the single - galaxy model , the stellar emission in the quasar spectrum comes from the outer part of the single galaxy and thus its flux would be much lower than that in the companion spectrum , which is extracted using the same aperture size . and it is not necessarily so according to the double - galaxy model . therefore a decomposition of quasar and stellar components in the quasar spectrum may help to distinguish the two models . we fit the quasar spectrum using models consisting of a stellar component and a quasar component . the stellar component is the addition of simple stellar population ( ssp ) templates of @xcite , which were broadened by convolving with a gaussian to match the stellar velocity dispersion and shifted to match the redshift . we used a single ssp and the best - fitting result was achieved with stellar population of @xmath13 gyr old , and redshift and velocity dispersion are @xmath10 and @xmath11 , respectively . we attempted to add one more ssp but after this the reduced @xmath12 only varied 0.02 and the second ssp contributes little flux , and therefore we did not adopt the second ssp . for the quasar component , we attempted to use a single power - law as the quasar continuum at first and the residual showed excess in spectral region redder than 6000 . the excess is also seen in quasar composite spectra @xcite and is interpreted as the radiation of a torus . therefore we used a broken power - law with broken point fitted to be @xmath15700 for the quasar continuum . we masked the spectral regions affected by nels , which are demonstrated in figure 2 . there are strong and wide bels , including balmer series and hei @xmath135876 , which influence some important stellar features , and therefore we fit the continuum and bels at the same time instead of masking the regions affected by bels in the fitting . the bel models are described as follows : 1 . we used double gaussian for all the bels since they are obviously asymmetric . 2 . the residual reveals a feature at around 4700 . though the quasar composite spectra show feii pseudo continuum here , it is unlikely that the feature is due to feii emission because the data were not well fitted by accordingly adding an optical feii template . thus we considered another possibility that the feature is heii @xmath134686 bel , which is seen in the spectra of feii weak quasars . we added a corresponding double gaussian for heii of which the profile is assumed to be identical to that of h@xmath14 , and yielded a good fit to the data . 3 . the residual shows an excess at @xmath15 , which is likely due to blended high order balmer bels and thus we added a component accordingly . the component consists of balmer lines from @xmath16 to @xmath17 , of which the profiles are fixed to that of h@xmath18 , and of which the relative strength are fixed using the line emissivity for case b @xcite under a typical circumstance for broad line regions of @xmath19 k and @xmath20 @xmath21 . we also used ssp templates to fit the companion spectrum . we tried single ssp and double ssp , and the results showed that double ssp is appropriate and the ages of the two ssps are @xmath22 gyr and @xmath110 myr . the redshift and velocity dispersion of the ssps are @xmath23 and @xmath24 @xmath2 , respectively . considering that the companion spectrum is contaminated by the quasar continuum and bels due to seeing effect , we added a contamination component which is @xmath14% of the corresponding flux in the quasar spectrum and the fraction was computed using psf . we overplot the best - fitting models for the two spectra in figure 2 . the fraction of starlight increases along direction of longer wavelength from 19% at 4000 to 46% at 7000 , in agreement with eh03 s result . the corresponding stellar mass in the quasar spectrum is @xmath25 , close to that in the companion spectrum of @xmath26 . therefore it is unlikely that the starlight underneath the quasar comes from the outer part of the single galaxy . the stellar mass is high thus it is also unlikely that the stellar component comes from stars which are ejected accompanied with the super - massive black hole when recoiling . we display the 1-d spatial brightness profile of the continuum from keck data in figure 3(a ) . the profile was extracted from a spectral region at around rest - frame 7000 , where the contribution of the quasar component is relatively small . as illustrated by figure 3(a ) , we obtained the minimum flux of the non - quasar component ( orange line ) . we also show the 2-d brightness profile from sdss i - band image in figure 3(c ) . this band ( rest - frame 54006500 ) was selected because it is not strongly affected by emission lines . using psf from a nearby star , we obtained a similar minimum flux ( figure 3(d ) ) and the inferred lower limit of the stellar luminosity is @xmath27 . such a bright galaxy would be either an elliptical or a disk galaxy if not a merger , and the radial brightness profiles of the both can be well described by srsic profile . thus this profile is used to fit the continuum brightness profile for both the 1-d ( keck ) and 2-d ( sdss ) data , and the later was done with the code * galfit * ( version 3.0 , * ? ? ? we used two models : a single - galaxy model consisting of a srsic profile for the single galaxy and a point source for the quasar component , and a double - galaxy model with one more srsic profile for the host . for 1-d data the double - galaxy model is significantly better than the single - galaxy model since the @xmath12/dof reduces from 65603/115 to 2505.2/110 , and for 2-d data it is marginally better since the @xmath12/dof are 11092.8/12304 and 11060.8/12297 . we adopted the conclusion from 1-d data that the single - galaxy model was excluded as the total signal to noise ratio of 1-d data is much higher than that of 2-d data . from 1-d spatial decomposition results , we obtained the contribution of the stellar component in an aperture which was used to extract quasar spectrum at around 7000 to be 44% , in accordance with that obtained from spectral decomposition results of 46% . the good agreement makes the two decompositions more robust . therefore we conclude that the off - nuclear agn character is not due to a recoiling super - massive black hole but due to a close companion . the distance between the centers of the two galaxies is @xmath28 arcsec and @xmath29 arcsec for 1-d and 2-d data , respectively , which are consistent . the little distance ( 12.5 kpc in projection at @xmath30 ) and the little radial velocity difference ( 120 @xmath2 ) indicate that the two galaxies are likely physically connected and rotating around each other , and going on a major merger considering that both the two galaxies are bright with @xmath31 and @xmath32 , which were obtained from 2-d spatial decomposition results . the srsic index of the companion galaxy is @xmath33 and @xmath34 for the 1-d and 2-d fit , respectively , being close to 1 and implying that the companion galaxy may be a disk galaxy . on the other hand , the srsic index of the host galaxy is @xmath35 for 1-d fit , and the corresponding value for 2-d fit is undetermined due to large uncertainty , and we are not sure if it is a disk galaxy or an elliptical galaxy . the blue - skewed profile of quasar bels of sbs 1421 + 511 implies a super - massive recoiling black hole . however , similar profiles are also seen in some agns with double - peaked bels . eh03 analyzed the h@xmath5 bel profile of sbs 1421 + 511 using spectrum obtained with mdm 2.4 m telescopes in 1998 . they fit the line with the relativistic keplerian disk model of @xcite , which is widely used for double - peaked bels . the model assumes that the line originates on the surface of a circular , keplerian disk whose axis is inclined by an angle @xmath36 relative to the line of sight , between radii @xmath37 and @xmath38 ( expressed in units of the gravitational radius , @xmath39 , where @xmath40 is the mass of the black hole ) . the disk has an axisymmetric emissivity of the form @xmath41 . local broadening of the line is represented by a gaussian rest - frame profile of velocity dispersion @xmath42 . the free parameters of the model are thus @xmath37 , @xmath38 , @xmath36 , @xmath43 and @xmath42 . eh03 found that the h@xmath5 bel could not be well fitted with a simple circular disk model , and the fit could be improved by either ( 1 ) applying an ad hoc blueshift to the model of 1370 km / s or ( 2 ) using an elliptical disk model , which has two more free parameters of eccentricity @xmath44 and the angle between the major axis of the elliptical disk and the line of sight @xmath45 @xcite . we accordingly fit the h@xmath5 profile of esi , which was obtained by subtracting continuum from original quasar spectrum , using both circular and elliptical disk models . we also attempted to apply a blueshift to the models . the statistics and best - fitting parameters of the models are listed in table 1 . without blueshift , neither circular nor elliptical disk model can well describe the h@xmath5 profile . after applying an ad hoc blueshift , both the two models yield good fit , and they are statistically just as good , for @xmath12/dof is 1013/843 and 1004/841 for the two models . the best - fitting circular disk model with a blueshift is displayed in the upper panel of figure 4 . the ad hoc blueshift applied to the circular disk model is 1450 @xmath2 , close to the value obtained by eh03 . we confirmed eh03 s conclusions that the double - peaked bels of sbs 1421 + 511 can not be well reproduced by a simple circular disk model and the fit was greatly improved by applying an ad hoc blueshift . and we excluded the elliptical disk model because : 1 . it can not reproduce the profile without a blueshift ; 2 . it is statistically not better than circular disk model if applying a blueshift . long term monitoring of agns with double - peaked bels @xcite reveal profile variations , which can be described by the changes of excess emissions in amplitude and projected velocity on timescales of years . since we have three spectra of sbs 1421 + 511 from mdm ( 1998 ) , sdss ( 2003 ) and keck ( 2013 ) , we plotted the normalized h@xmath5 profiles of the three spectra together in the lower panel of figure 4 to see the variability of the profile . the variance between the sdss and keck profiles is little , and the blue peak in mdm profile is weaker than those in sdss and keck profiles . we did not find velocity change of the blue peak among the three observations , though the change is common among agns with similar bel profiles . to quantify this , we estimated the change by comparing mdm and sdss profiles with shifted and normalized keck profile and by using minimum chi - square method . the change is @xmath46 @xmath2 between sdss and keck profiles , and @xmath47 @xmath2 between mdm and keck profiles , and the inferred velocity change rate is @xmath48 @xmath2 per year in rest frame . this constitutes an important constraint on the nature of the double - peaked bel of sbs 1421 + 511 , and we discuss the possible interpretations of the bels in section 4.2 . it can be seen from figure 2 that the companion spectrum shows detectable [ heii ] @xmath134686 and strong [ oiii ] @xmath94959 , 5007 . the line ratio of [ oiii]/h@xmath14 is @xmath49 measured from the residual between the original spectrum and the best - fitting model . these suggest that the gas in the companion is ionized by an accreting black hole and the companion may also be an agn . the [ oiii ] emission of the companion is so strong that a corresponding emission line feature can be easily seen northern to the quasar from the 2-d emission line spectral image ( figure 5(a ) ) . intriguingly , there also exists an emission line feature in the opposite direction southern to the quasar . as can be seen from the 2-d spectral image , the two features show symmetry in both spatial and velocity directions : the northern one is blueshifted relative to the quasar host galaxy and the southern one is redshifted , and the distances to the quasar position are both several arcsecs and the velocity differences are both @xmath50 @xmath2 . as the symmetry we investigated the two features together . for clarity , we referred to the northern one as `` feature n '' and the southern one as `` feature s '' hereafter . we extracted the spatial brightness profiles of feature n using a velocity range of @xmath51 @xmath2 , which are presented in figure 5(b ) . both the peak positions of profiles for [ oiii ] and [ nii ] locate @xmath13@xmath0 north to the quasar , roughly coincident with the center of the companion ( horizontal yellow dashed line ) , and for h@xmath5 and h@xmath14 the peak positions are @xmath14 north to the quasar . the disparity indicates that feature n may consist of two components : one has stronger [ oiii ] and [ nii ] emissions whose major emitting region is around the center of the companion ; the other has stronger balmer emissions whose major emitting region is in the outer region , and hence likely comes from an off - center hii region . considering the 1.4@xmath0 seeing , the two components are mixed with each other . to decompose them we extracted 1-d spectra ( figure 5(d ) ) using two 2@xmath0-width apertures , apern ( blue in figure 5(c ) ) and aperhii ( purple ) . we selected apern whose center is 3@xmath0 north to the quasar , in consistent with the major emitting region of the first component , and selected aperhii so that the emission most comes from the second component . we obtained the emission line spectra by subtracted the continua , and the velocity profiles of [ oiii ] and h@xmath5 are plotted in figure 5(e ) . we also classified the type of ionization source using the flux ratios of emission lines , which are listed in table 2 and plotted in figure 5(f ) , according to the diagnostic diagram of @xcite . for apern , the line ratios obtained by simply performing a gaussian fitting are seyfert - like ( blue star in figure 5(f ) ) according to the diagnostic diagram . considering the contamination of the second component , which contributes more to h@xmath5 and h@xmath14 than to [ oiii ] and [ nii ] , the line ratios of the first component would move upper right on the diagnostic diagram and still be agn - like . for aperhii , [ oiii ] profile is nearly the same as that of apern ( figure 5(e ) ) , suggesting that [ oiii ] emission is dominated by the first component . by contrast , h@xmath5 profile is much narrower than that of apern , and for both the two apertures h@xmath5 profiles are narrower than [ oiii ] profiles , implying that h@xmath5 emission of aperhii is dominated by the second component and h@xmath5 emission of apern is the mixture of the two components . by directly fitting the profiles using gaussian , the line ratios of aperhii ( purple star ) are close to the dividing line of agn and hii regimes in the diagnostic diagram . furthermore , the line ratios ( purple diamond ) are hii - like if calculated using fluxes which are integrated over a narrower velocity range of @xmath52 @xmath2 , which is less contaminated . thus it is likely that the second component has hii - like line ratios . we then decomposed the emission lines of the two apertures into two sets which are corresponding to the two components by assuming that the profiles are gaussian and are identical for all the emission lines for each set . the results show that 87% and 13% of the first set fall into apern and aperhii , respectively , and the corresponding fractions of the second set are 44% and 56% . the line ratios of the first set ( blue square ) are seyfert - like , and those of the second set ( purple square ) are hii - like . the velocity difference of the seyfert - like set of emission lines relative to the absorption line of the companion galaxy is @xmath53 @xmath2 and that of the hii - like set is @xmath54 @xmath2 , and the velocity dispersions are 75 and 150 @xmath2 for two sets , respectively . therefore we concluded that feature n is composed of a component in the companion center which has seyfert - like line ratios and an off - center hii region . the [ oiii ] luminosity of the seyfert - like component in a rectangular aperture which is set by 0.75@xmath0 slit and 2@xmath0 apern is @xmath55 . the contamination by quasar emission due to seeing effect is minor as the fraction is @xmath56 predicted by psf . to obtain the total luminosity of the source of the seyfert - like component , we should multiply an aperture correction which is computed to be 2.3 by using psf and assuming the source is point - like , and higher if the source is extended . thus the total [ oiii ] luminosity is l@xmath57 } } > 1.4\times10 ^ 8l_\odot$ ] , which may be underestimated considering an uncertain extinction correction . such a high [ oiii ] luminosity suggests an powerful agn in the companion galaxy . we also extracted the spatial brightness profiles of feature s using a velocity range of @xmath58 @xmath2 . feature s shows high [ oiii ] flux relative to h@xmath14 and extends to at least 6@xmath0 ( 25 kpc in projection ) away from the quasar where little starlight is seen , indicating an eelr . in order to compare the properties of feature s and feature n , we extract a 1-d spectrum in an aperture apers ( red in figure 5(c , d ) ) whose width and distance to the quasar are both the same with those of apern . the velocity profile of [ oiii ] shows a broad base , which may come from the contaminations by emissions of the quasar . to mask out the base , we fit [ oiii ] emission line using double gaussian and subtract the blueshifted broad gaussian component . by assuming that the rest emission comes from an eelr , we obtained the velocity difference and the velocity dispersion of the eelr of @xmath59 @xmath2 and @xmath60 @xmath2 . the line ratios of apers spectrum ( red star in figure 5(f ) ) are seyfert - like , agreeing with that the emission comes from an eelr . we will further discuss the natures of feature n and feature s in section 4.3 and 4.4 . for the discussion , we estimated the electron density of the two features using line ratios of [ sii ] @xmath136716/@xmath136731 and [ oii ] @xmath133729/@xmath133726 , and electron temperature using line ratio of [ oiii ] ( @xmath134959+@xmath135007)/@xmath134363 @xcite . for feature n , we obtained only an upper limit of the electron density of @xmath61 @xmath21 ( 95.3% confidence ) according to [ sii ] ratio of @xmath62 and [ oii ] ratio of @xmath63 , and the electron temperature is @xmath64 k according to [ oiii ] ratio of @xmath65 . for feature s , the electron density of feature s is @xmath66 @xmath21 according to the [ sii ] ratio of @xmath67 , and the uncertainty of [ oii ] line ratio is too large to limit the density . [ oiii ] @xmath134363 line is not detected in apers spectrum , and the inferred electron temperature is @xmath68 k ( 95.3% confidence ) using the upper limit . the electron density is higher in apers than in apern and the electron temperature is lower . in the quasar spectrum the profiles of nels show strong blue wings , which are generally thought to be correlated with an emission line outflow . figure 6(b ) displays the velocity profiles of three isolated nels ( h@xmath14 , [ oiii ] @xmath135007 and [ oi ] @xmath136300 ) . besides relatively narrow core components , strong blue wing components are seen in both high- and low - ionized lines , which extend for a velocity up to 1200 @xmath2 measured from the [ oiii ] velocity profile , suggesting a strong outflow . we tried to parameterize the nels to further investigate the two components . we used double gaussian for all the nels : one for the core component and the other for the blue wing component . some lines are blended , such as h@xmath5+[nii ] @xmath96548,6583 lines and [ sii ] @xmath96716,6731 doublet , so we tied the parameters of these lines as follows to reduce the uncertainty . for [ nii ] doublet , the profiles were assumed to be identical and the flux ratios were fixed to theoretical value 2.96 . and for [ sii ] doublet , the centroids were tied for both the core component and the wing component , and the widths were also tied , while the doublet flux ratios were set to be free parameters . the two gaussians to fit [ oiii ] are shown in figure 6(b ) ( purple lines ) . the redshift and the velocity dispersion of the core component are @xmath69 and @xmath70 @xmath2 , respectively , consistent with those of the starlight for the quasar spectrum , indicating that the kinematics of the core component is dominated by bulge gravitation and this component is likely to represent a traditional narrow line region ( nlr ) . the wing component contributes 70% of the total flux for [ oiii ] , and the velocity difference with the core component is @xmath71 @xmath2 , and the velocity dispersion is 340 @xmath2 . compared with the core component , the wing component is more complex , especially the near - zero - velocity part of this gaussian , which can be understood in two ways . one is that it comes from the traditional nlr whose velocity distribution is more extended than a gaussian profile . the other is that it does not come from the traditional nlr , though it has a radial velocity close to zero for some reason . in order to avoid confusion , we used a velocity range of @xmath72 @xmath2where the emission mostly comes from the outflow according to both interpretations , instead of the blueshifted gaussian , when referring to the blue wing component . we accordingly used an velocity range of @xmath73 @xmath2 when mentioning the core component . we noticed that the blue wing component shows lower ionization state than the core component as the line ratios of [ oiii]/h@xmath14 are @xmath74 and @xmath75 for the two component . the situation is different from blue wing outflows in most of other agns , which show blue wings stronger in high - ionized lines than in low - ionized lines . since the line ratios of [ oiii]/h@xmath14 are high , it is likely that both the outflow material and traditional nlr in sbs 1421 + 511 are mainly ionized by the quasar photons . if so , considering that the electron densities from [ sii ] are close for the two component , the lower ionization state suggests that the outflow region in sbs 1421 + 511 may be more extended than traditional nlr , which typically has a size of sub kpc to several kpc . this implies that the outflow region may have a size at an magnitude of order of @xmath1kpc , and thus may be resolvable under 1.4 arcsec seeing condition . to check this we extracted the spatial brightness profile of the blue wing component ( figure 6(c ) ) . although the northern half is influenced by feature n ( see figure 6(a ) ) , we can see from the uncontaminated southern half that the profile is broader than the psf . this is impossible to be caused by seeing variance because the profile is also broader than that of the core component . therefore we conclude that the outflow region is resolved . considering the seeing fwhm corresponds to @xmath16 kpc , the distinct disparity between the spatial profile and that of a point source indicates that the size of the outflow region is comparable to the seeing value , and thus is possibly at an order of magnitude of kpc . we then estimated the size of the blue wing outflow by fitting the southern half of the spatial profile . we tried several profile models , such as double gaussian , exponential and srsic , and found that a srsic profile model ( green lines in figure 6(d ) ) with index @xmath76 fit the profile best . the size of the outflow region is then estimated to be @xmath11 kpc using the half - light radius of the model profile . we estimated the mass rate of the blue wing outflow with a method used to estimate the parameters of nlr @xcite . the mass of emitting gas can be calculated as @xmath77 , where @xmath78 is the h@xmath14 luminosity in unit of @xmath79 @xmath80 , and @xmath81 is the electron density in unit of @xmath82 @xmath21 . the mass outflow rate is then @xmath83 , where @xmath84 is the mean velocity of the outflow material and @xmath85 is the size of outflow region . using the profile in the velocity range of @xmath72 @xmath2 , the h@xmath14 luminosity is l(h@xmath14)@xmath86 @xmath80 and the mean velocity is @xmath87 @xmath2 . the size of outflow region is estimated to be 1 kpc from the spatial profile and the electron density is @xmath88 from the flux ratio of [ sii ] @xmath136716/@xmath136731 of the wing component of @xmath89 . using these values , the mass of the emitting material is @xmath90 and the inferred mass outflow rate is @xmath91 . the kinetic energy outflow rate is @xmath92 @xmath80 . the bolometric luminosity and the black hole mass are key parameters to understand the quasar in sbs 1421 + 511 . we estimated the bolometric luminosity by applying bolometric correction to both uv - optical continuum luminosity and hard x - ray luminosity . the luminosity indicators of l@xmath93 and l@xmath94 were used since we could obtain the ratio of optical and x - ray @xmath95 to compare with other agns . because neither the sdss nor the keck spectrum covers a rest - frame wavelength of 2500 , we obtained l@xmath93 by extrapolating the continua in spectral regions bluer than 3500 . the aperture corrected l@xmath93 for both the two spectra are @xmath96 @xmath80 , and the variance between them is less than 10% . sbs 1421 + 511 was observed by swift x - ray telescope ( xrt ) in 2006 april and 2010 july ( target i d 35375 ) , and the effective exposure time was 5.2 and 2.7 ks for the two observations , respectively . we extract the spectra for pc mode using * xselect * ( version 2.4 ) through standard routine and combined the two spectra . we then grouped the combined spectrum so that it has at least 20 counts per bin to ensure the @xmath12 statistics , and fit it using * xspec * ( version 12.8 ) . the spectrum is well fitted ( @xmath97 ) by an absorbed power law with index @xmath98 and neutral absorption column density @xmath99 @xmath100 ( 90% confidence ) . the l@xmath94 is @xmath101 @xmath80 and the inferred @xmath95 is 1.1 for sbs 1421 + 511 , lower than 1.33 , the typical value of radio loud quasars @xcite . we checked whether this is due to variability in optical or in x - ray . the variance of l@xmath93 between 2003 ( sdss ) and 2013 ( keck ) is @xmath10210% , and the variance of l@xmath94 between 2006 and 2010 is @xmath130% , which was obtained by fitting the data of two swift observations individually . assuming these are the variability amplitudes of the quasar , the uncertainty of @xmath95 due to variability is only 0.06 . therefore the unusual shape of spectral energy distribution ( sed ) is barely attributed to variability . on the other hand , sbs 1421 + 511 is a flat spectrum radio quasar ( fsrq ) with spectral index @xmath103 ( defined as @xmath104 ) and p@xmath105 w hz@xmath106 , which is resolved in very large baseline array ( vlba ) observation . we show in figure 1(c ) the brightness distributions map taken during vlba imaging and polarimetry survey ( vips * ? ? ? the analysis of helmboldt et al . demonstrated that the radio source has a jet structure with a size of @xmath10730 pc and a position angle of @xmath108 degrees ( north by east ) . the power of the jet can be estimated with the radio luminosity @xcite to be @xmath109 @xmath80 , similar to the x - ray luminosity , and thus the excess in x - ray luminosity relative to uv - optical may be due to radio jet activity . accordingly , we used quasar 3c206 from the sample of elvis et al . whose @xmath95 ( 1.15 ) is the closest to that of sbs 1421 + 511 as sed template , and estimated the bolometric luminosity to be @xmath110 @xmath80 . the central black hole mass of the quasar in sbs 1421 + 511 could not be estimated from the bels because they are double - peaked . using the stellar velocity dispersion of the host galaxy ( @xmath111 @xmath2 ) and @xmath112@xmath113 relation @xcite , we estimated the black hole mass to be @xmath114 , and the inferred eddington ratio is 5 . the measurement of @xmath113 is robust considering that it is consistent with the velocity dispersion of gas in the bulge , which is measured to be 90 @xmath2 from the core components of nels . the uncertainty of black hole mass is dominated by the intrinsic scatter of the @xmath112@xmath113 relation , which is 0.5 dex for 1@xmath42 confidence . we also estimated the black hole mass to be @xmath115 using the bolometric luminosity and assuming that the black hole is accreting at eddington accretion rate . the kinetic energy outflow rate of the kpc - scale outflow is @xmath116 of the agn luminosity in sbs 1421 + 511 , and hence is rather strong . the main source driving the outflow can not be starburst because the maximum velocity of the outflow @xmath117 @xmath2 is higher than those driven by starbursts @xcite . it is also unlikely to be the radio jet because the scale of radio jet is one order of magnitude smaller than the size of outflow . ram pressure from agn wind and radiation pressure on the dust are also suggested as the possible origins of the agn driven outflow , and these mechanisms are effective when the agn luminosity is close to eddington luminosity @xcite . thus it is likely that agn wind or radiation pressure drives the kpc - scale outflow in sbs 1421 + 511 . in section 3.2 , we attempted to reproduce the blue - skewed h@xmath5 profiles of sbs 1421 + 511 with accretion disk models , which are the most successful models for double - peaked bels till now . a simple circular disk model of @xcite is unable to interpret the data . this can be easily understood , since a circular disk produces blue and red peaks with similar fluxes while in sbs 1421 + 511 only the blue peak is seen . thus there must be an excess emission to the circular disk which is blueshifted in velocity and contributes to the strong blue peak , and hence the disk is nonaxisymmetric . there are multiple nonaxisymmetric disk models such as elliptical disk @xcite , circular disk with an orbiting bright spot ( e.g. * ? ? ? * ) and circular disk with spiral emissivity perturbations @xcite . the elliptical disk model is first ruled out because it can not reproduce the data . although we did not directly examine other nonaxisymmetric models by fitting the data , we can test them using variability behavior of the bel profile because the excess emission from a nonaxisymmetric structure leads to long - term periodic profile variability . for the bright spot model the period is the dynamic time scale , which can be estimated using the peak velocity of h@xmath5 profile ( @xmath13000 @xmath2 ) and the black hole mass . the period is @xmath1024 yrs and @xmath10220 yrs for black hole masses of @xmath114 and @xmath115 , respectively , and 1.5 and 7 yrs assuming the unknown inclination angle to be 45@xmath118 . for spiral model the period is the precession time scale , which is limited by sound - crossing time scale and is typically several times higher than the dynamic time scale @xcite , and thus is at an order of magnitude of decades for sbs 1421 + 511 . these period values agree with those obtained from monitoring of agns with double - peaked bels , which are months to decades @xcite . on the other hand , the period @xmath119 can be limited observationally using the radial velocity of the excess emission @xmath120 and the velocity change rate @xmath121 : @xmath122 , where i is the inclination angle of the disk and @xmath123 is the phase angle in the disk of the source which produces the excess emission . using the peak velocity of h@xmath5 profile as @xmath120 and 3@xmath42 upper limit of @xmath121 and assuming the unknown angles to be 45@xmath118 , the observed lower limit of the period is @xmath1100 yrs . thus the bright spot model is unlikely and the possibility of spiral model remains open considering the uncertainty in black hole mass , unknown angles and structure of accretion disk . we showed that the data can be well reproduced using circular disk model if applying an ad hoc blueshift with a velocity of 1400 /kms relative to nels . this is unusual since the redshifts of disk models are typically fixed at those of host galaxies in previous research works . the large blueshift velocity can be physically interpreted either as resulting from orbiting around an unseen super - massive black hole companion or as resulting from recoiling at a large velocity . the orbiting model predicts that the blueshift velocity changes periodically and the period is @xmath1021 yr or @xmath1025 yrs using the equation of @xcite , which are inconsistent with the variability behavior of the bel profiles . the inconsistency was supported by long - term variability studies of agns with similar profiles @xcite which found their variability behaviors are also not consistent with orbital motion . in addition to the disk models , there are other models such as binary broad line region and emission from a bipolar outflow ( * ? ? ? * for a review ) . the binary broad line region model is also unlikely for sbs 1423 + 511 due to the disagreement between predicted and observed variability behaviours of the bel profile . although there are observational evidences against the bipolar outflow model in other agns with double - peaked bels ( see section 6.4.2 in eracleous & harpten 2003 ) which means that the model is not universal , this model is possible for sbs 1421 + 511 considering that the quasar has powerful radio jet and is accreting at a high rate , both of which are indicators of strong outflow . according to these discussions , the model that sbs 1421 + 511 hosts a super - massive black hole which is recoiling at a radial velocity of @xmath11400 @xmath2 and is accompanied by a circular disk which emits double - peaked bels can explain all the observed properties . and at the same time other possible models exist such as emissions from a circular disk with spiral emissivity perturbations and emissions from a bipolar outflow . follow - up spectral monitoring of sbs 1421 + 511 is needed to set a better constraint on the velocity change rate . ultraviolet spectra are also needed to obtain the profiles of high ionized bels and reveal possible broad absorption lines to examine the bipolar outflow model . the feature n consists of a seyfert - like component in the center of the companion galaxy and a hii region in the outer region . if we had not found feature s , we were likely to make the conclusion that the seyfert - like component of feature n comes from an obscured agn in the companion and hence sbs 1421 + 511 is a dual agn system . feature s is similar to the seyfert - like component in several aspects including size , velocity difference , emission line fluxes and flux ratios , implying they may have similar nature . furthermore , the high flux ratio of [ oiii]/h@xmath14 of feature s indicates that the quasar radiation is strong enough to illuminate the gas at such a distance , and thus the seyfert - like component may also be illuminated by the quasar . in addition , there are two facts which are unfavourable for an agn in the companion . one is that powerful agns are rare , for example the fraction of galaxies with a [ oiii ] luminosity as high as @xmath107 @xmath124 is less than 5% even in the most massive galaxies @xcite . the other is that the electron density of the seyfert - like component is @xmath102100 @xmath21 , lower than the typical value of nlr which is several @xmath125 @xmath21 . @xcite investigated a sample of seyfert 1 galaxies and found that nearly none of seyferts with bel fwhm(h@xmath14 ) @xmath107 2000 @xmath2 has nlr with such low density , and the fraction in narrow line seyfert 1 galaxies is only @xmath120% . though we can not ruled out the possibility of a dual agn in sbs 1421 + 511 , we should consider other interpretations for the seyfert - like component . one possible interpretation is that the seyfert - like component of feature n also comes from an eelr and just coincides with the companion galaxy . it is reasonable that feature s has an corresponding feature in the opposite direction considering that all possible origins of the eelrs , including tidal features , large scale radio jets and large scale outflows may have bipolar structure . the existence of eelrs with bipolar structure in other quasars , e.g. sdss j1356 + 1026 @xcite , gives weight to this speculation . in addition , the rareness of powerful agn and the low electron density can also be naturally explained . another possible interpretation is that the seyfert - like component of feature n comes from the ism in the companion galaxy illuminated by the quasar @xcite . the companion galaxy located in the ionization cone of a quasar would be ionized and look like a giant hii region for a distant observer if the quasar radiation is powerful enough . in order to test whether the quasar in sbs 1421 + 511 is capable of generating the observed line ratios and [ oiii ] luminosity , we ran a simulation using the photo - ionization code * cloudy * @xcite . we first attempted an oversimplified single - component model with input conditions as follows : 1 . the shape of the quasar spectrum was obtained by interpolating the sed of 3c206 and normalized to the bolometric luminosity of the quasar . the distance @xmath126 was estimated using the projected distance 12.5 kpc and assuming the unknown inclination angle to be the representative value @xmath127 . 3 . the density was assumed to be constant , and the value was set as the mean value of the ism for a disk galaxy of @xmath128 @xmath21 . 4 . the metallicity was assumed to be the solar one . the simulation shows that the quasar radiation can produce a kpc - thick hii region . the predicted line ratio of [ oiii ] @xmath135007/h@xmath14 @xmath1 10 . the [ oiii ] emission of apern comes from a region in a rectangular aperture which is set by 0.75@xmath0 slit and 2@xmath0 width apern , which has a cross - sectional area of @xmath129 . assuming the covering factor of the emitting gas , which depends on the orientation of the disk , is @xmath130 , the model predicts that [ oiii ] luminosity is @xmath131 . as both the [ oiii ] luminosity and the flux ratio of [ oiii]/h@xmath14 are high enough for the observed value , this oversimplified single - component model shows that the quasar radiation is strong enough to illuminated the ism of the companion . however , the model predicted that the line ratio of [ nii ] @xmath136584/h@xmath5 = 0.2 , which is significantly lower than the observed value . we attempted to vary the distance , density and metallicity in reasonable ranges , but failed to reproduce the data , for example a density of @xmath132 @xmath21 and a super - solar metallicity could generate a high enough [ nii]/h@xmath5 ratio but the [ oiii]/h@xmath14 ratio is too low . the inconsistency is common when one attempts to reproduce the spectrum of eelrs using photo - ionization models , and is removed by considering at least two kinds of clouds with significantly different physical properties ( e.g. * ? ? ? for example , though the parameters are not well limited , a double - component model consisting of a low density component with a density of @xmath133 @xmath21 and a solar or sub - solar metallicity , and a high density component with a density of @xmath132 @xmath21 and a super - solar metallicity can produce enough [ nii ] and [ oiii ] at the same time . and the two components could be explained as the warm ionized medium in the disk and high density medium in the galaxy center respectively for a typical galaxy . thus the interpretation that the seyfert - like component comes from gas in the companion illuminated by the quasar is supported by the simulation . this interpretation does not require that the companion galaxy is active and the low electron density of the seyfert - like component is more natural . however the 70 @xmath2 velocity shift between [ oiii ] emission line and the starlight absorption lines of the companion is difficult to be explained . by comparison , the shift can be easily explained if the seyfert - like component comes from an agn because the shift is common among agns or if it comes from an eelr because the companion and the eelr are not necessarily associated . in the survey of dual agns , separated [ oiii ] emission corresponding to the companion of a quasar is frequently used as indicator of an obscured agn in the companion . it is important to make clear whether sbs 1421 + 511 is a dual agn because the conclusion may influence the strategies of survey of dual agns . because [ oiii ] illuminated by an agn of the companion galaxy would concentrate on the center , while that illuminated by quasar would be more extended , [ oiii ] distribution map could be used to determine whether the companion is active . however , we can not make an unambiguous distinction based on esi data taken under 1.4@xmath0 ( @xmath16 kpc ) seeing condition . thus future narrow band imaging or ifs observations with high spatial resolution are required to map the [ oiii ] distribution , hence make a final conclusion . we first ruled out the possibility that the eelrs are generated by large scale radio structures because the radio source is compact and shows no large scale features . since we have seen a kpc - scale outflow from the blue wing of quasar nels , a natural question is that whether it is related with the eelrs . we checked whether the kpc - scale outflow can provided enough material to form a structure like feature s. the total mass of feature s is estimated to be @xmath134 using the electron density from [ sii ] doublet and h@xmath14 luminosity . however , the situation may be more complicated because a single - component model either can not reproduce the observed line ratios for feature s as what happened for the seyfert - like component of feature n : a similar * cloudy * simulation shows that a cloud with a density of @xmath135 @xmath21 would not reach a high enough ionization state to produce as much [ oiii ] relative to h@xmath14 as observed , and again lower densities which can produce enough [ oiii ] fail to produce as much [ nii ] and [ sii ] as observed . thus a double - component model is also needed for feature s and the electron density from [ sii ] doublet only applies for the high density component since [ sii ] emission mainly comes from it . if adopting this model , the total mass would be higher than what we estimated previously because the low density component would be much more massive assuming that the two components contribute comparable h@xmath14 . for example , @xcite showed that the eelr of quasar 4c 37.43 is better modeled as a two - phase medium , consisting of a matter - bounded diffuse component ( @xmath136 @xmath21 ) and an embedded small , dense component ( @xmath137 @xmath21 ) , and the fractions of the flux of h@xmath14 from the two components are @xmath11/3 and @xmath12/3 , respectively . if this is the case for sbs 1421 + 511 , the mass of low density component would be about two orders of magnitude higher than that of the high density component , and the total mass would be @xmath138 . the dynamic time scale of an outflow capable of producing an emission line feature like feature s is @xmath139 yr , which is estimated using the size and mean radial velocity of feature s. considering the mass rate of the kpc - scale outflow and the dynamic time scale , the blue wing outflow can provide enough material to form an eelr feature like feature s. however the model has a disadvantage : a cloud with a density of @xmath132 @xmath21 and a temperature of @xmath140 k in circum - galactic environment is unstable and would expand due to thermal pressure . for example , the pressure is @xmath132 higher than that of surrounding hot ionized medium , which has a typical density of @xmath141 @xmath21 and a temperature of @xmath142 k. the gas is also unlikely to be gravitational condensed because the minimum mass for a cloud to reach a equilibrium between gravity and thermal pressure is @xmath143 , which is unlikely to be true . using the strmgren radius of @xmath144 pc estimated from ionization parameter and density of feature s , and the sound speed @xmath145 @xmath2 , the life time of a high density cloud is estimated to be @xmath140 yr , far shorter than the dynamic time scale of an outflow which is expected to form feature s. thus there must be a mechanism to continuously regenerate the observed high density clouds if feature s indeed comes from the outflow . considering that sbs 1421 + 511 is likely to be a merging system , an alternative possible origin is that the eelrs are tidal tails which are illuminated by quasar . apart from the suggestion by the close projected distance and little velocity difference , there are also two pieces of evidence of strong interactions between the two galaxies . one is that the residual of sdss 2-d spatial decomposition ( figure 3(f ) ) shows a feature in northwest direction from the center of the companion galaxy . the other is that the companion galaxy shows an off - center hii region , which is rare among isolated galaxies and can be naturally interpreted as a tidal tail of the companion galaxy . one can imagine the natures of the two emission line features of sbs 1421 + 511 according to nearby mergers : feature n comes from the bridges between the two galaxies and the tidal tails of the quasar companion , and feature s comes from the tidal tails of the quasar host . thus the two features can be understand in a unified scenario . since eelrs associated with tidal features generally show correspondence between the continuum and emission - line morphologies while those ejected by outflow unnecessarily do , the distribution maps of emission line and continuum may help distinguish the two possible origins , tidal or outflow . however we can not make a detailed analysis based on keck data taken under 1.4@xmath0 seeing condition . new observations with higher spatial resolution are required , which would map the structure and dynamic of the system , and thus help to clarify the origin of the eelr in sbs 1421 + 511 . we selected sbs 1421 + 511 from sdss dr7 quasar catalog as a recoiling super - massive black hole candidate for that the bels show blue - skewed profile with a peak velocity of @xmath13300 @xmath2 blueshifted relative to nels and that there is offset between the quasar and the nearby extended source . we presented results of keck esi long slit spectrographic observation of sbs 1421 + 511 . by applying spectral and spatial decomposition technics , we identified an host galaxy at the quasar position and a companion galaxy 3@xmath0 north to the quasar with velocity difference of 120 @xmath2 relative to the host galaxy . thus the off - nuclear agn character is not due to a recoiling black hole , but due to a close companion of the quasar . we fitted the profile of h@xmath5 bel using disk models , which are the most successful model for agns with similar profiles . a circular disk model can well fit the data only if applying an ad hoc blueshift of @xmath11400 @xmath2 , and the elliptical disk model is ruled out . we detected no change of the peak velocity of the profile among mdm , sdss and keck observations , setting a 3@xmath42 upper limit of the velocity change rate of 14 @xmath2per year , and thus the period of the velocity change is @xmath107200 yrs . the non - detection of velocity change is inconsistent with the three models : emission from a circular disk with an orbiting bright spot , emission from a circular disk which orbiting around an unseen companion and emission from binary broad line regions . remained are three models : emission from a circular disk which is accompanied with a super - massive black hole recoiling at a velocity of @xmath11400 @xmath2 , emission from a circular disk with spiral emissivity perturbations and emission from a bipolar outflow , and the first is the most likely . thus sbs 1421 + 511 may hold a supermassive recoiling black hole . the companion galaxy shows seyfert - like line ratios and strong [ oiii ] emission with a luminosity of @xmath3 , suggesting that it may also be an agn and hence sbs 1421 + 511 is a dual agn system . intriguingly , an eelr is detected in the opposite direction of the companion southern to the quasar and thus the companion galaxy may only appear to be active . there are two possible interpretations . one is that the eelr has a bipolar structure and the northern half just coincide with the companion galaxy . the other is that the gas in the companion is illuminated by the quasar other than its own agn , which is supported by a simulation with photo - ionization model . the unsolved mysteries in sbs 1421 + 511 need follow - up observations . future optical spectral monitoring and ultraviolet spectrometry are needed to make a better understand on the nature of the bel of sbs 1421 + 511 and test whether it host a super - massive recoiling black hole . future narrow line imaging or ifs observations with sub - arcsec ( @xmath1021 kpc ) spatial resolutions are required to reveal possible merging features to determine whether the two galaxies are physically connected , and to map the [ oiii ] brightness distribution to tell whether the companion is active . 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 . we are grateful to the anonymous referee for his / her helpful comments and for pointing out possibilities of the emission line features . we also would like to thank j. x. prochaska for helping us learn keck observations and data reduction using his program , and thank g. li for offering language assistant . this work is supported by the nsfc grant ( 11421303 , 11473025 , nsf11033007 ) , national basic research program of china ( 973 program , 2013cb834905 ) , the soc program ( chinare2012 - 02 - 03 ) , and fundamental research funds for the central universities ( wk 2030220010 ) . brzan , l. , mcnamara , b. r. , nulsen , p. e. j. , et al . 2008 , apj , 686 , 859 bogdanovi , t. , eracleous , m. , & sigurdsson , s. 2009 , apj , 697 , 288 bruzual , g. , & charlot , s. 2003 , mnras , 344 , 1000 campanelli , m. , lousto , c. , zlochower , y. , et al . 2007 , apj , 659 , l5 chen , k. , halpern , j. p. , & filippenko , a. v. 1989 , apj , 339 , 742 elvis , m. , wilkes , b. j. , mcdowell , j. c. , et al . 1994 , apjs , 95 , 1 eracleous , m. , livio , m. , halpern , j. p. , et al . 1995 , apj , 438 , 610 eracleous , m. , halpern , j. p. , m. gilbert , a. , et al . 1997 , apj , 490 , 216 eracleous , m. , & halpern , j. p. , 2003 , apj , 599 , 886 eracleous , m. , boroson , t. a. , halpern , jules p. , et al . 2012 , apjs , 201 , 23 fabian , a. c. 2012 , ara&a , 50 , 455 ferland , g. j. , korista , k. t. , verner , d. a. , et al . 1998 , pasp , 110 , 761 filippenko , a. v. 1985 , aj , 90 , 1172 fu , h. , & stockton , a. 2009 , apj , 690 , 953 gezari , s. , halpern , j. p. , & eracleous , m. , 2007 , apjs , 169 , 167 greene , j. e. , & ho , luis c. 2006 , apj , 641 , l21 greene , j. e. , zakamska , n. l. , & smith , p. s. 2012 , apj , 746 , 86 helmboldt , j. f. , taylor , g. b. , tremblay , s. , et al . 2007 , apj , 658 , 203 kauffmann , g. , heckman , t. m. , tremonti , c. , et al . 2003 , mnras , 346 , 1055 kewley , l. j. , groves , b. , kauffmann , g. , et al . 2006 , mnras , 372 , 961 komossa , s. , burwitz , v. , hasinger , g. , et al . 2003 , apj , 582 , l15 komossa , s. , xu , d .- w . , zhou , h .- y . , et al . , 2008 , apj , 680 , 926 lousto , c. o. , zlochower , y. , dotti , m. , et al . 2012 , phrvd , 85 . l4015 lewis , k. t. , eracleous , m. , & storchi - bergmann , t. , 2010 , apjs , 187 , 416 mcgurk , r. c. , max , c. e. , rosario , d. j. , et al . 2011 , apj , 738 , 2 merloni , a. , & heinz , s. 2007 , mnras , 381 , 589 newman , j. a. , eracleous , m. , filippenko , a. v. , et al . 1997 , apj , 485 , 570 osterbrock , d. e. , & ferland , g. j. 2006 , astrophysics of gaseous nebulae and active galactic nuclei , second edition peng , c. y. , ho , l. c. , impey , c. d. , et al . 2010 , aj , 139 , 2097 peterson , b. m. 1997 , an introduction to active galactic nuclei , cambridge univ . press , cambridge rupke , d. s. , veilleux , s. , & sanders , d. b. , 2005 , apjs , 160 , 115 schneider , d. p. , hall , p. b. , richards , g. t , et al . , 2007 , aj , 134 , 102 shields , g. a. , rosario , d. j. , smith , k. l. , et al . 2009 , apj , 707 , 936 shih , h .- y . , & stockton , a. 2014 , apj , 794 , 117 stockton , a. , & mackenty , j. w. 1983 , natur , 305 , 678 stockton , a. , mackenty , j. w. , hu , e. m. , et al . 2002 , apj , 572 , 735 storchi - bergmann , t. , nemmen da silva , r. , eracleous , m. , et al . 2003 , apj , 598 , 956 storey , p. j. , & hummer , d. g. 1995 , mnras , 272 , 41 van dokkum , p. g. 2001 , pasp , 113 , 1420 vanden berk , d. e. , richards , g. t. , bauer , a. , et al . 2001 , aj , 122 , 549 viegas , s. m. , & prieto , m. a. 1992 , mnras , 258 , 483 vivek , m. , srianand , r. , noterdaeme , p. , et al . 2009 , mnras , 400 , 6 volonteri , m. , & madau , p. 2008 , apj , 687 , l57 wagner , a. y. , bicknell , g. v. , & umemura , m. 2012 , apj , 757 , 136 wang , j .- m . , chen , y .- m . , hu , c. , et al . 2009 , apj , 705 , 76 xu , d .- w . , komossa , s. , zhou , h .- y . , et al . 2007 , apj , 670 , 60 cccc & [ oiii]/h@xmath14 & [ nii]/h@xmath5 & [ sii]/h@xmath5 + + aperhii & @xmath146 & @xmath147 & @xmath148 + aperhii & @xmath149 & @xmath150 & @xmath151 + hii - like set & @xmath152 & @xmath153 & @xmath154 + + apern & @xmath155 & @xmath156 & @xmath157 + agn - like set & @xmath158 & @xmath159 & @xmath160 + + apers & @xmath161 & @xmath162 & @xmath163 +

the weak energy condition of general relativity , the requirement that there be no negative energy densities , is sufficient to prevent the appearance of exotic features such as compactly generated closed timelike curves @xcite and superluminal travel @xcite . quantum field theory appears to violate this condition , however . one example is the standard casimir system of parallel plates , for which there is a negative energy density between the plates . however , this is an idealized system , which assumes a perfect conductor with an infinitely sharp and flat edge . a real material will have a rough surface at the atomic scale and will also appear transparent to very high energy modes . since the casimir energy is a sum over all energies , it will always include modes for which these effects are relevant , so this idealization could affect the value of the sum . in addition , since the boundary condition is imposed externally , there is no measure of the energy that would be required to maintain it . although there are other ways to produce negative energy densities , for example a superposition of states with zero and two photons , these cases are constrained by averaged energy conditions , which require the energy to be positive when averaged along an entire geodesic . they are also constrained by quantum inequalities @xcite , which limit the total negative energy that can exist when averaging over a certain period of time . thus it is important to understand problems of the casimir type if we want to know whether quantum field theory protects general relativity against negative energies . in this paper , we reconsider the question of the energy density in such systems . to avoid the subtleties associated with the casimir problem , we consider a quantum field in the presence of a background potential ( i.e. , a field with a mass that depends on position ) . in such an approach , one can choose a potential that depends only on one spatial dimension , and simulate the parallel plates in the casimir problem . by generalizing the approach of @xcite to local densities , we can precisely cancel the divergences in the calculation in a definite renormalization scheme . @xcite considered similar problems for the special case of reflectionless potentials , such as the potentials for the supersymmetric kink and sine - gordon models . here we present a general approach suitable for numerical computation , in addition to analytic calculations in exactly solvable models @xcite . these techniques are also useful for the study of casimir forces and stresses @xcite . to illustrate our method , we will first consider a simple model . we take a real , massless scalar field in 2 + 1 dimensions in the background of a repulsive potential @xmath1 that depends on one spatial dimension but not the other . we start with the hamiltonian density = 12and expand the field @xmath2 in terms of small oscillations , giving ( x , y , t ) = _ -^ _ 0^_=+ , - ( _ k^(x ) e^ipy e^-i(p ) ta_k , p^+_k^(x)^e^-ipy e^i(p ) ta^_k , p^ ) . where @xmath3 . the @xmath4 are normal mode wave functions , which can be taken to be real . the sum is over the symmetric mode , @xmath5 , and the antisymmetric mode , @xmath6 . they satisfy [ eqn : eom ] -^_k(x ) + v(x)_k^(x ) = k^2_k^(x ) where prime denotes differentiation with respect to @xmath7 , with the normalization _ -^dx _ k^+(x ) _ k^-(x ) = 0 _ = + , - _ -^dx _ k^(x ) = 2(k - k ) , which gives @xmath8 and @xmath9 as the solutions in the free case , @xmath10 . the energy density is then = _ -^ _ 0^_=+ , - 12 . we write (x)^2 = 12d^2dx^2 ( x)^2-(x)(x ) and then use eq . ( [ eqn : eom ] ) to obtain = _ -^ _ 0^_=+ , - . the integral is highly divergent , but by using dimensional regularization and introducing counterterms into the hamiltonian , as discussed in sec . [ sec : method ] , we can render the integral finite . we then integrate out the transverse modes , giving _ = - _ 0^ _ = + , - \{&&^n+1 & & + n+14^n-1d^2dx^2(_k^(x)^2 ) } , where @xmath11 is the number of transverse dimensions , later to be set to 1 . the hamiltonian has been renormalized by subtracting a constant term for the vacuum energy and a term proportional to the potential , which is sufficient to render it finite in 2 + 1 dimensions . we can relate the norms of the mode wavefunctions to the green s function by ( see for example @xcite ) _ = + , -_k^(x)^2 = 2k g(x , x , k ) , where @xmath12 is the green s function , which satisfies -g(x , x,k ) + v(x)g(x , x,k ) - k^2g(x , x,k ) = ( x - x ) and has only outgoing waves ( @xmath13 ) at infinity . the green s function has the symmetry property @xmath14 , so for real @xmath15 we can write _ = + , -_k^(x)^2 = kig(x , x , k)-ki g ( x , x , -k ) . thus we can compute the energy by extending the range of integration to @xmath16 , and using @xmath17 , _ = - _ -^^n+1 dk , [ eqn : contourextend ] where , since we are taking the massless limit , @xmath18 . next we would like to convert this expression into a contour integral by closing the contour at infinity in the upper half plane . the contour at infinity does not contribute , because for large , positive @xmath19 , g(x , x , k ) 1+v(x)2k^2 + o ( k^-4 ) . singularities in the green s function in the upper half plane correspond to normalizable eigenfunctions of the hamiltonian , which represent bound states . since the hamiltonian is hermitian , the bound states must have real energies , so the singularities must lie on the imaginary axis and have @xmath20 where @xmath21 is the mass . in this example , we have a repulsive potential and a massless particle , each of which is sufficient to ensure that there are no bound states at all . thus the green s function has no singularities for @xmath22 , and the only contribution to the integral comes from the branch cut along the positive imaginary axis coming from @xmath23 . integrating around the branch cut and using eqs . ( [ id1 ] ) and ( [ id2 ] ) below gives _ = - _ 0^^n+1 d and then setting @xmath24 gives [ eqn : m = n=1special ] _ [ eqn : greenn=1 ] once one has computed the green s function , this integral is straightforward , though it may be necessary to resort to numerical techniques . we show the calculation for some example potentials in sec . [ sec : examples1 ] below . we will now consider the more general case of a real scalar field of mass @xmath21 in the background of a potential that is spherically symmetric in @xmath25 nontrivial coordinates , which we label by @xmath7 , and independent of the remaining @xmath11 trivial coordinates , which we label by @xmath26 . the energy density is = ( ^2 + ( ) ^2 + v(r ) ^2 + ^2 ^2 ) = ( ^2 + ^2(^2 ) - ^2 + v(r)^2 + ^2 ^2 ) , where @xmath27 . decomposing the quantum field @xmath2 in terms of modes gives ( r,,t ) & = & _ , _ z & & ( _ j ( ^_j(r)^y^m__z()^e^-ipy e^i_j^(p ) t a_j , p^_z^+ ^_j(r ) y^m__z ( ) e^ipy e^-i_j^(p ) t a_j , p^_z ) . & & + . _ 0^ ( ^_k(r)^y^m__z()^e^-ipy e^i(p ) t a_k , p^_z^+ ^_k(r ) y^m__z ( ) e^ipy e^-i(p ) t a_k , p^_z ) ) , where @xmath28 , the bound state energies are @xmath29 , and the sum over @xmath30 gives the partial wave expansion in the @xmath25 nontrivial dimensions . the degeneracy factor @xmath31 in each partial wave is given by the dimension of the space of symmetric tensors with @xmath30 indices , each running from @xmath32 to @xmath25 , with all traces ( contractions ) removed @xcite . by the symmetry of the indices , this dimension is given by the number of ways to make @xmath30 indices out of 0 or more 1 s , 0 or more 2 s , and so on , which is the number of distinct ways to place @xmath33 dividers into @xmath34 slots . removing all the traces requires subtracting the same quantity with @xmath30 replaced by @xmath35 . we thus obtain d^m_= - = ( m+2 - 2 ) . the wavefunctions @xmath36 are the eigenstates of the time - independent radial schrdinger equation ( - - + + v(r ) ) ^_k(r ) = k^2 ^_k(r ) , which in general comprise both bound and scattering states . the wavefunctions and creation and annihilation operators are normalized as follows . for the spherical harmonics , y^m__z()^y^m__z ( ) d= _ _ _ z _z , for continuum states , _ 0^r^m-1 ^_k(r)^^_k(r ) dr & = & ( k - k ) = [ a_k , p^_z , a_k,p^_z ] = 0 & & [ a_k , p^_z , a_k,p^_z^ ] = ( k - k ) ( p - p ) _ _ _ z _ z , and for bound states _ 0^r^m-1 ^_j(r ) ^_j(r ) dr & = & _ j j _ = [ a_j , p^_z , a_j,p^_z ] = 0 & & [ a_j , p^_z , a_j,p^_z^ ] = _ j j ( p - p ) _ _ _ z _ z . using these expressions , we obtain the vacuum expectation value of the hamiltonian , & = & _ d^m _ , where @xmath37 is the radial laplacian . for positive integer @xmath25 and @xmath11 , this quantity diverges , as we expect since we have not yet included the contribution of the counterterms . therefore we will calculate the result using analytic continuation in @xmath25 and @xmath11 from values where it is convergent . after introducing counterterms also depending on @xmath25 and @xmath11 , we will then let the dimensions go to their physical values while holding the renormalization conditions fixed . the first counterterm we will introduce renormalizes the cosmological constant . it is simply an overall constant in the hamiltonian , and is fixed by the renormalization condition that the energy density of the trivial background @xmath38 is zero . the free wavefunctions are given by _ k^^(0)(r ) = j_+-1(kr ) , which , by the bessel function identity _ = 0^ j_q+(z)^2 = ( ) ^2q , [ besselid ] satisfy the completeness relation _ d^m_|_k^^(0)(r)|^2 = , [ eqn : complete ] independent of @xmath39 . subtracting the energy in the trivial background we have - _ 0 & = & _ d^m _ . [ eqn : nolevsub ] by completeness , in each partial wave we have _ j |^_j(r)|^2 + _ 0^ ( |^_k(r)|^2 - |^_k^(0)(r)|^2 ) = 0 , so we can implement a local version of the levinson s theorem subtraction used in @xcite , giving - _ 0 & = & _ d^m _ . [ eqn : yeslevsub ] this subtraction is necessary to avoid the appearance of spurious infrared singularities in calculations in one space dimension . these singularities also appear in dimensions less than one , which we will need to consider as part of the dimensional regularization process . next we carry out the @xmath40 integral , using ( ) ^a = _ 0^p^n-1 dp ( ) ^a = , where we have done the integral by analytic continuation from values of @xmath41 and @xmath11 where it converges . we thus obtain - _ 0 & = & - _ d^m _ , where @xmath42 for the scattering states and @xmath43 for the bound states with @xmath44 . next , we must include the contribution of the counterterm proportional to @xmath45 , which is introduced to cancel the tadpole graph . in dimensional regularization , the contribution to the hamiltonian from this counterterm is _ 1 = ^m+n-1 v(r ) , so by using _ 0^ ( ^n+1-^n+1 ) k^m-3 = ^m+n-1 and eq . ( [ eqn : complete ] ) we have _ 1 = - _ d^m__0^ ( ^n+1- ^n+1)(2-m ) |^_k^(0)(r)|^2 so that _ & & - _ 0 - _ 1 & = & - _ d^m_[eqn : e1sub ] . we will use |^_k(r)|^2 = 2 k g_(r , r , k ) , [ eqn : normtogreen ] where the green s function is defined by -d_r^2 g_(r , r,k ) + ( v(r ) + - k^2 ) g_(r , r,k ) = ^(m)(r - r ) with the boundary conditions that it is regular at the origin and and has only outgoing waves ( @xmath46 ) at infinity . using @xmath47 , we can rewrite eq . ( [ eqn : e1sub ] ) as _ & & - _ 0 - _ 1 & = & - _ d^m_[eqn : e1subgreen ] . these subtractions are sufficient to render the theory finite for @xmath48 . however , it appears that if we set @xmath24 , the gamma function will cause eq . ( [ eqn : e1subgreen ] ) to diverge . in fact , as we will see in appendix [ sec : localsum ] , it does not , because the quantity in brackets vanishes . but here we will keep @xmath11 general and instead close the contour of integration in the upper half @xmath15 plane . for sufficiently small @xmath11 the contour at infinity does not contribute . there is a pole for each bound state at @xmath49 , where @xmath50 is the bound state energy , and the contributions from these poles exactly cancel the sum over bound states @xcite in eq . ( [ eqn : e1subgreen ] ) . thus the final result is just given by the contribution from the branch cut along the imaginary axis from @xmath21 to @xmath51 resulting from @xmath23 , which contributes ^n+1 ( i^n+1 - ( -i)^n+1 ) = 2i^n+1 , [ id1 ] where @xmath52 and @xmath53 . then using the identity z = - [ id2 ] we have _ & & - _ 0 - _ 1 & = & - _ d^m__^d 2^n+1 . [ eqn : e1subimag ] we can now put in integer values of @xmath25 and @xmath11 without any divergence , as long as @xmath48 . ( [ eqn : e1subimag ] ) can also be efficiently evaluated numerically @xcite . when we have @xmath54 space dimensions , we will need to introduce a second counterterm , @xmath55 . the first subtraction is particularly easy to define because there is a natural scheme , specified by the complete cancellation of the tadpole graph . higher subtractions require a definition in terms of a renormalization scale , which can be chosen arbitrarily . in choosing this scale , we must be able to relate it to physical inputs , such as masses and coupling constants , in order to define a predictive theory . to define the counterterm precisely , we consider the two - point function @xmath56 in dimensional regularization . it diverges as we approach the physical dimension . the divergence is canceled by the contribution of the counterterm to the two - point function , which is just @xmath57 . we define the renormalization scale @xmath58 by taking @xmath59 . with this definition , we have c & = & _ 0 ^ 1 d & = & _ 0^ dq , [ eqn : cterm ] where @xmath60 is the total momentum and we have integrated over the feynman parameter @xmath61 and the loop energy @xmath62 . typically we will choose @xmath63 , except in the case of massless theories , where to avoid infrared singularities we will choose a spacelike renormalization point @xmath64 . this regulated expression is defined precisely as an analytic function of the dimension . our goal is now to rewrite it in a way that allows us to incorporate it into our expression for the energy , eq.([eqn : e1subimag ] ) , which is also given as an analytic function of the dimension . we express eq . ( [ eqn : complete ] ) in terms of green s functions and analytically continue to express eq . ( [ eqn : cterm ] ) as c & = & _ ^^n+1 f(,m ) d & = & _ d^m__^^n+1 f(,m ) 2g_^(0)(r , r , i ) d , [ eqn : ctermrot ] where @xmath65 is given in terms of the hypergeometric function as f(,m ) = _ 2 f_1(,3 - , , ) [ eqn : fkappa ] as we show in appendix [ sec : fderivation ] . since we will eventually take the limit where @xmath25 becomes an integer , we note that f ( , m ) = & & f ( , m ) = ( 1 + ) & & f ( , m ) = & & [ eqn : fkappa1 ] eq . ( [ eqn : ctermrot ] ) is now in a form where we can include it under the integral sign in eq . ( [ eqn : e1subimag ] ) and obtain _ & & - _ 0 - _ 1 - _ 2 & = & - _ d^m__^d 2 . [ eqn : e2subimag ] before we can take the limit where @xmath54 , however , there is one more potential divergence in eq . ( [ eqn : e2subimag ] ) . our subtraction has cancelled the terms of order @xmath32 , @xmath66 and @xmath67 in the large-@xmath68 expansion of the norm of the wavefunctions . but there could also be a term of order @xmath69 , which will generate a divergence in this case . in the renormalization of the composite operator @xmath70 , we have a renormalization counterterm @xmath71 @xcite . since we are considering just @xmath72 here , this counterterm becomes @xmath73 , exactly the form needed to cancel the remaining divergence . in the original lagrangian , the divergent term would have cancelled automatically between the bulk term and the surface term and no renormalization would be necessary . however , conformally coupled theories have _ classical _ violations of the energy conditions @xcite . ] we fix this counterterm by subtracting the tadpole diagram with the composite operator carrying momentum @xmath74 . aside from this change , it is analogous to the tadpole subtraction above , with @xmath75 replaced by @xmath76 . the scale @xmath77 is then specified through the renormalization condition on the composite operator ( and would typically be chosen equal to @xmath58 ) . as with @xmath58 , a massless theory will require spacelike @xmath78 , while in a massive theory we may set @xmath79 . thus we obtain the contribution _ 2 = - _ d^m__^d ^n-1 2g_^(0)(r , r , i ) ( 2-m ) . this term is a total derivative , so it does not contribute to the total energy . we can split the contribution of this term beween the bulk and derivative terms so that it renders them both separately finite at integer dimensions , giving _ & = & - _ d^m__^d 2 . by eq . ( [ eqn : complete ] ) , the sum over @xmath30 of the free green s function weighted by the degeneracy factor is independent of @xmath39 , so we can pull the derivative outside in the last line , giving _ & = & - _ d^m__^d 2 . in this form , we see explicitly that the subtraction has cancelled the leading terms for large @xmath68 : in the surface term , inside the derivative we have implemented the same subtraction of the leading behavior of the green s function as we found for the tadpole graph in the bulk term ; in the bulk term , we have subtracted the leading term proportional to @xmath76 ( which is derived explicitly for @xmath80 in eq.([eqn : asymp1d ] ) ) . for the purposes of calculation , however , it is often easier to work with the combined expression _ & = & - _ d^m__^^n+1 d. [ e2subimag2 ] to illustrate our method , we would like to carry out some sample calculations in the case of @xmath80 and @xmath24 . since we are in @xmath81 dimensions , we only need one subtraction , proportional to @xmath82 . for @xmath80 , the sum over partial waves reduces to a sum over the symmetric and antisymmetric channels . the free wavefunctions thus become ^+_k ^(0)(x ) = kx ^-_k ^(0)(x ) = kx , so that |^+_k^(0)(x)|^2 + |^-_k^(0)(x)|^2 = 1 . we can sum over the two modes to get the overall green s function , g(x , x , k)=g^+(x , x , k)+g^-(x , x , k ) , with g^(0)(x , x , k)=. then from eq . ( [ eqn : e1subimag ] ) we have [ eqn : m=1 ] _ = - _ ^d ^n+1 and for @xmath24 we have _ = - _ ^d ^2 , which reduces to eq . ( [ eqn : m = n=1special ] ) when @xmath83 . for simplicity , we will consider the massless case for the remainder of this section . we next consider a potential that vanishes for all @xmath84 and calculate the energy density in this region . in this case , the only counterterm is the vacuum energy , and we get _ = -18_0^d . the green s function for @xmath85 is g(x , x , k ) = i2k(e^-ikx_<+r(k)e^ikx_<)e^ikx _ > , where @xmath86 is the reflection amplitude . thus [ eqn : goutside ] 2 g ( x , x , i ) = 1+r(i ) e^-2x and [ eqn : outside ] _ = 18_0^d^2 r(i ) e^-2x . in the large @xmath7 limit , only small @xmath68 contribute in the integral . as a result , the integral depends only on @xmath87 ( at @xmath88 we always have perfect reflection ) , so we can approximate [ eqn : outside - limit ] _ -18_0^d^2 e^-2x = -132x^3 . next we consider a square barrier with @xmath89 for @xmath90 and @xmath91 otherwise . in this case , we can compute the normal mode wave functions in closed form , but must do a numerical integration at the end . outside the barrier , the energy is given by eq . ( [ eqn : outside ] ) with [ eqn : r - square ] r = -v_0e^2a2 a2+ ( ^2+^2)2 a and @xmath92 . thus , outside the barrier , the energy is _ & = & -v_08_0^d ^2 e^-2(x- a)2 a2+ ( ^2+^2)2 a & = & -v_0 8a_0^dq q^2 e^-2q ( y- 1)2q2qq + ( q^2+q ^2)2q , where we have defined the dimensionless quantites @xmath93 , @xmath94 , and @xmath95 where @xmath96 . note that the integrand can not be less than 0 , so the energy outside the barrier is always negative . far from the potential , specifically where @xmath97 and @xmath98 , the contribution comes primarily from @xmath99 , and thus @xmath100 . the integral is then _ 0^ = 14 ( y-1)^3 v and [ eqn : outside - square - limit ] _ -132(x - a)^3 in agreement with eq . ( [ eqn : outside - limit ] ) . close to the potential , specifically when @xmath101 and @xmath102 , the contribution comes mostly from @xmath103 and @xmath104 . thus @xmath105 and @xmath106 . the integral becomes and _ -v_064(x- a ) . inside the barrier , we have g ( x , x , k ) = ik ( k k ( x_<+a ) - ikk ( x_<+a ) ) ( k k ( x_>-a ) + ikk ( x_>-a))2kk 2k a - i ( k^2+k ^2)2k a , where @xmath107 , so we can write g ( x , x , i ) = 12(^2+^2)2 a + 22 a+v_02 x22 a+ ( ^2+^2)2 a . we can then split the energy into two parts , _ = e_0+e_1 ( x ) , where @xmath108 depends on position , but @xmath109 does not . the position - independent part is e_0 & = & -18_0^d \ { ^3(^2+^2)2 a + 22 a 22 a+ ( ^2+^2)2 a- ^2+v_02}&= & -v_0 ^ 28_0^d & = & -v_0 ^ 2 a8_0^dq and is always negative . in the limit where @xmath110 , we can approximate @xmath111 to get e_0 = -v_0 ^ 3/212 . the position - dependent part is [ eqn : e1q ] e_1 ( x ) & = & 18_0^d (2^2-^2)v_02 x 2 2 a+ ( ^2+^2)2a & = & v_0 8a_0^ and is always positive . note that the the dominant term in the integrand in eq.([eqn : e1q ] ) is suppressed by @xmath112 . thus , far from the edge of the potential , where @xmath113 , @xmath108 is negligible . close to the edge of the potential , with @xmath114 and @xmath115 , the integral is @xmath116 and e_1 ( x ) -v_064(a - x ) which cancels , in a principal value sense , the divergence outside the barrier . the sign of the energy density at the center of the barrier depends on the competition between the position - dependent and position - independent parts . for large @xmath117 , the position - dependent part is suppressed in the center , and the energy density is negative . for small @xmath117 , it is positive . the total energy density is shown for several values of @xmath117 in fig . [ fig : square ] . it has long been known that the energy density near a perfectly reflecting boundary is zero if one uses the `` conformally coupled '' stress - energy tensor , but diverges if one uses the minimally coupled one , as we have done above . kennedy , critchley , and dowker @xcite argue that since the total energy is the same in the two cases , there must be a surface energy associated with the perfect conductor in the minimal case . ford and svaiter @xcite found that the surface energy could be seen by allowing the boundary to fluctuate . here , we can see the situation by approximating a perfect conductor by a square barrier with @xmath41 fixed and @xmath118 . we can produce the conformal hamiltonian density by including half the value of the total derivative term , _ = ^2 + ^2(^2 ) - ^2 + v(r)^2 . this choice gives zero energy outside the barrier and removes the divergence of the energy density everywhere inside . with the minimally coupled hamiltonian , the energy outside goes to eq.([eqn : outside - square - limit ] ) as @xmath119 becomes large , while the positive energy inside clusters ever closer to the boundary , as shown in fig . [ fig : square ] . since the change to the total derivative term does not affect the total energy , we can see that the `` surface energy '' located just inside the boundary cancels the divergent negative energy outside . to carry out calculations in @xmath120 dimensions , we now need subtractions proportional to @xmath82 , @xmath121 , and @xmath122 . we will use the renormalization scheme defined in section [ sec : method ] . for @xmath80 , using eq . ( [ eqn : fkappa1 ] ) and evaluating eq.([e2subimag2 ] ) with @xmath80 and @xmath123 gives _ & = & - _ ^d ^3 , [ eqn : m=1n=2 ] where the green s function has again been summed over the symmetric and antisymmetric channels . again , we will restrict our attention to massless fields for simplicity . the wave functions and green s functions are just as in sec . [ sec : examples1 ] . again , since the potential vanishes , the only counterterm is the vacuum energy . thus eq . ( [ eqn : m=1n=2 ] ) reduces to _ = -112 ^ 2_0^doutside the potential , and so from eq . ( [ eqn : goutside ] ) , _ = _ 0^d ^3r ( i ) e^-2x . in the large @xmath7 limit , we can again take @xmath124 , to get [ eqn : outside - limit3 ] _ = - _ 0^d ^3 e^-2x=- , a well - known result . outside a square barrier with width @xmath41 and height @xmath119 , the reflection coefficient is given by eq . ( [ eqn : r - square ] ) , and the energy is _ & = & -v_06 ^ 2_0^d ^3 e^-2(x- a)2 a2+ ( ^2+^2)2 a & = & -v_0 6 ^ 2 a^2_0^dq q^3 e^-2q ( y- 1)2q2qq + ( q^2+q ^2)2q . far from the potential , we approximate @xmath125 . the integral is _ 0^ = 38 ( y-1)^4 v and [ eqn : outside - square - limit3 ] _ -116 ^ 2 ( x - a)^4 , in agreement with eq . ( [ eqn : outside - limit3 ] ) . close to the potential , we approximate @xmath105 and @xmath126 . the integral becomes and _ -v_096 ^ 2(x- a)^2 . inside the potential , we need the renormalized form , _ = - _ 0^d , where @xmath64 is the spacelike renormalization point . for the square barrier , we get a position - independent part , e_0 & = & -112 ^ 2_0^d\ { ^4(^2+^2)2 a + 22 a 22 a+ ( ^2+^2)2 a & & - ^3+v_02 - } & = & -v_0 ^ 212 ^ 2_0^d\ { - } & = & -v_0 ^ 212 ^ 2_0^dq \ { - } , where @xmath127 and @xmath128 . we can isolate the dependence on the renormalization scale by using _ 0^dq\ { - } = + to obtain e_0 = -v_0 ^ 212 ^ 2 ( + + _ 0^dq \ { - } ) . in the limit where @xmath129 , we can approximate @xmath130 , the integral gives @xmath131 , and we obtain e_0 = , consistent with the result obtained from the effective potential @xcite . the position - dependent part is [ eqn : e1q3 ] e_1 ( x ) & = & 112 ^ 2_0^d ^2(3^2-^2)v_02 x 2 2 a+ ( ^2+^2)2a & = & v_012 ^ 2 a^2_0^dqq^2q ( 3q^2-q^2)2q y 2qq 2q+ ( q^2+q ^2)2q and is always positive . far from the edge of the potential , @xmath108 is negligible . close to the edge , where we can approximate @xmath132 and @xmath133 , the integral becomes @xmath134 , and e_1 ( x ) -v_096 ^ 2(a - x)^2 , which cancels the divergence outside the barrier . these results do not reflect any contribution from the @xmath122 counterterm . in this case it vanishes for all @xmath135 , since the potential is constant . furthermore , the contribution to the total energy from this term is also zero , since it is a total derivative . if we imagine that the square barrier represents the limit in which a smooth potential gets steeper and steeper , we will find large equal and opposite contributions to the energy localized in the tiny region on both sides of the boundary . as long as we average over larger distance scales , this contribution will always cancel out , so it can be ignored in the square barrier limit . finally , we consider the potential analyzed in @xmath81 dimensions in @xcite , v(x ) = c^2 ^ 2(x / a ) , [ eqn : potential ] which arises frequently in soliton models . it is exactly solvable in terms of associated legendre functions . for @xmath136 with integer @xmath30 it becomes reflectionless . the green s function at coincident points is g(x , x , i)= ( 1 + a + s)(a - s ) p^-a_s ( ( x / a ) ) p^-a_s ( -(x / a ) ) . [ eqn : green ] where @xmath137 is the associated legendre function as defined in @xcite for @xmath138 , and @xmath139 . plugging this into eq.([eqn : m=1n=2 ] ) , we have _ & = & - _ ^d ^3 which can then be computed numerically . figure [ fig : densityplot ] gives this energy density as a function of @xmath7 for particular values of the parameters . we have seen how to address the question of generation of negative energies through quantum fluctuations in the robust language of quantum field theory , where ambiguities associated with idealized boundary conditions are absent . this approach implements standard renormalization procedures and is applicable to generic background potentials that are spherically symmetric in some dimensions and independent of the rest . such potentials typically arise , for example , from topological defects or other extended objects . by using dimensional regularization , we have implemented a precise renormalization scheme , using only local subtractions for both the first- and second - order diagrams . we expect that this general formalism , together with fermion scattering theory in fractional dimensions developed in @xcite , will allow these results to be extended to fermions and gauge fields . in the case of the square barrier , we have recovered the negative energy associated with perfect reflection at large distances from the barrier , and we have seen that the divergent negative energy outside the barrier is canceled by positive energy immediately inside . in a realistic example in which one includes the energy associated with the background potential , such cancellations might lead the averaged null energy condition to be obeyed even though the weak energy condition is violated @xcite . finally , we have calculated the energy density for a smooth background representing a domain wall in @xmath120 dimensions . ref . @xcite has calculated the surface tension for a bosonic @xmath140 kink domain wall ( and also its supersymmetric generalization ) using an on - shell renormalization scheme , in space dimension one through four . in the language of the present paper , this calculation corresponds to the case of @xmath141 and @xmath142 in the potential of section [ sec : sech2 ] , renormalized with @xmath143 , setting @xmath80 and @xmath11 to zero through three . the surface tension is obtained by then integrating this result over the one nontrivial dimension . ( the choice of @xmath77 does not affect this calculation because the total derivative term integrates to zero . ) using the formulae in the present paper to carry out this calculation , we obtain results in agreement with the bosonic calculations in ref.@xcite . we would like to thank j. j. blanco - pillado , ruben cordero and delia perlov for assistance and larry ford , bob jaffe , vishesh khemani , markus quandt , tom roman , marco scandurra , xavier siemens , alexander vilenkin , herbert weigel and an anonymous referee for helpful conversations and suggestions . n. g. was supported in part by the u.s . department of energy ( d.o.e . ) under cooperative research agreement # de - fg03 - 91er40662 . k. d. o. is supported in part by the national science foundation . by comparing the last line of eq . ( [ eqn : cterm ] ) with the first line of eq . ( [ eqn : ctermrot ] ) , we require _ ^^n+1 ^m-1 d= _ 0^ dq , where @xmath145 and @xmath146 . let us change variables on the left from @xmath68 to @xmath147 and on the right from @xmath60 to @xmath148 to get _ 0^l^(n+1)/2 ^m-2 dl = _ 0^ dl with @xmath149 on the left and @xmath150 on the right . we can write [ eqn : goal ] _ 0^dl l^h ( l+^2 ) = _ 0^dl l^j ( l+^2 ) with & = & ( n+1)/2 + & = & ( n+m-2)/2 + h ( x ) & = & + j ( x ) & = & . denote the difference in the exponents as @xmath151 . if @xmath152 is a positive integer , the desired relationship is just integration by parts , and h ( x ) = ( -)^j ( x ) . to extend this formula to non - integer @xmath152 , we write @xmath153 in terms of the hypergeometric function , [ eqn : jf ] j ( x ) = ( ) ^d ( d ) _ 2f_1 ( , d ; ; ) , with @xmath154 . the operator @xmath155 just increments @xmath156 in eq.([eqn : jf ] ) , so we conjecture that the same relationship holds for all @xmath152 , and thus that the desired @xmath157 is given by eq . ( [ eqn : jf ] ) with @xmath158 . one can check that the conjecture is correct by explicitly performing the integrals in eq . ( [ eqn : goal ] ) , which both give ( ^2-)^-1/2 ( - ) _ 2f_1(,- ; ; ) . finally we find f ( , m ) & = & ( ) ( 3- ) ( ) ^3-m/2 ^ 2-m _ 2f_1(,3- ; ; ) & = & _ 2f_1(,3- ; ; ) . we used the analytic properties of the green s function as a mathematical tool , enabling us to carry out calculations efficiently on the imaginary axis . in so doing , we avoided the apparent singularity in the gamma function coefficient of eq . ( [ eqn : e1sub ] ) for odd @xmath11 . nonetheless , this expression should be a valid result , finite for @xmath48 . as in the case of the total energy @xcite , the quantity in brackets must vanish for @xmath24 in each partial wave individually . furthermore , the combination of the first two terms in brackets vanishes separately from the total derivative term . these cancellations depend on a local analog of the sum rules for the phase shift given in @xcite , which we demonstrate below . when similar apparent divergences arise in the calculation of the total energy , they are canceled according to generalizations of levinson s theorem @xcite . for a system with spherical symmetry , in each partial wave @xmath30 these sum rules take the form _ j ( -_j^2)^n + _ 0^k^2n ( _ ( k ) - _ s=1^n _ ^(s)(k ) ) dk = 0 , [ eqn : sumrule ] where the bound states have @xmath159 , @xmath160 is the scattering phase shift , and @xmath161 is the scattering phase shift computed at order @xmath162 in the born approximation . orders in the born approximation for any @xmath163 . however , there are some restrictions on such oversubtractions in the symmetric channel in one dimension @xcite . ] the @xmath164 case gives levinson s theorem . like levinson s theorem , these identities apply to general potentials in scattering theory and hold in each partial wave @xmath30 individually . also like levinson s theorem , they are modified for the case of the symmetric channel in one dimension , as discussed in @xcite . we have a relationship @xcite between the phase shift , the change in the density of states , and the norm of the wavefunction , = _ ( k ) - _ ^(0)(k ) = _ 0^dr r^m-1 ( |_k^(r)|^2 - |_k^^(0)(r)|^2 ) , where the zero superscript indicates a quantity evaluated in the free case . this equation also holds order by order in the born approximation . using these relations we can rewrite eq . ( [ eqn : sumrule ] ) as dr r^m-1 ( _ j ( -_j^2)^n |_j^(r)|^2 + _ 0^k^2n ( |_k^(r)|^2 - _ s=0^n |_k^^(s)(r)|^2 ) dk ) = 0 , [ eqn : intsum ] where @xmath165 is the born approximation to the wavefunction computed at order @xmath162 ( the free wavefunction is the order zero term ) . the identities we need for the present application are simply the slightly stronger condition that eq . ( [ eqn : intsum ] ) holds for each @xmath39 individually , rather than just as an integral . we can exploit the connection to the green s function that was used in @xcite to prove this result as well . the case of @xmath164 is particularly simple , because we know that _ j |_j^(r)|^2 + _ 0^ ( |_k^(r)|^2 - |_k^^(0)(r)|^2 ) dk = 0 [ eqn : completeness ] by completeness ; it is just the difference between the expectation value of a constant computed in the free and interacting bases . ( after summing over the spectrum , each term is independent of @xmath39 . ) for higher @xmath166 , we would like to show that _ j ( -_j^2)^n |_j^(r)|^2 + _ 0^k^2n ( |_k^(r)|^2 - _ s=0^n |_k^^(s)(r)|^2 ) dk is zero . ) when integrated over @xmath15 . it might be possible to find analogous results for the higher sum rules as well . ] we employ the relationship in eq.([eqn : normtogreen ] ) between the norm of the wavefunction and the green s function to rewrite this expression as _ j ( -_j^2)^n |_j^(r)|^2 + _ -^k^2n+1 ( g_(x , x , k ) - _ s=0^n g_^(s)(x , x , k ) ) dk , where we have extended the integral to the entire @xmath15 axis by the symmetry of the integrand . to show this expression is zero , we would like to do the @xmath15 integral as a contour , closed in the upper half plane . the singularities in the full green s function correspond to bound states , and will exactly cancel the explicit contribution from the bound states @xcite . the born approximation has no singularities ( since it does not see the bound states ) . thus we are left with the contour at infinity . however , it does not contribute because we have subtracted enough born approximations to ensure that the integrand falls like @xmath167 at large @xmath168 @xcite . in one dimension , we have to consider the symmetric channel , which can have additional singularities at @xmath88 . such singularities , for example , lead to an extra @xmath169 in levinson s theorem @xcite , relating the phase shift at @xmath88 to the number of bound states . we have _ s(0 ) = ( n_s - ) [ eqn : levs ] as opposed to the usual ( 0 ) = n . [ eqn : leva ] analogously in our problem , eq . ( [ eqn : completeness ] ) must be modified to _ j |_j^s(x)|^2 - + _ 0^ ( @xmath170 is the size of the system . subtracting @xmath171 reflects the contribution from the state @xmath172 in the free spectrum . this state is `` half - bound '' : while any potential will have a @xmath88 state in the symmetric channel , in this case the wavefunction goes to a constant at infinity . such states contribute to the spectrum with half the usual residue for a bound state , as the name indicates . ( generically a state with @xmath88 will approach a line with nonzero slope , in which case no special treatment is necessary . ) if a potential has a half - bound state , making the potential arbitrarily more attractive introduces a new bound state in the theory , and making it arbitrarily more repulsive eliminates the half - bound state . for all @xmath15 . if this equality is to hold at @xmath88 , to reconcile eqs . ( [ eqn : levs ] ) and ( [ eqn : leva ] ) there must be a half - bound state , which contributes only a half to the number of bound states . the half - bound state in the free case ( which is reflectionless ) is just a consequence of this requirement . ] there will be an analogous contribution to the energy density , so this term will cancel when we pass from eq . ( [ eqn : nolevsub ] ) to eq . ( [ eqn : yeslevsub ] ) and the rest of the derivation of the energy density is unchanged . for the other sum rules needed in our problem , however , we always multiply by enough powers of @xmath15 to cancel any anomalous effects coming from states at @xmath88 . we would have to be more careful if we do additional born `` oversubtractions , '' in which case we could encounter additional terms analogous to those found in @xcite . we can always avoid these problems as long as each ultraviolet born subtraction is preceded by a corresponding infrared levinson subtraction . for the first born subtraction , the corresponding levinson subtraction was done using eq . ( [ eqn : completeness ] ) . higher levinson subtractions would use local analog of the higher sum rules in @xcite . for casimir calculations it will be convenient to slightly modify the @xmath173 sum rule . our renormalization procedure subtracts not the full first born approximation , but rather just a local part of it . however , this replacement does not affect the sum rule . for example , to apply the results of section [ sec : onedirectsub ] for @xmath174 , we write _ = + , - |_k(x)|^2 & = & 1 + _ x^dy 2k(y - x ) + & = & 1 + + _ x^dy 2k ( y - x ) + & = & 1 + - _ x^dy 2k ( y - x ) + & = & 1 + - - _ x^dy 2k ( y - x ) + & = & 1 + - + _ x^dy 2k ( y - x ) + [ eqn : asymp1d ] and subtract only the term directly proportional to @xmath82 , rather than all terms that are first order in the strength of the potential . however , the additional terms , proportional to the derivatives of @xmath82 , do not introduce any singularities in the integral and do not affect the contour at infinity because they fall like @xmath175 or faster . therefore , this modification does not affect the proof of the sum rule . this result allows us to apply the sum rule to eq . ( [ eqn : e1sub ] ) . with the sum rules in hand , we can now extract a finite result from eq . ( [ eqn : e1sub ] ) . near @xmath176 we have ( - ) a^n-1 ( 1 + a^2 ) so that in the @xmath177 limit we obtain _ & = & - _ d^m_. [ eqn : sumrulen=1 ] the local sum rule ensures that the scale of the logarithm does not affect the final result . in addition , the limit @xmath178 is smooth ( except when @xmath179 and @xmath80 , where we have the usual infrared divergences of one - dimensional field theory ) . if we extend the range of integration in eq . ( [ eqn : sumrulen=1 ] ) as in eq . ( [ eqn : contourextend ] ) , and then close the contour in the upper half plane , the branch cut associated with @xmath180 will reproduce eq . ( [ eqn : greenn=1 ] ) . a. s. goldhaber , a. litvintsev and p. van nieuwenhuizen , phys . * d67 * ( 2003 ) 105021 . k. d. olum and n. graham , phys . * b554 * ( 2003 ) 175 . n. graham , r.l . jaffe , v. khemani , m. quandt , m. scandurra , and h. weigel , nucl . * b645 * ( 2002 ) 49 . n. graham and r.l . jaffe , nucl . * b544 * ( 1999 ) 432 ;

the ads / cft correspondence @xcite@xcite@xcite implies that type iib string theory in @xmath7 is equivalent to @xmath2 super - yang - mills in four dimensions . in the classical string theory limit , or equivalently in the planar gauge theory limit , integrable structures appear . this has lead to impressive progress in the understanding of this system ( see @xcite for a review ) . the ads / cft dictionary relates the energy of string states in the bulk to the conformal dimensions of operators on the boundary . a set of equations known as the y - system has been put forward in @xcite to solve the spectrum problem of planar @xmath2 sym , or equivalently of string theory in @xmath7 . the goal of the present article is to make a new step towards a definite proof of the validity of this set of equations . there is by now solid evidence in favor of the validity of the y - system . it reproduces the results of the asymptotic bethe ansatz @xcite , but it does not suffer from the same limitations . for instance it contains @xcite@xcite the spectrum of the quasi - classical string at large t hooft coupling ( see e.g. @xcite ) . even more impressively , it lead to correct predictions for the dimension of the konishi operator both at large @xcite and at small @xcite t hooft coupling . in order to claim that the spectrum problem for @xmath2 sym has been definitively solved , it would be comfortable to have a proof of the validity of the y - system . at that point the only known derivation of the y - system relies on the thermodynamic bethe ansatz @xcite ( see e.g. @xcite for a review ) . this approach was studied in @xcite@xcite@xcite . the thermodynamic bethe ansatz has been very successful and lead to numerous remarkable results . however this method relies on several crucial assumptions . in the first place , one has to assume quantum integrability of the theory . then one needs the `` string hypothesis '' : the spectrum of excitations that contribute to the thermodynamic limit of the theory essentially has to be guessed . most importantly , this method only gives the ground state energy . the spectrum of excited state can be obtained by analytic continuation , but the reason why this works is not understood . in this paper , we will initiate a different approach to derive the y - system from first principles . we will use only elementary tools of two dimensional conformal field theory ; in this aspect this article can be related to the seminal work of @xcite where the y - system was derived for the minimal models . we will be able to prove the validity of the y - system up to first non - trivial order at large t hooft coupling . [ [ the - idea - of - the - proof . ] ] the idea of the proof . + + + + + + + + + + + + + + + + + + + + + + up to a change of variables , the y - system can be rewritten as a t - system , also known as the hirota equation : [ tsystem ] _ a , s(u + 1 ) _ a , s(u - 1 ) = _ a+1,s(u+1)_a-1,s(u-1 ) + _ a , s+1(u-1)_a , s-1(u+1 ) in the above equation , @xmath8 is a spectral parameter and the indices @xmath9 are integers that label representations of the global symmetry group of the system . in the case at hand this group is @xmath10 . these labels take values in a t - shaped lattice . more details are given in section [ prooftsys ] . the t - functions are expected to be related to special line operators of the string worldsheet theory , namely the transfer matrices ( see e.g. @xcite ) . the transfer matrices play a central role in the study of classical integrability . indeed these operators code an infinite number of conserved charges . by definition , the transfer matrix is the supertrace of the monodromy matrix . the monodromy matrix itself is an element of the supergroup @xmath10 . thus the classical transfer matrix is a supercharacter . if the shifts of the spectral parameter are neglected , the t - system reduces to a character identity that is known to hold ( see e.g. @xcite@xcite ) . we deduce that the shifts of the spectral parameter come from some kind of quantum effects . in this paper we take the identification the t - functions to the transfer matrices seriously . the t - system is promoted to an operator identity between quantum transfer matrices . we postulate that the product appearing in the t - system is the fusion of line operators . the process of fusion of line operators involve quantum effects , that are responsible for the appearance of the shifts in the t - system . we will show that this picture is indeed correct up to first non - trivial order in the large t hooft coupling expansion . more precisely , we will compute the leading quantum correction in the fusion of two transfer matrices , and show that it correctly gives the shifts of the t - system at first order . the same strategy was previously successfully applied for the sigma - model on the supergroup @xmath11 in @xcite . [ [ organization - of - the - paper . ] ] organization of the paper . + + + + + + + + + + + + + + + + + + + + + + + + + + in section [ worldsheettheory ] we describe the features of the pure spinor string on @xmath3 that are relevant for our purposes . we also introduce the relevant line operators . in section [ fusion ] we present the central computation of this work : we study the fusion of line operators up to second order in perturbation theory . in section [ prooftsys ] we make good use of this computation to deduce the validity of the t - system up to first order in perturbation theory . section [ extensions ] contains a discussion of the extension of this method to other integrable models . eventually final remarks are gathered in section [ conclusion ] . in order to keep the bulk of the paper as readable as possible , most of the details of the computations are gathered in the appendices . appendix [ conventions ] contains the conventions . in appendix [ covcurrents ] we revisit the computation of the current - current opes in the pure spinor formalism using a novel and efficient method . in appendix [ computationdiv ] we study the uv divergences in line operators . eventually appendix [ comfusion ] contains the computations relevant for the fusion of line operators . to describe superstring theory in @xmath7 we will use the pure spinor formalism . this choice is a matter of convenience . indeed the computations of section [ fusion ] are simpler in the conformal gauge , where target - space covariance is preserved . in this section we introduce the pure spinor string on @xmath7 . we only discuss the features of this formalism that are relevant for the purpose of this paper . a more detailed discussion can be found for instance in @xcite@xcite@xcite . we introduce the flat connection , and show that the commutator of equal - time connections can be written in the canonical form of a @xmath12 system . we also introduce the line operators that are defined as the path - ordered exponential of the line integral of the flat connection . finally we discuss the uv divergences that appear in these line operators because of quantum effects . most of the results discussed in this section have appeared before in the literature . some new results are presented concerning the current algebra , the commutator of equal - time connections and the renormalization of the line operators . the target - space @xmath3 is embedded in a superspace with 32 supercharges . it is realized as the supercoset @xmath13 . the lie superalgebra @xmath14 admits the action of a @xmath15 automorphism . this automorphism induces a @xmath15 grading on the elements of the lie superalgebra . we can decompose the lie algebra @xmath16 according to this grading : [ z4decompo ] = _ 0 _ 1 _ 2 _ 3 where the subscript gives the @xmath15 grade . bosonic ( respectively fermionic ) generators of the lie superalgebra have an even ( respectively odd ) grade . [ [ the - action . ] ] the action . + + + + + + + + + + + let us introduce the currents @xmath17 and @xmath18 defined in terms of the group element @xmath19 as : [ j = gdg ] j = g^-1 g ; |j = g^-1 |g they take values in the lie superalgebra @xmath16 . we decompose the current @xmath17 according to the @xmath15 grading of the lie superalgebra : j = j_0 + j_1 + j_2 + j_3 and similarly for @xmath18 . let us also introduce the bosonic pure spinor ghosts @xmath20 as well as their conjugate momenta @xmath21 . they expand on the fermionic generators of the superalgebra with the following gradings : @xmath22 and @xmath23 . the ghosts satisfy the pure spinor constraint : @xmath24 , where the @xmath25 s are the @xmath26 gamma matrices . the pure spinor lorentz currents are : n=-\{w , } ; n = -\{w , } the action reads : @xmath27 the first line of the action contains a kinetic term both for the bosonic and fermionic target space coordinates . this implies in particular that the model does not exhibit kappa - symmetry , contrary to the green - schwarz string . the radius of the target space is denoted by @xmath28 in units of the string length . later on we will work perturbatively in a large radius expansion : the small parameter is @xmath29 . [ [ gauge - symmetry . ] ] gauge symmetry . + + + + + + + + + + + + + + + the action admits a @xmath30 gauge symmetry : g = g h_0 , = [ , h_0 ] , = [ , h_0 ] , w = [ w , h_0 ] , w = [ w , h_0 ] the holomorphic currents transform as : i0 : j_i = [ j_i , h_0 ] ; j_0 = h_0 + [ j_0 , h_0 ] ; n = [ n , h_0 ] and similarly for the anti - holomorphic currents . we introduce the associated covariant derivative : = + [ j_0 , ] ; |= |+ [ |j_0 , ] [ [ parity . ] ] parity . + + + + + + + the model enjoys a @xmath31 symmetry that exchanges holomorphic and anti - holomorphic worldsheet coordinates . it also flips the grade of the fermionic elements of the lie superalgebra : the subalgebras @xmath32 and @xmath33 are exchanged . [ [ the - maurer - cartan - equation . ] ] the maurer - cartan equation . + + + + + + + + + + + + + + + + + + + + + + + + + + + a consequence of is that the current satisfies the maurer - cartan equation : [ mc ] |j - |j + [ j , |j ] = 0 this is a crucial equation that is essentially responsible for the integrable properties of the model . [ [ equations - of - motion . ] ] equations of motion . + + + + + + + + + + + + + + + + + + + + the equations of motion combined with the maurer - cartan equation lead to : [ mc+eom ] lll |j_1 = [ j_3,|j_2 ] + [ j_2,|j_3 ] + [ n,|j_1 ] - [ j_1,n ] & & |j_1 = [ n,|j_1 ] - [ j_1,n ] |j_2 = [ j_3,|j_3 ] + [ n,|j_2 ] - [ j_2,n ] & & |j_2 = - [ j_1,|j_1 ] + [ n,|j_2 ] - [ j_2,n ] |j_3 = [ n,|j_3 ] - [ j_3,n ] & & |j_3 = - [ j_1,|j_2 ] - [ j_2,|j_1 ] + [ n,|j_3 ] - [ j_3,n ] |n = -[n , n ] & & n = [ n , n ] in this section we discuss the current - current opes that are the elementary input needed for the computations of section [ fusion ] . the set of currents we consider are the currents @xmath34 , @xmath35 , @xmath36 , @xmath37 as well as the ghost lorentz current @xmath38 , together with their anti - holomorphic partners @xmath39 , @xmath40 , @xmath41 , @xmath42 and @xmath43 . in order to simplify the expressions in the following computations , we introduce the generic notation @xmath44 , @xmath45 for the currents . the index @xmath46 takes the values in the set @xmath47 . for @xmath48 we define @xmath49 , @xmath50 . for the particular value @xmath51 we define @xmath52 , @xmath53 , and @xmath54 stands for ghost " . the index @xmath46 codes the @xmath15-grade of the current . the ghost currents have grade zero . the opes of the gauge covariant currents have been discussed in various papers . the opes at first - order in the @xmath29 expansion have been analyzed in @xcite@xcite@xcite@xcite . the @xmath55 corrections to the second - order poles have been computed in @xcite . for the purpose of the present article , the knowledge of the current algebra at order @xmath29 is enough . in appendix [ covcurrents ] we present a new and rather efficient way of computing the current algebra . the idea is to demand compatibility with the maurer - cartan equation and with the equations of motion ( more precisely with the reparametrization invariance of the path integral ) . notice that the opes involving @xmath34 and @xmath39 generically suffer from some ambiguities because of the gauge freedom . at the end of appendix [ covcurrents ] we compare the version of the current algebra we compute with the ones that appeared previously in the literature . it is convenient to expand the current on a basis of the lie superalgebra that is compatible with the @xmath15 grading . we write : k_m = k_m^a_m t_a_m the indices @xmath56 are adjoint indices$ ] , @xmath57 , @xmath58 , @xmath59 . although slightly less explicit , the notations used here allow for a much more compact writing . ] restricted to the subspace of the lie superalgebra of grade @xmath46 . the generators @xmath60 form a basis of the subspace @xmath61 . the current algebra takes the form : @xmath62 the tensors @xmath63 and @xmath64 are respectively the metric and the structure constants ( see appendix [ conventions ] for conventions ) . the non - trivial data in the current algebra is coded in the coefficients @xmath65 , @xmath66 . these coefficients should be read as follows : the coefficient @xmath67 give the coefficient of the identity operator in the ope between the currents @xmath35 and @xmath37 , the coefficient @xmath68 give the coefficient of the ghost current @xmath38 in the ope between @xmath36 and @xmath41 , and so on . we introduced a sum over all currents in the first - order poles in order to simplify the writing , but many of the @xmath69 s are clearly zero since they do not respect the @xmath15 grading . the coefficients @xmath65 , @xmath66 are symmetric in their two lower indices . also @xmath70 should be understood as @xmath71 . the non - zero coefficients are given below . the non - vanishing second - order poles are : c_13 = -1 , c_1 |3 = 1 , c_|1 3 = 1 , c_|1 |3 = -1 , c_22 = -1 , c_2|2 = 1 , c_|2 |2 = -1 the first - order poles involving only the currents of non - zero grade are : c c_11 ^ 2 = 2 , c_11^|2 = 1 , c_1 |1 ^ 2 = 1 , c_|1 |1^|2 = 1 c_33 ^ 2 = 1 , c_3|3^|2 = 1 , c_|3 |3 ^ 2 = 1 , c_|3 |3^|2 = 2 c_12 ^ 3 = 2 , c_12^|3 = 1 , c_|1 2 ^ 3 = 1 , c_1|2 ^ 3 = 1 , c_|1 |2^|3 = 1 c_32 ^ 1 = 1 , c_|3 2^|1 = 1 , c_3 |2^|1 = 1 , c_|3 |2 ^ 1 = 1 , c_|3 |2^|1 = 2 the first - order poles involving the ghosts currents are : c c_22^g = 1 , c_22^|g = -1 , c_2 |2^g = 1 , c_2 |2^|g = 1 , c_|2 |2^g = -1 , c_|2 |2^|g = 1 c_13^g = 1 , c_13^|g = -1 , c_1 |3^g = 1 , c_1 |3^|g= 1 , c_|1 3^g = 1 , c_|1 3^|g = 1 , c_|1 |3^g = -1 , c_|1 |3^|g = 1 c_gg^g = -1 , c_|g |g^|g = -1 eventually the first - order poles involving the currents @xmath34 , @xmath39 are : c c_10 ^ 1 = 1 , c_0 |1 ^ 1 = 1 , c_|0 1^|1 = 1 , c_|0 |1^|1 = 1 c_02 ^ 2 = 1 , c_0 |2 ^ 2 = 1 , c_|0 2^|2 = 1 , c_|0 |2^|2 = 1 c_30 ^ 3 = 1 , c_0 |3 ^ 3 = 1 , c_|0 3^|3 = 1 , c_|0 |3^|3 = 1 c_13 ^ 0 = 1 , c_13^|0 = 1 , c_|1 |3 ^ 0 = 1 , c_|1 |3^|0 = 1 c_22 ^ 0 = 1 , c_22^|0 = 1 , c_|2 |2 ^ 0 = 1 , c_|2 |2^|0 = 1 the method we are using to compute the current algebra does not fix completely the self - opes of @xmath34 and @xmath39 . we only obtain the following constraints : @xmath72 it turns out that these constraints are enough to perform explicitly the computations presented in this paper , @xmath39 appear in the flat connection with no dependence on the spectral parameter . this in turns implies that these coefficients cancel against each other in the computation of the fusion line operators presented in section [ fusion ] . notice however that the cancellation of some divergences in the line operators depends on the value of the coefficients , see . ] . similarly to the green - schwarz string @xcite@xcite , the pure spinor string on @xmath7 admits a one - parameter family of flat connections @xcite . this implies that the classical theory admits an infinite number of conserved charges . the flat connection @xmath73 is defined as : @xmath74 the flat connection is invariant under parity combined with the exchange of @xmath75 and @xmath76 . the equations of motion together with the maurer - cartan equation imply that the previous connection is flat for all values of the spectral parameter @xmath75 : d a(y ) + a(y ) a(y ) = 0 in the following we study line operators that are the path - ordered exponential of the integral of the flat connection along a given contour . we will only consider integration contours that lie at constant time . consequently only the spacelike component of the flat connection will appear . for simplicity , we use the same notation @xmath73 for the connection and for its spacelike component . the advantage of the version of the current algebra we are working with is that the commutator of two equal - time space - component of the flat connection can be written as a @xmath12 system : @xmath77 = & 2s \p_\sigma\delta^{(2)}(\sigma-\sigma ' ) + [ a_r(y;\sigma)+ a_{r'}(y';\sigma'),r ] \delta^{(2)}(\sigma-\sigma ' ) \cr & + [ a_r(y;\sigma)- a_{r'}(y';\sigma'),s ] \delta^{(2)}(\sigma-\sigma')\end{aligned}\ ] ] where @xmath28 and @xmath78 denote the representations the two connections are transforming in . the commutator transforms in the tensor product @xmath79 . only the terms explicitly written down in the opes contribute to the commutator . the infinite number of subleading singularities contained in the ellipses of do not contribute to the commutator of equal - time currents ( see e.g. @xcite ) . as shown in appendix [ apprsmatrices ] , the constant matrices @xmath80 and @xmath81 are given by : @xmath82 and : @xmath83 later it will be important that the @xmath80 matrix simplifies in the limit where the difference between the spectral parameters @xmath75 and @xmath84 is small : [ rlim ] y - y 0r ~ y(y^2+y^-2)^2 t_a^r t_b^r ^ba the @xmath12 matrices , first appeared in @xcite . a detailed study of the @xmath12 system for string theory in @xmath3 and its properties can be found in @xcite@xcite@xcite ( see appendix [ appcompkalgebras ] for more details ) . [ [ definitions . ] ] definitions . + + + + + + + + + + + + we are interested in studying line operators that are the path - ordered exponential of the integral of the flat connection on a given contour . when the contour is an interval @xmath85 $ ] , the line operator is called the transition matrix . we denote it as @xmath86 : t^b , a_r(y ) = p ( - _ a^b a_r(y ) ) the transition matrix is labelled by the representation @xmath28 in which the flat connection transforms . flatness of the connection implies that the classical transition matrix does not depend on the integration path chosen . this property has been argued to extend to the quantum theory in @xcite . for simplicity we consider only constant - time contours . for string theory purposes we are lead to define the theory on a cylinder . then we can define the monodromy matrix which is the line operator associated with a closed contour winding once around the cylinder : _ r(y ) = p ( -a_r(y ) ) flatness of the connection implies that the eigenvalues of the monodromy matrix are independent on time . consequently they code an infinite number of conserved charges . eventually the transfer matrix is the supertrace of the monodromy matrix : [ deftransfer ] _ r(y ) = str p ( -a_r(y ) ) [ [ regularization - of - uv - divergences . ] ] regularization of uv divergences . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in a quantum theory the line operators are generically ill - defined since the collisions of integrated connections lead to divergences . to properly define line operators one has to regularize these divergences , and then renormalize the line operators . in order to study the uv divergences , we first have to expand the exponentials in the line operators . we write the transition matrix as : t^b , a_r(y ) = _ m=0^(-1)^m t_r,(m)^b , a(y ) where the @xmath87-th term is the path - ordered integral of @xmath87 connections : t_r,(m)^b , a(y ) = p(_a^b a_r(y ) ) ^m = _ b>_1> ... >_m > ad_1 ... d_m a_r(y;_1) ... a_r(y;_m ) and similarly for the monodromy and transfer matrices . divergences occur when two integrated connections collide . it is clear from the current algebra that the collision of two connections leads to second- and first - order poles . in order to regularize these divergences , we introduce a uv cut - off @xmath88 . we use a principal - value regularization scheme as suggested in @xcite . the ope between two equal - time connections @xmath89 and @xmath90 is regularized by a small shift in time , in a symmetric way : [ defreg ] a ( ) a( ) ( a(+i ) a( ) + a ( ) a(+i ) ) for instance , a first - order pole is regularized as : [ reg1pole ] p.v . = ( + ) = this regularization scheme turns out to be very convenient to discuss the fusion of line operators , as explained in section [ fusion ] . [ [ divergences - at - order - r-2 . ] ] divergences at order @xmath29 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the first - order divergences in line operators in the pure spinor string on @xmath7 were first studied in @xcite . in this paper the authors used a different regularization scheme : the opes were regularized by imposing that the distance between two connections can not be smaller than the uv cut - off . the authors of @xcite also used a slightly different version of the current algebra . in appendix [ computationdiv ] we revisit the analysis of @xcite using the regularization scheme and the current algebra . the main difference we obtain with respect to @xcite is that the linear divergences do cancel thanks to our choice of regularization scheme . below we summarize the results derived in appendix [ computationdiv ] . the transition matrices contain logarithmic divergences . schematically , these divergences read : ~_i=0 ^ 3 # \ { t^a_i t_a_i , t^b , a(y ) } the precise expression for these divergences is given in equation . consequently the transition matrices need to be renormalized . these divergences are cancelled by a simple wave - function renormalization . the monodromy matrix also contains logarithmic divergences . these are given in . schematically , these divergences read : [ divomega ] ~_i=0 ^ 3 # ( t^a_i t_a_i ( y ) + ( y ) t^a_i t_a_i - 2 t^a_i ( y ) t_a_i ) the new divergences with respect to the transition matrices come from collisions between connections sitting on both sides of the starting point of the integration contour . these divergences are cancelled by a simple wave - function renormalization of the monodromy matrix . the most important result for the purpose of this paper is that the transfer matrix is completely free of divergences at order @xmath29 . this follows simply by taking the supertrace of equation . this remarkable property strongly relies on the vanishing of the dual coxeter number of the global symmetry group @xmath10 . for instance in generic wzw models , cancellation of divergences in the transfer matrices require both a wave - function renormalization and a renormalization of the spectral parameter @xcite . the divergences identified in @xcite also vanish if the dual coxeter number of the group is zero . in this section we study the fusion of two line operators . the fusion is the process of bringing the integration contours of two line operators on top of each other . we are interested in the quantum effects that occur in this process . the fusion of line operators for the pure - spinor string in @xmath7 was studied at first - order in perturbation theory in @xcite . in this section we will revisit and extend the first - order computations of @xcite . then we will further extend the computation of fusion up to second order in perturbation theory . the structure of the computations is similar to the ones presented in @xcite , where more details can be found . in @xcite the computations were performed in the sigma model on the supergroup @xmath11 . this theory is a good toy model for the pure spinor string on @xmath7 . indeed the complications coming from the coset structure and the pure spinor ghosts are absent . let us consider two transition matrices @xmath86 and @xmath91 that transform respectively in the representations @xmath28 and @xmath78 . the fusion of these two matrices transforms in the tensor product @xmath92 . in the following we will omit the symbol @xmath93 to lighten the formulas . we represent the fusion of these two transition matrices with the symbol @xmath94 . this process is defined as : [ deffusion ] t^b , a_r(y)t^d , c_r(y ) = _ 0^+ t^b+i , a+i_r(y)t^d , c_r(y ) assuming the integration contour of the transition matrices @xmath91 lies at constant time @xmath95 , then the integration contour of @xmath96 lies at constant time @xmath97 . if the intervals @xmath85 $ ] and @xmath98 $ ] do not overlap , the process of fusion is trivial . in the following we assume that the overlap of these intervals is non - zero . as the distance between the two contours goes to zero , the opes between integrated connections sitting on the two contours produce quantum corrections to the classical process of fusion . these are the corrections we will evaluate . let us consider the ope between two connections @xmath99 and @xmath100 integrated respectively on the first and on the second contour . we write this ope as : @xmath101 comparing with equation , we notice that the first term in the previous equation should be understood as regularized ope in the quantum line operator obtained after the fusion has been completed . on the other hand , the second term in should be understood as producing a quantum correction proper to the process of fusion . this are the corrections we want to compute . in order to understand better the meaning of , let us isolate a first - order pole in the ope between the two connections . under the decomposition , it is rewritten as : @xmath102 the term we focus on is the second term on the right - hand side . as the notation suggests , it is actually a regularization of the delta - function . once we integrate upon the free coordinates , it produces a finite quantum corrections to the fusion of line operators . we can perform a similar manipulation for all first- and second - order poles appearing in the ope between the two connections @xmath99 and @xmath100 . in order to isolate the contribution to the quantum corrections associated with fusion , we subtract the principal value from the singularities . we obtain that all the terms that contribute to the quantum corrections from fusion come with ( derivatives of ) regularized delta functions . the upshot is the following : in order to compute the quantum corrections in the process of fusion , we have to subtract the `` principal value '' piece from the ope between connections . what we are left with is essentially the commutator between connections that we computed in : [ 1-pvaa ] ( 1-p.v . ) a_r(y;+i ) a_r(y; ) [ a_r(y ; ) , a_r(y; ) ] we begin with the corrections of order @xmath29 . since all terms in the current algebra are of order @xmath29 , it is enough to perform one ope . the computation of the first - order corrections in the fusion of two transition matrices was performed in @xcite using ope techniques . this computation holds provided the commutator of connections can be written as a @xmath12 system , which is the case for the pure spinor string on @xmath3 ( see ) . the result obtained in @xcite matches the hamiltonian analysis of @xcite : @xmath103 the first term on the right - hand side is the zeroth - order result . the remaining terms are the first - order corrections . the function @xmath104 is the characteristic function of the interval @xmath105 $ ] which takes the value @xmath106 if @xmath107 and @xmath108 if @xmath109 or @xmath110 . for the special case where the integration intervals of the line operators have coinciding endpoints , that is for @xmath111 or @xmath112 , then the characteristic function @xmath104 has to be evaluated as @xmath113 . this prescription essentially had to be guessed in the hamiltonian formalism @xcite . in the ope formalism it is a consequence of the definition @xcite . notice that the first - order corrections in are anti - symmetric in the exchange of the two line operators . so they contribute only to the commutator of the line operators . this follows from the fact that the quantum corrections associated with fusion come from the anti - symmetric part in the opes ( see ) . from equation we can deduce the fusion of transfer matrices @xcite . we obtain that the fusion of transfer matrices at first order is trivial : @xmath114 indeed the first - order corrections associated with fusion in take the form of constant matrices inserted at the endpoints of the overlap of the integration intervals . since the transfer matrices have no endpoints , it is not surprising that these corrections vanish . this implies in particular that the commutator of the transfer matrices is zero at first order . remember that our main goal is to show that the leading quantum correction in the fusion of two transfer matrices gives the shifts in the t - system at first order . since the fusion of transfer matrices is trivial at first - order , we need to study the fusion of line operators at second - order . there are two different ways we can obtain @xmath55 corrections in the process of fusion : * the first way is to take one single ope between two integrated connections , and include @xmath55 corrections to the current algebra . as argued in section [ recipefusion ] , only the anti - symmetric part of the current - current opes contributes to the process of fusion . consequently at this order the @xmath55 corrections to the current algebra lead to @xmath55 corrections to the commutator of line operators . we postpone the computation of these corrections for future work since the current algebra is only partially known at order @xmath55 @xcite . * the second way is to perform two opes between integrated connections and use the current algebra at order @xmath29 . following the logic explained in section [ recipefusion ] , we again consider only the anti - symmetric part in each ope . since we perform an even number of opes this time the result will be symmetric under the exchange of the two line operators . more precisely we obtain a contribution to the symmetric product of the line operators of order @xmath55 . this are the terms that we will compute in this paper . more generally , the arguments of section [ recipefusion ] imply that any quantum correction in the process of fusion that involve an even ( respectively odd ) number of opes would contribute only to the symmetric product ( respectively commutator ) of the line operators . consequently the @xmath55 corrections to the current algebra would contribute to the symmetric product of line operators at order @xmath115 and higher . below we describe the different steps in the computation of the symmetric fusion of line operators at order @xmath55 . in an attempt to keep this section readable , the technical details of the computation have been gathered in appendix [ comfusion ] . more details can also be found in @xcite . let us consider two line operators with contours separated in time by a small distance @xmath88 . we have to take two opes between the connections integrated on the contours . the simplest way is to take two opes between two distinct pairs of connections . but we can also take one ope between two connections , and then take the ope of the resulting currents with a third connections . we will call this latter process a triple collision ( see figure [ triplecollision ] ) . ] the computation is conveniently decomposed in two steps . * a first part of the total answer is obtained in the way depicted in figure [ fuso2step1 ] . we start from the result of the fusion at first order . we pull the contours away , and then re - fuse the line operators . the result of this procedure was computed in @xcite for an arbitrary @xmath12 system . we obtain new insertion of constant matrices at the endpoints of the overlap of the contours of the line operators . roughly speaking , the first - order result exponentiate . what is important for our purposes is that this procedure gives once again a vanishing result for the fusion of transfer matrices . this simply follows from the fact that there is no first - order correction in the fusion of transfer matrices . * the procedure described above does not capture correctly the quantum corrections coming from triple collisions . indeed in this procedure , the intermediate currents in the triple collisions are distributed in an arbitrary way on the two integration contours so that they recombine into connections . this induces a source of errors . so we have to compute separately additional corrections coming from the triple collisions . this is done in appendix [ comfusion ] . ] the analysis of appendix [ comfusion ] shows that there are two types of corrections that we need to add on top of the result obtained by the procedure described in figure [ fuso2step1 ] . the first type of corrective terms contain the integration of an operator @xmath116 on the overlap of the contours : @xmath117\cap[c , d]}d\sigma \left \lfloor \int_\sigma^b a \right\rceil^i \left \lfloor \int_\sigma^d a ' \right\rceil^{i ' } \tilde k ( \sigma ) \left \lfloor \int_a^\sigma a \right\rceil^{m - i } \left \lfloor \int_c^\sigma a ' \right\rceil^{m'-i ' } \end{aligned}\ ] ] where we introduced the convenient notation @xmath118 to describe the path - ordered integral of @xmath87 connections on the interval @xmath85 $ ] . the precise expression for the operator @xmath116 is explicitly given in . schematically , the operator @xmath116 is a linear combination of the currents multiplied by three generators of the lie superalgebra in the representations @xmath28 or @xmath78 , and contracted with structure constants . the second type of corrective terms contain a constant matrix @xmath119 inserted in between the integrated connections on the overlap of the integration contours : @xmath120 the precise expression for the matrix @xmath119 is given in . schematically , it is a linear combination of the tensor product of two generators taken in the representation @xmath28 and @xmath78 and contracted with structure constants . notice that the integration over the regularized delta function squared produce a linear divergence when the uv regulator @xmath88 is sent to zero . it would be interesting to perform a complete analysis of the second - order divergences in the line operators along the lines of section 3 in @xcite , to see whether the transfer matrices as well as the result of the fusion of transfer matrices are free of divergences up to second - order . [ [ fusion - of - transfer - matrices - at - second - order - upshot . ] ] fusion of transfer matrices at second order : upshot . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the detailed expression for the symmetric fusion of transition and monodromy matrices at second order is quite indigestible , so we refrain from giving an explicit formula for those . on the other hand the symmetric fusion of transfer matrices , that is crucial for the purposes of this paper , turns out to be rather simple : @xmath121 where we denoted by @xmath122 the symmetrized fusion product . this result is schematically represented in figure [ fusionts ] . the detailed expressions for the operator @xmath116 and the constant matrix @xmath119 can be read from equations and . is integrated in between the connections . in the third term , a constant matrix @xmath119 is inserted in between the integrated connections.[fusionts ] ] in this section we use the previous computations to obtain a first - principle perturbative derivation of the t - system . as explained in the introduction , the idea is to promote the t - system to an operator identity , where the product between transfer matrices is understood as the fusion product : [ qtsystem ] _ a , s(u + 1 ) _ a , s(u - 1 ) = _ a+1,s(u+1)_a-1,s(u-1 ) + _ a , s-1(u+1 ) _ a , s+1(u-1 ) we expect the transfer matrices to commute in the quantum theory . this has been proven in section [ fusion ] at order @xmath29 . consequently we can equivalently use the symmetric fusion product to define the t - system . the integer label @xmath123 label unitary irreducible representation of @xmath10 . these labels take value in a t - shaped lattice ( see e.g. @xcite@xcite ) . the corresponding representations are associated with rectangular young tableaux which size is given by the value of the labels @xmath123 . it is known that these representations satisfy the following supercharacter identity ( see e.g. @xcite@xcite ) : [ charid1 ] ( a , s)^2 = ( a+1,s)(a-1,s)+(a , s+1)(a , s-1 ) in the limit where we neglect both the shifts of the spectral parameter as well as the quantum effects associated with fusion , the t - system reduces to the character identity . next we want to show that the t - system holds at first order . we consider : @xmath124 where we use the shorthand @xmath125 to denote the sum over representations that appears in the t - system . we look for a value of the shift of the spectral parameters @xmath126 such that the previous quantity does indeed vanish , and the t - system holds . then we can deduce the relationship between the spectral parameter @xmath75 used to define the flat connection , and the spectral parameter @xmath8 that appears in the t - system . we assume that @xmath126 is of order @xmath29 . we will now show that the terms of order @xmath29 in do vanish . more precisely , the terms of order @xmath29 coming from the derivative expansion of the transfer matrices cancel against the leading quantum correction coming from the process of fusion . [ [ fusion - of - transfer - matrices - when - the - difference - of - spectral - parameter - is - small . ] ] fusion of transfer matrices when the difference of spectral parameter is small . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in section [ fusion ] we obtained that the leading quantum correction in the fusion of transfer matrices is of order @xmath55 . however when the difference of the spectral parameter is of order @xmath29 , a piece of the leading quantum correction actually becomes of order @xmath29 . so it has the right order of magnitude to cancel the first term in the derivative expansion of the transfer matrices in . more details can be found at the end of appendix [ comfusion ] . the interesting term comes from the second line in . we assume @xmath127 . from equations we obtain that simplifies to : @xmath128 t_{b_n}^{r ' } + { f_{c_p}}^{b_n a_m } { f_{e_r}}^ { d_q c_p } t_{a_m}^r \ { t_{b_n}^{r ' } , t_{d_q}^{r'}]\right)\right)\cr & \quad \times \left . t_r^ { \sigma,0}(y ) t_{r'}^{\sigma,0}(y ' ) \right ) + \mathcal{o}(r^{-4})\end{aligned}\ ] ] we observe that the linear combination of the currents that appears is now proportional to the derivative of the flat connection . [ [ useful - character - identities . ] ] useful character identities . + + + + + + + + + + + + + + + + + + + + + + + + + + + + to further simplify the quantum corrections that appear in , we need to use some character identities that holds for the particular combination of representations that appear in the t - system . in @xcite the validity of the t - system was proven for transfer matrices associated to @xmath129 spin chains . then a large family of character identities was deduced by expanding the t - system as an infinite series . here we go the other way : we try to reconstruct the t - system from a perturbative expansion . thus it makes sense that we need to use the character identities of @xcite . in the appendix e of @xcite , it was shown that some of these character identities imply in particular that for any group element @xmath54 and for any function @xmath130 : @xmath131 g^r \otimes t_b^{r ' } g^{r ' } ) \cr & = 2 \sum_{r , r ' } str ( g^r \otimes k^e t_e^{r ' } g^{r ' } ) - str ( k^e t_e^{r } g^r \otimes g^{r ' } ) \end{aligned}\ ] ] and similarly : @xmath132 g^{r ' } ) \cr & = 2 \sum_{r , r ' } str ( g^r \otimes k^e t_e^{r ' } g^{r ' } ) - str ( k^e t_e^{r } g^r \otimes g^{r ' } ) \end{aligned}\ ] ] essentially , these character identities allow to replace the complicated combination of structure constants and generators appearing in by single generators , assuming we consider the sum of representation that appear in the t - system . so we obtain : @xmath133 [ [ the - t - system - at - first - order . ] ] the t - system at first order . + + + + + + + + + + + + + + + + + + + + + + + + + + + + performing a taylor expansion for the @xmath134 s in , we deduce : @xmath135 the first term in the previous expression vanishes because of the classical identity . in order for the first - order corrections to vanish as well , we have to take : [ defdelta ] = r^-2 y(y^2-y^-2)^2 thus have have shown that : [ tsys4 ] _ r , r _ r(y+ ) _ r(y- ) = 0 + ( r^-4 ) [ [ redefinition - of - the - spectral - parameter . ] ] redefinition of the spectral parameter . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in order to write the t - system in the canonical form we define : [ defu ] u = + cst such that @xmath136 . then equation is rewritten as : _ a , s(u + 1 ) _ a , s(u - 1 ) = _ a+1,s(u+1)_a-1,s(u-1 ) + _ a , s-1(u+1 ) _ a , s+1(u-1 ) + ... this is the canonical form of the t - system . [ [ comparison - with - the - t - system - obtained - via - the - thermodynamic - bethe - ansatz . ] ] comparison with the t - system obtained via the thermodynamic bethe ansatz . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we can now perform a consistency check with the thermodynamic bethe ansatz derivation of the @xmath3 t - system @xcite@xcite@xcite . in this context the t - system is obtained in a slightly different form : only the t - functions on the left - hand side have a shifted spectral parameter : [ tsystemtba ] ^tba_a , s(u + 1 ) ^tba_a , s(u - 1 ) = ^tba_a+1,s(u ) ^tba_a-1,s(u ) + ^tba_a , s-1(u ) ^tba_a , s+1(u ) this mismatch is easily cured by a redefinition of the t - functions . let us define the functions @xmath137 as : ^tba_a , s(u ) = _ a , s(u+a - s)then the functions @xmath138 s satisfy if and only if the functions @xmath134 s satisfy the t - system . however the previous redefinition does not change the magnitude of the shift on the left - hand side of the t - system . thus the matching of the shifts gives a quantitative check of the consistency between the approach taken in this paper , and the tba approach . next we perform this matching using the conventions of @xcite@xcite . the flat connection is written in terms of a spectral parameter @xmath139 as @xmath140 . comparing with we deduce that the spectral parameter @xmath75 that we use is related to @xmath139 as : @xmath141 . the variable @xmath142 that enters the tba t - system is linked to the spectral parameter @xmath139 via the zhukowsky map : @xmath143 , where @xmath54 is related to the t hooft coupling @xmath144 as @xmath145 . the parameter @xmath28 can be linked to the t hooft coupling @xmath144 by identification of the prefactor of the worldsheet action . this gives @xmath146 . consequently the parameter @xmath8 that we obtained is related to the parameter @xmath142 as : @xmath147 , assuming the free constant in takes the value @xmath148 . the analysis of @xcite gives a t - system where the parameter @xmath142 is shifted by @xmath149 . given that @xmath8 is shifted in our case by @xmath150 , there is an apparent mismatch by a factor of @xmath151 . this comes from the fact that we have been working on an euclidean worldsheet . if we wick - rotate the worldsheet to a minkowskian signature , then our analysis produce the t - system with imaginary shifts in e.g. would be replaced by a real shift @xmath88 . this implies that in the computation of fusion all opes would come with an additional factor of @xmath151 . consequently the second - order corrections from fusion would come with an additional minus sign . this would in turn induce a factor of @xmath151 in the shift of the t - system . ] in perfect agreement with the tba analysis . [ [ upshot . ] ] upshot . + + + + + + + as claimed previously , we have derived the t - system up to first order in the large radius expansion . more precisely , we have sown that the shifts of the spectral parameter in the t - system come from quantum effects in the fusion of transfer matrices . moreover we have checked that the shifts are the same than the ones obtained in the thermodynamic bethe ansatz derivation of the t - system . notice that the vanishing of the divergences in the transfer matrices is important . indeed if the transfer matrices would need to be renormalized , then the renormalization factor would most likely depend on the representation in which the transfer matrix is taken ( see e.g. @xcite ) . it implies that the different terms in the t - system would be renormalized differently , which would most likely destroy the balance needed for the previous computation to work . in this paper we have proven that the t - system is realized in the pure spinor string on @xmath7 up to first order in the large radius expansion . in @xcite a similar proof was given for the non - linear sigma model on the supergroup @xmath11 . it is natural to look for other theories where this derivation can be easily generalized . a close look at the computation leads to the following conclusion : there are only a few necessary and sufficient conditions that a given model has to fulfill in order for the derivation to apply . obviously the theory has to exhibit a one - parameter family of flat connections . we assume that the connection takes value in a lie algebra . the other conditions are the following : * the equal - time commutator of the spacelike component of the connection can be written as a @xmath12 system . this guarantees that formula can be directly reproduced . moreover the @xmath80 matrix must satisfy a property similar to equation : in the limit where the difference between the spectral parameter is small , the @xmath80 matrix needs to be proportional to the casimir @xmath152 . this condition is necessary for the simplification leading to equation to occur . * the lie group needs to possess an equivalent of the character identities , . given the role played by the results of @xcite , it is tempting to speculate that a sufficient condition is that there exists a spin chain with the same symmetry group that realizes the t - system . * eventually the transfer matrix has to be free of divergences at first order in perturbation theory . a crucial condition here is that the dual coxeter number of the symmetry group vanishes . this condition prevents the renormalization of the transfer matrices from destroying the balance needed for the computation to work . next we discuss several candidate theories that may fulfill these requirements . [ [ candidate - theories - relevant - for - the - adscft - correspondence . ] ] candidate theories relevant for the ads / cft correspondence . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us begin with the theories that describe superstrings in anti - de sitter backgrounds . these theories are built on sigma models on ( coset of ) supergroups , see e.g. @xcite for a classification of the relevant @xmath15 cosets . notice that all the supergroups involved have a vanishing dual coxeter number . this should not come as a surprise , since the vanishing of the dual coxeter number is tightly related to the vanishing of the spacetime supercurvature , and thus to the fact that the equations of motion of supergravity are satisfied . spacetime covariance was helpful in the previous analysis . consequently we will mostly discuss theories of the pure - spinor type that allow for a covariant quantization . obviously it would be interesting to reproduce the previous computations in green - schwarz - like theories that realize kappa - symmetry . the structure of the computations would be identical , but the computations themselves would be more tedious because the gauge - fixing of kappa symmetry usually comes with a breaking of the target space isometries . the first obvious candidate is string theory on @xmath153 . indeed in @xcite a y - system was conjectured to hold in this theory . the tba derivation of the y - system was performed in @xcite@xcite . in order to actually reproduce the computations described in the present paper , the pure spinor formulation of superstring theory in @xmath153 developed in @xcite is a natural starting point ( see also @xcite ) . the second candidate is string theory on @xmath154 . the analysis of @xcite applies to the sigma - model on @xmath155 . in @xcite the hybrid formalism was developed to describe superstrings in @xmath154 in a superspace with eight supercharges . in this formalism the worldsheet theory is the sigma model on @xmath155 , coupled to ghosts . this theory admits a consistent expansion in the ghosts . thus the analysis of @xcite implies that the hybrid string in @xmath154 realizes the t - system up to first order in the large radius expansion , and up to zeroth order in the ghosts expansion . this is valid for @xmath154 supported by rr fluxes , ns fluxes or by any mixing of these fluxes . it would be instructive to dress up the computation of @xcite with the hybrid ghosts there has been some interest in the question of integrability for string theory in @xmath154 , see e.g. @xcite . however the progress have been rather slower than in the case of @xmath7 , mostly because the dual conformal field theory is not as well understood . presumably the approach presented in @xcite and in the present article can lead to a faster road to the solution of this problem . other formulations of string theory on @xmath154 involve a @xmath15 coset of the supergroup @xmath156 . it is the case of the hybrid string with sixteen manifest supercharges @xcite . it is reasonable to expect that the t - system is also realized in this formalism . another natural candidate is the hybrid description of superstrings in @xmath157 discussed in @xcite . it is also based on a @xmath15 coset of the supergroup @xmath155 , and it is integrable @xcite@xcite . in the classification of @xcite we find other candidate string backgrounds that are @xmath15 coset of supergroups with vanishing dual coxeter number : @xmath158 , @xmath159 , @xmath160 and @xmath6 . quantum integrability is likely to show up at least in some of these examples . no formalism has been proposed to covariantly quantize string theory in these backgrounds yet . [ [ other - candidates . ] ] other candidates . + + + + + + + + + + + + + + + + + other theories that may not be directly relevant for string theory presumably also realize the t - system in the way described in this paper . these are the sigma models on ( cosets of ) supergroups with vanishing dual coxeter number , some of which play a role in condensed matter ( see e.g. @xcite@xcite@xcite ) . a first example is the sigma model on the supergroup @xmath161 . this model shares many of the remarkable properties of the sigma model on @xmath11 @xcite , see e.g. @xcite . next @xmath31 cosets of supergroups with vanishing dual coxeter number are classically integrable @xcite@xcite . some of them also display nice quantum features @xcite . in these model there is a current that is both flat and conserved . consequently the current algebra is very similar to the one found in sigma models on supergroups @xcite . this follows from the generic method introduced in @xcite and used in appendix [ covcurrents ] to compute the current algebra . classical integrability extends to @xmath15 and more generally to @xmath162 cosets @xcite . it would be interesting to understand these models better . in particular it may shed some new light on the question of the role of the pure spinor ghosts for quantum integrability of the pure spinor string in @xmath7 . [ [ summary - of - the - results . ] ] summary of the results . + + + + + + + + + + + + + + + + + + + + + + + we have studied the fusion of line operators in the pure spinor string on @xmath7 up to second order in perturbation theory . we deduced that the pure spinor string on @xmath7 realizes the t - system as an operator identity , with the fusion product , up to first order in the large t hooft coupling expansion . the quantum effects in the fusion of the transfer matrices give the shifts in the t - system . [ [ comparison - with - the - thermodynamic - bethe - ansatz . ] ] comparison with the thermodynamic bethe ansatz . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the t - system was previously derived using the tba machinery @xcite@xcite@xcite . here we will compare the advantages of both approaches . a weakness of the tba is that it relies on several assumptions that are notoriously difficult to check . in particular one has to assume quantum integrability to start with . moreover the spectrum of excitations that contribute in the thermodynamic limit essentially has to be guessed through the string hypothesis . the approach of the current article has the big advantage of starting from first principles . the other drawback of the tba is that the derivation of the t - system only applies to the ground state . the fact that the same set of equations also codes the spectrum of excited states upon analytic continuation is essentially an empirical observation . in this paper we have derived the t - system as an operator identity . thus there is no doubt that all states of the theory satisfy the t - system . on the other hand , the approach we are using here is intrinsically perturbative . the computations needed to derive the t - system at first order were already quite heavy . it would take a lot of efforts to go to the next order . the tba approach is free of this limitation since it produces the full t - system in one go . there are also some by - products of the tba approach that were not reproduced in the present work . in particular the tba gives a explicit formula to extract the spectrum from the t - functions . it also gives some informations about the analytic properties of these functions . it would be interesting to investigate these questions with the elementary techniques used in the present paper . we hope to come back to these questions in future work . the computation presented in this paper gives a very strong argument in favor of the validity of the t - system . it is not a definite proof since it is perturbative . however the previous discussion shows that the approach presented here is complementary with the tba analysis . indeed the weak points of the tba are the the strong points in our approach , and vice - versa . so the combination of both methods leaves little room for doubts . moreover this works sheds a new light on the t - system . the fact that it should be understood as an operator identity where the product is the fusion of line operators , may be helpful to understand better the integrable structures that appear in the ads / cft correspondence . [ [ the - role - of - the - pure - spinor - ghosts . ] ] the role of the pure spinor ghosts . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the pure spinor ghosts are expected to play an important role in the quantum worldsheet theory . we can ask the question of the role of the pure spinor ghosts in the computation described in this paper . interestingly , the same results would be obtained if we would set the pure spinor ghosts to zero from the start . the reason is that at tree level , the ghosts form a closed subsector . more precisely the ope of a ghost current with any other current can only produce ghost current . in other words , all the coefficients of the type @xmath163 and @xmath164 in the current algebra are zero if the index @xmath46 is not @xmath54 or @xmath165 . the fact that the computation does not relies on the pure spinor ghosts can be tracked back to the fact that we only needed the tree - level current algebra to compute the crucial term that produces the shifts in the t - system . presumably this is not going to be the case at higher order . continuing the computation of @xcite to get the full current algebra at second order would be interesting . already the pure spinor should play an important role . hopefully they allow for the cancellation of second - order divergences in the transfer matrices , and they also insure that transfer matrices commute up to second order in perturbation theory . in this paper we used the pure spinor formalism that allows for a covariant quantization . it may also be instructive to reproduce this computation in the green - schwarz formulation of @xcite . [ [ the - algebra - of - transfer - matrices . ] ] the algebra of transfer matrices . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the computations we performed allow to address the question of the algebra of transfer matrices for the pure spinor string on @xmath7 . naively , the fusion of two transfer matrices is rather complicated . it seems from that it does not even close on transfer matrices . however by selecting a particular combination of representations we managed to close the algebra . for these representations , the algebra of transfer matrices is nothing but the t - system . there might exists a generalization of the t - system that applies for other representations . it would be interesting to further explore this issue . [ [ generalization - to - other - integrable - field - theories . ] ] generalization to other integrable field theories . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it would be also interesting to try the approach advocated here in other integrable theories that play a role in the ads / cft correspondence . some examples were listed in section [ extensions ] . this approach may be more efficient than trying to reproduce the historical steps that were performed for @xmath7 . more generally , developing worldsheet technology for strings in rr background is certainly worthwhile . even if the progress in that direction have been rather slow , the results presented here together with other recent works ( see e.g. @xcite@xcite ) demonstrate that quantum string theory in some rr backgrounds can be studied with the tools that are currently available . the results presented here also suggest that the integrable models relevant for the ads / cft correspondence may belong to a special family . it is not clear that the interpretation of the t - system advocated here applies straightforwardly to generic integrable field theories . indeed generically the transfer matrices have to be renormalized when the dual coxeter number of the symmetry group is non - zero ( see e.g. @xcite ) . this would complicate a tentative derivation of the t - system from the fusion of transfer matrices . the author would like to thank gleb arutyunov , oscar bedoya , denis bernard , nikolay gromov , volodya kazakov , marc magro , valentina puletti , sakura schafer - nameki , joerg teschner , jan troost , benoit vicedo , dmitro volin and in particular pedro vieira for useful discussions and correspondence . the author is a postdoctoral researcher of fwo - vlaanderen . this research is supported in part by the belgian federal science policy office through the interuniversity attraction pole iap vi/11 and by fwo - vlaanderen through project g011410n . let @xmath166 be a basis of the generators of the lie superalgebra . the metric is defined as : _ ab = str(t_a t_b ) where the supertrace @xmath167 is a non - degenerate graded - symmetric inner product . we define the inverse metric as : ^ab _ ac = ^b_c the metric and its inverse are graded - symmetric : _ ab = ( -)^ab_ba where @xmath168 is a minus sign if and only if both indices @xmath169 and @xmath170 are fermionic . an element @xmath171 of the lie superalgebra is expanded as : j = j^a t_a we adopt ne - sw " conventions for the contraction of indices . indices are raised and lowered with the metric in the following way : x^a = ^abx_b ; x_a = x^b _ ba [ [ structure - constants . ] ] structure constants . + + + + + + + + + + + + + + + + + + + + the graded commutator for the generators is defined as : [ t_a , t_b } = t_a t_b - ( -)^ab t_b t_a we define the structure constants @xmath172 as : [ t_a , t_b } = f_ab^c t_c the identity @xmath173 follows from the graded - symmetry of the supertrace . it implies for the structure constants : ^d _ ad = f_ab^d _ dc in agreement with our conventions we define : f_abc = f_ab^d _ dc and so on . the structure constants are graded - antisymmetric in the 1 - 2 and 2 - 3 indices : f_abc = -(-)^abf_bac ; f_abc = -(-)^bc f_acb under the exchange of the first and third indices , we have : f_abc = -(-)^f_cba under cyclic permutation of their indices , the structure constant also satisfy : f_abc = ( -)^a f_bca [ [ tensor - product . ] ] tensor product . + + + + + + + + + + + + + + + the tensor product @xmath174 of two generators taken in different representation @xmath28 and @xmath78 is graded : t_a^r t_b^r = ( -)^ab ( 1^r t_b^r)(t_a^r 1^r ) in order to lighten the expressions in the bulk of the paper we often get rid of the @xmath93 symbol : t_a^r t_b^r t_a^r t_b^r = ( -)^ab t_b^r t_a^r in this appendix we give a new derivation of the tree - level gauge - covariant current algebra for the pure spinor string on @xmath7 . the method we use is inspired by the analysis of @xcite for the current algebra in sigma - models on supergroups . the different steps are the following . we make a natural ansatz for the current - current opes . then we demand that this ansatz is compatible with reparametrization invariance of the path integral , and the with maurer - cartan equation . this typically gives more constraints than the number of free coefficients in the ansatz . finally we solve these constraints to get the current algebra . this method can be generalized to compute the quantum corrections to the current algebra . this computation can be efficiently organized recursively @xcite . here we will only compute the tree - level coefficients since this is sufficient for the purpose of this article . we choose the ansatz for the current algebra , that we reproduce here for clarity : @xmath175 this ansatz is based on dimensional analysis and symmetry . we only wrote down the second- and first - order poles , but there is an infinite series of less and less singular terms that come with operators of ( classical ) dimension greater or equal to two . notice also that this ansatz is suitable for the tree - level current algebra , but it should be slightly modified if one is to take into account quantum corrections @xcite . in the following we will compute the coefficients @xmath69 s . many of these coefficient vanish trivially because the current algebra has to be compatible with the @xmath15 grading . parity also induces some redundancy in the remaining coefficients . these two symmetries leave 57 independent coefficients that we need to compute . [ [ equations - of - motion - and - path - integral - reparametrization - invariance . ] ] equations of motion and path - integral reparametrization invariance . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in this subsection we demand that the current algebra is compatible with the reparametrization invariance of the path integral . in particular this guarantees that the current algebra is compatible with the equations of motion . let us consider the action for the pure spinor string in @xmath7 . we consider a small variation of the group element @xmath54 parametrized by a element of the lie superalgebra @xmath171 : [ shiftg ] g = g x the variation of the currents is given by : @xmath176 & & \delta n = 0\cr & \delta \bar j = \bar \p x + [ \bar j , x ] & & \delta \hat n = 0 \end{aligned}\ ] ] we can decompose the infinitesimal shift @xmath171 on the @xmath15 subspaces of the lie superalgebra as @xmath177 . since @xmath178 generates gauge transformations that leave the action invariant , we set @xmath179 . we obtain the variation of the @xmath180 s : [ deltaj ] j_0 & = & [ j_1,x_3 ] + [ j_2,x_2 ] + [ j_3,x_3 ] j_1 & = & x_1 + [ j_0,x_1 ] + [ j_2,x_3 ] + [ j_3,x_2 ] j_2 & = & x_2 + [ j_0,x_2 ] + [ j_1,x_1 ] + [ j_3,x_3 ] j_3 & = & x_3 + [ j_0,x_3 ] + [ j_1,x_2 ] + [ j_2,x_1 ] and similarly for the @xmath181 s . we deduce the variation of the action under the infinitesimal shift of the group element : @xmath182- \frac{1}{2 } [ j_{2 } , \bar j_{1 } ] + 2[n,\bar j_3 ] - 2 [ j_3 , \hat n ] \right ) \right.\cr & + x_2 \left(- \bar \nabla j_{2 } - \nabla \bar j _ { 2 } - [ j_{1},\bar j_{1}]+ [ j_{3},\bar j_{3 } ] + 2[n,\bar j_2 ] - 2 [ j_2 , \hat n ] \right ) \cr & \left.+ x_3 \left ( -\frac{1}{2 } \bar \nabla j_{1 } - \frac{3}{2 } \nabla \bar j _ { 1 } + \frac{1}{2 } [ j_{2},\bar j_{3}]+ \frac{1}{2 } [ j_{3},\bar j_{2}]+ 2[n,\bar j_1 ] - 2 [ j_1 , \hat n]\right ) \right\ } \end{aligned}\ ] ] now we consider the following quantity : j_1(z ) = j_1(z ) e^-s where @xmath183 is the path integral measure over the fields . the previous one - point function , whatever its value is , is invariant under the reparametrization of the path integral . we further assume that the path - integral measure is also invariant under . let us mention at that point that we are simply following the method that would provide a path integral derivation of the ward identity for a global symmetry , if were indeed a global symmetry . we obtain : [ dj = jds ] j_1(z ) - j_1(z ) s = 0 it is convenient to rewrite the variation of the current as an integral over the worldsheet : j_1(z ) = d^2 w ( x_1(w ) (z - w ) + ( [ j_0(w),x_1(w)]+[j_2(w),x_3(w)]+[j_3(w),x_2(w ) ] ) ( z - w ) ) projecting equation on the @xmath15 subspaces , we obtain three operator identities : @xmath184 where the colons stand for normal ordering . next we plug the ansatz into these equations . more precisely , we use the ansatz to perform the opes on the left - hand side of the identities . since we are working at first - order in @xmath29 , the opes involving composite operators are easily dealt with : a single ope has to be taken with one or the other of the components of the composite operator . we use the equalities : _ |w = -2(z - w ) = _ w ; _ |w = 2(z - w ) = _ w we are left with some identities between operators multiplied by functions that are singular when @xmath185 . we demand that the operator identities do hold for the singular terms of order two and three : all the terms multiplying either a derivative of a delta function , a delta function , or a second - order pole shall cancel against each other . we obtain a set of linear equations that the free coefficients in the ansatz have to satisfy : @xmath186 we can play the same game replacing in equation @xmath35 by another current . for each current we obtain a new set of equations . to get more constraints for the opes involving the ghosts , we can also vary the ghosts variable instead of . in total we get 39 linear equations that the coefficients @xmath69 s in the ansatz have to satisfy . [ [ the - maurer - cartan - equation.-1 ] ] the maurer - cartan equation . + + + + + + + + + + + + + + + + + + + + + + + + + + + we can further constraint the coefficients in the ansatz by demanding compatibility with the maurer - cartan equation . projecting this equation according to the @xmath15 grading we obtain : @xmath187 + [ j_1 , \bar j_3 ] + [ j_2 , \bar j_2 ] + [ j_3 , \bar j_1 ] = 0 \cr & \p \bar j_1- \bar \p j_1 + [ j_0 , \bar j_1 ] + [ j_1 , \bar j_0 ] + [ j_2 , \bar j_3 ] + [ j_3 , \bar j_2 ] = 0 \cr & \p \bar j_2 - \bar \p j_2 + [ j_0 , \bar j_2 ] + [ j_1 , \bar j_1 ] + [ j_2 , \bar j_0 ] + [ j_3 , \bar j_3 ] = 0 \cr & \p \bar j_3 - \bar \p j_3 + [ j_0 , \bar j_3 ] + [ j_1 , \bar j_2 ] + [ j_2 , \bar j_1 ] + [ j_3 , \bar j_0 ] = 0 \end{aligned}\ ] ] the strategy is to demand that the maurer - cartan equation does hold as an operator identity . more precisely , we demand that the ope between a current and the left - hand side of does vanish . let us consider one example for illustrative purposes : we take the ope between the current @xmath35 and the left - hand side of the first line in equation : @xmath188 as previously we plug the ansatz in the previous equation , and demand that the singular terms of order three and two do vanish . we obtain the following equations : 0 = c_10^|1 + c_1 |0^|1 + c_13 0 = c_10 ^ 1 - c_1 the different lines of equation , we obtain in total 43 linear equations that the coefficients @xmath69 s in the ansatz have to satisfy . let us make a side remark here . there is no doubt that the maurer - cartan identity does hold at tree level . however it may get quantum corrections . in order to generalize the method described here to compute quantum corrections to the current algebra , one needs to assume that the maurer - cartan identity holds in the quantum theory as well . this may be interpreted as postulating quantum integrability of the model . this provides an efficient way to use quantum integrability of the model to compute the quantum current algebra . we leave it for future work . [ [ the - coefficients - of - the - current - algebra . ] ] the coefficients of the current algebra . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + using the maurer - cartan identity and reparametrization invariance of the path integral , we find in total 82 equations that constrain the 57 independent coefficients of the current algebra . this system of equation can be easily decomposed in subsystems of eight equations or less . it is remarkable that there exists a solution to this set of equations . the non - zero coefficients are given in section [ subkalgebra ] . there is however one exception for the opes between two of the currents @xmath34 and @xmath39 . in that case the equations we obtain only provide the constraints . [ [ associativity . ] ] associativity . + + + + + + + + + + + + + + the current algebra has to be associative . associativity of the current algebra can be tested in the following way . let us consider a 3-points function , for instance : j_1(x ) j_1(y ) j_2(z ) it can be computed by taking first the ope between @xmath189 and @xmath190 , and then take the ope between the resulting current and @xmath191 . but one can also start by taking the ope between @xmath192 and @xmath191 , and then take the ope of the result with @xmath189 . the two methods lead to the same result if the coefficients of the current algebra satisfy : c_11 ^ 2 c_22 = c_12 ^ 3 c_13 we can play the same game with any three - points functions . we find a large set of constraints that are all satisfied by the current algebra obtained previously . in this section we give some details on the derivations of equations , and . we want to compute the commutator of two equal - time connections @xmath193 and @xmath100 evaluated for different values of the spectral parameter @xmath75 and @xmath84 and taken in possibly different representation @xmath28 and @xmath78 . we can deduce this commutator from the current algebra . we define the commutator of equal - time operators as : = _ 0^+ ( a(+i ) b(0 ) - b(i ) a ( ) ) from this definition we extract an operative dictionary between opes and commutators . let us consider for instance the following ope : [ abope ] a(z ) b(0 ) = + + e ^(2)(z ) + + + + + ... we deduce the commutator : [ abcom ] [ a ( ) , b(0 ) ] = c ( ) - d ( ) - f(0 ) ( ) - g(0 ) ( ) + h(0 ) ( ) + i(0 ) ( ) this dictionary shows that the first - order computation of fusion presented in section [ fusionorder1 ] is equivalent to the computation of the poisson bracket of line operators in the hamiltonian formalism . notice that the ope generically contains additional sub - leading singularities , for instance @xmath194 , or even logarithms . they do not contribute to the commutator @xcite . generically , the ope contains more information than the commutator . in order to simplify the following expressions , we write the ( spacelike component of the ) flat connection @xmath73 defined in as : @xmath195 using the previous dictionary , we obtain for the commutator of two connections : @xmath196 } [ a_r&(y;\sigma ) , a_{r'}(y';\sigma ' ) ] = \cr & 2\pi i r^{-2 } \p_\sigma \delta(\sigma- \sigma ' ) \sum_{m , n } \kappa^{b_n a_m } t_{a_m}^r t_{b_n}^{r ' } ( f_m(y)f_n(y')c_{mn } - \bar f_m(y)\bar f_n(y')c_{\bar m \bar n } ) \cr & + 2\pi i r^{-2 } \delta(\sigma- \sigma ' ) \sum_{m , n , p } { f_{c_p}}^{b_n a_m } t_{a_m}^r t_{b_n}^{r ' } k_p^{c_p } ( -f_m(y)f_n(y ' ) c_{mn}^p \cr & \hspace{2.5 cm } + f_m(y)\bar f_n(y ' ) c_{m\bar n}^p + \bar f_m(y)f_n(y ' ) c_{\bar m n}^p + \bar f_m(y ) \bar f_n(y ' ) c_{\bar m \bar n}^p ) \cr & + 2\pi i r^{-2 } \delta(\sigma- \sigma ' ) \sum_{m , n , p } { f_{c_p}}^{b_n a_m } t_{a_m}^r t_{b_n}^{r ' } \bar k_p^{c_p } ( -f_m(y)f_n(y ' ) c_{mn}^{\bar p } \cr & \hspace{2.5 cm } - f_m(y)\bar f_n(y ' ) c_{m\bar n}^{\bar p } - \bar f_m(y)f_n(y ' ) c_{\bar m n}^{\bar p } + \bar f_m(y ) \bar f_n(y ' ) c_{\bar m \bar n}^{\bar p } ) \end{aligned}\ ] ] we wish to write this commutator as a @xmath12 system . from the terms coming with a derivative of the delta function in the commutator , we can read directly the @xmath81-matrix . we obtain : s = i r^-2 _ m , n ^b_n a_m t_a_m^r t_b_n^r ( f_m(y)f_n(y)c_mn - obtain . to obtain the @xmath80 matrix , we have to compare the terms coming with a delta function in and . this leads to the following equations for the components of the @xmath80 and @xmath81 matrices : @xmath197 in the previous equations , when the indices @xmath198 take the value @xmath108 or @xmath54 , one should understand@xmath199 " and @xmath200 " as being @xmath201 , etc . remarkably , this largely over - constrained system is solved by the @xmath80 and @xmath81 matrices and . [ [ appcompkalgebras ] ] comparison with previous analyses . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the current - current opes were previously discussed in the literature . in @xcite the opes for the currents of non - zero grade were computed using the background field methods . some of the opes involving the grade zero currents were further given in @xcite . the results we obtained here agree with these papers . in @xcite the current algebra was also computed using feynman diagram technology . the opes do match the ones we derived here except for those involving the currents @xmath34 , @xmath39 . this is not surprising given the gauge choice that was explicitly made for the coset element in @xcite . a consequence of this discrepancy is that the commutator of equal - time connections can not be written as a @xmath12 system with the opes of @xcite . however one should keep in mind that it is only an issue of gauge fixing . indeed in @xcite the opes of @xcite were used to compute the fusion of line operators at first order . then the @xmath80 and @xmath81 matrices were deduced by comparison with the expectations from the hamiltonian formalism . these matrices agree with the ones that we derived in this paper . in @xcite the hamiltonian formalism was used to compute the commutator of equal time connections . a careful treatment of the constraints was performed . it was argued that in the hamiltonian formalism , the flat connection realizes a @xmath12 system up to constraints generating gauge transformations . it was shown that one should add to the flat connection a term proportional to the constraints so that the commutator of connections take exactly the form of a @xmath12 system . the resulting @xmath12 system is slightly different than the one used here and in @xcite . the flat connection obtained in @xcite , including the additional term proportional to the constraints , was derived from first principles in @xcite in the green - schwarz formalism . it is remarkable that the analysis of @xcite leads to the pure spinor - like flat connections of @xcite ( without the pure spinor ghosts contribution ) and not to the bena - polchinski - roiban flat connections @xcite . this provides some evidence for the equivalence of the pure spinor and green - schwarz formulations of string theory on @xmath7 . in @xcite it was shown that the @xmath12 system of @xcite has a nice algebraic interpretation . for the purposes of this paper it is important that the @xmath80 matrix found in @xcite is identical to the one we worked with in the limit where the difference of spectral parameter is small . this guarantees that the results derived in the present paper would also hold if one were to work with the @xmath12 system of @xcite . in order to reproduce the @xmath12 system found in @xcite using opes technology , the first step would be to gauge - fix the @xmath202 gauge symmetry via a brst procedure . then one should generalize the analysis of @xcite by including in the flat connections additional terms written in terms of the ghosts resulting from the @xmath202 gauge - fixing . these new connections should realize the @xmath12 system of @xcite . in this appendix we give some details about the computations of the first - order divergences in line operators . as explained in section [ worldsheettheory ] , we use a principal - value regularization scheme . a first - order pole is regularized as : p.v . = ( + ) = and a second - order pole is regularized as : p.v . = ( + ) = ] there are three sources of divergences in the transition matrices . they are depicted in figure [ divergences ] . the first divergences come from the first order poles in the ope of two neighboring connections , say @xmath203 and @xmath204 ( case in figuredivergences ) . we evaluate the resulting currents at the point @xmath205 and perform the integration over @xmath206 . we obtain a logarithmic divergences : @xmath207 where the functions @xmath208 s were defined in . the second type of divergences come from the second - order poles in the ope between two neighboring connections ( case in figuredivergences ) . after performing the integration over the positions of the two connections , we obtain a logarithmic divergence : @xmath209 notice that there is no linear divergences . this is a pleasant feature of the regularization scheme that we are using . eventually the third type of divergences come from the second - order poles in the ope between two connections that are separated by a third one sitting in between ( case in figuredivergences ) . let us denote this third connection by @xmath210 . after performing the integrations , we obtain another logarithmic divergence : @xmath211 there is some freedom in how we write the last expression since we can commute the generators in different ways . we choose a writing that is symmetric with respect to the central connection @xmath210 . starting from a transition matrix , we compute all the different opes of the types described previously that lead to divergences . next we sum all these terms . most of the terms of the type cancel against the first terms in . we obtain : @xmath212 [ [ consequences - of - the - vanishing - of - the - dual - coxeter - number . ] ] consequences of the vanishing of the dual coxeter number . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + here we derive some identities that are useful to show the vanishing of some divergences in the line operators . these identities were first derived in @xcite . the vanishing of the dual coxeter number can be written as : [ ftt=0 ] f_c^ba[t_a , t_b } = 0 the super - jacobi identity together with the fact that @xmath213 implies : [ 13c=31c ] ^b_3 a_1[t_a_1,[t_b_3,t_c } } = ^a_1 b_3[t_b_3,[t_a_1,t_c } } the identities and further imply : @xmath214 @xmath215 @xmath216 [ [ cancellation - of - divergences . ] ] cancellation of divergences . + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us come back to the expression . we will now argue that the second piece of vanishes , as first shown in @xcite . using the identities and , we observe that the terms proportional to @xmath35 , @xmath40 , @xmath37 and @xmath42 vanish straight away . then using the identities together with the actual value of the coefficients of the current algebra , it is straightforward to check that the terms proportional to @xmath36 and @xmath41 also drop out . the vanishing of the terms proportional to @xmath34 , @xmath39 , @xmath38 and @xmath43 depends on the value of the simple poles in the opes @xmath217 , @xmath218 and @xmath219 . the method explained in appendix [ covcurrents ] to compute the current algebra does not fix completely these opes , but only gives the constraint . the identity combined with the value of the other coefficients of the current algebra implies the vanishing of all terms proportional to @xmath34 , @xmath39 , @xmath38 and @xmath43 provided we have : @xmath220 demanding consistency with the analysis of @xcite implies the previous equations . we deduce that only the first term in survives . we can write it as : @xmath221 so the first - order divergences in the transition matrix can be rewritten as : [ divtransition ] - \ { ^b_3 a_1 t_a_1 t_b_3 + ^b_2 a_2 t_a_2 t_b_2 + ^b_1 a_3 t_a_3 t_b_1 , t^b , a(x ) } in loop operators we have additional divergences coming from the collisions between two connections sitting on either side of the starting point of the integration contour . only the second - order pole in such a collision lead to a divergence . these divergences read : @xmath222 so the first - order divergences in the monodromy matrix add up to : @xmath223 taking the supertrace , we see that the transfer matrix is free of divergences at first order . in this appendix we give some details concerning the computation of the fusion of line operators at second order . in particular we describe the computation that leads to and . these terms are produced by triple collisions of connections . a triple collision means that we take one ope between two connections , and then take the ope of the resulting currents with a third connection . as explain in section [ recipefusion ] , one needs to disentangle two contributions from the opes . on one hand there is the contribution that gives a quantum correction associated with fusion . on the other hand there is the contribution that is interpreted as a regularized ope in the double line operator resulting from the process of fusion . in order to isolate the interesting part associated with fusion , we subtract the principal value from the singularities . we obtain : @xmath224 let us now face the computation of the individual quantum corrections that add up to and . in the first step of the computation of a triple collision we perform an ope between two connections sitting on different contours . we obtain intermediate currents that we evaluate on one of the two contours . the contour on which these intermediate currents are evaluated matters for the second step of the computation . the relevant ope for the first step of the computation is thus : @xmath225 where the index @xmath226 can take the values @xmath227 . in the following when the range of the sum for some index is not specified , it is understood that the sum runs over the set @xmath227 . the coefficients @xmath228 depends on the precise location where the currents are evaluated in the simple poles of the current algebra . from the coefficient given in section [ subkalgebra ] we can only deduce the sums @xmath229 and @xmath230 . it turns out that we do nt need more information about the coefficients @xmath228 for the purpose of this article ( see equation ) . actually what we want to compute here is not exactly the contribution of the triple collisions to the fusion of line operators . it is rather the part of this contribution that has not been taken into account by the first part of the computation described in section [ subfusion2nd ] and depicted in figure [ fuso2step1 ] . removing the piece of the triple collisions already taken into account amounts to perform the following replacement in the first ope : @xmath231 . in and the contribution from the first - order poles in the second ope is singled out . in the four other cases the contribution from the second - order poles in the second ope is considered.[alltriplecollisions ] ] now we can compute the individual terms that add up to and . these different terms are schematically depicted in figure [ alltriplecollisions ] . we compute separately the contribution of the first- and second - order poles in the second ope ( obviously only the first order poles have to be taken into account in the first ope ) . let us begin with the first - order poles in the second ope . the triple collision involves three connections , two of which are integrated on the same contour . there are two different cases . for a triple collision involving two neighboring connections on the first contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath232 t_{b_n}^{r ' } \cr & \times ( k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^r - { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^r - { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^r - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^r ) \cr & \quad + \bar k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^{\bar r } + { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^{\bar r } + { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^{\bar r } - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^{\bar r } ) ) \end{aligned}\ ] ] where we wrote @xmath233 as a shorthand for @xmath234 . similarly we will write @xmath235 for @xmath236 . for a triple collision involving two neighboring connections on the second contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath237 \cr & \times ( k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^r + { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^r + { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^r + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^r ) \cr & \quad + \bar k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^{\bar r } - { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^{\bar r } - { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^{\bar r } + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^{\bar r } ) ) \end{aligned}\ ] ] next we consider the second - order poles in the second ope . there are now four different cases . for a triple collision involving two neighboring connections on the first contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath238 and for a triple collision involving two neighboring connections on the second contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath239 we also obtain a non - zero contribution if the two connections that are on the same contour are separated by a third one . when this happens on the first contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath240 t_{b_n}^{r ' } \cr & \quad \times ( { \tilde{d'}}_{mn}^s f_p c_{sp } - { \tilde{d'}}_{mn}^{\bar s } \bar f_p c_{\bar s\bar p } ) ( f_r k_r^{e_r } + \bar f_r \bar k_r^{e_r})\cr + ( i\pi r^{-2})^2 & \delta'_{\epsilon}(\sigma_1-\sigma')\delta_{\epsilon}(\sigma_2-\sigma ' ) \frac{1}{2 } \sum_{n , p , q=0}^3\sum_{m , r , s } { f_{c_p}}^{b_n a_m } { f_{e_r}}^{c_p d_q } \ { t_{d_q}^r , t_{a_m}^r ] t_{b_n}^{r ' } \cr & \quad \times ( { \tilde{d'}}_{pn}^s f_m c_{sm } - { \tilde{d'}}_{pn}^{\bar s } \bar f_m c_{\bar s\bar m } ) ( f_r k_r^{e_r } + \bar f_r \bar k_r^{e_r})\end{aligned}\ ] ] finally for two connections separated by a third one in the second contour ( case in figure [ alltriplecollisions ] ) , we obtain : @xmath241 \cr & \quad \times ( { \tilde{d}}_{mn}^s f'_p c_{sp } - { \tilde{d}}_{mn}^{\bar s } \bar f'_p c_{\bar s\bar p } ) ( f'_r k_r^{e_r } + \bar f'_r \bar k_r^{e_r})\cr + ( i\pi r^{-2})^2 & \delta'_{\epsilon}(\sigma'_1-\sigma)\delta_{\epsilon}(\sigma'_2-\sigma ) \frac{1}{2 } \sum_{n , p , q=0}^3 \sum_{m , r , s } { f_{c_p}}^{b_n a_m } { f_{e_r}}^{d_q c_p } t_{a_m}^r \ { t_{b_n}^{r'},t_{d_q}^{r ' } ] \cr & \quad \times ( { \tilde{d}}_{pn}^s f'_m c_{sm } - { \tilde{d}}_{pn}^{\bar s } \bar f'_m c_{\bar s\bar m } ) ( f'_r k_r^{e_r } + \bar f'_r \bar k_r^{e_r})\\end{aligned}\ ] ] next we have to perform the integration over the free coordinates in the previous results . the integrals over the regularized delta functions provide a well - defined answer . this is an advantage of the ope formalism with respect to the hamiltonian formalism . the integrals needed are given below . the results are given in the limit @xmath242 . @xmath243 @xmath244 @xmath245 it is sometimes convenient to write @xmath246 as @xmath247 . notice that the integral over the squared regularized delta function is divergent in the limit @xmath242 : _ c^d d ^2_(_2- ) = |[a , b]| where we denoted by @xmath248\cap[c , d]|$ ] the length of the overlap of the intervals @xmath85 $ ] and @xmath98 $ ] . similarly we have : @xmath249 where we can also replace @xmath250 by @xmath251 . finally we can sum the various contributions from triple collisions . the terms of the form combined with the second and third terms of lead to : @xmath252\cap[c , d]}d\sigma \left \lfloor \int_\sigma^b a \right\rceil^i \left \lfloor \int_\sigma^d a ' \right\rceil^{i ' } \cr & \times ( i\pi r^{-2})^2 \tilde{\sum_{m , n , p , q , r } } { f_{c_p}}^{b_n a_m } { f_{e_r}}^{c_p d_q } \ { t_{d_q}^r , t_{a_m}^r ] t_{b_n}^{r ' } \cr & \quad \times ( k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^r - { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^r - { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^r - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^r \cr & \quad \quad\quad+ \frac{1}{2 } f_r \sum_s ( { \tilde{d'}}_{mn}^s f_p c_{sp } + { \tilde{d'}}_{pn}^s f_m c_{sm } - { \tilde{d'}}_{mn}^{\bar s } \bar f_p c_{\bar s\bar p } - { \tilde{d'}}_{pn}^{\bar s } \bar f_m c_{\bar s\bar m } ) ) \cr & \quad\quad + \bar k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^{\bar r } + { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^{\bar r } + { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^{\bar r } - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^{\bar r } \cr & \quad\quad\quad+ \frac{1}{2 } \bar f_r \sum_s ( { \tilde{d'}}_{mn}^s f_p c_{sp } + { \tilde{d'}}_{pn}^s f_m c_{sm } - { \tilde{d'}}_{mn}^{\bar s } \bar f_p c_{\bar s\bar p } - { \tilde{d'}}_{pn}^{\bar s } \bar f_m c_{\bar s\bar m } ) ) ) \cr & \times \left \lfloor \int_a^\sigma a \right\rceil^{m - i } \left \lfloor \int_c^\sigma a ' \right\rceil^{m'-i ' } \end{aligned}\ ] ] in order to shorten the previous expression we introduce the symbol @xmath253 with the following meaning : for each term in the expression , the lower indices of a coefficient @xmath254 have to be summed over the values @xmath255 , while all other indices have to be summed over the values @xmath227 . similarly the terms of the form combined with the second and third terms of lead to : @xmath256\cap[c , d]}d\sigma \left \lfloor \int_\sigma^b a \right\rceil^i \left \lfloor \int_\sigma^d a ' \right\rceil^{i ' } \cr & \times ( i\pi r^{-2})^2 \tilde{\sum_{m , n , p , q , r } } { f_{c_p}}^{b_n a_m } { f_{e_r}}^ { d_q c_p } t_{a_m}^r \ { t_{b_n}^{r ' } , t_{d_q}^{r ' } ] \cr & \quad \times ( k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^r + { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^r + { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^r + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^r \cr & \quad \quad\quad- \frac{1}{2 } f'_r \sum_s ( { \tilde{d}}_{mn}^s f'_p c_{sp } + { \tilde{d}}_{pn}^s f'_m c_{sm } - { \tilde{d}}_{mn}^{\bar s } \bar f'_p c_{\bar s\bar p } - { \tilde{d}}_{pn}^{\bar s } \bar f'_m c_{\bar s\bar m } ) ) \cr & \quad\quad + \bar k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^{\bar r } - { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^{\bar r } - { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^{\bar r } + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^{\bar r } ) \cr & \quad\quad\quad- \frac{1}{2 } \bar f'_r \sum_s ( { \tilde{d}}_{mn}^s f'_p c_{sp } + { \tilde{d}}_{pn}^s f'_m c_{sm } - { \tilde{d}}_{mn}^{\bar s } \bar f'_p c_{\bar s\bar p } - { \tilde{d}}_{pn}^{\bar s } \bar f'_m c_{\bar s\bar m } ) ) ) \cr & \times \left \lfloor \int_a^\sigma a \right\rceil^{m - i } \left \lfloor \int_c^\sigma a ' \right\rceil^{m'-i ' } \end{aligned}\ ] ] the sum of and leads to where the operator @xmath116 is given by : @xmath257 t_{b_n}^{r ' } \cr & \quad \times ( k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^r - { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^r - { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^r - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^r \cr & \quad \quad\quad+ \frac{1}{2 } f_r \sum_s ( { \tilde{d'}}_{mn}^s f_p c_{sp } + { \tilde{d'}}_{pn}^s f_m c_{sm } - { \tilde{d'}}_{mn}^{\bar s } \bar f_p c_{\bar s\bar p } - { \tilde{d'}}_{pn}^{\bar s } \bar f_m c_{\bar s\bar m } ) ) \cr & \quad\quad + \bar k_r^{e_r}({\tilde{d'}}_{mn}^p f_q c_{pq}^{\bar r } + { \tilde{d'}}_{mn}^p \bar f_q c_{p\bar q}^{\bar r } + { \tilde{d'}}_{mn}^{\bar p } f_q c_{\bar pq}^{\bar r } - { \tilde{d'}}_{mn}^{\bar p } \bar f_q c_{\bar p \bar q}^{\bar r } \cr & \quad\quad\quad+ \frac{1}{2 } \bar f_r \sum_s ( { \tilde{d'}}_{mn}^s f_p c_{sp } + { \tilde{d'}}_{pn}^s f_m c_{sm } - { \tilde{d'}}_{mn}^{\bar s } \bar f_p c_{\bar s\bar p } - { \tilde{d'}}_{pn}^{\bar s } \bar f_m c_{\bar s\bar m})))\cr % & + { f_{c_p}}^{b_n a_m } { f_{e_r}}^ { d_q c_p } t_{a_m}^r \ { t_{b_n}^{r ' } , t_{d_q}^{r ' } ] \cr & \quad \times ( k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^r + { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^r + { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^r + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^r \cr & \quad \quad\quad- \frac{1}{2 } f'_r \sum_s ( { \tilde{d}}_{mn}^s f'_p c_{sp } + { \tilde{d}}_{pn}^s f'_m c_{sm } - { \tilde{d}}_{mn}^{\bar s } \bar f'_p c_{\bar s\bar p } - { \tilde{d}}_{pn}^{\bar s } \bar f'_m c_{\bar s\bar m } ) ) \cr & \quad\quad + \bar k_r^{e_r}(-{\tilde{d}}_{mn}^p f'_q c_{pq}^{\bar r } - { \tilde{d}}_{mn}^p \bar f'_q c_{p\bar q}^{\bar r } - { \tilde{d}}_{mn}^{\bar p } f'_q c_{\bar pq}^{\bar r } + { \tilde{d}}_{mn}^{\bar p } \bar f'_q c_{\bar p \bar q}^{\bar r } ) \cr & \quad\quad\quad- \frac{1}{2 } \bar f'_r \sum_s ( { \tilde{d}}_{mn}^s f'_p c_{sp } + { \tilde{d}}_{pn}^s f'_m c_{sm } - { \tilde{d}}_{mn}^{\bar s } \bar f'_p c_{\bar s\bar p } - { \tilde{d}}_{pn}^{\bar s } \bar f'_m c_{\bar s\bar m})))\end{aligned}\ ] ] the combination of the terms and together with the first terms of and simplifies to : @xmath258 this can be rewritten as where the matrix @xmath119 is given by : @xmath259 for the purposes of this paper it is interesting to take the limit where the difference of spectral parameters is small . more precisely we assume that the difference @xmath261 is of order @xmath262 . in this limit one term dominates the previous result . this follows essentially from the observation that the @xmath80 matrix satisfies . in the limit @xmath260 , the @xmath80 matrix is no longer of order @xmath29 but rather of order @xmath263 . consequently the coefficients @xmath264 introduced in behave like : @xmath265 using equation , we deduce that simplifies to : @xmath266\cap[c , d]}d\sigma \left \lfloor \int_\sigma^b a \right\rceil^i \left \lfloor \int_\sigma^d a ' \right\rceil^{i ' } \cr & \times ( i\pi r^{-2})^2 \sum_{m , n , p , q=0}^3\sum_{r } { f_{c_p}}^{b_n a_m } { f_{e_r}}^{c_p d_q } \ { t_{d_q}^r , t_{a_m}^r ] t_{b_n}^{r ' } \cr & \quad \times \left ( -\frac{1}{16}\frac{y^2(y^2+y^{-2})^4}{y - y'}\right ) \left(\p_y f_r(y ) k_r^{e_r } + \p_y \bar f_r(y ) \bar k_r^{e_r}\right ) \cr & \times \left \lfloor \int_a^\sigma a \right\rceil^{m - i } \left \lfloor \int_c^\sigma a ' \right\rceil^{m'-i ' } + \mathcal{o}(r^{-4})\end{aligned}\ ] ] remarkably the combination of currents that factors out is the derivative of the components of the flat connection with respect to the spectral parameter . similarly simplifies to : @xmath267\cap[c , d]}d\sigma \left \lfloor \int_\sigma^b a \right\rceil^i \left \lfloor \int_\sigma^d a ' \right\rceil^{i ' } \cr & \times ( i\pi r^{-2})^2 \sum_{m , n , p , q=0}^3\sum_{r } { f_{c_p}}^{b_n a_m } { f_{e_r}}^ { d_q c_p } t_{a_m}^r \ { t_{b_n}^{r ' } , t_{d_q}^{r ' } ] \cr & \quad \times \left ( -\frac{1}{16}\frac{y^2(y^2+y^{-2})^4}{y - y'}\right ) \left(\p_y f_r(y ) k_r^{e_r } + \p_y \bar f_r(y ) \bar k_r^{e_r}\right ) \cr & \times \left \lfloor \int_a^\sigma a \right\rceil^{m - i } \left \lfloor \int_c^\sigma a ' \right\rceil^{m'-i ' } + \mathcal{o}(r^{-4})\end{aligned}\ ] ] so the operator @xmath116 becomes : @xmath268 t_{b_n}^{r ' } + { f_{c_p}}^{b_n a_m } { f_{e_r}}^ { d_q c_p } t_{a_m}^r \ { t_{b_n}^{r ' } , t_{d_q}^{r ' } ] \right ) + \mathcal{o}(r^{0})\end{aligned}\ ] ] eventually the term remains of order @xmath55 , since : [ tttildelim ] -_mn^p f_q c_pq + _ mn^|p |f_q c_|p|q -_rm^p f_q c_pq + _ rm^|p |f_q c_|p|q = 0 + ( r^-4 ) consequently the matrix @xmath119 defined in remains of order one . actually it might be that the term and thus the matrix @xmath119 cancel exactly . this could be decided by computing the coefficients of the derivatives of the currents in the current - current opes . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' adv . * 2 * ( 1998 ) 231 [ int . j. theor . * 38 * ( 1999 ) 1113 ] [ arxiv : hep - th/9711200 ] . s. s. gubser , i. r. klebanov , a. m. polyakov , `` gauge theory correlators from noncritical string theory , '' phys . * b428 * ( 1998 ) 105 - 114 . [ hep - th/9802109 ] . n. gromov , v. kazakov and p. vieira , `` exact spectrum of anomalous dimensions of planar n=4 supersymmetric yang - mills theory , '' phys . * 103 * , 131601 ( 2009 ) [ arxiv:0901.3753 [ hep - th ] ] . n. beisert , b. eden , m. staudacher , `` transcendentality and crossing , '' j. stat . mech . * 0701 * ( 2007 ) p01021 . [ hep - th/0610251 ] . n. gromov , v. kazakov , p. vieira , `` exact spectrum of planar @xmath269 supersymmetric yang - mills theory : konishi dimension at any coupling , '' phys . * 104 * ( 2010 ) 211601 . [ arxiv:0906.4240 [ hep - th ] ] . n. gromov , v. kazakov , a. kozak , p. vieira , `` exact spectrum of anomalous dimensions of planar n = 4 supersymmetric yang - mills theory : tba and excited states , '' lett . * 91 * ( 2010 ) 265 - 287 . [ arxiv:0902.4458 [ hep - th ] ] . v. v. bazhanov , s. l. lukyanov , a. b. zamolodchikov , `` integrable structure of conformal field theory , quantum kdv theory and thermodynamic bethe ansatz , '' commun . * 177 * ( 1996 ) 381 - 398 . [ hep - th/9412229 ] v. kazakov and p. vieira , `` from characters to quantum ( super)spin chains via fusion , '' jhep * 0810 * ( 2008 ) 050 [ arxiv:0711.2470 [ hep - th ] ] . r. benichou , `` fusion of line operators in conformal sigma - models on supergroups , and the hirota equation , '' jhep * 1101 * ( 2011 ) 066 . [ arxiv:1011.3158 [ hep - th ] ] n. berkovits , `` super - poincare covariant quantization of the superstring , '' jhep * 0004 * ( 2000 ) 018 [ arxiv : hep - th/0001035 ] . n. berkovits , `` quantum consistency of the superstring in ads(5 ) x s**5 background , '' jhep * 0503 * ( 2005 ) 041 [ arxiv : hep - th/0411170 ] . l. mazzucato , `` superstrings in ads , '' [ arxiv:1104.2604 [ hep - th ] ] . v. g. m. puletti , `` operator product expansion for pure spinor superstring on ads(5 ) x s**5 , '' jhep * 0610 * ( 2006 ) 057 [ arxiv : hep - th/0607076 ] . v. g. m. puletti , `` aspects of quantum integrability for pure spinor superstring in ads(5 ) x s**5 , '' jhep * 0809 * ( 2008 ) 070 . [ arxiv:0808.0282 [ hep - th ] ] . o. a. bedoya , d. z. marchioro , d. l. nedel and b. c. vallilo , `` quantum current algebra for the @xmath7 superstring , '' arxiv:1003.0701 [ hep - th ] . r. r. metsaev and a. a. tseytlin , `` type iib superstring action in ads(5 ) x s(5 ) background , '' nucl . b * 533 * ( 1998 ) 109 [ arxiv : hep - th/9805028 ] . i. bena , j. polchinski and r. roiban , `` hidden symmetries of the ads(5 ) x s**5 superstring , '' phys . d * 69 * ( 2004 ) 046002 [ arxiv : hep - th/0305116 ] . b. c. vallilo , `` flat currents in the classical ads(5 ) x s**5 pure spinor superstring , '' jhep * 0403 * ( 2004 ) 037 [ arxiv : hep - th/0307018 ] . a. mikhailov , s. schafer - nameki , `` algebra of transfer - matrices and yang - baxter equations on the string worldsheet in ads(5 ) x s(5 ) , '' nucl . * b802 * ( 2008 ) 1 - 39 . [ arxiv:0712.4278 [ hep - th ] ] . k. zarembo , `` strings on semisymmetric superspaces , '' arxiv:1003.0465 [ hep - th ] . d. bombardelli , d. fioravanti , r. tateo , `` tba and y - system for planar ads(4)/cft(3 ) , '' nucl . * b834 * ( 2010 ) 543 - 561 . [ arxiv:0912.4715 [ hep - th ] ] . n. berkovits , c. vafa and e. witten , `` conformal field theory of ads background with ramond - ramond flux , '' jhep * 9903 * ( 1999 ) 018 [ arxiv : hep - th/9902098 ] . a. pakman , l. rastelli , s. s. razamat , `` a spin chain for the symmetric product cft(2 ) , '' jhep * 1005 * ( 2010 ) 099 . [ arxiv:0912.0959 [ hep - th ] ] . n. berkovits , `` quantization of the type ii superstring in a curved six - dimensional background , '' nucl . b * 565 * ( 2000 ) 333 [ arxiv : hep - th/9908041 ] . n. berkovits , m. bershadsky , t. hauer , s. zhukov and b. zwiebach , `` superstring theory on ads(2 ) x s(2 ) as a coset supermanifold , '' nucl . b * 567 * ( 2000 ) 61 [ arxiv : hep - th/9907200 ] . c. a. s. young , `` non - local charges , z(m ) gradings and coset space actions , '' phys . b * 632 * , 559 ( 2006 ) [ arxiv : hep - th/0503008 ] . i. adam , a. dekel , l. mazzucato and y. oz , `` integrability of type ii superstrings on ramond - ramond backgrounds in various dimensions , '' jhep * 0706 * ( 2007 ) 085 [ arxiv : hep - th/0702083 ] . g. parisi , n. sourlas , `` self avoiding walk and supersymmetry , '' j. phys . * 41 * ( 1980 ) 403 . m. r. zirnbauer , `` conformal field theory of the integer quantum hall plateau transition , '' arxiv : hep - th/9905054 . m. bershadsky , s. zhukov and a. vaintrob , `` psl(n|n ) sigma model as a conformal field theory , '' nucl . b * 559 * ( 1999 ) 205 [ arxiv : hep - th/9902180 ] . a. babichenko , `` conformal invariance and quantum integrability of sigma models on symmetric superspaces , '' phys . b * 648 * , 254 ( 2007 ) [ arxiv : hep - th/0611214 ] . c. candu , t. creutzig , v. mitev , v. schomerus , `` cohomological reduction of sigma models , '' jhep * 1005 * ( 2010 ) 047 . [ arxiv:1001.1344 [ hep - th ] ] . s. k. ashok , r. benichou and j. troost , `` conformal current algebra in two dimensions , '' arxiv:0903.4277 [ hep - th ] . r. benichou and j. troost , `` the conformal current algebra on supergroups with applications to the spectrum and integrability , '' arxiv:1002.3712 [ hep - th ] . s. k. ashok , r. benichou and j. troost , `` asymptotic symmetries of string theory on ads3 x s3 with ramond - ramond fluxes , '' jhep * 0910 * ( 2009 ) 051 [ arxiv:0907.1242 [ hep - th ] ] . b. c. vallilo , l. mazzucato , `` the konishi multiplet at strong coupling , '' [ arxiv:1102.1219 [ hep - th ] ] .

the aim of this paper is to critically reconsider the physical origin of the onset of dynamical arrest and the associated crossovers between distinct dynamical regimes displayed by liquids supercooled through their melting temperature towards the glass transition @xcite . we do this by extending the real space theoretical framework based on dynamic facilitation of refs.@xcite to the moderately supercooled regime corresponding to the region where mode coupling theory ( mct ) @xcite supposedly applies , as reviewed in @xcite . this novel approach takes directly into account the spatial aspects of the dynamics , in particular those related to dynamic heterogeneity @xcite , in contrast with many other theories @xcite . our analysis shows that the onset of slowing down can be understood in a simple physical way in terms of the dynamical properties of effective excitations , or defects , as a progressive crossover from a regime of fast dynamics dense in defect clusters , to one of slow heterogeneous dynamics dominated by isolated localized defects . we demonstrate that this real space picture explains the observed crossover temperatures , challenges the idea that these crossovers are related to changes in the topography of the energy surface or to mct singularities , and is able to account for the apparent correlations observed between `` landscape '' and dynamical properties . the paper is organized as follows . in the rest of the introduction we review the mct and energy landscape points of view , discuss their problems and limitations , and describe the alternative real space perspective we will pursue . in section [ physics ] we develop the physical picture of the onset of slowing down and dynamical crossovers which emerges from our theoretical approach . in section [ consequence ] we discuss its quantitative consequences and compare them to published numerical results . in section [ landscape ] we show how our approach also enables to derive the observed properties of the potential energy landscape of supercooled liquids . finally , in section [ conclusion ] we discuss our results and state our conclusions . it is often assumed that the initial slowing down of the dynamics of supercooled liquids can be rationalized by mct @xcite . numerical simulations are now able to investigate the first five decades in time of this slowing down @xcite , so this is also the regime which has been studied in greatest microscopic detail . the degree of success of mct is still a matter of debate . this is due to the fact that the central mct prediction , a complete dynamical arrest at a temperature @xmath0 where the alpha - relaxation time diverges as a power law , @xmath1 , is actually never observed , but a power law fit to the data apparently works on a restricted time window @xcite . the appearance of new mechanisms for relaxation , often termed `` activated processes '' , but seldom described in any detail , is then invoked to explain the discrepancy between observations and mct predictions . in fact , activated processes are actually quantitatively defined , within mct , by deviations between data and predictions @xcite . it is believed that activated processes become relevant close to the dynamical singularity @xmath0 , their main effect being to prevent the predicted transition . from @xmath0 downwards it is assumed that the physics is dominated by activated processes , which determine also the canonical features of glass transition phenomena @xcite : non - exponential relaxation , strong and fragile liquid behaviours , decoupling between transport coefficients , etc . it is sometimes said that the relevant physics for the glass transition sets in at @xmath0 , and is therefore out of reach of numerical simulations @xcite . crossovers into the activated dynamics regime are also reported to occur at temperatures @xmath2 @xcite or @xmath3 @xcite , depending on which aspect of the physics is considered . it is believed that all these temperatures are close enough to be taken as equivalent , @xmath4 @xcite . the above scenario is apparently corroborated by the study of the statistical properties of the potential energy landscape of model liquids @xcite . from the properties of the landscape two temperatures seem to emerge , @xmath5 and @xmath0 @xcite . the onset of slowing down of the dynamics takes place at @xmath5 , and coincides with the temperature below which the average energy of inherent structures ( is ) , i.e. , local minima of the potential energy @xcite , @xmath6 , starts to decrease markedly , see fig . [ crossoverlj ] . this has been interpreted as the sign that the landscape starts to `` influence '' the dynamical behaviour @xcite . at @xmath0 , it is further argued , a second change in the landscape properties takes place , which is indicated by several observations @xcite . for example , the mean - square displacement from an equilibrated configuration to its corresponding inherent structure , @xmath7 , is proportional to @xmath8 below @xmath0 , as expected from pure vibrations in quadratic wells , but the temperature dependence changes above @xmath0 , revealing `` anharmonicities '' in the landscape @xcite , see fig . [ crossoverlj ] . another indication of a topological change in the energy landscape was discussed in ref . @xcite , in analogy with what happens in mean field models @xcite : the vanishing as @xmath8 approaches @xmath0 of the mean intensive number of negative directions ( intensive index ) of stationary points of the potential energy , @xmath9 . numerical simulations @xcite found that @xmath9 decreases with decreasing @xmath8 , and fits were performed to show that @xmath10 @xcite . the physical interpretation of this result is the apparent existence at @xmath0 of a `` geometric transition '' between a `` saddle dominated regime '' above @xmath0 and a `` minima dominated regime '' below @xmath0 @xcite . @xmath0 would then really coincide with the appearance of activated processes , described in a topographic language as `` hopping '' between minima of the landscape . analogous findings had been previously reported from instantaneous normal mode analysis of equilibrium configurations @xcite , also supporting a qualitative change in the landscape topology close to @xmath0 @xcite . figure [ crossoverlj ] summarizes this mct / landscape scenario with published numerical data @xcite obtained for the standard supercooled liquid model of ref . @xcite . at first sight fig . [ crossoverlj ] appears as convincing evidence in favour of the mct / landscape interpretation of the dynamics . on closer inspection , however , the above scenario is less robust . several qualitative and quantitative observations do not fit into the picture presented above . _ ( i ) activated dynamics above @xmath0 . _ the main idea behind the landscape approach @xcite is that vibrations and structural relaxations take place on very different time scales , so that the system is `` trapped '' and vibrates in one minimum before `` hopping '' to another minimum . this is indeed observed using the mapping from trajectories to is @xcite in simulations of sufficiently small systems @xcite . we have also discussed theoretically this issue in a recent work @xcite . given that numerical studies were performed much above @xmath0 , a crucial conclusion of refs.@xcite is that `` activated dynamics '' is indeed present in this temperature regime . _ ( ii ) heterogeneous dynamics above @xmath0 . _ it is now well - documented that the dynamics of supercooled liquids , even above the mode coupling temperature @xmath0 , is heterogeneous in the sense that the local relaxation time has non - trivial spatial correlations @xcite . this phenomenon is not very different from what happens experimentally close to the glass transition @xmath11 @xcite . besides , the decoupling between transport coefficients , which is also interpreted in terms of dynamical heterogeneity @xcite , is observed in numerical simulations above @xmath0 @xcite , although the effect is quantitatively less pronounced than in experiments near @xmath11 @xcite . _ ( iii ) presence of saddles below @xmath0 . _ despite claims based on numerical results that @xmath0 marks a real change in the topology of the landscape @xcite , there are strong indications that this is at best only a crossover @xcite , and at worst a biased interpretation of numerical data @xcite . for instance , careful numerical studies have shown that the saddle index @xmath9 remains positive even below @xmath0 @xcite . moreover , ref . @xcite argues convincingly that the data for @xmath9 can be described by an arrhenius law , @xmath12 , with @xmath13 an energy scale , which means that @xmath0 does not mark any particular change in the saddle index . the problems described above can be overcome through an alternative perspective on glass transition phenomena which puts the real space aspects of the dynamics at its core . this is the approach developed in refs.@xcite . interestingly , several of its central concepts , like the relevance to the dynamics of localized excitations @xcite and the importance of effective kinetic constraints @xcite , have been present in the literature for many years . moreover , one of the original key observations of dynamic heterogeneity in glass formers was made by harrowell and coworkers @xcite in the models of ref.@xcite . see ref . @xcite for an exhaustive review . our approach relies on only two basic observations . ( i ) at low temperature mobility within a supercooled liquid is sparse and very few particles are mobile . this is somewhat equivalent to the statement that particles are `` caged '' for long period of times , as reflected by a plateau in the mean - square displacement of individual particles . ( ii ) when a microscopic region of space is mobile it influences the dynamics of neighbouring regions , enabling them to become mobile , and thus allowing mobility to propagate in the system . this is the concept of dynamic facilitation @xcite . the observation that very mobile particles in a supercooled liquid move along correlated `` strings '' @xcite is a confirmation of this fundamental idea . from these two concepts it is possible to build effective microscopic models for glass formers by means of a coarse - graining procedure . this procedure can be schematically described as follows @xcite . spatially , the particles are coarse - grained over a length scale @xmath14 of the order of the static correlation length given by the pair correlation function . this removes any static correlations between coarse grained regions of linear size @xmath14 . cells are then identified according to their mobility by performing a coarse - graining on a microscopic time scale @xmath15 . in its simplest version cells are identified by a scalar `` mobility field '' , @xmath16 , the values @xmath17 corresponding to an immobile / mobile cell at position @xmath18 and time @xmath19 . the next step is to replace continuous space by a lattice , @xmath20 . mobile or excited cells carry a free energy cost , so when mobility is low it is reasonable to describe their static properties with a non - interacting hamiltonian @xcite , @xmath21 for a lattice of @xmath22 sites . the link between mobility and potential energy @xcite has also been observed in numerical simulations @xcite . the coarse - graining procedure described above will generate local dynamical rules for the mobility field . the prominent feature of this dynamics will be dynamic facilitation , which in its simplest version states that a cell at site @xmath23 is allowed to move only if it has an excited nearest neighbour @xcite , @xmath24 { } \\ \end{array } n_i=1 , \label{rates}\ ] ] where @xmath25 , and @xmath26 indicates nearest neighbour , and @xmath27 represents the average concentration of excited cells easily deduced from ( [ ham ] ) : @xmath28 explicit examples where dynamic facilitation is generated under coarse - graining can be found in ref . clearly , different models are defined simply by changing the kinetic rules , e.g. , the number or directionality of mobile neighbours required to move @xcite . also , a more complex mobility field may be required to account quantitatively for all glass transition features @xcite . crucially , we will show that the physical mechanisms which explain the onset of slowing down and crossovers between different dynamical regimes in supercooled liquids are generic to this class of models . this means that we can use the simplest of them , the fredrickson - andersen ( fa ) model defined by eqs . ( [ ham],[rates ] ) in one spatial dimension ( hereafter 1d fa model ) to make detailed predictions and calculations . in order to understand the physics captured by the coarse - grained facilitated models defined above , it is useful to look at trajectories , that is , space - time representations of the dynamics @xcite . we show in fig . [ traj ] three representative trajectories for the 1d fa model , where mobile cells ( defects ) are black , and immobile ones are white . from the trajectories , the principal observation is the appearance at low temperatures of non - trivial spatio - temporal correlations , seen as spatially and temporally extended domains of immobile cells delimited by isolated defects @xcite . in 1d they look like `` bubbles '' @xcite , and trajectories are dense assemblies of these slow bubbles . this nanoscopic ordering in trajectory space is the cause of the phenomenon of dynamic heterogeneity observed experimentally and in simulations @xcite . dynamic heterogeneity is the central aspect of the physics of supercooled liquids : it is naturally captured by our approach . the statistical mechanics of trajectories , rather than configurations , determines the dynamical behaviour . for example , due to the non - interacting hamiltonian ( [ ham ] ) , static correlations are trivial . however , when trajectories are considered , it is clear that cells become dynamically correlated . in other words , these models naturally predict the existence of a dynamical correlation @xmath29 which grows when the dynamics slows down . this statement can be quantified @xcite by studying multi - point functions , for example , @xmath30 , where @xmath31 is a dynamical correlator at site @xmath23 ( below we will consider the persistence of site @xmath23 ) . the spatial decay of a function like @xmath32 defines unambiguously the dynamical correlation length @xmath29 , as already discussed theoretically @xcite and measured in numerical simulations @xcite . furthermore , the joint distributions of time and length scales give rise to the canonical features of glass formers , like stretched relaxation , decoupling between transport coefficients , and ( kinetic and thermodynamic ) strong and fragile behaviours @xcite . let us take a closer look at the temperature evolution of the trajectories in fig . [ traj ] . starting from the very low temperatures where trajectories consist of a mixture of slow bubbles , the dynamics can be understood in terms of the opening and closing of bubbles , that is , the branching of an excitation line or the coalescence of two . as shown in @xcite , these events are the `` hopping between minima '' described in ref . therefore , we have a clear understanding of `` activated processes '' and of their statistical properties @xcite . as temperature is increased more and more defects are present . this has several consequences . first , the typical spatial and temporal extension of bubbles reduces , that is , the system becomes faster and less heterogeneous . second , clusters of defects become more common . these objects are important because their dynamics is completely different from that of isolated defects . in a cluster , defects do not have to diffuse and create or annihilate other defects but can instantaneously relax in a much faster process . at high temperature , the dynamics is fast because almost no bubbles are present , and the dynamics is governed by clusters of defects . interestingly , at some intermediate temperature ( middle frame in fig . [ traj ] ) a coexistence between clusters and isolated defects is observed , so that the dynamics has a `` mixed '' character . clusters of defects disappear much faster with decreasing temperature than the overall concentration of defects . the probability to have a cluster of @xmath33 defects is indeed @xmath34 , so that at low @xmath8 we have @xmath35 . the coexistence of fast and slow processes with different temperature behaviour has a direct influence on the distribution of local relaxation times , which we present in fig . [ distrib ] for various temperatures . at high temperature , @xmath36 , where fast processes are dominant , the distribution is exponential , with a mean which depends weakly on temperature ( below we discuss in detail its temperature dependence ) . around @xmath37 , a shoulder develops in the large time tail of the distribution , corresponding to the appearance of the bubbles in the trajectories of fig . this marks the increasing relevance of slow processes and the growth of the dynamic correlation length beyond the microscopic high temperature value . this temperature corresponds therefore to the onset temperature , @xmath38 . decreasing further the temperature , we clearly see a regime of mixed dynamics . at @xmath39 , for instance , there are two peaks in the distribution , reflecting the coexistence of clusters and isolated defects . at this temperature , the time scale has already increased by several orders of magnitude , and the dynamic correlation length is about @xmath40 . finally , further decrease in temperature makes clusters of defects very rare and we are left only with the contribution of isolated defects . this low temperature distribution is the one discussed in refs . @xcite , which in turn implies the stretched exponential decay of dynamical correlators . the contribution of clusters becomes negligible beyond a second crossover temperature , here @xmath41 . while this crossover temperature is not linked in any way to the mode coupling singularity @xmath0 , this choice of notation will become clear shortly . from these distributions , it is possible to propose an empirical but quantitative determination of @xmath5 and @xmath0 . at each temperature @xmath8 , the distribution is composed of fast processes , @xmath42 , and slow processes , @xmath43 , where @xmath44 can be defined , e.g. , as in ref . @xcite . requiring that slow processes are a significant fraction ( say 90% ) of the distribution leads to the definition of @xmath0 : @xmath45 where in an abuse of notation we also call @xmath46 the distribution of persistence times . requiring that slow processes , while not dominant in number , still contribute to a significant fraction of the mean relaxation time ( say again 90% ) leads to the definition of @xmath5 : @xmath47 in fig . [ regime ] , we summarize the physical picture that emerges from the considerations of this section . from the distributions of fig . [ distrib ] , we recognize that the dynamics becomes slow when the temperature is decreased below the onset temperature @xmath5 . this distinguishes the trivial liquid and the slow glassy regimes . from the trajectories of fig . [ traj ] , we were also able to distinguish between fast , non - activated processes ( clusters ) , and slow , activated processes ( isolated defects , bubbles ) . the former , typical of the liquid high-@xmath8 physics , become negligible below @xmath0 , while the latter , typical of the low-@xmath8 physics , appear at the onset temperature @xmath5 . as a consequence , the crossover region @xmath48 contains traces of both high-@xmath8 and low-@xmath8 physics , as observed in the time distributions of fig . [ distrib ] . the physical picture we have presented for the onset of slowing down , based on the increasing relevance of a dynamically heterogeneous evolution of the system , leads to quantitative predictions which are in good agreement with previous numerical and experimental studies , as we discuss in this section . the basic dynamical quantities recorded in experiments and simulations of supercooled liquids are spatially averaged two - time functions . simulations usually focus on the time domain and typically consider density - density correlation functions , while experimental results are often expressed in the frequency domain , measuring for instance dielectric susceptibilities . we will only consider systems in equilibrium so that the information content of both kinds of measurements is equivalent . from the distributions of times , fig . [ distrib ] , it is easy to derive dynamical correlators for the 1d fa model considered here . the spatially averaged persistence function reads , @xmath49 the behaviour of @xmath50 as a function of time for various temperatures is shown in fig . [ persistence ] . at very low temperatures , the persistence function is known exactly due to the diffusion properties of isolated defects , and one gets @xmath51 , \label{stretched}\ ] ] where @xmath52 is the relaxation time discussed in the next section . for the 1d fa model , @xmath53 , but the stretching exponent might be temperature dependent in more elaborated ( fragile ) models @xcite , as is indeed observed in experiments @xcite . we see from fig . [ persistence ] that for @xmath54 and @xmath55 @xmath50 is well approximated by eq . ( [ stretched ] ) on the whole time window . for higher temperatures , @xmath56 , the mixed character of the correlators is evident from the fact that eq . ( [ stretched ] ) only describes the long time behaviour of the correlator , as expected . in this temperature regime , short times are best described by a simple exponential . in the high temperature regime , @xmath57 , relaxation is just exponential for all times . we conclude that the appearance of isolated defects at @xmath5 is reflected in the long time behaviour of dynamical correlators . in the crossover region , @xmath58 , more and more of the decorrelation is due to isolated defects when @xmath8 decreases . below @xmath0 the entire decorrelation is due to these slow processes . a confirmation of the progressive domination of slow over fast processes described above can be found in the numerical results of ref . the similarity of fig . 4 of ref.@xcite and our fig . [ persistence ] is in fact quite striking . in particular , ref . @xcite calculated density - density correlations from both real configurations and their corresponding is in a binary lennard - jones mixture . the latter , where thermal energies were removed in the quenching procedure , are the ones which have to be compared with fig . [ persistence ] , since fast vibrations are also removed in our coarse - grained approach . an important conclusion is that the long - time decay of dynamical correlators , sometimes referred to as the alpha - relaxation , is due to the presence of isolated defects and therefore of heterogeneous dynamics , even in the @xmath59 regime . this means that activated dynamics , in the language of mct , or hopping events in topographic terms , are responsible for the alpha - relaxation , even at temperatures well above @xmath0 . this conclusion is unavoidable in view of the numerical data of refs . @xcite . the next natural quantity to consider , the relaxation time @xmath52 , is readily obtained from the dynamical correlators discussed in the previous subsection . from the discussion of section [ physics ] , we expect a crossover from high-@xmath8 to low-@xmath8 at the onset temperature @xmath5 , see eq . ( [ to ] ) . our results for the 1d fa model are presented in fig . [ talpha ] , where @xmath52 is defined as the time where the persistence function has decayed to the value @xmath60 . the simplest mean - field approximation to the dynamics of the fa model consists in a hartree like decoupling of spatial correlations , @xmath61 , in the dynamical equation for @xmath62 . this amounts to replacing the actual neighbourhood of site @xmath23 by an average neighbourhood , and spins are always facilitated with an average rate equal to @xmath27 . this gives a mean - field estimate of the relaxation time @xcite , @xmath63 fig . [ talpha ] shows that this simple approximation accounts for the dependence of the relaxation time at high temperatures , @xmath64 . the exact result for the relaxation time of the 1d fa model is obtained by realizing that isolated defects undergo diffusion with a temperature dependent diffusion constant , @xmath65 . the system relaxes when defects have diffused over a distance given by the mean separation between defects , @xmath66 , so that @xmath67 this mechanism relies on the notion of dynamic facilitation which implies that local fluctuations of the mobility determine the dynamics , and is essentially beyond reach of any mean - field type of approximation @xcite . we see from fig . [ talpha ] that eq . ( [ tex ] ) accounts for the behaviour at low temperatures , @xmath68 . figure [ talpha ] also presents a fit to the data with an mct power law form for the relaxation time @xcite , @xmath69 similar to the one obtained in ref . @xcite for the 2-spin facilitated , two dimensional version of the fa model . from fig . [ talpha ] , we draw the following conclusions . the behaviour of the relaxation time changes from the high-@xmath8 to low-@xmath8 behaviour close to the onset temperature @xmath5 . the combination of simple mean - field at high-@xmath8 with the exact form at low-@xmath8 allows to describe the temperature dependence of the relaxation time over the whole temperature range . however , given that @xmath52 smoothly interpolates between these two different functional forms , the mct power law form , eq . ( [ tmct ] ) , appears to works reasonably well in a time window of about three decades ( see inset in fig . [ talpha ] ) . this range of apparent power - law behaviour is in fact larger than the corresponding one in the canonical binary lennard - jones mixture of ref . @xcite , where extensive tests of mct have been performed . remarkably , we also find that the @xmath0 extracted from the power law fit to the relaxation time coincides well with the temperature where fast processes cease to contribute in a significant manner to the distribution of relaxation times , fig . [ distrib ] . this explains our choice of notation for the lower crossover temperature , @xmath0 . following the standard mct reading of the data @xcite , we would erroneously conclude that activated processes only appear close to @xmath70 , since these processes are often tautologically defined by the breakdown of the power law behaviour of the relaxation time @xcite . [ distrib ] , [ persistence ] , and [ talpha ] , prove instead that activated dynamics starts to be relevant at @xmath5 , much above @xmath0 , dominating the alpha - relaxation of the correlators , and hence the relaxation time of the system . the results of this section considerably weaken the possibility of the existence of a temperature regime in supercooled liquids where the relaxation time is correctly described by a power law behaviour . it follows that the standard determination of the location of the mct `` singularity '' @xmath0 in experiments and simulations is physically unjustified @xcite . in fact , the issue of the location of @xmath0 has been recently addressed in ref . @xcite , where it was found that for a variety of systems , the temperature @xmath0 obtained from the actual mct equations systematically coincides with the onset temperature @xmath5 discussed above . let us now consider the quantities shown in fig . [ crossoverlj ] as evidence in favour of the mct / landscape picture in lennard - jones mixtures , from the perspective of dynamically facilitated models . in fig . [ crossoverfa ] we show for the 1d fa model a plot analogous to fig . [ crossoverlj ] . the first quantity presented in fig . [ crossoverfa ] is the logarithm of the relaxation time , @xmath71 , as a function of temperature , which we discussed in detail in the previous subsection . the second quantity shown in fig . [ crossoverfa ] is the average energy of inherent structures , @xmath72 . for the 1d fa model it can be computed analytically by solving the zero - temperature dynamics of the model @xcite , @xmath73 where the concentration of defects , @xmath27 , is defined in eq . ( [ ct ] ) . at high temperature @xmath72 changes very slowly . when @xmath8 is reduced below @xmath5 the concentration of defects starts to decrease markedly , and @xmath72 follows the same trend , as can be seen in fig . [ crossoverfa ] . this change in behaviour at the onset temperature @xmath5 is due to the appearance of isolated defects , and therefore of a heterogeneous dynamics , and not to any special change of the potential energy surface . this is a very different interpretation of the physics from that of ref.@xcite . the third quantity shown in fig . [ crossoverfa ] is analogous to the distance between a configuration and its nearest is ( see fig.[crossoverlj ] ) . in the lattice models we are considering the natural quantity to compute is the concentration of sites which change during the descent towards the inherent structure , @xmath74 . since only excited sites can change during this procedure , we get , @xmath75 clearly , at low temperature @xmath76 . this behaviour is physically natural . contributions to @xmath74 come from clusters of defects , which are the only objects that can relax during the descent to an is . since the probability for a cluster of @xmath33 defects , @xmath77 , goes as @xmath78 , the main non trivial contribution to @xmath74 at low @xmath8 comes from the smallest clusters , @xmath79 . these relax only one defect in the descent , so that @xmath80 . moreover , our defect interpretation is consistent with real space observations in simulations of silica @xcite , where it was found that during the descent to the is the major contributions to the distance comes from annihilation of localized topological defects of the amorphous structure . the similarity of fig . [ crossoverfa ] to fig . [ crossoverlj ] is striking . the emerging physical picture is however completely different from the one of the mct / landscape scenario . for example , while it may appear from the behaviour of @xmath81 above @xmath0 that this quantity extrapolates to zero when @xmath82 ( see fig.[crossoverfa ] ) , the exact temperature dependence of @xmath74 is purely arrhenius . this means that @xmath0 has no particular importance [ @xmath0 would not look special in a plot of @xmath74 versus @xmath83 . even if one accepts @xmath0 as delimiting two regimes with high and low concentrations of clusters of excitations , this apparent crossover is completely irrelevant as far as the long - time dynamics is concerned . these observations suggest that the crossover at @xmath0 from `` landscape - influenced '' to `` landscape - dominated '' of ref . @xcite is not physically significant for the alpha - relaxation . in recent years , the potential energy landscape of supercooled liquids has become an object of study _ per se _ @xcite . in ref . @xcite , we have developed the idea that the main motivation behind these works was the observation of the separation between fast vibrations and slow hopping processes if sufficiently small systems are considered . this apparently harmless statement on the system size , we argued in ref . @xcite , results in fact from the central feature of the dynamics of supercooled liquids : `` sufficiently small '' really means `` if the system size is of the order of the dynamical correlation length @xmath29 '' @xcite . however , in a purely topographic description of the physics based on the statistical properties of minima , the relevance of the dynamical correlation length is not obvious @xcite . in that sense , a topographic description of the glass transition misses a central aspect of the physics . we shall show below that using the very simple spatial approach described in previous sections , we can trivially derive the statistical properties of the landscape reported in recent years . this successful confrontation to such an amount of apparently non - trivial and detailed numerical results is again a strong indication of the validity of our approach . reference @xcite described in detail the connection between the nanoscopic ordering in the trajectories of dynamically facilitated models and the dynamics between is , `` metabasins '' , or `` traps '' observed in numerical simulations or experiments of supercooled liquids . this same approach can be extended to account for the properties of `` saddles '' , i.e. , configurations related to transitions between is . figure [ saddle ] zooms on the lowest temperature trajectory of fig . the top panel of fig . [ saddle ] shows diffusion of isolated defects . it also shows coalescence and branching events , i.e. , closing and opening of bubbles . these two kinds of processes correspond to `` hopping '' events between dynamical `` traps '' @xcite . lets consider in detail one of these events , for example the coalescence process enlarged in the bottom left panel of fig . [ saddle ] . isolated defects diffuse by first facilitating one of their neighbours , for instance , @xmath84 for two defects to coalesce the minimum number of excitations that have to be present when they merge is three : @xmath85 in this sequence , the total number of defects is 2 at the beginning , 3 at the transition , and 1 at the end . this process is schematically described in the bottom right panel of fig . [ saddle ] . in topographic terms , an isolated defect corresponds to a local minimum of the energy , since such a configuration can only evolve by an energy increase , as in ( [ diff ] ) . on the other hand , a cluster of three excitations corresponds locally to a saddle point , since it is the transition configuration between two minima , as in ( [ trans ] ) and fig . [ saddle ] . larger clusters thus correspond to higher order saddles , since the larger the cluster the larger the number of possible moves into minima . the case @xmath79 is particular : it is not a minimum since it can relax one defect to decrease its energy , but it is not a saddle either since it does not correspond to a hopping event like that of fig . [ saddle ] . clusters with @xmath79 are just ordinary points ( i.e. , not stationary points ) of the landscape . the previous discussion generalizes in a natural way to the whole class of dynamic facilitated systems . the above identification between the relevant dynamical objects , isolated defects and clusters of defects , and `` landscape properties '' allows one to compute quantities such as the mean saddle index , @xmath9 , and the corresponding mean energy of stationary points , @xmath86 . the quantities @xmath87 and @xmath88 were estimated numerically in simulations of supercooled liquids @xcite . it was found that both functions decrease when @xmath8 decreases , and extrapolations were performed that indicated @xmath10 . also , plotting the dependence of @xmath87 on @xmath89 parameterized by the temperature , a simple linear relation was obtained , @xmath90 @xcite . in the case of the 1d fa model , it is very simple to devise a procedure to go from an equilibrium configuration to the `` nearest '' stationary point . isolated defects and clusters of defects with @xmath91 are locally such points , so we only have to deal with @xmath79 clusters . from these we can either reach a `` minimum '' @xmath92 or a saddle with @xmath93 . we respectively assign the probabilities @xmath94 and @xmath95 to these two possibilities . we then have , @xmath96 and @xmath97 where @xmath98 is the probability to have a cluster of size @xmath33 . from the discussion above we know that @xmath99 , @xmath100 , @xmath101 , @xmath102 , @xmath103 . putting all together we obtain , @xmath104 , \label{index}\ ] ] and @xmath105 at low @xmath8 both quantities scale as @xmath106 and @xmath107 . three important conclusions can be drawn : \(i ) it is obvious from eq . ( [ index ] ) that @xmath108 for @xmath109 . this means that there is no `` geometric transition '' to a regime with vanishing saddle index . \(ii ) the saddle index has a temperature dependence which follows closely that of the distance @xmath74 discussed in the previous subsection . this is expected because they both receive their principal contribution , at low @xmath8 , from clusters with @xmath79 . in other words , the main objects for the low temperature dynamics , the isolated defects , do not contribute to these quantities . therefore , as for the distance @xmath74 in fig . [ crossoverfa ] , the rapid decrease of @xmath9 when the temperature decreases can easily be confused with a vanishing of the saddle index close to @xmath0 . \(iii ) the linear relation between @xmath87 and @xmath89 becomes exact at low temperatures , see eqs.([eis],[index],[ens ] ) . again , the difference between @xmath88 and @xmath72 comes from the clusters of defects which are relaxed during the descent to the inherent structure . in fig.[geometric2 ] , we show the behaviour of @xmath87 versus @xmath89 for the entire range @xmath110 . note that in ref.@xcite a linear behaviour between @xmath87 and @xmath88 was also reported . this is true at relatively high temperature , given that above @xmath5 the energy of inherent structure is almost constant while @xmath87 and @xmath88 change with temperature in the same way . these results are valid beyond the fa model which we have used to illustrate them . the inexistence of the geometric transition where the saddle index vanishes follows from the observation that the low temperature behaviour of @xmath87 is given by the smallest cluster of defects necessary to make a transition , in the sense described in the previous section . since these objects are spatially localized , their energy cost is @xmath111 , and they exist with non zero probability at finite temperature , @xmath109 . this argument is close in spirit to stillinger s argument for the inexistence of an entropy crisis at the kauzmann temperature involving point defects @xcite . moreover , as discussed in the introduction , careful numerical simulations both confirm that @xmath112 @xcite , and report an arrhenius behaviour @xmath9 @xcite , in agreement with our results . the relation between saddle index and energy , @xmath113 which was first observed numerically @xcite , is also a general result for dynamically facilitated systems . this relation contains two different pieces of information . first , it shows that the intensive saddle index is a number of @xmath111 . this is again a trivial consequence of the existence of the dynamical correlation length @xmath29 , so that a large sample can in fact be thought of an assembly of independent subsystems of linear size @xmath29 @xcite . second , and more interesting , is a connection between energy and saddle index , presented as a `` general feature of the potential energy landscape of supercooled liquids '' @xcite , for which no theoretical explanation was however available . this feature is in fact almost a tautology in the context of facilitated models : the more defects are present , the more available directions to move , the higher the energy above that of the is . the fact that relation ( [ lin ] ) holds in different model liquids is another confirmation that dynamical facilitation is a key generic feature of the dynamics of supercooled liquids . another common procedure of the landscape approach is to decompose configurations into vibrational and configurational components . stillinger and weber @xcite suggested to perform this decomposition at the level of the partition function , @xmath114 where the sum is over energies of is , @xmath115 , their number is indicated by @xmath116 , and @xmath117 is the `` basin free energy '' which takes into account fluctuations within an is due to vibrations and possible `` anharmonicities '' ( i.e. , all the rest ) . it is instructive to consider the calculation of the partition function in the case of the 1d fa model using the stillinger and weber decomposition . the thermodynamics of the fa model is that of a non - interacting gas of binary excitations . this simple thermodynamics , however , can be obtained with any dynamics obeying detailed balance with respect to the hamiltonian ( [ ham ] ) , the actual fa dynamics defined by ( [ rates ] ) being just one possibility . in this sense , the stillinger and weber prescription for thermodynamics is an approximation for the way a thermodynamic quantity would be calculated using a particular choice of dynamics to sample the configuration space . inherent structures , basin free energies , etc . , have no thermodynamic meaning , they only have a dynamical meaning associated with a particular choice of dynamics . for the 1d fa model we can evaluate the stillinger and weber partition function exactly . we have : @xmath118 \right ) } . \label{zn}\end{aligned}\ ] ] the first factor counts the number of configurations of energy @xmath115 with only isolated defects in a system of @xmath22 sites . due to the coarse - grained nature of facilitated models the only contribution to the basin free energy comes from anharmonicities . performing the sum in eq . ( [ zn ] ) without the anharmonic contribution gives , @xmath119 , \ ] ] where @xmath120 is the @xmath22-th chebyshev polynomial of second kind . in the thermodynamic limit the above expression simplifies to give the free energy in the `` harmonic '' approximation , @xmath121 in fig . [ figsw ] , we compare the approximation ( [ fharm ] ) to the exact expression for the free energy , @xmath122 as observed numerically in supercooled liquids @xcite , both thermodynamic evaluations , eqs . ( [ fharm ] ) and ( [ fex ] ) , apparently coincide below @xmath0 when anharmonicities become negligible . from the previous sections we know that anharmonicities are just a consequence of the existence of clusters of defects . at low temperature the difference between ( [ fharm ] ) and ( [ fex ] ) is therefore proportional to @xmath123 , reflecting the fact that : ( i ) anharmonicities do not disappear below @xmath0 , which again is no particular temperature in this context ; ( ii ) anharmonicities are due to clusters of defects , @xmath79 being the leading term at low temperatures . in the particular case of the fa model we can formulate an exact expression for the anharmonic free energy @xmath124 . it is easy to check that the choice @xmath125 in eq . ( [ zn ] ) yields the exact expression for the free energy in the thermodynamic limit . this exact expression for the anharmonic part of the free energy is simple to understand . in an is all defects are isolated . the contribution of clusters of defects is obtained from the probability that an isolated defect in the is was a cluster in the original configuration , which is given by eq.([fanh ] ) . interestingly , a numerical procedure to evaluate the anharmonic contributions can be proposed . first , the harmonic expression ( [ fharm ] ) is evaluated . then , the approximate expression @xmath126 can be used as an educated guess for the anharmonic contributions . this gives in turn a first order free energy , @xmath127 . the improvement on the harmonic evaluation can be judged in fig . [ figsw ] , where we see that @xmath127 coincides with the exact free energy up to @xmath128 . this evaluation can then be improved iteratively using @xmath129 to get @xmath130 , and so on . it is easy to show that , in our particular case , @xmath131 . although these results could lead to an improvement on present evaluations of anharmonic contributions in studies of the thermodynamics of supercooled liquids , they also show that topographic concepts are very far from the physical objects they pretend to describe . we end this section with a remark on the adam - gibbs relation , which is an attempt to connect dynamical properties to thermodynamic ones . the adam - gibbs formula relates the relaxation time of a glass former , @xmath132 , to the configurational entropy , @xmath133 [ which would correspond to @xmath134 in the is formalism ] : @xmath135 $ ] . apparently , this relation has been seen to hold both in numerical simulations and in experiments of various systems @xcite . a careful look at the published data reveals however that the correlation between relaxation time and entropy does not quantitatively satisfy the adams - gibbs relation . this important observation is often not clearly stated @xcite . our analysis shows indeed that thermodynamic properties do not fully determine dynamical behaviours . clearly , almost by definition , @xmath132 increases and @xmath133 decreases as temperature is lowered , but that is where the connection ends . it is easy to check that the adam - gibbs formula fails completely when applied to dynamic facilitated systems . we find instead that time scales are broadly distributed , the distribution of times being the result of an integral over a distribution of length scales scale , @xmath136 , imposed by thermodynamic equilibrium , eq . ( [ ham ] ) . crucially however , dynamics also enters the integral in the form of the conditional probability of time and length , @xmath137 , which can be described , in a topographic language as containing informations on the relevant `` barriers '' , which have a priori no obvious link with the statistics of minima . as a consequence , thermodynamics alone can not be used to predict the dynamical behaviour . again , we find in the literature an excellent numerical confirmation of this statement . in refs @xcite , using a purely topographic description of a supercooled liquid , it was shown that the diffusion constant could be computed by a combination of thermodynamic _ and _ dynamical quantities , well in line with the above discussion . in this paper , we have developed a spatial description of the physics of the progressive slowing down of supercooled liquids . the only ingredients in our method have been the notions of localized mobility excitations and facilitated dynamics @xcite . our results were illustrated explicitly for the simplest case of the 1d fa model , but are generic for this theoretical approach , and are in very good agreement with experimental and numerical observations in supercooled liquids . at high temperatures , @xmath138 , the dynamics is fast and liquid - like , corresponding to the relaxation of large clusters of defects . dynamic facilitation plays no major role , and a simple mean - field hartree - like decoupling of the equations of motion yields predictions in good agreement with numerical results . when @xmath139 , the dynamics becomes heterogeneous , in the sense that local relaxation times are spatially correlated in a non trivial way . this can be seen in the trajectories of fig . [ traj ] as the appearance of slow bubbles @xcite . the long - time dynamics of the system results from the wide joint distribution of length scales and time scales , and the relaxation becomes stretched . this dynamic heterogeneity , which can be thought of as the activated dynamics invoked , but never described , by mct , determines the alpha - relaxation and its temperature dependence for @xmath140 . also , dynamic heterogeneity implies that decoupling of transport coefficients actually starts at @xmath5 , as confirmed by the simulations . from a theoretical point of view , local fluctuations of mobility crucially influence the dynamical behaviour . any mean - field like approach , no matter how involved , is most probably doomed to fail . at @xmath5 not all trace of high-@xmath8 physics ( clusters of defects ) disappears . the dynamics has a mixed character in the range @xmath141 , as seen for example in the behaviour of dynamical correlators like in fig . [ persistence ] . the temperature @xmath0 is just a crossover . it is the temperature below which isolated defects not only dominate the long time dynamics ( as for @xmath48 ) but are also the most numerous dynamical objects , see eq . ( [ tc ] ) . clusters of defects , whose dynamics is homogeneous and non - activated , are responsible for the temperature dependence of several quantities , such as distance between configurations and is , saddle index @xmath9 , and anharmonic contributions to the free energy . in numerical simulations , @xmath0 has been interpreted as a key temperature , in accordance with mct for which it represents a dynamical singularity . we have shown , however , that all of these quantities have a smooth temperature dependence , as has been recently observed numerically @xcite . this means that @xmath0 does not correspond to a transition or singular point , but is at most a crossover . crucially , the objects which display a crossover close to @xmath0 are also irrelevant for the long - time dynamics , so that the inexistence of the singularity @xmath0 is anyway not a physically important issue for the alpha - relaxation . below @xmath0 , isolated defects are the only remaining objects and the dynamics is dominated by the nanoscopic demixing of slow and fast regions so that trajectories look like a dense mixture of slow bubbles , which in turn gives a natural theoretical interpretation of the canonical features of glass transition phenomena @xcite . * the dynamics is heterogeneous and activated well above @xmath0 . * the dynamical slowing - down of supercooled liquids is due to the growth , below @xmath5 , of a dynamic correlation length @xmath29 , or more precisely , of a whole distribution of length and time scales . * the long - time dynamics , and therefore the relaxation time @xmath132 of the liquid is dominated by heterogeneous `` activated '' dynamics below @xmath5 . * the mct definition of activated processes as deviations from the ideal theory is incorrect . it is unlikely that the power law behaviour predicted by mct correctly describes the temperature dependence of @xmath132 . the practical definition of the temperature @xmath0 can not be used . * no topological change of the potential energy landscape takes place close to @xmath0 . quantities such as the saddle index and anharmonicities do not vanish close to @xmath0 and have a smooth temperature behaviour . at best , they undergo a crossover from large to small which remains to be quantified . * even if one accepts @xmath0 as a crossover temperature , as in eq . ( [ tc ] ) , quantities related to this crossover are unimportant for the long - time dynamics . * knowledge of thermodynamic properties is not enough to predict dynamical behaviour , which explains the quantitative failure of relations like the adams - gibbs formula . the approach we developed in this paper , which is an extension of previous efforts @xcite , is generic . it can be applied both to systems like lennard - jones liquids or to hard sphere systems . it gives a perspective on the physics of glass formers which is clearly distinct to , and in many respects much more natural than , that of mct or topographic approaches . there are many important and interesting open questions which need to be addressed from this new perspective . this include , among others , understanding properly the origin of mobility excitations , and the breakdown of stokes - einstein - debye relations and associated decouplings between transport coefficients . a general conclusion that can be drawn from this work and our previous ones is that , in many respects , glass transition phenomena is more standard than often assumed , in the sense that it is determined by the interplay between growing dynamic lengthscales and timescales . this is obviously reminiscent of critical phenomena @xcite , meaning that it should be possible to adapt renormalization group techniques to study the dynamics of the glass transition . we are grateful to j .- p . bouchaud , g. biroli , a. buhot , a. heuer , d.r . reichman , g. tarjus , and especially d. chandler for useful discussions . we acknowledge financial support from cnrs ( france ) , epsrc grant no . gr / r83712/01 , eu marie curie fellowship no . hpmf - ct-2002 - 01927 , the glasstone fund , and worcester college oxford . some of the numerical results were obtained on oswell at the oxford supercomputing center , oxford university . for extension of the landscape ideas in the glass phase see : c.a . angell , y. yue , l.m . wang , j.r.d . copley , s. borick , and s. mossa , j. phys . c condens . matter * 15 * , s1051 ( 2003 ) ; t. s. grigera , v. martin - mayor , g. parisi , p. verrocchio , nature * 422 * , 289 ( 2003 ) ; for alternative real space interpretations , see e.g. e. duval , l. saviot , l. david , s. etienne , j. f. jal , to appear in europhys . lett . , cond - mat/0306666 . p. shah and c. chakravarty , j. chem . phys . * 115 * , 8784 ( 2001 ) ; t.s . grigera , a. cavagna , i. giardina , and g. parisi , phys . * 88 * , 055502 ( 2002 ) ; l. angelani , r. di leonardo , g. ruocco , a. scala , and f. sciortino , j. chem . phys . * 116 * , 10297 ( 2002 ) ; l. angelani , g. ruocco , m. sampoli , and f. sciortino , cond - mat/0305280 . s. butler and p. harrowell , j. chem . phys . * 95 * , 4454 ( 1991 ) ; s. butler and p. harrowell , j. chem . phys . * 95 * , 4466 ( 1991 ) ; p. harrowell , phys . rev . e * 48 * , 4359 ( 1993 ) ; m. foley and p. harrowell , j. chem . phys . * 98 * , 5069 ( 1993 ) .

the lowest singlet and triplet states of a two electron system are split by the exchange energy . this is a direct consequence of the pauli exclusion principle and the coulomb interaction . as a result , a spin structure may appear even without explicit spin dependent interactions.@xcite in quantum dot spin qubits@xcite the exchange interaction implements a fundamental two qubit gate.@xcite compared to single qubit gates,@xcite the exchange - based gates are much faster@xcite and easier to control locally , motivating the solely exchange - based quantum computation.@xcite the control is based on the exponential sensitivity of the exchange energy on the inter - particle distance . manipulation then can proceed , for example , by shifting the single particle states electrically @xcite or compressing them magnetically . @xcite the practical manipulation schemes require quantitative knowledge of the exchange energy . the configuration interaction,@xcite a numerically exact treatment , serves as the benchmark for usually adopted approximations . the simplest one is the heitler - london ansatz , in which one particle in the orbital ground state per dot is considered . the exchange asymptotic in this model differs from the exact@xcite and the method fails completely in finite magnetic fields . extensions of the single particle basis include the hund - mullikan , @xcite molecular orbital,@xcite or variational method.@xcite other approaches , such as the hartree - fock,@xcite random phase approximation@xcite and ( spin-)density functional theory@xcite were also examined . none of them , however , is reliable in all important regimes,@xcite which include weak / strong interdot couplings , zero / finite magnetic field and symmetric / biased dot . the spin - orbit interaction , a non - magnetic spintronics workhorse,@xcite is a generic feature in semiconductor quantum dots.@xcite although it is usually weak , it may turn out of major importance as , for example , for the spin relaxation , @xcite , or , more positively , a handle for the electrical spin manipulation.@xcite it is natural to expect that the presence of the spin - orbit interaction will influence the exchange hamiltonian.@xcite the resulting corrections to the rotationally symmetric exchange hamiltonian are referred to as the anisotropic exchange ( we do not consider other sources than the spin - orbit interaction@xcite ) . stringent requirements of the quantum computation algorithms motivate studies of the consequences of the anisotropic exchange of a general form on quantum gates . @xcite usually , the anisotropic exchange is viewed as a nuisance to be minimized.@xcite on the other hand , it was considered as a possible way of implementing the quantum gates . @xcite in both views , it is of utter importance to know the strength and the form of the anisotropic exchange . since the spin - orbit interaction is weak , it is enough to answer the following question : what is the anisotropic exchange in the leading order ? surprisingly , arriving at the answer was not straightforward at all . the dzyaloshinskii - moriya@xcite interaction is of the first order in spin - orbit coupling . however , since it couples only states split by the isotropic exchange , it is necessary to consider also the second order anisotropic exchange terms to arrive at correct energies . @xcite ref . suggested such a hamiltonian , which was unitarily equivalent to the isotropic exchange hamiltonian , with the exchange energy renormalized in the second order . this was later revisited,@xcite with the following conclusion : in zero magnetic field , the two qubit hamiltonian is , up to the second order in the linear - in - momenta spin - orbit interaction , unitarily equivalent to the isotropic exchange hamiltonian in the weak coupling limit , with the unchanged exchange energy . further corrections appear in the third order . in the unitary operator providing the change of the basis , the spin - orbit interaction appears in the linear order . these results are a consequence of the special form of the spin - orbit interaction , which in the leading order leads to a spatially dependent spin rotation.@xcite in the short version of this article,@xcite we derived the leading order anisotropic exchange terms which appear in a finite magnetic field . we derived all anisotropic exchange parameters in a form valid for arbitrary interdot coupling . we also compared the results obtained using the first order versus the second order treatment of the spin - orbit interactions . the main goal of the present work is a detailed assessment of the quantitative reliability of the presented anisotropic exchange model comparing with exact numerical results . specifically , we examine the model in the strong and weak coupling regimes [ corresponding to single ( sec . [ singledot ] ) and double ( sec . [ doubledot ] ) dots , respectively ] and in zero and finite perpendicular magnetic field . we also study the role of the cubic dresselhaus term ( sec . [ cubicdresselhaus ] ) , whose action does not correspond to a spatial texture ( in the leading order ) and could potentially become dominant over the linear terms , changing the picture considerably . in addition to that , we supply the derivations , not presented in the short version ( sec . [ derivation ] ) and a detailed account of our numerical method ( app . [ appendix ] ) . the analytical pitfalls in evaluating the isotropic exchange are well known . @xcite on top of that , the anisotropic exchange is a ( very ) small correction to the exponentially sensitive isotropic exchange , and therefore it is involved to extract even numerically . our main conclusion here is that the presented analytical model is valid in all studied regimes . quantitatively , the effective parameters are usually within a factor of 2 from their counterparts derived from the numerically exact spectra . the main source of the discrepancy is the cubic dresselhaus term . surprisingly , in the most important regime for quantum dot spin qubits , namely the weak coupling , the heitler - london approximation works great for the anisotropic exchange , even though it fails badly for the isotropic one . this finding justifies using simple analytical formulas for the anisotropic exchange parameters . our system is a two - dimensional electron gas confined in a [ 001 ] plane of a zinc - blende semiconductor heterostructure . an additional lateral potential with parabolic shape defines the double quantum dot . we work in the single band effective mass approximation . the two - electron hamiltonian is a sum of the orbital part and the spin dependent part , @xmath0 where the subscript @xmath1 labels the two electrons . the orbital hamiltonian is @xmath2 here , @xmath3 is the kinetic energy with the effective mass @xmath4 and the kinetic momentum @xmath5 ; @xmath6 is the proton charge and @xmath7 is the vector potential of the magnetic field @xmath8 . the potential @xmath9 describes the quantum dot geometry @xmath10 here @xmath11 is the confinement length , @xmath12 measures the distance between the two potential minima , the vector @xmath13 defines the main dot axis with respect to the crystallographic axes and @xmath14 is the confinement energy . the coulomb interaction between the two electrons is @xmath15 where @xmath16 is the vacuum dielectric constant and @xmath17 is the dielectric constant of the material . the lack of the spatial inversion symmetry is accompanied by the spin - orbit interaction of a general form @xmath18 where the vector @xmath19 is kinetic momentum dependent . in the semiconductor heterostructure , there are two types of spin - orbit interactions . the dresselhaus spin - orbit interaction , due to the bulk inversion asymmetry of the zinc - blende structure , consists of two terms , one linear and one cubic in momentum@xcite @xmath20 here h.c . denotes the hermitian conjugate . the interaction strength @xmath21 is a material parameter , the angular brackets in @xmath22 denote the quantum averaging in the @xmath23 direction . since both electrons are in the ground state of the perpendicular confinement , we have @xmath24 , the value depending on the confinement details . a confinement asymmetry along the growth direction ( here @xmath23 ) gives rise to the bychkov - rashba term@xcite @xmath25 the coupling @xmath26 of the interaction is structure dependent and can be , to some extent , experimentally modulated by the top gates potential . equations ( [ dresselhaus]-[rashba ] ) are valid for a coordinate system where the x and y axes are chosen along [ 100 ] and [ 010 ] directions , respectively . below we use the effective spin - orbit lengths defined as @xmath27 and @xmath28 . the spin is coupled to the magnetic field through the zeeman interaction @xmath29 where @xmath30 is the effective gyromagnetic factor , @xmath31 is the bohr magneton ( alternatively , we use a renormalized magnetic moment @xmath32 ) and @xmath33 is the vector of the pauli matrices . in lateral quantum dots the coulomb energy @xmath34 is comparable to the confinement energy and the correlation between the electrons strongly influence the states.@xcite one can compare the energies considering @xmath35 where the coulomb length @xmath36 is a material parameter and @xmath37 is the mean distance between the electrons . in gaas @xmath38 nm , while a typical lateral dot has @xmath39 nm , corresponding to @xmath40 mev . the mean length @xmath37 is of the order of the confinement length , if the two electrons are on the same dot , and of the interdot distance , if the electrons are on different dots . in the first case , the coulomb energy is typically @xmath41 mev . in the second case ( one electron per dot ) the coulomb interaction is typically at least @xmath42 mev . the strength of the coulomb interaction precludes the use of perturbative methods . therefore , to diagonalize the two electron hamiltonian eq . ( [ ham_tot ] ) , we use the exact numerical treatment , the configuration interaction method . details are given in app . [ appendix ] . below we consistently use the notation of @xmath43 for spinor and @xmath44 for orbital wavefunctions . they fulfill the equations @xmath45 and @xmath46 , respectively . we use the gaas realistic parameters : @xmath47 ( @xmath48 is the free electron mass ) , @xmath49 , @xmath50 and @xmath51 . the coupling of the linear dresselhaus term is @xmath52 and of the bychkov - rashba term is @xmath53 , corresponding to the effective spin - orbit lengths @xmath54 m and @xmath55 m , according to the recent experiments.@xcite we use the confinement energy @xmath56 , which corresponds to the confining length @xmath57 , in line with an experiment.@xcite analytically , we will analyze the role of the spin - orbit interactions in the two - electron spectrum using the perturbation theory . this approach is appropriate since the spin - orbit energy corrections are small compared to the typical confinement energy . for a gaas quantum dot the ratio between the confinement length and the spin - orbit length @xmath58 . furthermore , for a magnetic field of @xmath42 tesla , the ratio between the zeeman energy and the confinement energy is @xmath59 . therefore the spin - orbit interactions are small perturbations , comparable in strength to the zeeman term at @xmath60 tesla . we consider the perturbative solution of the hamiltonian eq . ( [ ham_tot ] ) . we transform the hamiltonian to gauge out the linear spin - orbit terms , @xcite ( we neglect the cubic dresselhaus term in the analytical models ) @xmath61 using the operator @xmath62 where @xmath63 keeping only terms up to the second order in the spin - orbit and zeeman couplings , we get the following effective spin - orbit interactions @xmath64 , where @xmath65 here , @xmath66 , and @xmath67 equation ( [ eq : hzprime ] ) describes the mixing between the zeeman and spin - orbit interactions , which is linear in the spin - orbit couplings . it disappears in zero magnetic field , where only the terms in eq . ( [ eq : hsoprime ] ) survive a sum of an overall constant shift of @xmath68 and the spin - angular momentum operators . both of these are quadratic in the spin - orbit couplings . the point of the transformation , which changes the form of the spin - orbit interactions , is that the transformed interactions are much weaker ( being the second , instead of the first order in the spin - orbit / zeeman couplings ) . of course , both hamiltonians are equivalent , giving the same exact energies . however , a perturbative expansion of the transformed hamiltonian converges much faster . the symmetry of the two - electron wavefunctions @xmath44 has important consequences , for example , in the form of selection rules for the couplings between the states due to the spin - orbit interactions . the choice of the potential in eq . ( [ confinement ] ) is motivated by the fact that for small ( @xmath69 ) and large ( @xmath70 ) interdot distance the eigenstates of the single particle hamiltonian converge to the single dot solutions centered at @xmath71 and @xmath72 , respectively . for zero magnetic field , since the double dot potential does not have the rotational symmetry around the @xmath73 axis , the inversions of the coordinate along axes of the confinement potential ( @xmath74 and @xmath75 ) are the symmetries involved . indeed , the orbital hamiltonian eq . ( [ ham_orb ] ) commutes with the inversion operator @xmath76 and @xmath77 , @xmath78=0 $ ] . furthermore @xmath79=0 $ ] , where @xmath80 is the inversion of both axes simultaneously . all these operations belong to the @xmath81 group . accordingly , the wavefunctions transform as the functions 1 , x , xy , and y , which represent this group . if a perpendicular magnetic field is applied , only the total inversion operation , @xmath80 , commutes with the hamiltonian and the wavefunction is symmetric or antisymmetric with respect to the total inversion this is due to the lack of @xmath76 and @xmath77 symmetry of the kinetic energy operator . the slater determinants ( the two - electron basis that we use in the diagonalization procedure see app . [ appendix ] ) have also definite symmetries , if they are built from single particle states of definite symmetry ( see app . [ sdsymmetries ] ) . we define the functions @xmath82 to be the lowest eigenstates of the orbital part of the hamiltonian , @xmath83 with the following symmetry , @xmath84 where @xmath85 is the particle exchange operator . we observe that @xmath82 have , in addition to the particle exchange symmetry , also a definite spatial symmetry . in further we assume they fulfill @xmath86 we point out that while eq . ( [ eq : part_ex ] ) is a definition , eq . ( [ eq : symmetry1 ] ) is an assumption based on an observation . in zero magnetic field @xmath87 follows from the mattis - lieb theorem.@xcite for the validity of eq . ( [ eq : symmetry1 ] ) we resort to numerics we saw it to hold in all cases we studied . figure [ fig:2dot_spectrum ] shows the calculated double dot spectrum at zero magnetic field without the spin - orbit interactions . the two lowest states @xmath82 are split by the exchange energy @xmath88 . in the single dot case ( @xmath71 ) , the ground state is non degenerate , while the first exited state is doubly degenerate . increasing the interdot distance , this degeneracy is removed , as the two states have different spatial symmetry ( x and y ) . the energy of the states @xmath82 is separated from the higher states by an energy gap @xmath89 . this gap allows us to consider only the two lowest orbital states when studying the spin - orbit influence on the lowest part of the two - electron spectrum . indeed , in the double dot @xmath89 is of the order of @xmath42 mev , while the spin - orbit interactions are two orders of magnitude smaller . in the case of @xmath90 , the two orbital states approximation can be improved including more states ( although we show below this is not in fact necessary for a qubit pair in a circular dot ) . without the spin - orbit interactions , the eigenstates of the hamiltonian eq . ( [ ham_tot ] ) are separable in the spin and orbital degrees of freedom . we get the four lowest states by supplementing @xmath82 with spinors , forming the singlet and triplets : @xmath91 here @xmath92 is a singlet spinor built out of two spin-1/2 spinors , @xmath93 , @xmath94 , @xmath95 are the three possible triplets ; the quantization axis is chosen along the magnetic field . the symmetry leads to selection rules for the matrix elements between two electron states . in zero perpendicular magnetic field , because the @xmath96 operator transforms as xy , the singlet and triplets are not coupled , up to the second order in the spin - orbit interactions , @xmath97 . the only contribution is due to the constant @xmath98 . for non zero perpendicular magnetic field , the singlet and a triplet are coupled only if their orbital parts have the opposite spatial symmetry , due to the term in eq . ( [ eq : hzprime ] ) . the non - vanishing matrix elements are listed in table [ table 1 ] . . conditions on the orbital symmetries for the matrix elements @xmath99 to be non - zero . the orbital symmetries are defined by @xmath100 . [ cols="^,^,^",options="header " , ] here we derive effective four level hamiltonians , which provide understanding for the numerical results . we follow two different approaches : ( i ) restriction of the total hamiltonian , eq . , to the basis in eq . ( [ eq : basis4 ] ) ; ( ii ) including higher excited states through a sum rule using the schrieffer - wolff transformation with the unitary operator , eq . . then we compare the two models , including their simplifications using the heitler - london approximation , to demonstrate the quality of their description of the two - qubit subspace . we restrict the hilbert space of the double dot to the four lowest functions eq . ( [ eq : basis4 ] ) to describe the qubit pair . we start with the case of zero spin - orbit interactions . in the external magnetic field , the two triplets @xmath101 and @xmath102 are split by twice the zeeman energy @xmath103 . the restriction of the hamiltonian eq . ( [ ham_tot ] ) to the basis eq . ( [ eq : basis4 ] ) produces a diagonal matrix @xmath104 the standard notation is to refer only to the spinor part of the basis states . the matrix eq . ( [ eq:4x4 ] ) can be rewritten in a more compact way using the basis of the sixteen sigma matrices , @xmath105 ( index 0 denotes a unit matrix ; for explicit expressions see app . [ spinmatrices ] ) . the result is the so - called isotropic exchange hamiltonian ( where the constant @xmath106 was subtracted ) @xmath107 where the singlet and triplets are separated by the isotropic exchange energy @xmath108 , the only parameter of the model . the hamiltonian eq . ( [ eq : isotropic exchange ] ) describes the coupling of the spins in the heisenberg form . with this form , the swap gate can be performed as the time evolution of the system , assuming the exchange coupling @xmath88 is controllable . the isotropic exchange has already been studied analytically , in the heitler - london , hund - mulliken , hubbard , variational and other approximations , as well as numerically using the finite - difference method . usually analytical methods provide a result valid within certain regime of the external parameters only and a numerical calculation is needed to assess the quality of various analytical models . when the spin - orbit interactions are included , additional terms in the effective hamiltonian appear , as the matrix elements due to the spin - orbit interactions @xmath109 . selection rules in tab . [ table 1 ] restrict the non - zero matrix elements to those between a singlet and a triplet , @xmath110 here @xmath111 and @xmath112 where vector @xmath19 is defined by the spin - orbit interactions eq . ( [ spinorbit ] ) . using the sigma matrix notation , eq . ( [ 4x4 ] ) can be written as ( see app . [ spinmatrices ] ) @xmath113 where the @xmath114 and @xmath115 are the spin - orbit vectors defined as [ eq : spin orbit vectors ] @xmath116 the standard exchange hamiltonian follows as @xmath117 and we refer to it in further as the first order ( effective model ) to point the order in which the spin - orbit interactions appear in the matrix elements . note that we repeated the derivation of ref . additionally including the external magnetic field . as we will see below , comparison with numerics shows that treating the spin - orbit interactions to the linear order only is insufficient . to remedy , we generalize the procedure of ref . to finite magnetic fields . this amounts to repeating the derivation that lead to eq . [ eq:4x4 ] , this time starting with the unitarily transformed hamiltonian eq . ( [ eq : unit_transf ] ) . in this way , the linear spin - orbit terms are gauged out and the resulting effective hamiltonian treats the spin - orbit interactions in the second order in small quantities ( the spin - orbit and the zeeman couplings ) . the transformation asserts that the original schrdinger equation @xmath45 can be equivalently solved in terms of the transformed quantities @xmath118 , with the hamiltonian @xmath119 . the transformed hamiltonian @xmath120 is the same as the original , eq . , except for the linear spin - orbit interactions , appearing in an effective form @xmath121 . we again restrict the basis to the lowest four states and for the spin - orbit contributions we get @xmath122 using the selection rules and the algebra of the pauli matrices , we get the exchange hamiltonian ( for obvious reasons , we refer to it as the second order model ) @xmath123 compared to the first order model eq . , the functional form of the second order model hamiltonian is the same , except for the effective spin - orbit magnetic field @xmath124 which appears due to an inversion symmetric part of @xmath121 , eq . ( [ eq : hsoprime ] ) . the spin - orbit vectors , however , are qualitatively different [ eq : spin orbit vectors 2 ] @xmath125 we remind that the second order effective model hamiltonian eq . ( [ eq : final result ] ) refers to the four functions in eq . ( [ eq : basis4 ] ) unitarily transformed @xmath126 . the agreement between the second order effective model and the numerical data is very good , as we will see below . in this section we give @xmath127 explicitly for zero @xmath128 and diagonalize it . this is the only case for which is possible to give an analytical solution . for zero magnetic field , one can choose the functions @xmath82 to be real . then the matrix elements of the spin - orbit operator @xmath129 in eq . ( [ spinorbit ] ) are purely imaginary and @xmath130 . with the spin quantization axis chosen along the vector @xmath131 , the 4x4 matrix , eq . , takes the form of @xmath132 the upper left @xmath133 block of this matrix is a hamiltonian of a spin @xmath134 particle in a fictitious magnetic field @xmath135 . the eigenstates of this hamiltonian are spins oriented along the magnetic field @xmath136 . since the matrix in eq . ( [ kavokin ] ) is block diagonal , it is easy to see it can be diagonalized with the help of the following matrix @xmath137 the hamiltonian eq . can be diagonalized by @xmath138 , @xmath139 where @xmath140 . in the notation of the pauli matrices , ( see app . [ spinmatrices ] ) , @xmath141 where @xmath142 and @xmath143 . the unitary transformation @xmath144 in eq . performs the rotation of the two spins in the opposite sense . the hamiltonian can be interpreted as a rotation of the electron around a spin - orbit field when transferred from one dot to the other.@xcite the spectrum given by eq . ( [ diagonal ] ) qualitatively differs from the numerics , which shows there is no influence on the exchange in the second order of the spin - orbit couplings . we start with the single dot case , corresponding in our model to @xmath71 . the analytical solution of the single particle hamiltonian @xmath145 is known as the fock - darwin spectrum . the corresponding wave functions @xmath146 and the energies @xmath147 are @xmath148 where @xmath149 and @xmath150^{-1/4}$ ] is the magnetic length ; @xmath151 and @xmath152 are the radial and the angular quantum numbers , @xmath153 is the normalization constant and @xmath154 are the associated laguerre polynomials . let us consider now the orbital two electron states @xmath44 , eigenstates of @xmath155 , eq . ( [ ham_orb ] ) . the coulomb operator @xmath156 commutes with the rotation of both electrons around the @xmath73 axis , that is , the coulomb interaction couples only states with the same total angular momentum . this allows us to label the states with the quantum number @xmath157 , the total angular momentum . furthermore , the hamiltonian @xmath155 commutes with any spin rotation of any of the electrons , which expresses the fact that the coulomb interaction conserves spin . therefore we can consider the full two electron wavefunctions obtained by supplementing the orbital part @xmath44 with a spinor , respecting the overall wavefunction symmetry , similarly as in eq . ( [ eq : basis4 ] ) . the two - electron spectrum , without the zeeman and the spin - orbit interactions , is shown in fig . [ fig:2elec_spectrum ] . at zero magnetic field the ground state is a non - degenerate singlet state with total angular momentum zero @xmath158 . the next two degenerate states are triplets with @xmath159 and their degeneracy is split by the magnetic field . focusing on the two lowest states , most relevant for the qubit pair , they cross at @xmath160 t , so one can turn the ground state from the singlet to the triplet by applying an external magnetic field . in the presence of spin - orbit interactions , the crossing is turned into anticrossing , as described below . suppose some parameter , such as the magnetic field , is being changed . it may happen at some point that the states of the opposite spin become degenerate . such points are called spin hot spots . here , because of the degeneracy , weak spin - orbit interactions have strong effects . for the spin relaxation , spin hot spots play often a dominant role.@xcite we are interested in the changes to the spectrum due to the spin - orbit interactions . let us neglect the cubic dresselhaus in this section . to understand the spin - orbit influence , it is important to note the following commutation relations for the linear spin - orbit terms @xmath161 = 0 , \\ & [ { \bf w}_{d,1}\cdot \boldsymbol{\sigma}_{1}+{\bf w}_{d,2}\cdot \boldsymbol{\sigma}_{2},\hat{j}_- ] = 0 , \end{split}\ ] ] where @xmath162 . these commutation rules hold for any magnetic field @xmath128 . since the hamiltonian eq . ( [ ham_orb ] ) commutes with the operator @xmath163 , we can label the states using the quantum numbers @xmath164 and @xmath165 . the spin - orbit interactions couple only the states with the same quantum numbers @xmath166 and @xmath167 , for bychkov - rashba and dresselhaus term , respectively . let us focus on the part of the spectrum close to @xmath168 and on the states with @xmath169 , fig . [ fig:2elec_spectrum ] . the degeneracy of the states is removed by the spin - orbit interactions , as shown in fig . [ fig : closeb0 ] . let us now use the hamiltonian eq . ( [ eq : unit_transf ] ) , to understand the influence of the spin - orbit interactions . the degeneracy of the states with angular momenta @xmath169 makes the description with the lowest two orbital states questionable . therefore now we take @xmath41 orbital states and repeat the derivation of the second order effective hamiltonian , obtaining a @xmath170 matrix . the basis functions are @xmath171 where @xmath172 is the electron wavefunction with angular momentum @xmath158 , and @xmath173 and @xmath174 have angular momentum @xmath175 , and @xmath176 , respectively . since the magnetic field is negligible with respect to the spin - orbit couplings , the hamiltonian eq . ( [ eq : hzprime ] ) is negligible . because of the selection rules , tab . [ table 1 ] , the contributions from eq . ( [ eq : hsoprime ] ) in the basis eq . ( [ eq : basis7 ] ) , gives non zero matrix elements only for the following pairs , @xmath177 , and @xmath178 . for the gaas parameters , @xmath179 ev . in the region of small magnetic field , the states with @xmath180 are coupled by the spin - orbit interactions and the lifting is in the second order in the spin - orbit couplings . the other states are not coupled since they have different values of @xmath181 . therefore we conclude that the @xmath182-orbital state approximation can be used also for the single dot case ( or strongly coupled double dots ) , because the spin - orbit interactions do not mix the states @xmath44 and @xmath183 in the basis eq . ( [ eq : basis7 ] ) . note that as the coupling is forbidden by the inversion symmetry , the claim holds for an arbitrary oriented magnetic field . let us now discuss the second degeneracy region marked in fig . [ fig:2elec_spectrum ] , magnified in fig . [ fig : crossing ] . the spin - orbit interactions induce two anticrossings . the first is due to the bychkov - rashba term , since the crossing states have different @xmath167 , but the same @xmath184 and couples the singlet @xmath185 and triplet @xmath101 . the second is due to the dresselhaus term which couples states with @xmath186 , the singlet @xmath185 and the triplet @xmath102 . the central point is a crossing point , because the crossing state differ in both @xmath166 and @xmath167 . the splitting energy can be evaluated using the unitarily transformed hamiltonian eq . ( [ eq : unit_transf ] ) . using the degenerate perturbation theory , one can estimate analytically , using eq . ( [ eq : hsoprime ] ) and eq . ( [ eq : hzprime ] ) , the value of the two gaps to be @xmath187 and @xmath188 . these values are consistent with the numerical values . the double dot denotes the case when the interdot distance is of the order of the confinement length . in the next sections we discuss our effective models , eq . ( [ eq : final result ] ) and eq . ( [ eq : standard model ] ) in the double dot regime and compare them with numerics . the analytical solution for the two electron wavefunctions in a double dot potential is not known . we consider here the heitler - london ansatz since it is a good approximation at large interdot distances and we can work out the spin - orbit influence on the spectrum analytically . for this purpose , we compute the spin - orbit vectors , eq . ( [ eq : spin orbit vectors ] ) and eq . ( [ eq : spin orbit vectors 2 ] ) , for our models . in the heitler - london ansatz , the two electron eigenfunctions are given by @xmath189 where @xmath190 is a single electron fock - darwin state centered in the left ( right ) dot occupied by the @xmath1-th electron . below , in eqs . - , we skip the particle subscript @xmath1 , as the expressions contain only single particle matrix elements ( all @xmath146 , @xmath191 , @xmath151 , @xmath96 would have the same subscript , say @xmath192 ) . with this ansatz , the spin - orbit vectors , eq . , follow as [ spin orbit vectors lr ] @xmath193 similarly we get the spin - orbit vectors , eq . ( [ eq : spin orbit vectors 2 ] ) , as [ so vectors standard ] @xmath194 the explicit formulas for the vectors in eqs . ( [ spin orbit vectors lr])-([eq : aux ] ) are in app . [ heitlerlondon ] . differently from the spin - orbit vectors in eq . , the vectors in eq . reveal explicitly the anisotropy with respect to the magnetic field and dot orientation@xcite ( note that @xmath74 and @xmath75 in the definition of @xmath195 , eq . ( [ direction ] ) are the crystallographic coordinates ) . in the previous sections , we have derived two effective hamiltonians , @xmath127 , and @xmath196 , given by eqs . - and eqs.- , respectively . we now compare the energy spectrum given by these models with exact numerics . we present the spin - orbit induced energy shift , the difference between a state energy if the spin - orbit interactions are considered and artificially set to zero . for each model we examine also its heitler - london approximation , which yields analytical expressions for the spin - orbit vectors , as well as the isotropic exchange energy ( given in sec . iv.a and appendix c ) . thus , the effective models in the heitler - london approximation ( we denote them by superscript hl ) are fully analytic . the non - simplified models ( we refer to them as `` numerical '' ) require the two lowest exact double dot two - electron wavefunctions , which we take as numerical eigenstates of @xmath155 . apart from the energies , we compare also the spin - orbit vectors . since they are defined up to the relative phase of states @xmath197 and @xmath198 , the observable quantity is @xmath199 and analogously for @xmath200 . we refer to these quantities as the anisotropic part of the exchange coupling . figure [ fig : socont_b0 ] shows the spin - orbit induced energy shift as a function of the interdot distance for each of the four states . the exact numerics gives a constant and equal shift for all @xmath201 spin states , with value @xmath202ev . let us consider the second order model , eq . ( [ eq : final result ] ) . for zero magnetic field , all spin - orbit vectors are zero , as is the effective magnetic field . the only contribution comes from the constant term @xmath203ev that is the same for all states . our derived spin - model , eq . ( [ eq : final result ] ) , accurately predicts the spin - orbit contributions to the energy . on the other hand , the first order models @xmath204 and @xmath127 are completely off on the scale of the spin - orbit contributions . the exchange hamiltonian @xmath127 does not predict the realistic spin - orbit influence on the spectrum , even in the simple case when the magnetic field is zero . figure [ fig : param_b0 ] shows the non zero parameters for all four models . the exact isotropic exchange @xmath88 decays exponentially with the interdot distance . the same behavior is predicted in the heitler - london approximation . it decays exponentially , but deviates from the numerical results . as for the anisotropic exchange , the first order model @xmath127 gives an exponentially falling spin - orbit parameter @xmath205 , an order of magnitude smaller than @xmath88 . in contrast , the second order model @xmath196 predicts zero spin - orbit anisotropic exchange . first main result , proved numerically and justified analytically by the hamiltonian @xmath196 , is that _ at zero magnetic field the spin - orbit vectors vanish _ , up to the second order in spin - orbit couplings at any interdot distance . in the transformed basis , there is no anisotropic exchange at the zero magnetic field due the spin - orbit interactions , an important result for the quantum computation . indeed , since the exchange energy can be used to perform a swap operation , this means that the spin - orbit interactions do not induce any significant errors on the gate operation . the only difference is the computational basis , which is unitarily transformed with respect to the usual singlet - triplet basis . in the presence of a perpendicular magnetic field the structure of the spin - orbit contributions are quite different with respect to the zero field case . first of all , anticrossing points appear , where the energy shift is enhanced . figure [ fig : socont_b1 ] shows the spin - orbit contributions in a finite magnetic field . we plot only the anticrossing states , the singlet @xmath185 and the triplet @xmath206 . the prediction of the first order model is shown in the left panels of fig . [ fig : socont_b1 ] . as in the case of zero magnetic field , this model is off from the numerical results . in particular , it still predicts a zero contribution , except close to the anticrossing point . we note that the discrepancy is not connected to ( a failure of ) the heitler - london approximation , as using the exact numerical two electron wavefunctions does not improve the model predictions . in the right panels of fig . [ fig : socont_b1 ] , the comparison between the second order model and the numerics is provided . we observe that the model is very close to the numerics , even though the heitler - london approximation predicts the crossing point in a different position . the predictions of the numerical second order model @xmath196 is consistent with the exact numerics . the only discrepancy is due to the influence of the cubic dresselhaus term , as we will see in the next section . to get more insight , in fig . [ fig : param_b1 ] we have plotted the parameters of the models . fig . [ fig : param_b1]a shows the anisotropic exchange strengths in the two models . the first order model @xmath127 predicts the anisotropic exchange decreasing with the interdot distance , similar to the isotropic exchange energy . for large interdot distance the anisotropic exchange @xmath205 disappears . this means there is no influence on the energy due to the spin - orbit interactions . on the other hand , for the second order model @xmath196 the conclusion is different . for large interdot distances @xmath207 and @xmath208 are linear in @xmath209 . furthermore , the anisotropic exchange computed in the heitler - london ansatz is very close to the numerical one . we make a very important observation here : surprisingly , concerning the anisotropic exchange the heitler - london is quite a good approximation for all interdot distances even in a finite magnetic field . therefore , despite its known deficiencies to evaluate the isotropic exchange @xmath88 , it grasps the anisotropic exchange even quantitatively , rendering the spin - orbit part of the second order effective hamiltonian @xmath196 fully analytically . one can understand this looking at eqs . ( [ eq : spin orbit vectors 2 ] ) . the anisotropic exchange vectors are given by the dipole moment of the matrix element between the left and right localized state ( see app . c for explicit formula ) . this dipole moment is predominantly given by the two local maxima of the charge distribution ( the two dots ) and is not sensitive to the interdot barrier details , nor on the approximation used to estimate the lowest two orbital two - electron states . this is in strong contrast to the isotropic exchange , which , due to its exponential character , depends crucially on the interdot barrier and the used approximation . figure [ fig : param_b1]b shows the isotropic exchange @xmath88 , and the effective magnetic field induced by the spin - orbit interactions @xmath210 compared to the zeeman energy @xmath211 . we see the failure of the heitler - london approximation for @xmath88 . although the numerical calculation and the analytical prediction have the same sign ( this means that the ground state is the triplet ) they differ by an order of magnitude . the zeeman energy is constant and always much larger than the effective spin - orbit induced magnetic field @xmath210 . consequently , the effective field can be always neglected . the point where the zeeman energy equals to the isotropic exchange ( close to @xmath212 nm ) is the anticrossing point , where the spin - orbit contributions are strongly enhanced , as one can see in fig . [ fig : socont_b1 ] . let us consider a double dot system at fixed interdot distance of @xmath213 nm , corresponding to zero field isotropic exchange of @xmath214ev . in fig . [ fig : socont_d55 nm ] the spin - orbit contributions versus the magnetic field are plotted for the second order model @xmath196 and the exact numerics . we can conclude that to describe the spin - orbit influence on the states in a double - dot system it is important to use the second order hamiltonian @xmath196 . in fig . [ fig : param_d55 nm ] the spin - orbit parameters versus the magnetic field are plotted . the main influence on the spin is due to the zeeman interaction in the whole range of @xmath128 , since @xmath210 is several orders of magnitude smaller than the zeeman energy . at the ground state anticrossing point , the isotropic exchange crosses zero , while the anisotropic parameter @xmath208 is finite , leading to spin hot spots . apart from these , since the anisotropic exchange is two orders of magnitude smaller than the zeeman energy , the spin - orbit induced energy shifts are minute . finally we consider the role of the cubic dresselhaus term . the schrieffer - wolff transformation does not remove it in the linear order . figure [ fig : cubic ] shows the energy shifts induced by the spin - orbit interactions also in the case where we do not take into account the cubic dresselhaus term . one can see a very good agreement between the second order model @xmath196 and the exact numerics where the cubic dresselhaus term was omitted . therefore we can conclude that the main part of the discrepancy we see in the spin - orbit induced energy shifts are due to the cubic dresselhaus term . we analyzed the spin - orbit influence on two electrons confined in a lateral double quantum dot . we focused on the lowest part of the hilbert space , which corresponds to a qubit pair . in ref . a hamiltonian for such pair was proposed , with the spin - orbit interactions giving rise to an anisotropic exchange interaction . within a unitarily transformed basis , this interaction is encoded into two real three dimensional spin - orbit vectors . these , together with the isotropic exchange energy and the magnetic field vector , completely parametrize an effective two qubit hamiltonian . in this work , we examined the quantitative validity of this effective hamiltonian . in addition to a numerical study , we also provided the details of the effective hamiltonian derivation , which were skipped in ref . . we noted that it can be diagonalized analytically if the effective spin - orbit vectors are all aligned with the external magnetic field , the only exactly solvable case ( apart from the trivial case of no spin - orbit interactions present ) . we also evaluated the spin - orbit vectors in the heitler - london approximation and compared the analytical results with their exact numerical counterparts . there are three possible sources for a discrepancy between the model and the exact data : the higher excited orbital states of the quantum dot , the higher orders of the effective ( unitary transformed ) spin - orbit interactions and the cubic dresselhaus term . elucidation of their importance is one of the main results of this work . i ) we find the cubic dresselhaus term is the main source of the discrepancy . in a typical double dot regime and a moderate field of 1 tesla , it brings an error of @xmath215ev for the energies , while the two other mentioned corrections have an order of magnitude smaller influence . ii ) we find the effective hamiltonian describes both the weak and the strong coupling regimes ( the single dot represents the strongest possible coupling ) . iii ) surprisingly , the spin - orbit vectors obtained within the heitler - london approximation are faithful even at a finite magnetic field . overall , we find the anisotropic exchange hamiltonian to be generally reliable , providing a realistic and yet simple description for an interacting pair of spin qubits realized by two coupled quantum dots . we would like to thank guido burkard for useful discussion , martin gmitra and andrea nobile for numerical advice . this work was supported by dfg grk 638 , spp 1285 , sfb 689 , nsf grant dmr-0706319 , rpeu-0014 - 06 , erdf op r&d project `` qute '' , ce sas qute and daad . here we describe the numerical method we use to diagonalize the two electron hamiltonian eq . ( [ ham_tot ] ) . we proceed in three steps.@xcite we first diagonalize the single electron hamiltonian @xmath216 , using the numerical finite differences method with the dirichlet boundary condition ( vanishing of the wave function at boundaries ) . we do not consider the spin dependent part ( spin - orbit , zeeman ) at this step . this allows us to exploit the symmetries of the confinement potential . the single electron hamiltonian is diagonalized by the lanczos method@xcite . the typical number of points in the grid we use is @xmath217 , giving relative precision of the energy of order @xmath218 . in the second step , using the obtained single electron eigenstates @xmath219 , we construct the two electron states . we use them as a basis in which the two electron orbital hamiltonian eq . ( [ ham_orb ] ) is diagonalized . the two - electron states are constructed as symmetric @xmath220 with respect to the particle exchange . we choose @xmath221 single electron orbitals , typically @xmath222 . the total number of the two particle states is then @xmath223 . the spatial symmetry allows us to reduce the dimension of the two electron hamiltonian matrix to diagonalize . namely , the matrix is block diagonal , with the basis functions grouped according to the spatial symmetry ( @xmath42 , @xmath74,@xmath75 , @xmath224 ) and particle exchange symmetry ( @xmath225 ) . this results in @xmath226 blocks and holds for zero perpendicular magnetic field . in a finite field , we get @xmath201 blocks , as there are only two spatial symmetries possible ( @xmath42 , and @xmath74 ) . each block is diagonalized separately . the matrix element of the two - electron hamiltonian , eq . ( [ ham_orb ] ) , in our basis is @xmath227 the last term in eq . ( [ coulomb_int ] ) is due to the coulomb interaction and it leads to off diagonal terms in the hamiltonian . we diagonalize the matrix defined in eq . ( [ coulomb_int ] ) to get the eigenspectrum @xmath228 . in the third step , we add the spin dependent parts to the hamiltonian . we construct a new basis by expanding the wavefunctions obtained in the previous step by the spin . the orbital wavefunction @xmath229 gets the spinor according to its particle exchange symmetry . the symmetric function gets the singlet @xmath185 while the antisymmetric appears in three copies , each with one of the three triplets @xmath230 , @xmath206 and @xmath102 . we denote the new states by @xmath231 corresponds to one of the 4 spin states . the matrix elements of the total hamiltonian eq . ( [ ham_tot ] ) are @xmath232 where the last term is the matrix element of the spin - orbit interactions . the resulting matrix is diagonalized to get the final eigenstates . we choose a certain number @xmath233 of lowest @xmath234 states , depending on the required precision . in our simulations @xmath235 , resulting to the accuracy of the order of @xmath236 mev for the energy . computationally most demanding are the coulomb integrals . indeed , the typical size of the hamiltonian matrix , in the second step , is @xmath237 , requiring at least @xmath238 coulomb integrals . writing functions involved in the eq . ( [ coulomb_int ] ) as slater determinants , we can express the integral as a sum of terms such as the following @xmath239 where @xmath240 . the symmetry of the coulomb integral @xmath241 reduces the number of needed matrix elements to a half . for the single dot , @xmath242 are the fock - darwin functions and it is possible to derive an analytical formula for @xmath243 . in our case , since the single particle functions are given numerically , we have performed a numerical integration . using the fourier transform , we can reduce the @xmath201-dimensional integration to two dimensional @xmath244 where @xmath245 for the evaluation of the fourier transforms , we use the discrete fourier transform algorithm with the attenuation factors , as described in ref .. we compute ( [ fourier_integral ] ) according to the perturbative formula @xmath246 where @xmath247 is calculated in the point @xmath248 , @xmath249 is the perturbative order ( the order of the taylor expansion ) , @xmath250 and @xmath251 are the number of grid points in the @xmath74 and in the @xmath75 direction , respectively . the coefficients @xmath252 depend only on the geometry of the grid and are defined as @xmath253 here @xmath254 is the integration region and @xmath255 is the grid spacing along @xmath74 . similarly for the @xmath75 direction . in our simulations we use the previous formula up to the 4-nd order in the taylor expansion . the achieved relative precision is @xmath236 , with the computational time for one coulomb element @xmath256 ms . suppose the single particle hamiltonian commutes with certain set of operators @xmath257 , and therefore the single particle states @xmath242 can be chosen such that they have definite symmetries forming a representation of the group @xmath258 of the symmetry operators @xmath259 for example , since the double dot potential has inversion symmetry along @xmath74 axis , @xmath76 is in the group @xmath258 , while @xmath260 the states are symmetric or antisymmetric with respect to @xmath74 inversion . now consider the two electron states @xmath261 , eq . ( [ symm_fun]-[antisymm_fun ] ) . these states also have definite symmetry if a certain operator from @xmath258 acts simultaneously on both particles @xmath262 for our case of the symmetry group @xmath81 , since @xmath263 , the set of all possible products of two characters is the same as the set of characters for a single particle , @xmath264 . this means the two particle states will form the same symmetry classes as single particle states with the same characters . in the heitler - london approximation , the exchange energy is calculated as @xmath265 with the functions @xmath82 given in eq . ( [ states pm ] ) . the single particle ground state wavefunction of the fock - darwin spectrum is @xmath266,\ ] ] where @xmath267 is the effective confinement length defined by @xmath268 . the wavefunctions @xmath269 are obtained shifting the fock - darwin ground state to @xmath270 . in the presence of the magnetic field we have to add a phase factor because of the gauge transformation @xmath271 ; we have @xmath272}\psi_{00}(x\pm l_0d , y),\\ \zeta=\left(\frac{l_0}{l_b}\right)^2,\;\;\ ; \vartheta=\frac{bel_b^2}{2\hbar } , \;\;\;l_b = l_0(1-\vartheta^2)^{1/4}. \end{split}\ ] ] the overlap between the left and right functions is @xmath273},\ ] ] and the exchange energy is @xmath274 } \big\{c_s\sqrt{\zeta}\big(\exp{[-\zeta d^2]}i_0(\zeta d^2)+\\ & -\exp{[\zeta d^2\vartheta^2 ] } i_0(\zeta d^2\vartheta^2)\big)+ \frac{2d}{\sqrt{\pi \zeta}}\left ( 1-\exp{[-\zeta d^2]}\right)+ \\ & + 2d^2 \left ( 1-\mbox{erf}(d\sqrt{\zeta } ) \right ) \big\ } , \end{split}\ ] ] where @xmath275 is the zeroth - order modified bessel function of the first kind . the factor @xmath276 is the ratio between the coulomb strength and the confinement energy , @xmath277 . similar formula can be found in ref . for a quartic confinement potential . the formula ( [ eq : iso - exchange ] ) has been derived in ref . ( in the original paper there is a trivial typo that we correct ) . the two electron energies for the states @xmath198 and @xmath197 are @xmath278 where @xmath279}i_0(\zeta d^2),\\ e_{w_{ri } } & = & \hbar\omega_0 \left[2d^2(1-\mbox{erf}(d\sqrt{\zeta } ) ) -\frac{2d}{\sqrt{\zeta \pi}}\exp{[-\zeta d^2]}\right],\\ e_{ce } & = & \hbar\omega_0 c_s \sqrt{\zeta } \exp{\left[-\zeta d^2(2+\vartheta^2)\right]}i_0(\zeta d^2\vartheta^2),\\ e_{w_{ce } } & = & -\hbar\omega_0\frac{2d}{\sqrt{\pi \zeta } } \exp{[-2\zeta d^2(1+\vartheta^2)]}.\end{aligned}\ ] ] the components of the vectors @xmath280 and @xmath281 are @xmath282 where @xmath283 and @xmath284 are the spin - orbit lengths for the rashba and dresselhaus respectively . the matrix elements of the vector @xmath195 are @xmath285 and @xmath286.\ ] ] in the singlet and triplet basis , one can evaluate the sixteen matrices which can be formed as the direct product of two pauli matrices and the identity . here we list only the matrices needed for our purposes , and we regroup them to combinations in which they appear in the text .

the theory of finite type invariants allows us to interpret many important knot and link invariants in purely combinatorial terms , as functionals on spaces of chord diagrams . for knots , there is an obvious intersection graph associated with these diagrams , first studied ( in this context ) by chmutov , duzhin and lando @xcite . in many cases , these graphs contain all the relevant information for the functionals coming from knot invariants @xcite . some of the most important knot invariants , such as the conway , jones , homfly and kauffman polynomials , can be interpreted in terms of these intersection graphs @xcite . however , it is not obvious how to extend the idea of the intersection graph to finite type invariants of other objects , such as braids , links and string links . the goal of this paper is to introduce a reasonable definition of the intersection graph for chord diagrams associated with string links . as evidence that this is the `` right '' definition , we use these intersection graphs to construct weight systems for string links , and show that these weight systems are related to ( though weaker than ) the weight systems arising from the conway and homfly polynomials . in future work we will show that in some cases the intersection graph contains all the relevant information in the string link chord diagram @xcite , and that it can be used to give a new interpretation of milnor s homotopy invariants @xcite . in section [ s : prelim ] we will review the necessary background , including finite type invariants , chord diagrams , intersection graphs and lando s graph bialgebra @xcite . in section [ s : igstring ] we will define intersection graphs for string links , and provide the space of graphs with a bialgebra structure similar to lando s . in section [ s:2term ] we look at chord diagrams modulo the 2-term relations introduced by bar - natan and garoufalides @xcite , and use the adjacency matrices of their intersection graphs to construct weight systems related to those arising from the conway and homfly polynomials . finally , in section [ s : questions ] we pose some questions for further research . + + _ acknowledgement : _ the author thanks loyola marymount university for supporting this work via a summer research grant in 2003 . an _ ( oriented ) knot _ is an embedding of the ( oriented ) circle @xmath0 into the 3-sphere @xmath1 . knot invariant _ is a map from these embeddings to some set which is invariant under isotopy of the embedding . we will also consider invariants of _ regular _ isotopy , where the isotopy preserves the _ framing _ of the knot ( i.e. a chosen section of the normal bundle of the knot in @xmath1 ) . a _ link _ with @xmath2 components is simply an embedding of the disjoint union of @xmath2 copies of @xmath0 into @xmath1 ; each component is a knot . a _ string link _ can be defined as follows : ( habegger and lin @xcite ) let d be the unit disk in the plane and let i = [ 0,1 ] be the unit interval . choose k points @xmath3 in the interior of d , aligned in order along the the x - axis . a * string link * @xmath4 of k components is a smooth proper imbedding of k disjoint copies of i into @xmath5 : @xmath6 such that @xmath7 and @xmath8 . the image of @xmath9 is called the ith string of the string link @xmath4 . link and string link invariants are defined in the same way as for knots . note that any string link can be closed up to a link in a unique way by joining the top and bottom of each component by an arc which lies outside of @xmath5 ( and which is unlinked with the other such arcs ) . so any link invariant gives rise to a string link invariant by evaluating it on the closure of the string link . our treatment of finite type invariants will follow the combinatorial approach of birman and lin @xcite . we will give a brief overview of this combinatorial theory , and its natural extensions to links and string links . for more details , see bar - natan @xcite . we first note that we can extend any link invariant to an invariant of _ singular _ links , where a singular link is an immersion of a disjoint union of copies of @xmath0 in 3-space which is an embedding except for a finite number of isolated double points . given a link invariant @xmath10 , we extend it via the relation : @xmath11 an invariant @xmath10 of singular links is then said to be of _ finite type _ , specifically of _ type n _ , if @xmath10 is zero on any link with more than @xmath12 double points ( where @xmath12 is a finite nonnegative integer ) . we denote by @xmath13 the vector space over @xmath14 generated by ( framing - independent ) finite type invariants of type @xmath12 . we can completely understand the space of finite type invariants by understanding all of the vector spaces @xmath15 . an element of this vector space is completely determined by its behavior on links with exactly @xmath12 singular points . in addition , since such an element is zero on links with more than @xmath12 singular points , any other ( non - singular ) crossing of the knot can be changed without affecting the value of the invariant . this means that elements of @xmath15 can be viewed as functionals on the space of _ chord diagrams _ : a * chord diagram of degree n with k components * is a disjoint union of k oriented circles , together with @xmath12 chords ( line segments with endpoints on the circles ) , such that all of the @xmath16 endpoints of the chords are distinct . the circles represent a link of k components and the endpoints of a chord represent 2 points identified by the immersion of this link into 3-space . the diagram is determined by the orders of the endpoints on each component . for string links , chord diagrams involve oriented line segments rather than circles . functionals on the space of chord diagrams which are derived from finite type link invariants will satisfy certain relations : a * weight system of degree n * is a linear functional @xmath17 on the space of chord diagrams of degree @xmath12 ( with values in an associative commutative ring @xmath18 with unity ) which satisfies the 1-term and 4-term relations , shown in figure [ f:4-term ] . for knots , the three arcs all belong to the same circle ; for links ( resp . string links ) , they may belong to the same or different circles ( resp . line segments ) . ( 1-term relation ) @xmath19 ( 4-term relation ) @xmath20 the natural map from elements of @xmath15 to functionals on chord diagrams is a homomorphism into the space of weight systems @xcite . kontsevich proved the much more difficult fact that these spaces are isomorphic @xcite ( the inverse map is the famous _ kontsevich integral _ ) . for convenience , we take the dual approach , and simply study the space of chord diagrams of degree @xmath12 modulo the 1-term and 4-term relations . the 1-term relation is occasionally referred to as the `` framing - independence '' relation , because it arises from the framing - independence of the invariants in @xmath13 ( essentially , from the first reidemeister move ) . since most of the interesting structure of the vector spaces arises from the 4-term relation , it is common to look at the more general setting of invariants of regular isotopy , and consider the vector space @xmath21 of chord diagrams of degree @xmath12 on links with @xmath2 components modulo the 4-term relation alone . linear functionals on @xmath21 are called _ regular weight systems _ of degree @xmath12 . similarly , we define the vector space @xmath22 for chord diagrams of degree @xmath12 on _ string links _ of @xmath2 components . it is useful to combine all of these spaces into graded modules @xmath23 and @xmath24 via direct sum . for @xmath25 , we can define a product on @xmath26 via connect sum @xcite ; however , this does not extend to @xmath27 . but we _ can _ give the module @xmath28 a bialgebra ( or hopf algebra ) structure for any @xmath2 by defining an appropriate product and co - product : * we define the ( noncommutative ) product @xmath29 of two chord diagrams @xmath30 and @xmath31 as the result of placing @xmath31 on top of @xmath30 ( joining the components so the orientations agree ) , as shown below : @xmath32 * we define the co - product @xmath33 of a chord diagram @xmath34 as follows : @xmath35 where @xmath36 is a subset of the set of chords of @xmath34 , @xmath37 is @xmath34 with all the chords in @xmath36 removed , and @xmath38 is @xmath34 with all the chords _ not _ in @xmath36 removed . @xmath39 it is easy to check the compatibility condition @xmath40 . for knots , there is a very natural notion of intersection among the chords of a chord diagram - two chords intersect if their endpoints alternate around the bounding circle . the intersection graph then has a vertex for each chord , and connects two vertices by an edge if the corresponding chords intersect . the difficulty with extending this idea to links and string links is how to deal with chords which have their endpoints on different components of the diagram - specifically , if they each have one endpoint on a given component , and their other endpoints are on different components . in what sense , if any , do these chords intersect ? the essential value of the intersection graph for knots is that it can detect when the order of two endpoints for different chords along the bounding circle is switched ( as happens in the 4-term relation ) , since this changes the pair of chords from intersecting to non - intersecting or vice - versa . to usefully extend the notion of intersection graphs to links or string links , we need to retain this ability . for string links , the existence of a `` bottom '' and `` top '' for each component allows us to give a linear ( rather than cyclic ) ordering to the endpoints of the chords on each component , and so the notion of one endpoint being `` below '' another is well - defined . let @xmath34 be a string link chord diagram with @xmath2 components ( oriented line segments , colored from 1 to @xmath2 ) and @xmath12 chords . the _ intersection graph _ @xmath41 is the labeled , directed multigraph such that : * @xmath41 has a vertex for each chord of @xmath34 . each vertex is labeled by an unordered pair @xmath42 , where @xmath43 and @xmath44 are the labels of the components on which the endpoints of the chord lie . * there is a directed edge from a vertex @xmath45 to a vertex @xmath46 for each pair @xmath47 where @xmath48 is an endpoint of the chord associated to @xmath45 , @xmath49 is an endpoint of the chord associated to @xmath46 , @xmath48 and @xmath49 lie on the same component of @xmath34 , and the orientation of the component runs from @xmath48 to @xmath49 ( so if the components are all oriented upwards , @xmath48 is below @xmath49 ) . we count these edges `` mod 2 '' , meaning that if two vertices are connected by two directed edges with the same direction , they cancel each other . if two vertices are connected by a directed edge in each direction , we will simply connect them by an undirected edge . examples of chord diagrams and their associated intersection graphs are given in figure [ f : igstring ] . @xmath50 note that when the two chords have both endpoints on the same component @xmath43 , our definition of intersection graph corresponds to the usual intersection graph for knots . our definition also matches our intuition in the case of chord diagrams of two components , as shown in figure [ f : igstring ] . note also that the total number of directed edges between a vertex @xmath10 labeled @xmath42 and a vertex @xmath51 labeled @xmath52 is given by the sum of the number of occurrences of @xmath43 in @xmath52 and the number of occurrences of @xmath44 in @xmath52 . in particular , if a vertex @xmath10 has a label @xmath53 , this number will be even ( 0 , 2 or 4 ) . since we count directed edges modulo 2 , this implies there is an ( uncancelled ) directed edge from @xmath10 to another vertex @xmath51 if and only if there is also an ( uncancelled ) directed edge from @xmath51 to @xmath10 . we will say that labeled directed multigraphs which have this property are _ [ d : semisym ] a directed multigraph g , with each vertex labeled by a pair \{i , j } , is * semisymmetric * if for every vertex v labeled \{i , i } , and any other vertex w , there is a directed edge from v to w if and only if there is a directed edge from w to v. following lando s work for knots @xcite , we can define a bialgebra structure on the space of intersection graphs for string links so that @xmath54 becomes a bialgebra homomorphism from the space of chord diagrams for string links to the space of intersection graphs . the key is to define the analogue of the 4-term relation for intersection graphs . consider the graded vector space ( over @xmath55 ) of formal linear combinations of labeled semisymmetric directed multigraphs , with vertices labeled by unordered pairs \{i , j } , @xmath56 , graded by the number of vertices in the graphs . for any graph g and vertices a and b in v(g ) , with labels \{i , j } and \{i , l } respectively ( j and l may be equal , or equal to i ) , we impose on the vector space the relation : @xmath57 here @xmath58 is the result of complementing the edge ab in g ( i.e. adding a directed edge between the vertices in each direction , and then cancelling `` mod 2 '' ) . @xmath59 is the result of changing the label on a to \{j , l } , complementing the edge ab if @xmath60 ( and leaving it unchanged if i = l ) , and adding a directed edge from a to c ( respectivley c to a ) for every vertex c in v(g ) for which g has a directed edge from b to c ( respectively c to b ) , and then cancelling mod 2 . finally , @xmath61 is the result of complementing the edge ab in @xmath59 . here is an example of such a relation : @xmath62 the bialgebra @xmath63 is defined as this graded vector space , together with a product and a coproduct . the product is a map @xmath64 , defined as follows . given graphs @xmath65 and @xmath66 , @xmath67 is the disjoint union of the graphs , together with a directed edge from @xmath68 to @xmath69 for each color in the label \{i , j } for @xmath45 which is also in the label for @xmath46 . an example is shown below : @xmath70 the coproduct is a map @xmath71 , defined as follows . for any graph g , and subset @xmath72 of its vertices , let @xmath73 denote the subgraph induced by @xmath36 . then : @xmath74 an example is shown below : @xmath75 [ t : bialgebra ] the product and coproduct defined above induce the structure of a co - commutative ( but not commutative ) bialgebra on the space f of labeled , directed multigraphs modulo the 4-term relation . proof : it is easy to check that @xmath63 is a bialgebra ( i.e. that @xmath76 ) , and that it is co - commutative , but not commutative . we also need to check that the product and coproduct respect the 4-term relation , i.e. that the product of a graph and a 4-term relation yields a 4-term relation , and that the coproduct of a 4-term relation yields a sum of 4-term relations . to begin with , consider a product @xmath77 . it is clear that @xmath78 , since complementing the edge @xmath79 is independent of any edges between @xmath80 and @xmath81 . moreover , @xmath82 , since in each case there is an directed edge from a vertex @xmath83 to @xmath84 for each occurrence of @xmath44 or @xmath85 in the label of @xmath10 ( with cancellation mod 2 ) . it is then clear that @xmath86 , and so the product of the 4-term relation and another graph is another 4-term relation . for the coproduct , consider @xmath87 . the sum in the coproduct splits into two groups - terms where both @xmath84 and @xmath88 belong to either @xmath36 or its complement , and terms where one of the two vertices is in @xmath36 and the other is in its complement . the first group gives a sum of 4-term relations , while the terms of the second group already sum to zero in @xmath89 and @xmath90 . this finishes the proof of the theorem . @xmath91 it is now easy to check that @xmath54 is a bialgebra homomorphism from the bialgebra @xmath28 of chord diagrams for string links with @xmath2 components modulo the 4-term relation to the bialgebra @xmath92 of directed , labeled graphs with labels 1 , ... , @xmath2 , since the 4-term relation for graphs was defined to mimic the 4-term relation for the chord diagrams . as a result , any functional on @xmath92 will induce , by composition with @xmath54 , a regular weight system . we call such a functional a _ regular graph weight system_. any particular weight system will satisfy relations in addition to the 1-term and 4-term relations , and it can be useful to look at weight systems which lie in the subspaces determined by these additional relations . in particular , bar - natan and garoufalides @xcite noted that the weight system associated with the conway polynomial for knots satisfies a set of _ 2-term _ relations which induce the 4-term relations . we can extend this notion to define a set of 2-term relations for string links , shown in figure [ f:2-term ] . [ h ] @xmath93 clearly , these relations imply that the weight system satisfies the 4-term relation as well . as a result , the product and coproduct of section [ ss : finitetype ] are still well - defined . so we can give the vector space of chord diagrams for string links with @xmath2 components modulo the 2-term relations the structure of a bialgebra . we will denote this bialgebra ( and the underlying vector space ) by @xmath94 . there is a natural projection from @xmath28 to @xmath94 . similarly , we can look at the quotient space of @xmath63 by the 2-term relations @xmath95 since @xmath96 , the 2-term relations imply the 4-term relations , and the quotient space is a bialgebra @xmath97 with the product and coproduct induced from @xmath63 . @xmath54 is then a bialgebra homomorphism from @xmath94 to @xmath98 . bar - natan and garoufalides @xcite analyzed the space of chord diagrams for knots ( equivalently , string links of one component ) modulo the 2-term relations . they found that @xmath99 is generated ( as a vector space ) by _ ( @xmath100)-caravans _ of @xmath101 `` one - humped camels '' ( isolated chords which intersect no other chords ) and @xmath102 `` two - humped camels '' ( pairs of chords which intersect each other , but no other chords ) . an example of such a caravan is shown in figure [ f : caravan ] . [ h ] @xmath103 our goal in this section is to find a similar `` normal form '' for string link chord diagrams with more than one component ; this will be useful in the remainder of the paper . we will prove the following theorem : [ t : normalform ] any connected string link chord diagram is equivalent , modulo the 2-term relations , to a chord diagram with components numbered 1 , ... , n ( possibly after renumbering ) such that : * every chord either has both endpoints on component 1 , or endpoints on components i and i+1 for some i. * the chords with both endpoints on component 1 are arranged in a caravan . moreover , all of their endpoints lie below the endpoints of any chords between components . * there are at most 2 chords between components i and i+1 , and if there are 2 they do not cross ( so the endpoints of one chord lie above the endpoints of the other on both components ) . * on component i , the endpoints of chords connecting component i to component i+1 lie below the endpoints of chords connecting component i to component i-1 . an example is shown in figure [ f : normal ] . [ h ] @xmath104 if a diagram is not connected , we simply put each connected component into normal form . the proof proceeds in several steps . we will describe an algorithm to put any connected chord diagram into the normal form described above via 2-term relations . + step 1 : we first deal with chords which have both endpoints on the same component . to begin with , any pair of intersecting chords which have all four endpoints on the same component can be slid to the bottom of the component , away from any other chords , to form a two - humped camel as in @xcite . an isolated chord with both endpoints on the same component can also be slid down to the bottom , forming a one - humped camel . see figure [ f : factor ] for examples of these moves . [ h ] @xmath105 we are left with chords with both endpoints on the same component which only intersect chords which lie between two different components . by sliding this chord over one of the chords it intersects , its endpoints are now on two different components , as shown in figure [ f : step1 ] . [ h ] @xmath106 step 2 : our next step is to put every chord connecting two different components on a different horizontal `` level '' ; in particular , we will remove intersections between chords between the same two components . say that we have @xmath12 chords between components . select a chord arbitrarily , which we will denote chord @xmath107 . without loss of generality , assume @xmath107 connects components 1 and 2 . we will move @xmath107 to the top level . as we move it up , if we come to a chord with an endpoint on components 1 or 2 which is above @xmath107 we move that endpoint below @xmath107 via a 2-term relation , as shown in figure [ f : step2 ] . [ h ] @xmath108 ultimately , the endpoints of @xmath107 will lie above the endpoints of any other chord . during this process , some chords may have had both their endpoints moved to the same component - remove or modify these chords as in step 1 ( notice that this only reduces the number of chords between components from @xmath12 ) . we are left with _ at most _ @xmath109 chords connecting components , all below @xmath107 . so we can continue the process inductively , ending with all chords between components horizontal , at different levels ( and with a caravan at the bottom of each component ) . + step 3 : now we consider one component of the diagram , say component 1 , and move all the chords with an endpoint on component 1 to a higher level than all the other chords in the diagram ( leaving chords with both endpoints on the same component at the bottom ) . we will denote by an _ ( i , j)-chord _ a chord with endpoints on components @xmath43 and @xmath44 . if a ( 1,@xmath43)-chord lies directly below an ( @xmath110)-chord along component @xmath43 , we can use a 2-term relation to move the ( 1,@xmath43)-chord above the other chord , in the process transforming it into a ( 1,@xmath44)-chord , as in figure [ f : step3 ] . [ h ] @xmath111 since there are only a finite number of chords , we can successively move all chords with one endpoint on component 1 above all other chords ( while the other endpoint may move , one will always remain on component 1 ) . + step 4 : let @xmath107 be the top chord at this point ; without loss of generality , @xmath107 is a ( 1,2)-chord . let @xmath112 be the number of ( 1,@xmath43)-chords , and @xmath113 . now move @xmath107 down component 1 , stripping other chords off of component 1 by 2-term relations , transforming them from ( 1,@xmath43)-chords to ( 2,@xmath43)-chords , reversing the move from step 2 shown in figure [ f : step2 ] . continue this until @xmath107 encounters another ( 1,2)-chord , @xmath114 . now move @xmath107 back up as in step 3 ( see figure [ f : step3 ] ) ; when it is again at the top , it will be a ( 1,@xmath43)-chord for some @xmath43 . repeat the process for @xmath114 , moving it up to just below @xmath107 , and with all the other ( 1,2)-chords . this reduces @xmath115 to @xmath116 , though the chords may no longer be ( 1,2)-chords . we can relabel the components so that @xmath107 is once again a ( 1,2)-chord and repeat the process . each repetition reduces @xmath115 , until we reach a point where @xmath117 ( i.e. all chords with an endpoint on component 1 are ( 1,2)-chords ) , and the ( 1,2)-chords lie above all other chords in the diagram . finally , we can reduce @xmath116 to 1 or 2 by noticing that whenever @xmath118 , we can factor out a 2-humped camel and reduce @xmath116 by 2 , as in figure [ f : reduce ] . + [ h ] @xmath119 step 5 : we can now repeat steps 3 and 4 for component 2 , and then for each component in turn . we are left with a diagram which is almost in normal form - the final step is to move all of the caravans to component 1 . we can move a caravan from the @xmath43th component to the @xmath120th component by sliding it over a chord between the two components , as in figure [ f : step5 ] . [ h ] @xmath121 any other chords between the two components can be slid over the caravan in turn by 2-term relations , leaving the caravan below the chords connecting components @xmath122 and @xmath123 . continuing in this way , the caravans can be moved to component 1 , and finally pushed to to bottom of component 1 as in step 1 . this completes the proof of theorem [ t : normalform ] . @xmath91 in this section we will define the adjacency matrix for an intersection graph , and show that its rank and determinant ( over @xmath124 ) are regular graph weight systems . [ d : adj ] given a directed graph g with n vertices , @xmath125 , such that each vertex @xmath126 has a label @xmath127 , the * adjacency matrix of g * , or adj(g ) , is the @xmath128 matrix defined by : @xmath129 @xmath130 these matrices can be viewed as bilinear forms over @xmath124 , but unlike the case for knots @xcite , they are not generally symmetric , so their classification is much more difficult ( and still , to my knowledge , an open question ) . we are interested in the notion of _ congruence _ of such matrices . we are particularly interested in the adjacency matrices of _ semisymmetric _ graphs ; which have the property ( which we will also denote by semisymmetry ) that if the @xmath43th diagonal element is 0 , then the @xmath43th row and column are the same . we say that two @xmath131 matrices a and b are * congruent * ( denoted @xmath132 ) if there is an invertible matrix p such that @xmath133 . one of our main results is that the congruence class of the adjacency matrix ( over @xmath124 ) satisfies the 2-term relation , and so induces a regular graph weight system . for any graph g in e , @xmath134 . proof : reordering the vertices of @xmath80 changes @xmath135 by a congruence ; in this case , @xmath136 is the result of doing a corresponding reordering of the rows of the identity matrix . so we can assume that the first two rows and columns of @xmath135 and @xmath137 correspond to the vertices @xmath84 and @xmath88 . the matrices are identical except for the first row and column . say that @xmath84 has label @xmath138 and @xmath88 has label @xmath139 , so in @xmath59 @xmath84 has label @xmath140 . an entry @xmath141 ( where @xmath142 ) is equal to @xmath143 ( mod 2 ) , and similarly for the first column . @xmath144 if and only if @xmath145 , i.e. if @xmath146 , so @xmath147 . finally , @xmath148 if @xmath149 , and 1 otherwise . we have five cases . + + case 1 : @xmath150 . in this case @xmath151 and @xmath152 ( due to the semisymmetry of the graph ) . so , mod 2 , @xmath153 . + + case 2 : @xmath154 . in this case @xmath155 and @xmath152 . so , mod 2 , @xmath156 . + + case 3 : @xmath157 . in this case , @xmath158 and @xmath159 , and @xmath160 . so , mod 2 , @xmath161 . + + case 4 : @xmath162 . in this case , @xmath163 and @xmath164 , and @xmath160 . so , mod 2 , @xmath161 . + + case 5 : @xmath165 all different . in this case @xmath155 and @xmath166 . so , mod 2 , @xmath161 . + we conclude that @xmath167 . therefore @xmath135 and @xmath137 are congruent by a matrix @xmath136 , where @xmath136 is the elementary matrix constructed from the identity matrix by adding the second row to the first row . in other words , @xmath136 is the same as the identity , except for the 2 by 2 matrix in the upper left corner , which is @xmath168 $ ] . @xmath91 the congruence class of the adjacency matrix of the intersection graph is invariant under the 4-term relations . the determinant ( mod 2 ) and rank ( over @xmath124 ) of the adjacency matrix of the intersection graph are regular graph weight systems . proof : if @xmath136 is invertible over @xmath124 , the rank of @xmath169 is the same as the rank of @xmath84 . in addition , @xmath170 ( mod 2 ) , so the determinant is also a congruence invariant mod 2 . @xmath91 for knots , the conway and homfly weight systems can be interpreted via intersection graphs @xcite . to what extent can we do the same for string links ? in general , this is still an open problem , but we can offer a few initial results . let us recall the definitions of the conway polynomial and weight system . the conway polynomial @xmath171 of a link is a power series @xmath172 . it can be computed via the skein relation ( where @xmath173 are as in figure [ f : skein ] ) : [ b ] @xmath174 @xmath175 @xmath176 we define the conway polynomials of a string link as the conway polynomial of the link formed by closing the string link in the natural way . the coefficient @xmath177 is a finite type invariant of type @xmath12 @xcite , and therefore defines a weight system @xmath178 of degree @xmath12 . the collection of all these weight systems is called the conway weight system , denoted @xmath179 . consider a chord diagram @xmath34 , together with a chord @xmath10 . let @xmath180 be the result of _ surgery on v _ , i.e. replacing @xmath10 by an band which preserves the orientation of the components , and then removing the interior of the band and the intervals where it is attached to @xmath34 , as shown in figure [ f : surgery ] [ h ] @xmath181 ( so @xmath180 may have multiple boundary circles ) . the skein relations for the conway polynomial give rise to the following relations for @xmath179 : @xmath182 @xmath183 it is easy to show @xcite that this weight system satisfies the 2-term relations of section [ s:2term ] . simply surger the two chords ; the 2-term relation then says just that one band can be `` slid '' over the other , which does nt change the topology of the diagram . bar - natan and garoufalides also showed that the weight system for the conway polynomial for knots is just the determinant of the adjacency matrix for the intersection graph , mod 2 . the next theorem extends this result to string links with two components . [ t : conway2 ] for any chord diagram d on a string link with two components , @xmath184 ( mod 2 ) . proof : since both of these weight systems satisfy the 2-term relations , it suffices to show that they agree on diagrams in normal form ( see section [ ss : chord2term ] ) . for string links with two components , normal form consists of a caravan on the first component , and 0 , 1 , or 2 parallel chords between the two components . say that the caravan has @xmath101 1-humped camels and @xmath102 2-humped camels , and there are @xmath85 chords between the components ( @xmath185 ) . then @xmath186^l \oplus [ 0]^{m_1 } \oplus \left[{\matrix{0 & 1 \cr 1 & 0}}\right]^{m_2}$ ] . so @xmath187 ( mod 2 ) and @xmath188 ( mod 2 ) . the product of the rank and determinant is 1 ( mod 2 ) when @xmath189 and @xmath190 , and 0 otherwise . on the other hand , if we surger all the chords of the diagram , and close the string link , we obtain an unlink with @xmath191 components when @xmath192 , and @xmath193 when @xmath194 , which means that @xmath195 only when @xmath196 and @xmath194 . so the two weight systems agree . @xmath91 with more than two components , the situation is more complicated ; in fact , the conway weight system is not always determined by the adjacency matrix of the intersection graph . figure [ f : counterexample ] shows two chord diagrams @xmath30 and @xmath31 on 4 components which illustrate this . [ h ] @xmath197 it is easy to see that these intersection graphs for these diagrams differ only in the labeling of the vertices , so : @xmath198\ ] ] however , surgering the chords in @xmath30 yields a string link whose closure has one component , and surgering the chords in @xmath31 yields a string link whose closure has 3 components . therefore , @xmath199 , but @xmath200 . however , we can at least give a sufficient condition for @xmath201 to be trivial which depends only on the adjacency matrix . [ p : suff ] let d be a string link chord diagram of degree k on n components . if either of the following conditions holds , then c(d ) = 0 . * @xmath202 . * @xmath203 is even . proof : the first condition means that , in the normal form , there will be an isolated chord in the caravan , so @xmath201 will be trivial . the second condition comes from observing that every surgery of a chord in @xmath34 the number of components in the closure of @xmath34 by 1 ( either adding or removing a component ) . since we begin with @xmath12 components , the number of components after all the chords are surgered is congruent to @xmath203 modulo 2 . so @xmath201 can be non - trivial only if @xmath204 mod 2 . hence , if @xmath203 is even , then @xmath205 . @xmath91 now we will consider the ( framed ) homfly weight system for string links . the homfly invariant is the laurent polynomial @xmath206 $ ] defined by the following skein relations @xcite ( @xmath207 is the result of adding a positive kink to the link @xmath208 ) : @xmath209 @xmath210 @xmath211 @xmath212 if we make the substitutions @xmath213 and @xmath214 , and expand the resulting power series , we transform the homfly polynomial into a power series in @xmath215 , whose coefficients are finite type invariants ( of regular isotopy ) . these invariants give rise to regular weight systems which we can collect together as the homfly regular weight system @xmath81 . the skein relations above give rise to the following relations for @xmath81 , by looking at the first terms of the power series ( as before , @xmath180 is the result of surgering the chord @xmath10 in @xmath34 ) : @xmath216 @xmath217 @xmath218 so if @xmath34 is an unlink of @xmath2 components , @xmath219 . since the first of these relations is almost the same as for the conway weight system @xmath179 , the same argument shows that @xmath81 satisfies the 2-term relations . we can now consider string link diagrams with two components ( as with the conway weight system , the homfly weight system is not necessarily determined by the adjacency matrix of the intersection graph for chord diagrams on more than two components ) . [ t : homfly2 ] for any chord diagram d of degree k on a string link with 2 components , let @xmath220 . if r is odd , then @xmath221 ; if r is even , then @xmath222 . proof : as with theorem [ t : conway2 ] , it suffices to show that the weight systems agree on diagrams in normal form . let @xmath34 be the diagram with @xmath101 one - humped camels and @xmath102 two - humped camels on component 1 , and @xmath85 chords between the two components ( @xmath185 ) ( so the degree of @xmath34 is @xmath223 ) . as before , @xmath186^l \oplus [ 0]^{m_1 } \oplus \left[{\matrix{0 & 1 \cr 1 & 0}}\right]^{m_2}$ ] , so the rank is @xmath224 . @xmath225 is even when @xmath192 and odd when @xmath194 . on the other hand , if we surger all the chords ( each time multiplying @xmath81 by @xmath226 ) , the resulting link has @xmath191 components when @xmath192 ( and @xmath225 is even ) , and @xmath193 components when @xmath194 ( and @xmath225 is odd ) . so when @xmath225 is even , @xmath227 , and when @xmath225 is odd , @xmath228 . @xmath91 + + remark : if we let @xmath229 and @xmath230 , then @xmath231 . according to theorem [ t : homfly2 ] , if @xmath34 is a string link chord diagram with two components , we will have @xmath232 in this case _ unless _ @xmath233 is odd ( in which case @xmath234 ) . but this is exactly the case when @xmath235 ( since the matrix has full rank ) and @xmath236 mod 2 , so @xmath195 . so the formulas for the conway and homfly weight systems agree in this case . there are several obvious questions and directions for further research . figure [ f : counterexample ] gives an example of chord diagrams on 4 components where the adjacency matrix of the intersection graph is insufficient to determine the conway and homfly weight systems . however , this example seems to depend strongly on there being 4 components - it is not clear how to find connected examples with , say , 5 components where the adjacency matrix is insufficient . and certainly the adjacency matrix ( as shown by the example in figure [ f : counterexample ] does not contain all the information of the intersection graph - in particular , the information contained in the vertex labels . this leads to our next question : this question is still open for knots , as well , though some progress has been made @xcite . some work has been done by the author @xcite for the case when the intersection graph is a tree , but this is only a bare beginning . 10 bar - natan , d. : on the vassiliev knot invariants , _ topology _ 34 , no . 2 , pp . 423 - 472 , 1995 bar - natan , d. and garoufalidis , s. : on the melvin - morton - rozansky conjecture , _ invent . _ 125 , pp . 103 - 133 , 1996 birman , j. and lin , x.s . : knot polynomials and vassiliev s invariants , _ invent . _ 111 , pp . 225 - 270 , 1993 bouchet , a. : circle graph obstructions , _ journal of combinatorial theory , series b _ , 60 , pp . 107 - 144 , 1994 chmutov , s.v . and duzhin , s.v . : the kontsevich integral , _ acta applicandae math . 155 - 190 , 2001 chmutov , s.v . , duzhin , s.v . and lando , s.k . : vassiliev knot invariants i , ii , iii , _ advances in soviet mathematics _ 21 , pp . 117 - 145 , 1994 habegger , n. and lin , x.s . : the classification of links up to link - homotopy , _ j. of the amer . 3 , no . 2 , 1990 , 389 - 419 kontsevich , m. : vassiliev s knot invariants , _ advances in soviet mathematics _ , vol . 16 , part 2 , pp . 137 - 150 , 1993 lando , s. : on a hopf algebra in graph theory , _ journal of combinatorial theory , series b _ , v.80 , p. 104 - 121 , 2000 le , thang : unpublished , 1996 lickorish , w.b . : _ an introduction to knot theory _ , gtm vol . 175 , springer , new york , 1997 mellor , b. : the intersection graph conjecture for loop diagrams , _ journal of knot theory and its ramifications _ , vol . 9 , no . 2 , 2000 . also see arxiv : math.gt/9807033 mellor , b. : a few weight systems arising from intersection graphs , _ michigan math . j. _ , vol . 509 - 536 , 2003 . also see arxiv : math.gt/0004080 mellor , b. : tree diagrams for string links , preprint , may 2004 , arxiv : math.gt/0405537 mellor , b. : tree diagrams for string links ii : determining chord diagrams , preprint , august 2004 , arxiv : math.gt/0408286 mellor , b. : weight systems for milnor invariants , in preparation meng , g. : bracket models for weight systems and the universal vassiliev invariants , _ topology and its applications _ 76 , 47 - 60 , 1997 morton , h.r . and cromwell , p.r . : distinguishing mutants by knot polynomials , _ journal of knot theory and its ramifications _ 5 , no . 225 - 236 , 1996 milnor , j. and husemoller , d. : _ symmetric bilinear forms _ , springer - verlag , new york , 1973 stanford , t. : finite - type invariants of knots , links , and graphs , _ topology _ 35 , no . 4 , pp . 1027 - 1050 , 1996 vassiliev , v.a . : cohomology of knot spaces , in _ theory of singularities and its applications _ arnold ) , _ advances in soviet mathematics _ , vol . 1 , pp . 23 - 69 , 1990

the disciplined software engineering relies on software architectures to describe systems @xcite . for modeling a software architecture , the academy proposed the architecture description language ( adl ) and their toolsets @xcite . also , the industry has used adls to develop systems . industry examples are ( i ) @xmath0-adl has been used to architect and refine federated knowledge management systems in engineering ingegneria informatica - italy @xcite ; ( ii ) and aadl ( architecture analysis & design language ) is a standard language for the society of automotive engineers @xcite . according to the state - of - the - art @xcite , the following concepts are relevant to describe software architectures : components and connectors ( respectively computational and communication elements ) , architectural constraints , non functional properties , and behaviour . software systems evolve over their life time @xcite . dynamic reconfiguration is when the evolution is performed at runtime with no service disruption . the dynamic reconfiguration can be handled by architectural concepts in an adl . however among many existing adls , only few allow modeling dynamic reconfiguration . @xmath0-adl @xcite , acme / plastik @xcite , c2sadl @xcite , daop - adl @xcite , darwin @xcite , dynamic wright @xcite , rapide @xcite , weaves @xcite , and xadl @xcite are typical examples . nevertheless , there is no current consensus about how adls should address reconfiguration , e.g. , what language constructs should be provided . in this work , the goal is twofold : ( i ) to investigate the adls support for handling dynamic reconfiguration in the literature ; and , ( ii ) to illustrate how four well - known adls support dynamic reconfiguration : @xmath0-adl , acme , c2 sadl and dynamic wright . we chose these four languages because they rely on different paradigms : the higher order typed @xmath0-calculus @xcite , first order predicate logic @xcite , compo - nent- and event - based @xcite , and , graph grammars and communicating sequential processes ( csp ) @xcite , respectively . they also complement each other as @xmath0-adl models the behavior of architectures , acme focuses on the structure , c2sadl make the attention for components and their concurrent events , and dynamic wright supports the definition of structure and behavior . this work is structured as follows . section [ section : motivation ] motivates this work thanks to the example of a tcp / ip stack system . three scenarios illustrate three facets of dynamic reconfiguration . section [ section : concepts ] presents the basics concepts of software architecture and dynamic reconfiguration . section [ section : related - works ] details the related works about dynamic reconfiguration at the architecture level . section [ section : comparison ] compares the four above - mentioned adls in the light of our example of section [ section : motivation ] . section [ section : conclusion ] concludes the paper with open issues that we note in the adl support for dynamic reconfiguration . dynamic reconfiguration aims at modifying the software systems at runtime with no service disruption . critical systems usually want to benefit of a dynamic reconfiguration because any service disruption may have substantial consequences . examples of critical systems are stock market quotation systems , telecommunications systems , safety systems , and air traffic control systems . however , there is no current consensus about how the software architecture should address dynamic reconfiguration at architecture level . does regular syntax for dynamic architectures allow the same reconfigurations as specific language constructs ? should the state of components and connectors be taken into account at the architecture level ? to support this work , we rely on a simplified version of the tcp / ip stack ( fig . [ fig : tcpip - stack ] ) . each of the four layers ( application , transport , internet , and link ) is modeled as a component ( fig . [ fig : tcpip - system ] ) . we consider the three following reconfigurations : 1 . _ switch the application component . _ assume two alternatives : mpeg - decoder and h263-decoder . mpeg - decoder is used if _ bandwidth _ is high ( fig . [ fig : tcpip - mpeg ] ) . otherwise h263-decoder is selected ( fig [ fig : tcpip - h263 ] ) . switching between mpeg and h263 starts a new stream . therefore , no state in the decoder needs be kept . we can therefore replace one component by another . insertion of a new component type . _ presume that the internet protocol version 6 ( ipv6 ) replace the version 4 ( ipv4 ) ( fig . [ fig : tcpip - ipv6 ] ) . the ipv6 component has backward compatibility with the ipv4 component . so it has the ports for both ipv4 ( for compatibility reason ) and for ipv6 ( new feature ) . therefore , the type of the component at the internet layer is changed : it has 2 additional ports . _ update a component type . _ assume that the algorithm of error control of the transport component is improved . when the error control algorithm is changed , the new behaviour needs to know the status of each packet ( acknowledged , sent , timeout , ... ) as well as the content of non - acknowledged packets ( in order to resend them ) . therefore , only the behaviour part of the component should be replaced . the state of the component should be unchanged . these scenarios cover a wide spectrum . they can either be programmed at design - time or be discovered at run - time . they feature reconfiguration at both type and instance level . they target both the structure and the behaviour of the architecture . we investigate the adls literature in order to illustrate how reconfiguration is supported in these scenarios . in this paper , we focus on four adls that are representative of the main trends . each adls complement the other . the complement at the sense for describing an initial architecture as well as dynamic reconfiguration . we choose @xmath0-adl because it provides both description with behaviours thanks to its @xmath0-calculus roots . whilst the acme / plastik is a declarative language with focus on the architecture structure for both descriptions . while c2sadl is used to compose architectures based on component and dynamic reconfiguration based on architectural events . finally , dynamic wright provides support for architectural description based in graph - grammars and dynamic reconfiguration specification with a variant of communicating sequential processes ( csp ) . [ [ software - architecture - concepts ] ] software architecture concepts + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + software architectures + describe systems by specifying their elements and the interactions between them @xcite . in this work , we use the basics concepts defined in @xcite : a _ component _ is an unit of computation and storage ; a _ connector _ is an unit of communication ; and , a _ configuration _ is a specification of a software architecture in terms of components , connectors , and relationship between them . ports , roles , behaviours , constraints , and non - functional properties are other concepts defined in @xcite used to describe these architectural elements . ports and roles are used to define the way of interaction , ports for configurations and components , and roles for connectors . behaviours are the architectural element internal computation . and the last two concepts , constraints and non - functional properties , are used to denote the assertions , invariants , quality of service that architectural elements are expected to meet . also , we consider types and instances of configurations , components , and connectors for modelling a software architectures @xcite . the types are abstractions like classes in the object - oriented paradigm which encapsulate both the structure and the behaviours that can be performed on its structure . these types will be used to build the instances and these instances will be used to describe software architectures . fig . [ code : tcpip - acme ] illustrates in acme the basics concepts with our example tcp / ip stack system ( sec . [ section : motivation ] fig . [ fig : tcpip - system ] ) at instance and type level . the types are defined in the ` family ` statement block ( lines 1 - 18 ) while the instance level in ` system ` ( lines 20 - 30 ) . .... family tcpip_mf { . . connector type tcpip_conn2layers { providedrole source ; requiredrole sink ; } ; component type tcpip_component { port type datafrom extends providedport with { } ; port type datato extends revidedport with { } ; property type layer = enum{application , transport , internet , link } } ; component type tcpip_application extends tcpip_component with { port datareceivedfromtransport : datafrom ; port requiredservice : datato ; property layer = " application " ; } ; . . } system decoderstream : tcpip_mf = new tcpip_mf extended with { component application : tcpip_application ; component transport : tcpip_transpot ; component internet : tcpip_internet ; component link : tcpip_link ; connector application2transport , transport2internet , internet2link : conn2layers ; attchments { application.requiredservice to application2transport.source ; application2transport.sink to transport.service ; transport.requiredservice to transport2internet.source ; transport2internet.sink to internet.service ; internet.requiredservice to internet2link.source ; internet2link.sink to link.service ; } . . } .... [ [ dynamic - reconfiguration - concepts ] ] dynamic reconfiguration concepts + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + a dynamic reconfiguration is a set of operations to modify an existing configuration at runtime . at the software architecture level , the operation is defined in terms of the architectural elements at type or instance level . scenario 1 is an example of reconfiguration at the instance level ( mpeg and h263 share the same type ) ; and scenario 2 is an example of reconfiguration at the type level ( ipv6 extends the type of ipv4 ) . also , the dynamic reconfiguration can be foreseen or unforeseen @xcite . the foreseen is a programmed dynamic reconfiguration specified at design time . while the unforeseen is an ad - doc defined at runtime . example of foreseen and unforeseen is the scenario 1 and 3 , respectively . in the scenario 1 , the architect specify at design time that the system modify itself when the bandwidth change . whereas the behaviour improvement is built at runtime as defined in scenario 3 . medvidovic and taylor @xcite created a classification framework based on architectural elements and illustrated them with the following adls : acme , aesop , c2 , darwin , metah , rapide , sadl , unicon , weaves , and wright . however , this work has no focus on the dynamic reconfiguration . hence , its discussion about dynamic reconfigurations is limited , e.g. all adls were assessed only at the instance level , not at the type level . kacem _ et al . _ @xcite described the capabilities of darwin , archjava , olan , rapide , wright , and acme to support dynamic reconfiguration . they classify the adls as configuration and description language . the configuration language supports the description of a software system and limited dynamic reconfigurations , while the description language supports the specification of modifications to be applied to an existing architecture . despite this work focus is on dynamic reconfiguration , it also presents some limitations such as the fact that the evolution is applied only at instance level and do not take into account the behaviour of the architectural elements . bradbury @xcite and bradbury _ et al . _ @xcite proposed a framework to compare fourteen formal specification approaches which support dynamic software architectures . amongst the criteria , bradbury _ et al . _ considered the basic operations , how they can be composed and whether they apply to variable sets of architectural element types . about the basic operations ( addition and removal of components and connectors ) , bradbury et al . showed that most approaches support them . also , they considered the criterion of operations composition ( basic , sequence , choice , and iteration ) . only 3 approaches ( community , dynamic wright , and gerel ) provide full support while the other only support basic composition . finally , bradury _ et al . _ mentioned that all approaches support only the types of architectural elements defined prior runtime , fixed sets of architectural elements . while adl surveys have reached good understanding of the concepts underlying software architectures ( medvidovic and taylor @xcite ) , this is not true regarding dynamic reconfiguration at the architecture level . even bradbury @xcite and bradbury _ et al . _ @xcite , probably the most complete surveys , do not consider operations such as insertion of configuration and behaviour modification . they do not recognize _ link _ and _ unlink _ as basic operations . however , they present the examples of these operations on some approaches , like gerel and c2sadl . considering the scenarios of section [ section : motivation ] , these related works address only the first scenario ( partially ) but not the two last . consequently , it is clear that to discuss the three scenarios presented in section [ section : motivation ] is essential to have a better understanding of the dynamic reconfiguration issues . these scenarios consider both instance and type levels , the basic operations include configurations , components , and connectors , as well as their structures and behaviours . in this subsection , we describe the scenarios defined in sec . [ section : motivation ] in all four adls : @xmath0-adl ( subsection [ sec : pi - adl ] ) , acme / plastik ( subsection [ sec : acme ] ) , c2sadl ( subsection [ sec : c2sadl ] ) , and dynamic wright ( subsection [ sec : wright ] ) . for each adl we show the purpose , the tcp / ip stack system , a solution for three scenarios , and discussion about this implementation . [ [ purpose - of - the - adl ] ] purpose of the adl + + + + + + + + + + + + + + + + + + @xmath0-adl is based on @xmath0-calculus and it is designed to describe the concurrent and mobile systems @xcite . the description of architectural elements is mainly represented by ports and behaviour . [ [ dynamic - reconfiguration - support ] ] dynamic reconfiguration support + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + this adl provides + support for foreseen and unforeseen dynamic reconfiguration . the foreseen can be described in the behaviour of any architectural elements . while unforeseen needs some support , e.g. , in the virtual machine ( vm ) , in order to obtain a root reference to the reconfigured system . in archware , this issue is addressed by the deep intrication between the toolset ( including the visual and textual editors ) and the vm thanks to the hyper - code technology @xcite . [ [ initial - architecture ] ] initial architecture + + + + + + + + + + + + + + + + + + + + the implementation of tcp / ip stack system is shown in fig . [ code : tcpip - piadl ] . the configuration is composed at lines 1 - 18 basically with a behaviour , statement ` behaviour ` is ... where ... ! ( lines 2 - 17 ) . this architectural behaviour defines the instances of components ( lines 3 - 6 ) and connectors ( lines 7 - 9 ) . the statement ` where ` in behaviour specifies the connections between components and connectors ( lines 11 - 16 ) . the example of component ` tcpip_application ` is partially showed at lines 20 - 35 : two ports and a behaviour . the component behaviour is implemented with two internal methods ( lines 24 - 25 ) and the definition of communication between its ports ( lines 27 - 33 ) . .... archictecture tcpip is abstraction ( ) { behaviour is compose { application is tcpip_application ( ) and transport is tcpip_transport ( ) and net is tcpip_internet ( ) and link is tcpip_link ( ) and app2transp is conn2layers ( ) and transp2net is conn2layers ( ) and net2link is conn2layers ( ) } where { appliation::request unifies app2transp::source and app2transp::sink unifies transport::service and transpost::request unifies transp2net::source and transp2net::sink unifies net::service and net::request unifies net2link::source and net2link::sink unifies link::service } } component tcpip_application is abstraction ( ) { port service is { ... } . port request is { ... } . behaviour is { processrequest is function(d : datatype):datatype { unobservable}. processresponse is function(d : datatype):datatype { unobservable}. choose { via service::wait receive entrydata : datatype . via request::call send processrequest(entrydata ) . or via request::response receive replydata : datatype . via service::reply send processresponse(replydata ) . } } } component tcpip_transport is abstraction ( ) { ... } component tcpip_internet is abstraction ( ) { ... } component tcpip_link is abstraction ( ) { ... } connector conn2layers is abstraction ( ) { ... port source is { ... } port sink is { ... } behaviour is { ... } } .... [ [ scenario-1 ] ] scenario 1 + + + + + + + + + + the implementation of scenario 1 in @xmath0-adl is composed by a component that performs the dynamic reconfiguration ( fig . [ code : scenario1-piadl ] ) . this component has two parameters : the application component and the system component . both are represented as a behaviour . the application component is a mpeg - decoder or h263-decoder . still , this component is structured with seven `` variables '' and a behaviour , lines 2 - 4 and 5 - 27 , respectively . the variables are used for behaviours of components and connectors . to perform the modification , the behaviour component use the following steps : decomposition the system behaviour ( line 6 ) ; assign the behaviours to variables ( lines 8 - 14 ) ; compose a new system behaviour ( lines 16 - 26 ) with a replaced application component ( line 20 ) . .... component reconfiguration is abstraction(application : behaviour , system : behaviour ) { behaviours : sequence[behaviour ] . transport : behaviour . net : behaviour . link : behaviour . app2transp : behaviour . transp2net : behaviour . net2link : behaviour . behaviour is { behaviours : = decompose system . transport : = behaviours::1::bhvr . net : = behaviours::2::bhvr . link : = behaviours::3::bhvr . app2transp : = behaviours::4::bhvr . transp2net : = behaviours::5::bhvr . net2link : = behaviours::6::bhvr . compose { application and transport and net and link and app2transp and transp2net and net2link } where { appliation::request unifies app2transp::source and app2transp::sink unifies transport::service and transpost::request unifies transp2net::source and transp2net::sink unifies net::service and net::request unifies net2link::source and net2link::sink unifies link::service } } } .... [ [ scenario-2 ] ] scenario 2 + + + + + + + + + + the implementation for the unforeseen scenario 2 is shown in fig . [ code : scenario2-piadl ] . this code is applied in a system at runtime to deploy the two new component `` type '' . first , the component type is a new protocol _ ipv6 _ ( lines 1 - 3 ) , and the second is a component to perform dynamic reconfiguration ( lines 4 - 32 ) . in order to execute this dynamic reconfiguration two steps are applied : ( i ) to obtain the root behaviour of the system , and ( ii ) to invoke the computation of the reconfiguration component . both operations are aided by toolset and hyper - code . the code of the reconfiguration component is described using the following steps : ( i ) to decompose the behaviours of the system in components and connectors ( lines 9 ) ; ( ii ) to assign to `` variables '' the instances of components that represent application , transport , and link ( lines 11 - 13 ) ; ( iii ) to attribute connectors to `` variables '' ( lines 15 - 17 ) ; ( iv ) to compose a new system behaviour ( lines 19 - 30 ) with a new instance of the protocol ipv6 component ( line 21 ) . as in the scenario 1 , when performing a new composition , the behaviour of the system is automatically updated . .... component tcpip_ipv6 is abstracion ( ) { ... } component reconfiguration is abstraction(system : behaviour ) { behaviours : sequence[behaviour ] . application : behaviour . transport : behaviour . link : behaviour . app2transp : behaviour . transp2net : behaviour . net2link : behaviour . behaviour is { behaviours : = decompose system . application : = behaviours::0::bhvr . transport : = behaviours::1::bhvr . link : = behaviours::3::bhvr . app2transp : = behaviours::4::bhvr . transp2net : = behaviours::5::bhvr . net2link : = behaviours::6::bhvr . compose { application and transport and link and net is tcpip_v6 and app2transp and transp2net and net2link } where { appliation::request unifies app2transp::source and app2transp::sink unifies transport::service and transpost::request unifies transp2net::source and transp2net::sink unifies net::service and net::request unifies net2link::source and net2link::sink unifies link::service } } } .... [ [ scenario-3 ] ] scenario 3 + + + + + + + + + + in the scenario 3 it is also used the process of decomposition , modification , and composition , as showed above . we use the archware virtual machine facility to get a reference to the root of the target transport component . after , we make a modification into the behaviour of the component . at last , the archware virtual machine synchronize the system with this modification . [ [ summary - of - pi - adl ] ] summary of @xmath0-adl + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + with these implementations , some issues are identified on @xmath0-adl support for dynamic reconfiguration : 1 . changing types needs external help , e.g. toolset , hyper - code and @xmath0-arl . toolset and hyper - code are used for capturing , specifying the changes , and applying the changes . while @xmath0-arl is a language to specify refinements in the system ; 2 . instances modifications can be specified in architectural elements or also with toolset and hyper - code help . at first , usually the dynamic reconfiguration is tangled with nominally behaviour . however , we used other approach with a specific component to specify dynamic reconfiguration ; 3 . for the third scenario , the @xmath0-adl do not provide a mechanism to design the intermediate states for updating the instances with the improvement of behaviour . however , if we use the @xmath0-arl approach , we can have better control ; 4 . also with @xmath0-arl is possible build constraints and non - functional properties for architectural elements . however , it s not used to assess dynamic reconfiguration . [ [ purpose - of - the - adl-1 ] ] purpose of the adl + + + + + + + + + + + + + + + + + + acme / armani is a declarative language based on the first - order predicate logic @xcite . its initial purpose was to define a common interchange language for architecture design tools . also , its statements are designed to describe the architectural structures at instance and type levels . the extensions @xcite to support dynamic reconfiguration preserves these initials purpose , and are named as acme / plastik . thereafter , any configuration and dynamic reconfiguration specified using this extension ( acme / plastik ) are based on the structure of an architecture . [ [ dynamic - reconfiguration - support-1 ] ] dynamic reconfiguration support + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + acme / plastik permits to describe foreseen and unforeseen dynamic reconfiguration . to unforeseen reconfiguration it is possible to describe as a specific behaviour of a configuration with the ` on ( $ conditional\_expression$ ) do { $ operations$ } ` statement . unforeseen reconfiguration can be expressed in a script that will be applied in the system at runtime aided by a toolset provided by the plastik framework . both , @xmath1 and script are composed with acme statements and its acme / plastik extension . the @xmath2 is composed in the armani language @xcite . [ [ initial - architecture-1 ] ] initial architecture + + + + + + + + + + + + + + + + + + + + the example of the tcp / ip stack system in acme is showed in fig . [ code : tcpip - acme ] . the types are specified in the ` family ` statement while the instances are in the ` system ` statement . for types , a connector ( lines 3 - 6 ) and two components ( lines 7 - 11 and 12 - 16 ) are implemented as example . the configuration is composed by the component instances ( lines 21 - 24 ) , connector instances ( lines 26 - 27 ) , and the connections between them ( lines 28 - 37 ) . [ [ scenario-1 - 1 ] ] scenario 1 + + + + + + + + + + the implementation in acme / plastik of scenario 1 is showed in fig . [ code : scenario1-acme ] . the configuration is named with ` streamdecodersystem ` and it extends the ` tcpip_mf ` and ` plastikmf ` . the ` tcpip_mf ` is showed in fig . [ code : tcpip - acme ] while ` plastik_mf ` in @xcite . such configuration consists of : the insertion of two new components ( type and instance ) in lines 2 - 5 and line 6 ; the instantiation of components and connectors ( omitted with comment and @xmath3 in lines 7 - 8 ) ; the linking of components and connectors ( omitted with comment and @xmath3 in lines 7 - 8 ) ; and the specification of two dynamic reconfiguration situations , lines 12 - 19 and 21 - 28 . both dynamic reconfigurations use the same similar description . first , the conditional expressions are ` link.bandwidth = = ' low ' ` and ` link.bandwidth = = ' high ' ` . second , the @xmath1 are described as two unlinking ( lines 13 - 14 , 22 - 23 ) and one linking ( lines 15 - 18 , 24 - 27 ) statements . these @xmath1 specified the replacement of an instance of the component . .... system streamdecodersystem : tcpip_mf , plastikmf { component mpeg - decoder : tcpip_mpegdecoder = new tcpip_application extend with { property decoder - type = " mpeg " ; property algorithm = ... ; } component h263-decoder : tcpip_h263decoder = new tcpip_application extend with { ... } // other instances of component and connector ... // initial attachments ... // programmed dynamic reconfiguration to low bandwidth on ( link.bandwidth = = ' low ' ) do { detach mpeg-decoder.requiredservice to application2transport.source ; detach transport2application.sink to mpeg-decoder.datareceivedfromtransport ; attachments { h263-decoder.requiredservice to application2transport.source ; transport2application.sink to h263-decoder.datareceivedfromtransport ; } } // programmed dynamic reconfiguration to high bandwidth on ( link.bandwidth = = ' high ' ) do { detach h263-decoder.datato to application2transport.source ; detach transport2application.sink to h263-decoder.datareceivedfromtransport ; attachments { mpeg-decoder.requiredservice to application2transport.source ; transport2application.sink to mpeg-decoder.datareceivedfromtransport ; } } } .... [ [ scenario-2 - 1 ] ] scenario 2 + + + + + + + + + + the scenario 2 , an unforeseen insertion of a new component type , is describe with an acme / plastik script in fig . [ code : scenario2-acme ] . for this scenario , the script is composed by : the insertion of a type and instance of a component ( lines 1 - 3 ) , the unlinking of the old component instance ( lines 5 - 6 ) and the linking of the new component ( lines 7 - 10 ) . .... component ipv6 : tcpip_ipv6 = new tcpip_internet extended with { ... } detach ipv4.requiredservice to internet2link.source ; detach link2internet.sink to ipv4.datareceivedfromlink ; attachments { ipv6.requiredservice to internet2link.source ; link2internet.sink to ipv6.datareceivedfromlink ; } .... [ [ scenario-3 - 1 ] ] scenario 3 + + + + + + + + + + the unforeseen upgrade of a behaviour in scenario 3 can be described by the script illustrated in fig . [ code : scenario3-acme ] . this script defines a new version of a component with an improvement on its behaviour . the behaviour in acme / plastik is expressed as a property and it can be also described in other languages , such as external formal languages , see lines 3 - 5 in fig . [ code : scenario3-acme ] . .... component type tcpip_transport extends tcpip_component with { ... property behaviour = { \\ description of behaviour in other languages , e.g. csp } ... } .... [ [ summary - of - acmeplastik ] ] summary of acme / plastik + + + + + + + + + + + + + + + + + + + + + + + the dynamic reconfiguration issues for acme can be summarized as follows : 1 . dynamic reconfiguration in terms of structure is supported by acme / plastik . in terms of behaviour , dynamic reconfiguration is described using any external language embedded inside the property element . usually , this is used for foreseen reconfiguration . this approach is limited in the sense that components and connectors usually needs the dynamic reconfiguration of their behaviour but as acme / plastik does not provides elements for behaviour specification , the architect has to rely on an external language . e.g. if the ` tcpip_transport ` component needs to decide if it has a buffer or not , and if it uses a error detecting or not , and so on . these situations have to be described using an external language ( in general formal languages such as csp ) . 2 . acme / plastik does not provide a mechanism to control the intermediate states of a reconfiguration . e.g. the implementation of the scenario 3 , the architect can not define the approach used to update the instances . 3 . for unforeseen reconfiguration acme / plastik relies on external scripts . this approach requires an external interpreted associated to the adl to interpret the external script in order to reconfigure the system . [ [ purpose - of - the - adl-2 ] ] purpose of the adl + + + + + + + + + + + + + + + + + + initially , the purpose of c2 was to describe the architecture of a software based on a graphical user interface ( gui ) @xcite . therefore , the adl is based on hierarchically concurrent components . c2 also allows the use of messages to notify the architectural elements . c2 is subdivided in 2 languages : c2 idl ( interface description language ) for describing the components types , and c2 adl to specify the configurations . for components it is possible to specify top and bottom ports , to declare internal methods , and to specify the behaviour of its ports . the implementation of the internal methods is done by the developer in the source code generated by a toolset . with c2 adl it is possible to define instances of components and connectors , as well as links between them . [ [ dynamic - reconfiguration - support-2 ] ] dynamic reconfiguration support + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the architectural + modification language ( aml ) was created to extend c2 for supporting dynamic reconfiguration @xcite . aml is a declarative language that uses the structure view of an existing configuration and messages to notify this configuration about a dynamic reconfiguration . this language provides statements to build the scripts that specify dynamic reconfiguration . however , in these scripts it is not possible to determine the moment to apply a dynamic reconfiguration . because that , human intervention is needed via a toolset to trigger dynamic reconfiguration . [ [ initial - architecture-2 ] ] initial architecture + + + + + + + + + + + + + + + + + + + + the tcp / ip stack system is described in c2 idl in fig . [ code : tcpip - c2idl ] and c2 adl in fig . [ code : tcpip - c2adl ] . .... ... component tcpip_transport is interface // define the component ports top_domain is // ports to link high level connectors out ... in receivedata(package : applicationpackage ) ; bottom_domain is // ports to link low level connectors out sendtointernt(package : transportpackage ) ; in ... methods // define interfaces of internal behaviours function pack(data : applicationpackage ) : transportpackage ) ; ... behaviour // describe the external behaviour ... received_messages receivedata ; invoke_methods pack ; always_genarate sendtointernet ; ... end tcpip_transport ... .... .... architecture decoderstream is component_intances { mpeg instantiates tcpip_mpegdecoder ; h263 instantiates tcpip_h263decoder ; transport instantiates tcpip_transport ; internet instantiates tcpip_internet ; link instantiates tcpip_link ; } connectors { application2transport ; transport2internet ; internet2link ; } topology { connector application2transport { top_ports { mpeg filter no_filtering ; } bottom_ports { transport filter no_filtering ; } } connector transport2internet { top_ports { transport filter no_filtering ; } bottom_ports { internet filter no_filtering ; } } connector internet2link { top_ports { internet filter no_filtering ; } bottom_ports { link filter no_filtering ; } } } end decoderstream ; .... [ [ scenario-1 - 2 ] ] scenario 1 + + + + + + + + + + for scenario 1 , for changing the component according to the low or high bandwidth state , it is necessary two c2sadl scripts . in fig . [ code : scenario1-c2sadl ] is described the script for dynamic reconfiguration when bandwidth is low . this script uses two statements , a ` unweld ` for unlinking the @xmath4 instance and ` weld ` for linking the @xmath5 instance . other aml s statements are ` addcomponent ` , ` addconnector ` , ` removecomponent ` , and ` removeconnector ` . all these statements are used with a defined configuration , e.g. line 2 and 3 in fig . [ code : scenario1-c2sadl ] with the configuration named as ` decoderstream ` . despite of the fact that aml statements are based on services of the archstudio tool suite , there are services that are not provided by the aml language , e.g. ` start ( ) ` and ` stop ( ) ` . .... \\ bandwidth is low decoderstream.unweld(mpeg , application2transport ) ; decoderstream.weld(h263 , application2transport ) ; .... [ [ scenario-2 - 2 ] ] scenario 2 + + + + + + + + + + the second scenario , the unforeseen insertion of component instance and type , is inferred as shown in fig . [ code : scenario2-c2sadl ] . inference because both work about aml @xcite stated that it is possible to make insertion and deletion of types . however , they only show examples how to build dynamic reconfiguration at the instance level . et al . _ @xcite mention types and subtypes but at design level . dynamic reconfiguration in fig . [ code : scenario2-c2sadl ] specifies : the insertion of a new type of component ( lines 2 - 5 ) , the creation a new instance of component ( line 9 ) , the unlinking of the instance of ipv4 ( lines 12 - 13 ) , and the linking of the instance of ipv6 ( lines 16 - 17 ) . the component type name is inferred by a toolset before performing dynamic reconfiguration . .... \\ creata a new type of component component tcpip_ipv6 is subtype all < = all tcpip_internet ( ... ) ... end tcpip_ipv6 // create a new instance of component // the name of the type is implicitly discovered decoderstream.addcomponent(tcpip_ipv61 ) ; // unlink the instance of component ipv4 decoderstream.unweld(trasport2internet , tcpip_ipv41 ) ; decoderstream.unweld(tcpip_ipv41 , internet2link ) ; // link the instance of component ipv6 decoderstream.weld(trasport2internet , tcpip_ipv61 ) ; decoderstream.weld(tcpip_ipv61 , internet2link ) ; .... [ [ scenario-3 - 2 ] ] scenario 3 + + + + + + + + + + the third scenario is an improvement of the behaviour that can be modeled by the ` behaviour ` statement via a sequence of invocations to internal methods . example of this change , see line 6 in fig . [ code : scenario3-c2sadl ] , the invocation is modified to verify the received and packed data before sending a transport package to the internet layer . if the change is an improvement of an internal method , it is not possible to describe with c2sadl . .... component tcpip_transport is ... behaviour // describe the external behaviour ... received_messages receivedata ; invoke_methods verifydata , pack ; always_genarate sendtointernet ; ... end tcpip_transport ... .... [ [ summary - of - c2sadl ] ] summary of c2sadl + + + + + + + + + + + + + + + + + after the c2sadl implementation of the three scenarios , we identified these issues : 1 . archstudio tool provides some services for supporting dynamic reconfiguration , e.g. to start and to stop architectural elements . however , these services are not available in aml . thus , the dynamic reconfiguration specification in aml is limited . 2 . foreseen dynamic reconfiguration needs human intervention because c2sadl do not provide statements for the system apply dynamic reconfiguration . 3 . it does not support the specification of dynamic reconfiguration inside of the architectural elements . because that , the foreseen dynamic reconfiguration is implemented only at the implementation level ; 4 . it does not provide a mechanism for controlling and monitoring the intermediate states of a dynamic reconfiguration ; 5 . it also does not support the assessment of dynamic reconfiguration . [ [ purpose - of - the - adl-3 ] ] purpose of the adl + + + + + + + + + + + + + + + + + + dynamic wright has focus on the structure and behaviour of an architecture @xcite . dynamic wright supports the description of the architectural elements types structures and behaviours , and initial configuration . structure , initial configuration , instances , and links are described in a declarative form , while behaviours are specified in a graph - based and a variant of the communicating sequential process ( csp ) form . [ [ dynamic - reconfiguration - support-3 ] ] dynamic reconfiguration support + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + dynamic wright only supports foreseen dynamic reconfiguration . it is built as a special behaviour of the initial configuration . it has the same base of a `` nominally '' behaviour of architectural elements . the extension described in @xcite proposes additional statements and control events . the statements are ` new ` , ` remove ` , ` attach ` , and ` detach ` . the control events are used to define the specific moment to perform a dynamic reconfiguration . as dynamic wright does not support unforeseen reconfiguration , the scenarios 2 and 3 can not be implemented . [ [ initial - architecture-3 ] ] initial architecture + + + + + + + + + + + + + + + + + + + + the initial configuration of the tcp / ip stack system is illustrated in fig . [ code : scenario1-wright ] . the component and connector types are specified in the ` style ... endstyle ` ( lines 1 - 16 ) statements . the implementation of the ` tcpip_``mpeg - decoder ` component is described with two ports ( lines 3 - 4 ) and a behaviour ( lines 6 - 7 ) . the configuration is described with the ` configuror ... endconfiguror ` statement . the initial configuration instances and connections are specified at lines 18 - 27 . [ [ scenario-1 - 3 ] ] scenario 1 + + + + + + + + + + fig . [ code : scenario1-wright ] also illustrates the specification of scenario 1 . it uses the ` configuror [ initial configuration ] where [ behaviours ] ` statement . _ behaviours _ are special named behaviours of the configuration for specifying dynamic reconfiguration . in this case , there are two specification : lines 29 - 36 for low bandwidth and lines 37 - 44 for high bandwidth . the following steps were used to define the first and the second dynamic reconfiguration situations : 1 . to define a name ( lines 29 and 37 ) . 2 . to specify the control events that define the moment for performing the operations of dynamic reconfiguration ( lines 30 and 38 ) . 3 . to determine the type of configuration to use ( lines 31 and 39 ) . dynamic wright use the term ` style ` for this purpose . 4 . to specify the operations to perform dynamic reconfiguration ( lines 32 - 35 and 40 - 43 ) . 5 . to use the statements of the external choice ( @xmath6 ) and successfully terminate ( @xmath7 ) , lines 36 and 44 . .... style tcpip - style component tcpip_mpeg - decoder port datareceivedfromtransport = ... port requiredservice = ... [ describe the port behaviour using a variant of csp ] computation = ... [ describe the component behaviour using a variant of csp ] ... connector conn2layers role source = ... role sink = ... glue = ... [ describe the connector behaviour using a variant of csp ] constraints ... endstyle configuror decoderstream style tcpip - style new.mpeg : tcpip_mpeg - decoder $ \rightarrow$ new.h263 : tcpip_h263-decoder ... [ other component instances ] $ \rightarrow$ new.app2trans : conn2layers $ \rightarrow$ new.trans2net : conn2layers $ \rightarrow$ new.net2link : conn2layers $ \rightarrow$ attach.mpeg.requiredservice.to.app2trans.source $ \rightarrow$ attach.app2trans.sink.to.transport.service ... [ other attachments ] where waitforbandwidthlow = ( link.control.bandwidthdown $ \rightarrow$ mpeg.control.off $ \rightarrow$ h263.control.on $ \rightarrow$ style tcpip - style detach.mpeg.requiredservice.to.app2trans.source $ \rightarrow$ attach.h263.requiredservice.to.app2trans.source $ \rightarrow$ detach.trans2app.sink.to.mpeg.datareceivedfromtransport $ \rightarrow$ attach.trans2app.sink.to.h263.datareceivedfromtransport ) $ \box$ $ \s$ waitforbandwidthhigh = ( link.control.bandwidthup $ \rightarrow$ h263.control.off $ \rightarrow$ - > mpeg.control.on $ \rightarrow$ style tcpip - style detach.h263.requiredservice.to.app2trans.source $ \rightarrow$ attach.mpeg.requiredservice.to.app2trans.source $ \rightarrow$ detach.trans2app.sink.to.h263.datareceivedfromtransport $ \rightarrow$ attach.trans2app.sink.to.mpeg.datareceivedfromtransport ) $ \box$ $ \s$ endconfiguror .... [ [ summary - of - dynamic - wright ] ] summary of dynamic wright + + + + + + + + + + + + + + + + + + + + + + + + + the following issues are identified after the implementation of the scenario 1 : 1 . it does not provide operations to define dynamic reconfiguration for the type level of architectural elements ; 2 . this adl provides a refined mechanism of control events to determine the moment to perform a dynamic reconfiguration . although , it does not provide a mechanism to control and monitor the intermediate states . e.g. if the application of the scenario 1 is a banking system , the state must be copied from the instance of the replaced component to the new component . dynamic wright does not provide operations for this purpose . 3 . despite of providing the specification of behaviour for components and connectors , it is not possible to define dynamic reconfiguration for this architectural elements ; 4 . similarly to the other adls , dynamic wright does not provide a mechanism for the assessment of dynamic reconfiguration . the summary of the four adls support for dynamic reconfigurations is shown in tab . [ tab : reconfiguration ] . @xmath0-adl and acme / plastik are the only adls that support both foreseen and unforeseen dynamic reconfigurations . as c2sadl with the aml language do not provide a way to specify an internal initiation of dynamic reconfiguration , foreseen dynamic reconfiguration can not be automatically triggered . dynamic wright supports only foreseen reconfiguration by defining the special behaviour of a configuration . .foreseen and unforeseen dynamic reconfiguration support in adls . [ cols="<,<,<",options="header " , ] none of the four adls has a mechanism for assessing the system . @xmath0-adl can rely on @xmath0-aal for defining and analyzing the architectural constraints and non - functional properties . while acme / plastik has armani . even though it has no formal semantics , and therefore , it is not liable to rigorous analysis @xcite . c2sadl provides type verification for events communication . dynamic wright provides a refined mechanism of control events to determine the moment to perform a dynamic reconfiguration . although , the assessment of dynamic reconfiguration is not provided on four adls . in this paper we analysed the support of adls for handling dynamic reconfigurations . we started with a motivation example and some reconfiguration scenarios and we described the example and the scenarios using four well - known adls . we used the example and the specifications to discuss the following issues : ( i ) how each adl addresses the issue of consistently reconfiguring both instances and types ; ( ii ) how each adl takes into account the behaviour of the architectural elements during reconfiguration ; and ( iii ) how each adl supports the assessing of dynamic reconfiguration . in comparison to related works , we analysed some important issues for the dynamic reconfiguration support at the architectural level : foreseen and unforeseen changes ; instance and type level modifications ; structure and behaviour for all architectural elements ; definition of nominally and dynamic reconfiguration behaviour for all architectural elements ; and , the analysis of dynamic reconfiguration . we can conclude that some issues still remain open as no adl provides support for all of them together : how to apply the changes made in type level to their respective instances ? how to control the set of intermediate states of a software system during a dynamic reconfiguration ? how to assess dynamic reconfiguration ? special thanks to collge doctoral international ( cdi)/uni - versit europenne de bretagne ( ueb ) for financial support this work . r. allen , r. douence , and d. garlan . specifying and analyzing dynamic software architectures . in _ proceedings of the 1998 conference on fundamental approaches to software engineering ( fase98 ) _ , lisbon , portugal , march 1998 . an expanded version of a the paper `` specifying dynamism in software architectures , '' which appeared in the proceedings of the workshop on foundations of component - based software engineering , september 1997 . t. v. batista , a. t. a. gomes , g. coulson , c. chavez , and a. f. garcia . on the interplay of aspects and dynamic reconfiguration in a specification - to - deployment environment . in _ proceedings of the 2nd european conference on software architecture _ , ecsa 08 , pages 314317 , berlin , heidelberg , 2008 . springer - verlag . k. h. bennett and v. t. rajlich . software maintenance and evolution : a roadmap . in _ proceedings of the conference on the future of software engineering _ , icse 00 , pages 7387 , new york , ny , usa , 2000 . acm . d. garlan , r. monroe , and d. wile . acme : an architecture description interchange language . in _ proceedings of the 1997 conference of the centre for advanced studies on collaborative research_. ibm press , 1997 . a. gomes , t. v. batista , a. joolia , and g. coulson . . in r. de lemos , c. gacek , and a. romanovsky , editors , _ architecting dependable systems iv _ , volume 4615 of _ lecture notes in computer science _ , pages 237261 . springer berlin / heidelberg , berlin , heidelberg , 2007 . j. magee and j. kramer . self organising system structuring . in _ joint proceedings of the second international software architecture workshop ( isaw ) and international workshop on multiple perspectives in software development ( viewpoints 96 ) on sigsoft 96 workshops _ , isaw 96 , pages 3538 , new york , ny , usa , 1996 . acm press . n. medvidovic , p. oreizy , j. e. robbins , and r. n. taylor . using object - oriented typing to support architectural design in the c2 style . in _ proceedings of the 4th acm sigsoft symposium on foundations of software engineering _ , sigsoft 96 , pages 2432 , new york , ny , usa , 1996 . acm . p. oreizy , m. m. gorlick , r. n. taylor , d. heimbigner , g. johnson , n. medvidovic , a. quilici , d. s. rosenblum , and a. l. wolf . an architecture - based approach to self - adaptive software . , 14(3):5462 , may / june 1999 . p. oreizy , n. medvidovic , and r. n. taylor . architecture - based runtime software evolution . in _ proceedings of the 20th international conference on software engineering _ , icse 98 , pages 177186 , washington , dc , usa , 1998 . ieee computer society . p. oreizy and r. n. taylor . on the role of software architectures in runtime system reconfiguration . in _ configurable distributed systems , 1998 . fourth international conference on _ , pages 6170 , may 1998 . m. pinto , l. fuentes , and j. m. troya . daop - adl : an architecture description language for dynamic component and aspect - based development . in _ proceedings of the 2nd international conference on generative programming and component engineering _ , gpce 03 , pages 118137 , new york , ny , usa , 2003 . springer - verlag new york , inc . r. n. taylor , n. medvidovic , k. m. anderson , j. whitehead , e. james , j. e. robbins , k. a. nies , p. oreizy , and d. l. dubrow . a component- and message - based architectural style for gui software . , 22(6):390406 , june 1996 . j. vera , l. perrochon , and d. c. luckham . event - based execution architectures for dynamic software systems . in _ proceedings of the tc2 first working ifip conference on software architecture ( wicsa ) _ , wicsa , pages 303318 , deventer , the netherlands , the netherlands , 1999 . kluwer , b.v . p. waewsawangwong . a constraint architectural description approach to self - organising component - based software systems . in _ proceedings of the 26th international conference on software engineering _ , icse 04 , pages 8183 , washington , dc , usa , 2004 . ieee computer society .

we are motivated by studying equivariant pullbacks of @xmath6-algebras exemplified , for instance , by a join construction for compact quantum groups . insisting on equivarince to be given by free actions " leads to galois extensions of @xmath6-algebras by hopf algebras , and generalising pullbacks to multi - pullbacks " brings up distributive lattices as a fundamental language . as a by - product of our considerations , we obtain an equivalence between the distributive lattices generated by @xmath3 ideals intersecting to zero and flabby sheaves of algebras over a certain @xmath7-projective space . comodule algebras provide a natural noncommutative geometry generalisation of spaces equipped with group actions . less evidently , principal extensions @xcite appear to be a proper analogue of principal bundles in this context ( see section [ prelimi ] for precise definitions ) . principal extensions can be considered as functors from the category of finite - dimensional corepresentations of the hopf algebra ( replacing the structure group ) to the category of finitely generated projective modules over the coaction - invariant subalgebra ( playing the role of the base space ) . the aim of this article is to establish a viable concept of locality of comodule algebras and analyse its relationship with principality . the notion of locality we use herein results from decomposing algebras into `` pieces '' , meaning expressing them as multiple fibre products ( called multirestricted direct sums in @xcite ) . if @xmath8 is a compact hausdorff space and @xmath9 form a finite closed covering , then @xmath10 can be expressed as such a multiple fibre product of its quotient @xmath6-algebras @xmath11 . this leads to a @xmath6-algebraic notion of a covering of a quantum space " given by a finite family of algebra surjections @xmath12 with @xmath13 ( see @xcite , cf . @xcite ) . recall that not all properties of group actions are local in nature : there is a natural example of a locally proper action of @xmath14 on @xmath15 that is not proper ( see @xcite , cf . * example 1.14 ) for more details ) . on the other hand , a group action is free if and only if it is locally free . therefore , since for compact groups all actions are proper , the principal ( i.e. , free and proper ) actions of compact groups are local in nature . our main result ( theorem [ main ] ) is a noncommutative analogue of this statement : a comodule algebra @xmath0 which is covered by `` pieces '' @xmath5 is principal if and only if so are the pieces . in particular , a smash product of an @xmath1-module algebra @xmath16 with the hopf algebra @xmath1 ( with bijective antipode ) is principal , so that gluing together smash products is a way of constructing principal comodule algebras . however , it was pointed out in @xcite that the aforementioned coverings can show a certain incompleteness when going beyond the @xmath6-setting . this is related to the fact that the lattice of ideals generated by the @xmath17 s is in general not distributive . ( this problem does not arise for @xmath6-algebras . ) hence we analyse a stronger notion of covering that includes this nontrivial assumption as part of the definition ( see definition [ covdef ] ) . if all @xmath5 s are smash products , we arrive at a concept of piecewise trivial comodule algebras . they appear to be a good noncommutative replacement of locally trivial compact principal bundles . our motivation for going beyond @xmath6-algebras comes from the way we consider compact principal bundles ( the hausdorff property assumed ) . we aim to use at the same time algebraic techniques of hopf - galois theory and analytic tools coming with @xmath6-algebras . to this end , we look at the total space of such a bundle in terms of the algebra of functions continuous along the base and polynomial along the fibres @xcite . then the base space algebra is always a @xmath6-algebra , but , unless the group is finite , the total space algebra is not @xmath6 . the data of a covering by @xmath3 pieces can be equivalently encoded into a flabby sheaf of algebras over @xmath18 . this is the 2-element field @xmath7-projective space whose topology subbasis is its affine covering . it is a finite space encoding the combinatorics " of an @xmath3-covering , and is non - hausdorff unless @xmath19 . the lattice of all open subsets of @xmath18 turns out to be isomorphic to a certain lattice of antichains in the set of all subsets of an @xmath3-element set . combining this with the chinese remainder theorem for distributive lattices of ideals in an arbitrary ring , we prove that distributive lattices generated by @xmath3 ideals are equivalent to flabby sheaves over @xmath18 . in particular , consider a compact hausdorff space @xmath8 with a covering by @xmath3 closed subsets @xmath9 . then we have the soft sheaf of continuous functions with @xmath3 distinguished @xmath6-algebras @xmath20 . however , the soft sheaf of complex - valued continuous functions on @xmath8 is not a sheaf of @xmath6-algebras . therefore , there seems to be no evident way to use soft sheaves in the noncommutative setting . to overcome this difficulty , we declare the closed sets open and consider @xmath8 with the new topology generated by these open sets . this leads us to flabby sheaves over @xmath18 . this way we obtain a covering version of gelfand s theorem : there is an equivalence between the category of compact hausdorff spaces with @xmath3-coverings by closed subsets and the category of flabby sheaves of unital commutative @xmath6-algebras over @xmath18 . in the noncommutative setting , this sheaf - theoretic reformulation of coverings allows us to view piecewise trivial comodule algebras as introduced in this paper as what is called locally trivial quantum principal bundles " in @xcite . throughout , we work over a field @xmath21 and all considered algebras , coalgebras etc . are over @xmath21 . an unadorned @xmath22 denotes the tensor product of @xmath21-vector spaces . for coproduct and coactions we adopt the sweedler notation with the summation sign suppressed : @xmath23 , @xmath24 . we recall here elementary facts concerning pullback diagrams that will be used in what follows . to focus attention , we consider the category of vector spaces , which suffices for our applications . let @xmath25 and @xmath26 be linear maps of vector spaces . the pullback ( fibre product ) @xmath27 of @xmath28 and @xmath29 is defined by a universal property , and turns out to be isomorphic to @xmath30 as a consequence of this description , we obtain : @xmath31 this is because for any linear map @xmath32 of vector spaces one has @xmath33 . next , let us consider the following commutative diagram of linear maps @xmath34_\eta \ar[ddll]_{\phi_1 } \ar[ddrr]^{\phi_2 } & & \\ & & \overset{\phantom{\pi}}{v_1}\!\underset{\pi_1,\pi_2}{\times } \!v_2 \ar[dll]^{p_1 } \ar[drr]_{p_2 } & & \\ v_1 \ar[drr]_{\pi_1 } & & & & v_2 \ar[dll]^{\pi_2}\ , \\ & & v_{12 } & & } \ ] ] and show : [ fib]assume that the @xmath35 s and @xmath36 s in the diagram ( [ hunz ] ) are surjective . then @xmath37 is surjective if and only if @xmath38 . assume first that @xmath37 is surjective , and @xmath39 . then both @xmath40 and @xmath41 belong to @xmath42 , and there exist @xmath43 and @xmath44 such that @xmath45 and @xmath46 . clearly , @xmath47 . therefore , as @xmath48 , @xmath49 , and @xmath50 we conclude that @xmath51 . conversely , if @xmath52 is in the fibre product , then @xmath53 , so that @xmath54 for some @xmath55 and @xmath56 . hence , for @xmath57 , we have @xmath58 . we need a method yielding a presentation of elements in finitely generated distributive lattices . we denote by @xmath59 the set of antichains in @xmath60 . for any @xmath61 we define @xmath62 it is easy to see that @xmath63 . if @xmath64 is a lattice generated by @xmath65 , then we define map @xmath66 , @xmath67 conversely , we define map @xmath68 , @xmath69 by the definition of @xmath70 it follows that for any @xmath71 , @xmath72 . on the other hand , as @xmath73 is a distributive lattice generated by the @xmath74 s , there exits @xmath61 such that @xmath75 . note that for any @xmath76 , @xmath77 , and hence there exists @xmath78 such that @xmath79 . it follows that @xmath80 and therefore @xmath81 . thus , we have proven that for any finitely generated distributive lattice @xmath73 @xmath82 let us define two binary operations @xmath83 : @xmath84 it is immediate by the distributivity of @xmath73 that for all @xmath85 , @xmath86 let @xmath87 be a hopf algebra with bijective antipode . a right @xmath1-comodule algebra @xmath0 is a unital associative algebra equipped with an @xmath1-coaction @xmath88 that is an algebra map . for a comodule algebra @xmath0 , we call @xmath89 the subalgebra of coaction - invariant elements in @xmath0 . the assumed existence of the inverse of the antipode allows us to define a left coaction @xmath90 by the formula @xmath91 . this makes @xmath0 a left @xmath1-comodule and a left @xmath92-comodule algebra . let @xmath0 be a right comodule algebra over a hopf algebra @xmath1 with bijective antipode , and let @xmath93 be the coaction - invariant subalgebra . the comodule algebra @xmath0 is called _ principal _ if the following conditions are satisfied : 1 . the coaction of @xmath1 is galois , that is , the map @xmath94 ( called the canonical map ) is bijective , 2 . the comodule algebra @xmath0 is right @xmath1-equivariantly projective as a left @xmath16-module , i.e. , there exists a right @xmath1-colinear and left @xmath16-linear splitting of the multiplication map @xmath95 . this splitting can always be chosen to be unital @xcite . also , in this setting , one can show that the @xmath1-equivariant projectivity of @xmath0 over @xmath16 is equivalent to the faithful flatness of @xmath0 as a @xmath16-module @xcite . if @xmath0 is a principal comodule algebra , then the extension of algebras @xmath96 is a special case of principal extensions defined in @xcite . a cleft hopf - galois extension ( e.g. , a smash product @xmath97 of an @xmath1-module algebra @xmath16 by @xmath1 ) is always principal . indeed , by @xcite , cleft hopf - galois extensions always enjoy the normal basis property , and the latter can be viewed as equivariant freeness , a special case of equivariant projectivity . for more details and an introduction to hopf - galois theory , see , e.g. , @xcite . the inverse of the canonical map defines a monomorphism @xmath98 , @xmath99 called the translation map . it turns out that lifts of this map to @xmath100 that are both right and left @xmath1-colinear yield an equivalent approach to principality @xcite : [ strongdef ] let @xmath1 be a hopf algebra with bijective antipode . then a strong connection ( cf . @xcite ) on a right @xmath1-comodule algebra @xmath0 is a unital linear map @xmath101 satisfying @xmath102 here @xmath103 is the lift of @xmath104 to @xmath100 . the last property of the strong connection ( splitting of @xmath105 ) gives rise to the commutative diagram @xmath106 ^ -{\ell } \ar[d]_{1\otimes{\operatorname{\rm i d } } } & p \otimes p \ar[dl]_{{\widetilde{\mathsf{can } } } } \ar[d]^{\text{canonical surjection}}\\ p \otimes h & \;\;p \otimes_b p \,.\ar[l]^{\,{\mathsf{can}}}}\ ] ] using the sweedler - type notation @xmath107 ( summation suppressed ) , we can write the bicolinearity and splitting property of a strong connection as follows : @xmath108 applying @xmath109 to the last equation yields a very useful formula : @xmath110 one can prove that an @xmath1-comodule algebra @xmath0 is principal if and only if it admits a strong connection @xcite . given a strong connection @xmath111 , one can show that the formula @xmath112 defines the inverse of the canonical map @xmath104 , so that the coaction of @xmath1 is galois . next , one can also show that @xmath113 is a splitting whose existence proves the equivariant projectivity . much as above , one argues that the formula @xmath114 provides a left @xmath1-colinear and right @xmath16-linear splitting of the multiplication map@xmath115 . the functions continuous along the base and polynomial along the fibre on a principal bundle with compact structure group have an analogue in the noncommutative setting : let @xmath116 be the @xmath6-algebra of a compact quantum group in the sense of woronowicz @xcite and @xmath1 its dense hopf @xmath117-subalgebra spanned by the matrix coefficients of the irreducible unitary corepresentations . let @xmath118 be a unital @xmath6-algebra and let @xmath119 be a @xmath6-algebraic right coaction of @xmath116 on @xmath118 . ( see ( * ? ? ? * definition 0.2 ) for a general definition and ( * ? ? ? * definition 1 ) for the special case of compact quantum groups . ) then the subalgebra @xmath120 of elements for which the coaction lands in @xmath121 ( algebraic tensor product ) , @xmath122 is an @xmath1-comodule algebra . it follows from results of @xcite and @xcite that @xmath0 is dense in @xmath118 . we call @xmath0 the comodule algebra associated to the @xmath6-algebra @xmath118 . it is straightforward to verify that the operation @xmath123 commutes with taking fibre products . note also that @xmath124 . to show the piecewise nature of principality , we begin by proving lemmas concerning quotients and fibre products of principal comodule algebras . [ fer ] let @xmath125 be a surjection of right @xmath1-comodule algebras ( bijective antipode assumed ) . if @xmath0 is principal , then : 1 . the induced map @xmath126 is surjective . 2 . there exists a unital @xmath1-colinear splitting of @xmath127 . it follows from the colinearity of @xmath127 that @xmath128 . to prove the converse inclusion , we take advantage of the left @xmath2-linear retraction of the inclusion @xmath129 that was used to prove ( * ? ? ? * theorem 2.5(3 ) ) : @xmath130 here @xmath111 is a strong connection on @xmath0 and @xmath131 is any unital linear functional on @xmath0 . if @xmath132 , then @xmath133 is a desired element of @xmath2 that is mapped by @xmath127 to @xmath134 . indeed , since @xmath135 , using the unitality of @xmath111 , @xmath127 and @xmath131 , we compute @xmath136 concerning the second assertion , one can readily verify that the formula @xmath137 defines a unital colinear splitting of @xmath127 . here @xmath138 is any @xmath21-linear unital splitting of @xmath139 , e.g. , @xmath140 , and @xmath111 is again a strong connection on @xmath0 . + [ strocopuba ] let @xmath0 be a fibre product in the category of right @xmath1-comodule algebras : @xmath141 \ar[rd ] & \\ p_1 \ar[dr]_{\pi^1_2 } & & p_2 \ar[dl]^{\pi^2_1}\\ & p_{12}\ , . & } $ } \ ] ] then , if @xmath142 and @xmath143 are principal and @xmath144 and @xmath145 are surjective , @xmath0 is a principal comodule algebra . given strong connections @xmath146 and @xmath147 on @xmath142 and @xmath143 , respectively , we want to construct a strong connection on @xmath0 . a first approximation for such a strong connection is as follows : @xmath148 here @xmath149 , and @xmath150 is a unital colinear splitting of @xmath145 , which exists by lemma [ fer](2 ) . the map @xmath151 is unital and bicolinear , but it does not split the lifted canonical map : @xmath152 now , let @xmath153 be the lifted canonical map on @xmath154 . applying the splitting of @xmath153 given by @xmath147 ( cf . ( [ ei0 ] ) ) to the right hand side of the above equation , gives a correction term for @xmath151 : @xmath155 this defines a bicolinear map into @xmath154 which annihilates 1 . considering @xmath151 as a map into @xmath156 , we can add these two maps . the map @xmath157 is still unital and bicolinear and splits the lifted canonical map on @xmath156 . remembering ( [ compa ] ) and ( [ elco ] ) , it is clear from the formula for @xmath158 that to make it taking values in @xmath159 we only need to add the term @xmath160 much as above , here @xmath161 and @xmath162 is a unital colinear splitting of @xmath163 , which exists by lemma [ fer](2 ) . the formula ( [ t ] ) defines a bicolinear map into @xmath164 which annihilates 1 . since the lifted canonical map on @xmath156 vanishes on @xmath164 , the sum @xmath165 splits the lifted canonical map , takes values in @xmath159 and is unital and bicolinear . thus it is as desired a strong connection on @xmath0 . + let us now consider a family @xmath166 , @xmath167 , of surjections of right @xmath1-comodule algebras with @xmath168 . denote @xmath169 . by lemma [ fib ] and formula [ kereta ] , for any @xmath170 there is a fibre - product diagram of right @xmath1-comodule algebras @xmath171+<3mm,3 mm > \ar[rd ] & \\ \save-<-3mm,0mm>*{p/(j_1 \cap \ldots \cap j_k ) } \restore \mbox{\huge \ } \save < 3mm,-14 mm > * { } \ar[dr]_(.15){\pi^1_2 } \restore & & p / j_{k+1 } \ar[dl]^(.35){\pi^2_1}\\ & p/((j_1\cap\ldots\cap j_k)+j_{k+1 } ) & .}$}\ ] ] assume that all the @xmath172 are principal . then lemma [ strocopuba ] implies by an obvious induction that @xmath173 is principal for all @xmath174 . in particular ( @xmath175 ) , @xmath0 is principal . on the other hand , if @xmath0 is principal then all the @xmath5 s are principal : if @xmath176 is a strong connection on @xmath0 , then @xmath177 is a strong connection on @xmath5 . thus we have proved the following : [ main ] let @xmath166 , @xmath167 , be surjections of right @xmath1-comodule algebras such that @xmath168 . then @xmath0 is principal if and only if all the @xmath5 s are principal . our next step is a statement about a relation between the ideals of a principal comodule algebra @xmath0 that are also subcomodules , and ideals in the subalgebra @xmath16 of coaction - invariant elements . both sets are obviously lattices with respect to the operations @xmath178 and @xmath179 . [ diens ] let @xmath0 be a principal right @xmath1-comodule algebra and @xmath93 the coaction - invariant subalgebra . denote by @xmath180 the lattice of all ideals in @xmath16 and by @xmath181 the lattice of all ideals in @xmath0 which are simultaneously subcomodules . then the map @xmath182 is a monomorphism of lattices . the only non - trivial step in proving that @xmath183 is a homomorphism of lattices is establishing the inclusion @xmath184 . to this end , we proceed along the lines of the proof of lemma [ fer](1 ) . since @xmath185 is a comodule and an ideal , from the formula ( [ ei1 ] ) we obtain : @xmath186 now , let @xmath187 , @xmath188 , @xmath189 . then ( [ stropro ] ) implies that @xmath190 applying any unital linear functional @xmath191 to the second tensor component implies @xmath192 . finally , since @xmath193 is a splitting of the multiplication map , it follows from ( [ stropro ] ) that @xmath194 . this , in turn , proves the injectivity of @xmath195 . + much as the formula ( [ ei1 ] ) implies ( [ stropro ] ) , the formula ( [ ei2 ] ) implies @xmath196 therefore , since @xmath197 is a splitting of the right multiplication map , @xmath198 . combining this with the above discussed left - sided version yields @xmath199 note that the homomorphism @xmath195 is not surjective in general . a counterexample is given by a smash product ( trivial principal comodule algebra ) of the laurent polynomials @xmath200 $ ] with the hopf algebra @xmath201 $ ] of laurent polynomials ( @xmath202 ) . the action is defined by @xmath203 , @xmath204 . viewing @xmath205 and @xmath206 as generators of @xmath0 , it is clear that @xmath207 . it is straightforward to verify that , if @xmath208 is the two - sided ideal in @xmath16 generated by @xmath209 , then the right ideal @xmath210 is not a two - sided ideal of @xmath0 . hence the map @xmath195 can not be surjective by ( [ linv ] ) . in @xcite , families @xmath12 of algebra homomorphisms as in theorem [ main ] were called coverings . however , it was explained therein that such coverings are well - behaved when the kernels @xmath17 generate a distributive lattice of ideals ( with respect to @xmath178 and @xmath179 as lattice operations ) . hence we adopt in the present paper the following terminology : [ covdef ] a finite family @xmath211 , @xmath212 , of surjective algebra homomorphisms is called a weak covering if @xmath213 . a weak covering is called a _ covering _ if the lattice of ideals generated by the @xmath214 s is distributive . the above definition can obviously be extended to the case when the @xmath36 s are algebra and @xmath1-comodule morphisms . then the @xmath214 s are ideals and @xmath1-subcomodules . the next claim is concerned with the distributivity condition from definition [ covdef ] for coverings of principal comodule algebras . it follows from proposition [ diens ] and lemma [ fer](1 ) . [ auchmain ] let @xmath12 , @xmath212 , be surjective homomorphisms of right @xmath1-comodule algebras . assume that @xmath0 is principal . then @xmath215 is a covering of @xmath0 if and only if @xmath216 is a covering of @xmath2 . the above corollary is particularly helpful when @xmath2 is a @xmath6-algebra because lattices of closed ideals in a @xmath6-algebra are always distributive ( due to the property @xmath217 ) . we are now ready to propose a noncommutative - geometric replacement of the concept of local triviality of principal bundles . since these are the closed rather than open subsets of a compact hausdorff space that admit a natural translation into the language of @xmath6-algebras , we use finite closed rather than open coverings to trivialize bundles . as is explained in ( * ? ? ? * example 1.24 ) , there is a difference between these two approaches . we reserve the term locally trivial " for bundles trivializable over an open cover , and call bundles trivializable over a finite closed cover piecewise trivial " . it is the latter ( slightly more general ) property that we generalize to the noncommutative setting . [ pipri ] an @xmath1-comodule algebra @xmath0 is called _ piecewise principal ( trivial ) _ if there exist comodule algebra surjections @xmath12 , @xmath212 , such that : 1 . the restrictions @xmath218 form a covering . the @xmath5 s are principal . ( the @xmath5 s are isomorphic as @xmath1-comodule algebras to a smashed product @xmath219 . ) while not every compact principal bundle is piecewise ( or locally ) trivial ( ( * ? ? * example 1.22 ) ) , every piecewise principal compact @xmath220-space ( i.e. , covered by finitely many compact principal @xmath220-bundles ) is clearly a compact principal @xmath220-bundle . the second statement becomes non - trivial when we replace compact @xmath220-spaces by comodule algebras . however , it is an immediate consequence of theorem [ main ] and corollary [ auchmain ] : let @xmath1 be a hopf algebra with bijective antipode and @xmath0 be an @xmath1-comodule algebra that is piecewise principal with respect to @xmath221 . then @xmath0 is principal and @xmath221 is a covering of @xmath0 . finally , let us consider the relationship between piecewise triviality and a similar concept referred to as local triviality " in @xcite . therein , sheaves @xmath222 of comodule algebras were viewed as quantum analogues of principal bundles . they were called locally trivial provided that the space @xmath8 on which @xmath222 is defined admits an open covering @xmath223 such that all @xmath224 s are smash products . if we assume such a sheaf to be flabby ( that is , for all open subsets of @xmath225 , @xmath226 , the restriction maps @xmath227 are surjective ) , then we can use theorem [ main ] to deduce the principality of all @xmath228 s : [ niema ] let @xmath1 be a hopf algebra with bijective antipode and @xmath222 be a flabby sheaf of @xmath1-comodule algebras over a topological space @xmath8 . if @xmath223 is a finite open covering such that all @xmath224 s are principal , then @xmath228 is principal for any open subset @xmath229 . in this section , we focus entirely on a flabby - sheaf interpretation of distributive lattices of ideals defining coverings of algebras ( see definition [ covdef ] ) . we will explain that for a flabby sheaf in corollary [ niema ] , the underlying topological space plays only a secondary role and can be replaced by a certain space that is universal for all @xmath3-element coverings . this space is the 2-element field @xmath230-projective space @xmath231 whose topology subbasis is its affine covering , i.e. , it is the topology generated by the subsets @xmath232 consider now an arbitrary space @xmath8 with a finite covering @xmath233 . define on @xmath8 the topology generated by the @xmath234 s ( considered as open sets ) and pass to the quotient by the equivalence relation @xmath235 obviously , @xmath236 depends on the specific features of the covering @xmath223 . however , for a fixed @xmath3 , it can always be embedded into @xmath237 : [ embedding ] let @xmath238 be any set equipped with the topology generated by the @xmath234 s . let @xmath239 be the quotient map defined by the equivalence relation ( [ paul ] ) . then @xmath240 is an embedding of topological spaces . it is immediate that @xmath241 is well defined and injective . next , since @xmath242 is open for each @xmath243 , all @xmath244 s are open in the quotient topology on @xmath245 . now the continuity of @xmath241 follows from the fact that @xmath246 s form a subbasis of the topology of @xmath237 . the key step is to show that images of open sets in @xmath245 are open in @xmath247 . first note that by the definition of the relation ( [ paul ] ) , @xmath248 therefore , as preimages and images preserve unions and any open set in @xmath8 is the union of intersections of @xmath234 s , @xmath249 is an open map . on the other hand , by the surjectivity of @xmath249 , we have @xmath250 for any subset @xmath251 . hence it follows that a set in @xmath245 is open if and only if it is an image under @xmath249 of an open set in @xmath8 . finally , by the definition of @xmath241 , @xmath252 and the claim follows from the distributivity of @xmath179 with respect to @xmath253 . + note that the map @xmath254 is a homeomorphism precisely when the @xmath234 s are in a generic position , that is when all intersections @xmath255 such that @xmath256 are non - empty . thus we have shown that if we consider @xmath8 and the @xmath234 s as in corollary [ niema ] , then the composition @xmath257 is continuous . hence we can produce flabby sheaves over @xmath237 by taking direct images of flabby sheaves over @xmath8 . they will have the same sections globally and on the covering sets . in this sense , they carry an essential part of the data encoded in the original sheaf . let @xmath258 . denote by @xmath259 $ ] the class of @xmath260 in @xmath261 . define a family of closed sets @xmath262\in { \mathbb p}^{n-1}({\mathbb c})\;|\ ; @xmath263 s cover @xmath8 , and moreover , for all @xmath264 such that @xmath265 , @xmath266 , the element @xmath267\in x , \text{\ where\ } x_i=\left\{{{\!\!\!1 \text{\ if\ } i\in\lambda}\atop { 0 \text{\ otherwise}}}\right.\ ] ] belongs to @xmath268 . it follows that the @xmath263 s are in generic position and the map @xmath269 ( proposition [ embedding ] ) is a homeomorphism ( if @xmath270 is considered with finite topology as in proposition [ embedding ] ) . let @xmath271\in{\mathbb p}^{n-1}({\mathbb c})\;|\ ; x_i\neq 0\;\},\ \ \ i=1,\ldots , n\ ] ] be the standard ( open ) affine cover of @xmath261 and @xmath272 be the homeomorphism @xmath273 \mapsto ( x_1/x_i,\ldots,\widehat{x_i / x_i},\ldots , x_{n}/x_i).\ ] ] observe that for all @xmath274 , @xmath275 and latexmath:[\[\psi_i(x_i)=\{(y_1,\ldots , y_{n-1})\in{\mathbb c}^{n-1}\;|\;\forall_j\ ; sets . our next aim is to demonstrate that flabby sheaves over @xmath237 are just a reformulation of the notion of covering introduced in definition [ covdef ] . it turns out that the distributivity condition discussed in the previous section is the key property needed to reconcile the results from @xcite with those from @xcite . in particular , our results imply the principality of pflaum s noncommutative instanton bundle ( see the last section for details ) . let us consider the lattice @xmath277 of open subsets of @xmath18 generated by the @xmath246 s ( [ ai ] ) . [ upperin ] if @xmath278 , for some @xmath61 then @xmath279 . suppose that @xmath280 . therefore , for all @xmath281 there exists @xmath282 , such that @xmath283 . then @xmath284 as @xmath246 s are in generic position ( see the remark after proposition [ embedding ] ) . hence @xmath285 can not be contained in @xmath286 . by lemma [ upperin ] , it is clear that @xmath287 as @xmath288 , it follows that for any @xmath61 , @xmath289 thus , for any @xmath290 , using ( [ idcone ] ) , ( [ latticeid ] ) and ( [ mordot ] ) , @xmath291 similarily one can show that @xmath292 hence , we have proven [ isomlattice ] the map @xmath293 is an isomorphism of distributive lattices ( with @xmath294 ) where @xmath59 is a distributive lattice with meet and join operations defined in ( [ meetjoin ] ) . in the following proposition , we use the above results for the lattice @xmath277 of open subsets of @xmath18 and the lattice @xmath295 of ideals generated by the kernels of surjections forming an @xmath3-covering . for the first lattice , we consider the category of flabby sheaves over @xmath18 , and for the second , the category of @xmath3-coverings of algebras . here `` @xmath3-covering '' means a covering by @xmath3 surjections , and a morphism between @xmath3-coverings @xmath215 and @xmath296 consists of morphisms @xmath297 and @xmath298 such that @xmath299 . [ nagut ] let @xmath300 be the category of @xmath3-coverings of algebras , and @xmath301 be the category of flabby sheaves of algebras over @xmath18 . then the following assignments @xmath302 yield an equivalence of categories . suppose we are given a flabby sheaf @xmath222 of algebras over @xmath18 . let @xmath303 denote the restriction map for any open @xmath304,@xmath305 , @xmath306 . for brevity , we write @xmath307 instead of @xmath308 , if @xmath309 . by the flabbiness of @xmath222 , the morphisms @xmath310 ( see ( [ ai ] ) ) are surjective . the property @xmath311 follows from the sheaf condition . it remains to prove the distributivity of the lattice generated by the kernels of @xmath310 s . lattices of sets are always distributive , so that it is enough to show that the assignment @xmath312 defines a surjective morphism from the lattice of open subsets of @xmath18 onto the lattice of ideals generated by @xmath313 s . here by a morphism of lattices we mean a map that transforms the union and intersection of open subsets to the intersection and sum of ideals , respectively . to show this , let @xmath314 , @xmath315 be open subsets of @xmath316 . since @xmath222 is a sheaf , we know that @xmath317 is the fibre product of @xmath318 and @xmath319 . now it follows from ( [ kereta ] ) that @xmath320 , as needed . similarly , since the sheaf @xmath222 is flabby , lemma [ fib ] implies that @xmath321 . thus we have shown that ( [ gga ] ) assigns coverings to flabby sheaves . conversely , assume that we are given a covering . for brevity , let us denote for any @xmath322 , @xmath323 . let @xmath304 , @xmath314 be open subsets of @xmath18 such that @xmath324 . for all @xmath325 , @xmath326 which implies that there exists a subset @xmath327 such that @xmath328 . it is then clear that @xmath329 and one can define restriction map @xmath330 hence @xmath222 is a presheaf . let @xmath304 be an open subset of @xmath18 and let @xmath331 be an open covering of @xmath304 ( i.e. @xmath332 ) . suppose that @xmath333 is a family of elements , where for each @xmath334 , @xmath335 , and @xmath336 . by the distributivity of lattice of ideals generated by @xmath214 and generalised chinese remainder theorem , ( see e.g. @xcite , theorem 18 on p. 280 ) there exists an element @xmath337 such that , for all @xmath243 , @xmath338 . it is easy to see that if @xmath339 , for any open subsets @xmath304 , @xmath314 of @xmath18 then this element is unique . but this follows by lemma [ isomlattice ] and ( [ mordot ] ) . denote assignement in ( [ ggb ] ) by @xmath340 and functor defined by ( [ gga ] ) by @xmath220 . the functoriality of @xmath220 is immediate . let @xmath341 , @xmath342 be the morphism of @xmath3-coverings @xmath215 and @xmath296 . note that from @xmath299 it follows that for all @xmath243 , @xmath343 . accordingly , for any open subset @xmath304 of @xmath18 , @xmath344 , and therefore for all open @xmath304 , the following family defines a map of corresponding sheaves @xmath345 that @xmath346 is a functor naturally isomorphic to identity functor in the category of @xmath3-coverings is obvious , as @xmath347 and @xmath348 . on the other hand suppose that we are given a flabby sheaf @xmath222 . by the sheaf property , for any open sets @xmath304 , @xmath314 , @xmath349 is a fibre product of @xmath228 and @xmath318 over @xmath350 . then using flabbines , we can apply lemma [ fib ] and formula [ kereta ] to conclude that @xmath351 then by the property that assignement @xmath352 is a morphism of lattices and again using flabbiness one sees that for all open @xmath304 , @xmath353 , but this immedietaly shows that @xmath354 is naturally isomorphic to identity functor on @xmath301 , which ends the proof . + since the intersection of closed ideals in a @xmath6-algebra equals their product , lattices of closed ideals in @xmath6-algebras are always distributive . thus we immediately obtain : [ claco ] compact hausdorff spaces with a fixed covering by @xmath3 closed subsets are equivalent to flabby sheaves of commutative unital @xmath6-algebras over @xmath18 . here , the zero algebra is allowed as a unital @xmath6-algebra . this is needed if the closed sets are not in generic position . the set of unital morphisms from the zero @xmath6 algebra to any other is understood to be empty . in this last section we recall from @xcite the construction of examples for the above concepts that illustrate possible areas of applications . if @xmath220 is a compact group , then the join @xmath355 becomes a @xmath220-principal fibre bundle over the unreduced suspension @xmath356 of @xmath220 , see e.g. @xcite , proposition vii.8.8 or @xcite . for example , one can obtain the hopf fibrations @xmath357 and @xmath358 in this way using @xmath359 and @xmath360 , respectively . recall that @xmath355 is obtained from @xmath361 \times g \times g$ ] by shrinking to a point one factor @xmath220 at @xmath362 $ ] and the other factor @xmath220 at @xmath363 . @xmath364{pict215.eps } \nonumber\ ] ] alternatively , one can shrink @xmath365 at @xmath366 to the diagonal . this picture is generalised in @xcite . our aim in this first part of section [ examples ] is to describe a noncommutative analogue of this construction that nicely fits into our general concepts and will be studied in greater detail in @xcite . to this end , let @xmath1 be the hopf algebra underlying a compact quantum group @xmath116 ( see @xcite or chapter 11 of @xcite for details ) . we define @xmath367,\bar h ) \otimes h\,|\ , f(0 ) \in \delta(h)\},\nonumber\\ & & p_2:=\{f \in c([0,1],\bar h ) \otimes h\,|\ , f(1 ) \in \mathbb{c } \otimes h\}\nonumber\end{aligned}\ ] ] which will play the roles of the two trivial pieces of the principal extension . here we identify elements of @xmath368,\bar h ) \otimes h$ ] with functions @xmath361 \rightarrow \bar h \otimes h$ ] . the @xmath5 s become @xmath1-comodule algebras by applying the coproduct of @xmath1 to @xmath1 , @xmath369,\bar h ) } \otimes \delta$ ] , and the subalgebras of @xmath1-invariants can be identified with @xmath370,\bar h)\,|\,f(0 ) \in \mathbb{c}\ } , \nonumber\\ & & b_2:=\{f \in c([0,1],\bar h)\,|\,f(0 ) \in \mathbb{c}\}. \nonumber \end{aligned}\ ] ] furthermore , @xmath371 , @xmath372 , where @xmath1 acts on @xmath373 via the adjoint action , @xmath374 , @xmath375 $ ] , see @xcite . now one can define @xmath0 as a glueing of the two pieces along @xmath376 , that is , as the pull - back @xmath377 of the @xmath5 s along the evaluation maps @xmath378 theorem [ main ] implies that @xmath0 is principal . based on the idea of a heegaard splitting of @xmath379 into two solid tori , a noncommutative deformation of @xmath379 was proposed in @xcite . on the level of @xmath6-algebras , it can be presented as a fibre product @xmath380 of two @xmath6-algebraic crossed products @xmath381 and @xmath382 of the toeplitz algebra @xmath383 by @xmath384 . we denote the isometries generating @xmath383 in the two algebras by @xmath385 . the @xmath384-actions are implemented by unitaries @xmath386 , respectively , in the following way : @xmath387 the fibre product is taken over @xmath388 with action @xmath389 , where @xmath390 is the generator of @xmath391 and @xmath392 is the unitary giving the @xmath384-action in this algebra . the corresponding surjections defining the fibre product are @xmath393 there is a natural @xmath394-action on @xmath380 corresponding classically to the action in the hopf fibration , see @xcite . its restriction to the two crossed products is not the canonical action of @xmath394 viewed as the pontryagin dual of @xmath384 . however , to obtain the canonical actions one can identify @xmath380 with a fibre product of the same crossed products , but formed with respect to the surjections @xmath395 the identification is given by @xmath396+<-3mm,3mm>_{\pi^1_2 } \ar@/^2pc/[rrrrr]^{\phi_1 } & & { \mathcal{t}}\rtimes_{-\theta } \mathbb{z } \ar[ld]+<3mm,3mm>^{\pi^2_1 } \ar@/^2pc/[rrrrr]^{\phi_2 } & & & { \mathcal{t}}\rtimes_{\theta } \mathbb{z } \ar[rd]+<-3mm,3mm>_{\hat\pi^1_2 } & & { \mathcal{t}}\rtimes_{-\theta } \mathbb{z}\ar[ld]+<3mm,3mm>^{\hat\pi^2_1}\\ & \save[]*\txt{$c(s^1 ) \rtimes_{\theta } \mathbb{z}$\mbox{\ } } \ar[rrrrr]+<-12mm,0mm>^-{\phi_{12}}\restore & & & & & \save[]*\txt{$c(s^1 ) \rtimes_{\theta } \mathbb{z}$}\restore & . } } \ ] ] here isomorphisms @xmath397 are given on respective generators by @xmath398 the @xmath6-subalgebra of @xmath394-invariants is the @xmath6-algebra of the mirror quantum 2-sphere from @xcite . as mentioned in the introduction , we can pass from @xmath380 to the associated principal extension , and this procedure commutes with taking fibre products . in this way , we obtain a subalgebra @xmath399 which is a piecewise trivial @xmath400-comodule algebra , so that it fits the setting of this paper . the invariant subalgebra @xmath2 is again the @xmath6-algebra of the mirror quantum 2-sphere . on the other hand , there is a second natural hopf - like @xmath394-action on @xmath401 described in @xcite ( see also @xcite ) . again , its restriction to the two crossed products making up the fibre product @xmath401 is not the canonical action of @xmath394 . this fibre product can be transformed into an isomorphic one ( carrying the canonical @xmath394-action ) constructed by gluing two copies of @xmath382 over @xmath402 ( with generators @xmath403 ) using the gluing maps @xmath404 the identifying maps are now given by @xmath396+<-3mm,3mm>_{\pi^1_2 } \ar@/^2pc/[rrrrr]^{\tilde\phi_1 } & & { \mathcal{t}}\rtimes_{-\theta } \mathbb{z } \ar[ld]+<3mm,3mm>^{\pi^2_1 } \ar@/^2pc/[rrrrr]^{\tilde\phi_2 } & & & { \mathcal{t}}\rtimes_{-\theta } \mathbb{z } \ar[rd]+<-3mm,3mm>_{\check\pi^1_2 } & & { \mathcal{t}}\rtimes_{-\theta } \mathbb{z}\ar[ld]+<3mm,3mm>^{\check\pi^2_1}\\ & \save[]*\txt{$c(s^1 ) \rtimes_{\theta } \mathbb{z}$\mbox{\ } } \ar[rrrrr]+<-13mm,0mm>^-{\tilde\phi_{12}}\restore & & & & & \save[]*\txt{$c(s^1 ) \rtimes_{-\theta } \mathbb{z}$}\restore & . } } \ ] ] here isomorphisms @xmath397 are given on generators by @xmath405 the subalgebra of @xmath394-invariants is now the @xmath6-algebra of the generic podle quantum 2-sphere from @xcite . however , note that it is not possible to obtain the algebraic podle sphere in this way by replacing @xmath406 by the coordinate algebra of a quantum disc with generator @xmath407 satisfying @xmath408 @xcite . this is related to the fact that already in the commutative case the algebra of polynomial functions on a sphere has no covering corresponding to two hemispheres there are no nontrivial polynomials vanishing on a hemisphere . therefore to be in this setting of fibre products we use more complete algebras , e.g. , @xmath6-algebras . the authors are very grateful to pawe witkowski for producing the join picture , and to gabriella bhm , tomasz brzeziski , george janelidze , dorota marciniak , tomasz maszczyk , fred van oystaeyen , marcin szamotulski and elmar wagner for discussions . this work was partially supported by the european commission grants eif-515144(uk ) , mktd - ct-2004 - 509794 ( rm ) , the polish government grants1 p03a 036 26 ( pmh , rm ) , 115/e-343/spb/6.pr ue / die 50/2005 - 2008 ( pmh , bz ) , and the university of d grant 795 ( bz ) . pmh is also grateful to the max planck institute for mathematics in the sciences in leipzig for its hospitality and financial support .

communication is a recently developed communication paradigm whose communication process involves `` transmission and reception of information encoded in molecules '' @xcite . it is born in the study of nano - scale communication in nanonetworks @xcite where the applicability of classical electromagnetic communication is limited by several constraints and a novel solution is called for . inspired by the biological communication process such as the intra- and inter - cellular communication by cells @xcite and the pheromone diffusion by insects @xcite , the researchers have created various message - carrying molecules and their corresponding receptors . since the molecules are themselves at the nano - scale , the difficulty of nano - scale communication is solved by building up communication media between the transmitter and the receptor . moreover , since the major potential application of nanonetworks is implemented in the organisms body ( e.g. , organ monitoring in human body ) , the compatibility of the bioinspired communication process makes it a perfect choice to consider the encoding , transmission , and decoding of messages in molecules . excited by the reasons above ( and beyond ) , the researchers have treated the molecular communication as one of the mainstream subjects of nanotechnology nowadays . among the various proposed system blue maps for molecular communication , a specific class of system called the `` diffusion - based molecular system '' draws lots of interest due to the universality of _ mass transport phenomena _ in molecular communication . the mass transport phenomena refer to the propagation of encoded molecules in the communication media . in most cases , the communication media of molecular communication are modeled as fluid . as is well - known , the motion of molecules in fluid media is governed by the `` diffusion process '' . the universality of mass transport phenomena not only make the study of diffusion - based system natural , but at the same time guarantee the widest possible generality of research results . to this end , most efforts in molecular communication are devoted to diffusion - based systems , both in practical system design and information - theoretic analysis point of view . for example , for the practical system design , the physical phenomena of bacteria chemotaxis , pheromone diffusion @xcite , or calcium diffusion @xcite are adopted , and there are many existing systems operating on these phenomena such as @xcite , @xcite , and @xcite . the information - theoretic analysis of diffusion - based molecular system are also conducted in @xcite , @xcite , and @xcite . despite of this much effort for diffusion - based molecular communication , two important things are still lacking . first , _ an unified channel model for diffusion - based molecular communication is missing . _ the existing works on information - theoretic analysis tend to be system - dependent ; that is , they are treating different kinds of diffusion - based systems independently and giving their own capacity analysis . so far , the most general result concerning the information rate capacity of diffusion - based molecular communication is probably @xcite where the authors have cleverly utilized the thermodynamics to derive the information rate capacity in closed form , but the result is only valid for receiver with poisson noise . second , and the more important one , _ no mathematical foundations have been put down for information theory in diffusion - based molecular communication_. in particular , the channel coding theorem is not proven for diffusion - based systems . to state more directly , no one has ever shown that , for the diffusion - based molecular channel , the information rate capacity ( maximum mutual information ) is equal to the code rate capacity ( supremum achievable code rate ) , where the former is a system statistic of mathematical importance and the latter is the real practical concern . furthermore , the diffusion - based molecular channels exhibit non - classical randomness ( which is explained later in section iv.a ) and _ no existing channel model apply to diffusion - based molecular channel_. therefore , the lacking of proof for equivalence between the information rate capacity and code rate capacity renders the purpose of calculating information rate capacity ambiguous , or at least in a shaky situation . this paper aims at solving the above two problems once and for all . let us first put molecular communication aside for a moment and recall the history of information theory . ever since the advent of the paper @xcite by shannon in 1948 , the `` channel capacity '' has been the primary subject of every communication systems . in particular , the way that shannon intervenes the information - theoretic and coding - theoretic quantities for discrete memoryless channels is the one that communication engineers imitate when treating the communication system under consideration . despite its great novelty and ingenuity , a complete and rigorous mathematical setting for information theory is not laid down until a series of papers by mcmillan @xcite , feinstein @xcite , and finally khinchin @xcite , where the definition of `` finite memory channel without anticipation '' is made precise and for such channels the equivalence between the information rate capacity and code rate capacity is established . a key step of @xcite is the _ measure - theoretic formulation for sources and channels_. khinchin noticed that except for some simple classes of channel ( e.g. , i.i.d . , markov , ... ) , the mathematical formulations of sources and channels are most conveniently made in the measure - theoretic form to provide rigorousness and applicability . later , works in @xcite are generalized by many others ( e.g. , @xcite , @xcite , @xcite , ... ) to include classes of channels other than finite memory channel without anticipation , all under the same proof structure . it is now a communication engineers consensus that the measure - theoretic formulations of source and channel have struck a perfect balance between the mathematical rigorousness of information theory and the applicability thereof to practical communication systems . therefore , for any communication system , once its channel abstraction is made in the measure - theoretic form and the corresponding capacity theorems are derived , then a solid mathematical foundation is laid down and applicability of information theory to the system is guaranteed . in this paper , we shall follow the path of khinchin @xcite to provide the mathematical foundations for information theory in diffusion - based molecular communication . fortunately , there are already many building blocks from the works of predecessors @xcite , @xcite , and @xcite . from these works , the two crucial concepts of `` channels with asymptotically decreasing input memory and anticipation '' and `` @xmath0-continuous channels '' will be used heavily in our paper . we will give a channel model for diffusion - based molecular communication in the most rigorous language , namely the measure - theoretic one . the channel model we create captures the essential features of all existing diffusion - based molecular systems and all diffusion - based molecular systems to come , thus allowing our results the maximum possible generality . for our channel , we first prove its @xmath0-continuity and then utilize the @xmath0-continuity to establish various equivalence relationships between the statistical and operational definitions of channel capacity , including the equivalence of information rate capacity and the code rate capacity . in particular , the channel coding theorems for diffusion - based molecular channel is given . a solid foundation is therefore laid down and we can use the usual sense of `` channel capacity '' without worry thereafter . the rest of paper is organized as follows . we state the essential terminologies and notations in section ii . section iii introduces `` channels with asymptotically decreasing input memory and anticipation '' and `` @xmath0-continuous channels '' as well as their properties for our use . section iv is the main section in which we give the definition of diffusion - based molecular channel and prove capacity theorems , along with some discussions . finally , we conclude this paper in section v. we begin by setting up terminology and notation for sources . if a set @xmath1 is finite , the symbol @xmath2 will denote the set of all doubly infinite sequences @xmath3 , @xmath4 for all @xmath5 . the notation @xmath6 denotes the finite sequence @xmath7 and @xmath8 stands for the set of all such sequences . given a sequence @xmath9 , there corresponds a cylinder set ( or rectangle set ) : @xmath10 the borel field generated by all cylinder sets in @xmath2 will be denoted by @xmath11 . a source is a probability measure @xmath12 on the measurable space @xmath13 and @xmath1 is called the alphabet of the source . the probability measure @xmath12 induces a sequence of random variables @xmath14 on the probability space @xmath15 by the formula @xmath16 for all @xmath17 . for integers @xmath18 and @xmath19 define @xmath20 as the borel field generated by random variables @xmath21 and we allow @xmath22 and @xmath23 . the induced sequence @xmath24 is viewed as the transmitted signal for each time @xmath19 , and @xmath20 represents all possible events in time @xmath18 to @xmath25 . besides @xmath24 , @xmath12 also defines a probability measure @xmath26 on @xmath8 by the formula @xmath27 when @xmath28 , @xmath26 is abbreviated @xmath29 . whenever we need to emphasize the alphabet or the induced random variables , we will also use the notation @xmath30 $ ] or @xmath31 $ ] to denote a source . the ( left ) shift operator @xmath32 on @xmath2 is defined by @xmath33 , where @xmath34 is the element at the @xmath19th position . a source @xmath12 is said to be stationary if @xmath35 a source is @xmath19-stationary for some positive integer @xmath19 if @xmath36 for all @xmath37 . in other words , @xmath12 is stationary if it is shift - invariant with respect to all events , and it is @xmath19-stationary if it is @xmath19-times shift - invariant with respect to all events . obviously stationarity implies @xmath19-stationarity . if @xmath12 is @xmath19-stationary for some @xmath19 , we say that @xmath12 is block stationary . a source @xmath12 is ergodic if for every @xmath37 such that @xmath38 , @xmath39 is either @xmath40 or @xmath41 . an important property of the ergodic sources is the following : for a given source @xmath30 $ ] , denote the characteristic function of the set @xmath37 by @xmath42 . then , we have for all @xmath37 @xmath43 if and only if for every @xmath44 such that @xmath38 , @xmath39 is either @xmath40 or @xmath41 . the necessity is provided by the well - known `` ergodic theorem '' of birkhoff and the sufficiency is a direct verification of the definition . for complete proof , see , e.g. , @xcite . the summation appeared in the theorem represents the `` ensemble mean '' . * theorem 1 * therefore states that the ergodic sources are those whose features can be approximated well by ensemble statistics on observable samples . a discrete channel is characterized by an input alphabet @xmath1 , an output alphabet @xmath45 , and a set of probability measures @xmath46 on @xmath47 . the probability measure @xmath48 is determined for each @xmath17 such that for each @xmath49 the mapping @xmath50 is @xmath11-measurable . each @xmath48 is recognized as the output distribution conditioning on the input @xmath51 . we shall refer @xmath52 $ ] to such structure and say that @xmath52 $ ] is a channel . the channel is stationary if for every @xmath53 and @xmath49 we have @xmath54 now , consider the case of connecting a source @xmath30 $ ] to a channel @xmath52 $ ] . we are naturally led to consider the probability measure @xmath55 on the product space @xmath56 characterized by @xmath57 the event @xmath58 is interpreted as `` source emits event @xmath44 and the channel output receives @xmath59 '' . if we are interested in the output process , we shall consider the probability measure @xmath60 on the output space @xmath61 ) : @xmath62 the sequence of random variables induced by @xmath60 on @xmath61 ) will be denoted by @xmath63 . elementary analysis shows that if @xmath30 $ ] and @xmath52 $ ] are both stationary , then so is @xmath55 and @xmath60 . in this paper we will encounter the `` cascade '' of channels . suppose that we are given two channels @xmath64 $ ] and @xmath65 $ ] . if we have these two channels in cascade , the overall channel @xmath52 $ ] is described by @xmath66 another important notion of channel is the ergodicity . the channel @xmath67 is ergodic if for any input sequence @xmath51 , @xmath48 is ergodic . the following is an useful criterion for examining the ergodicity of a channel : a stationary channel @xmath52 $ ] is said to be strongly mixing if for any input sequence @xmath51 and any two cylinders @xmath68 , we have @xmath69 where @xmath32 is the time shift operator . if a channel is strongly mixing , then it is ergodic . . intuitively speaking , a channel is strongly mixing if for any two events separated by a long time at the output , the two events are `` almost '' independent , regardless of the input . we have seen that the source @xmath12 can be viewed as a sequence of random variables @xmath24 . similarly , we can view @xmath60 as @xmath63 and @xmath55 as @xmath70 . it is well - known that if the processes represented by these sequences of random variables are stationary then their entropy rates exist . since we shall only consider stationary ergodic sources and stationary channels in this paper , we are free to write @xmath71 @xmath72 @xmath73 the mutual information rate between the input and output is defined as @xmath74 or , equivalently , @xmath75 in the rest of this paper we shall refer to these quantities interchangably in their measure forms and random variable forms . + there are many defintions of channel capacity . in the followings we state and itemize them for ease of reference . see @xcite , @xcite , and @xcite for detail descriptions of these channel capacities . given a channel @xmath52 $ ] , there are several choices of source which we can extremizing the mutual information over . the most common ones are the follows : * the ergodic capacity @xmath76 $ } } i(\mu\nu),\ ] ] * the stationary capacity @xmath77 $ } } i(\mu\nu),\ ] ] and * the block stationary capacity @xmath78 } i(\mu\nu).\ ] ] we now state coding - theoretic terminologies . a blocklength @xmath19 channel block code @xmath79 for a noisy channel @xmath52 $ ] is a collection of @xmath80 distinct codewords @xmath81 and @xmath82 disjoint decoding sets @xmath83 for each @xmath5 . the rate @xmath84 of a code is @xmath85 . a code @xmath86 has probability error @xmath87 if @xmath88 where @xmath89 is the complement of @xmath90 . such a channel block code is called an @xmath91 channel code . we say @xmath84 is a permissible transmission rate if there exist @xmath92 channel codes such that @xmath93 as @xmath94 . the operational channel block coding capacity @xmath95 is defined to be the supreme of all permissible rate : * the operational channel block coding capacity @xmath96 an alternate operational definition of capacity , the operational source / channel block coding capacity @xmath97 , is defined as follows . we consider the transmission of an ergodic source @xmath98 $ ] over the channel @xmath52 $ ] . define a blocklength @xmath19 source / channel block code as a pair of mappings @xmath99 and @xmath100 . the @xmath101 is called the encoder and @xmath102 the decoder . the block encoder @xmath101 implements a mapping on infinite sequences @xmath103 defined by @xmath104 . all such encoded sequences can be viewed as another channel input process @xmath105 $ ] characterized by @xmath106 for all @xmath37 . the block error probability for the code is @xmath107 where @xmath108 is the random variable induced at the channel output , or , presented in the measure from , @xmath109 a source / channel code of blocklength @xmath19 and of probability error @xmath110 is called an @xmath111 source / channel code . a source @xmath112 $ ] is said to be admissible if there exist @xmath113 source / channel codes for the source such that @xmath114 as @xmath115 . the operational source / channel block coding capacity @xmath97 is defined as the supreme of the entropy over all admissible stationary ergodic sources : * the operational source / channel block coding capacity@xmath116 } h(\tau).\ ] ] the last operational capacity definition concerns another function of samples defined over input / output alphabets . the @xmath19th order sample mutual information @xmath117 is defined by @xmath118 if @xmath55 is stationary then there exist a measurable shift - invariant nonnegative bounded function @xmath119 such that @xmath120 with convergence in @xmath121 and @xmath122 if , furthermore , @xmath55 is ergodic , then the shannon - mcmillan theorem states that @xmath123 @xmath55-almost everywhere . therefore , for stationary ergodic sources and channels the @xmath19th order sample mutual information @xmath117 is a good approximation of the `` true '' mutual information @xmath124 provided @xmath19 is large enough . define for a stationary channel @xmath67 and for @xmath125 the quantile @xmath126 where @xmath127 and the outer supreme is taken over all block stationary input process @xmath30 $ ] . our last operational capacity definition , the information quantile capacity @xmath128 , is defined by * the information quantile capacity @xmath129 the relation between all these capacities ( namely , @xmath130 , and @xmath128 ) will be given in section iii , where we need the notion of @xmath131-continuity to unify these concepts . a channel @xmath67 is said to have asymptotically decreasing input memory if given @xmath132 there is an integer @xmath133 such that , for any @xmath19 and any output event @xmath134 , @xmath135whenever @xmath136 for @xmath137 . a channel @xmath67 is said to have asymptotically decreasing input anticipation if given @xmath132 there is an integer @xmath138 such that for any @xmath19 and any output event @xmath139 , @xmath135 whenever @xmath136 for @xmath140 . if a channel has both asymptotically decreasing input memory and input anticipation then we call such channels the adima ( asymptotically decreasing input memory and anticipation ) channels or channels with adima . therefore , a channel @xmath67 with adima is one for wich given @xmath141 , there exist integers @xmath133 and @xmath138 such that for all @xmath142 and all @xmath143 we have @xmath135 whenever @xmath136 , @xmath144 . intuitively speaking , a channel has asymptotically decreasing input memory if for any two input sequences @xmath51 and @xmath145 , the conditional channel output distributions @xmath48 and @xmath146 in the interval @xmath147 are `` close '' provided the two input sequences coincide on the interval @xmath148 . similar interpretation works for channel with asymptotically decreasing input anticipation . the measure of closeness can be made precise in the language of the _ variational distance_. given two probability measures @xmath149 and @xmath150 on @xmath151 , the variational distance between the measures is defined by @xmath152 if we consider stationary channels only ( so that the definition of adima can be made to require having only all @xmath153 with the desired property , not with all @xmath154 for all @xmath155 ) , then we can rephrase the definition of the adima channel by saying that a stationary channel is with adima if , given @xmath132 , there are integers @xmath133 and @xmath138 such that for all n , @xmath156 whenever @xmath136 , @xmath157 . given two probability measures @xmath158 and @xmath159 on @xmath151 , let @xmath160 denote the set of all joint probability measures @xmath161 on @xmath162 satisfying the property that for all @xmath163 , @xmath164 and @xmath165 in other words , @xmath161 has @xmath149 and @xmath150 as marginals . define the coordinate functions @xmath166 and @xmath167 , @xmath168 by @xmath169 @xmath170 let @xmath171 be the hamming distance and let @xmath172 to be the @xmath19th order normalized hamming distance @xmath173 the @xmath19th order @xmath0-distance between @xmath149 and @xmath150 is defined as @xmath174 a channel @xmath67 is @xmath0-continuous if given @xmath141 there is an @xmath175 such that for @xmath176 we have @xmath177 whenever @xmath136 , @xmath178 . equivalently , @xmath67 is @xmath0-continuous if @xmath179 here we list several important properties of @xmath0-continuous channel . we begin by establishing the relation between the adima channel and @xmath0-continuous channel . a channel @xmath67 is strongly @xmath131-continuous if given @xmath141 there are positive integers @xmath133 and @xmath138 such that for all @xmath19 , @xmath177 whenever @xmath136 , @xmath180 . an adima channel is strongly @xmath131-continuous . a strongly @xmath131-continuous channel is @xmath0-continuous . see @xcite . next , we summarize information - theoretic results . recall the following notions : @xmath181 \a ) for stationary channels we have @xmath182 hence , for stationary channels , we can simply call @xmath183 the information rate capacity ( or transmission rate capacity or shannon capacity ) . + + b ) if a channel is stationary and @xmath0-continuous then @xmath184 + c ) let @xmath67 be a discrete stationary @xmath0-continuous channel . if @xmath185 and @xmath132 , then for sufficiently large @xmath175 there exist @xmath186 block codes for all @xmath176 . if @xmath187 there exists an @xmath188 such that there do not exist @xmath189 block code for any @xmath190 . + + d ) let @xmath67 be a discrete stationary @xmath0-continuous channel and @xmath12 a discrete stationary ergodic source . if @xmath191 , then @xmath12 is admissible . if @xmath192 , then @xmath12 is not admissible . + + e ) if , in addition , @xmath67 is also ergodic in b ) , c ) , and d ) , then @xmath193 . see @xcite . * theorem 4 * lists the most important channel coding theorems for discrete stationary @xmath0-continuous channels . for a discrete stationary @xmath0-continuous channel * theorem 4.c ) * says that if we transmit message at a rate @xmath194 , then @xmath84 can be achieved by some block codes with arbitrarily small block error probability ; conversely , if we transmit at a rate @xmath195 , then the block error probability is always bounded away from zero . similarly , * theorem 4.d ) * says that , for a discrete stationary @xmath0-continuous channel and any discrete stationary ergodic source @xmath12 , if @xmath191 , then @xmath12 can be transmitted via block coding with arbitrarily small block error probability ; conversely , if @xmath192 , then no such block code exists . the contents of * theorem 4.c ) * and * theorem 4.d ) * therefore collectively tell us that @xmath128 is the ultimate limitation of the system performance over the given channel , where the performance is measured in terms of the maximal code rate with vanishing block error probability . * theorem 4.e ) * says that if , in addition , @xmath67 is also ergodic , then the shannon capacity @xmath196 equals @xmath128 . the last property we need for @xmath0-continuous channel is its preservation under the @xmath0-limit @xcite : the @xmath0-limit of a sequence of @xmath0-continuous channels is @xmath0-continuous . see @xcite . this is the main section of our paper . in this section , we first state the general properties of diffusion - based molecular channel in section iv.a . the abstract diffusion - based molecular channel model is given in section iv.b and capacity theorems and channel coding theorems are proven in section iv.c . section iv.d discusses the implications and assumptions of our results . the theory of molecular communication studies the process of `` transmission and reception of information encoded in molecules '' @xcite . in other words , in a molecular communication system , we utilize features of certain molecules as the information carrier , transmit the molecules through the communication media , and design a receiver which is capable of recognizing the features encoded over . the diffusion - based systems refer to those systems whose communication media are described by the `` diffusion processes '' . in the following we introduce the essential components of diffusion - based molecular communication systems : in molecular communication , the transmitter has many choices of information carrier . for example , we can embed the information in the molecule concentration magnitudes @xcite , the inter - transmission times @xcite , the relative transmission times @xcite , the molecule types @xcite , or hybrids of each other @xcite , @xcite . after the messages are encoded , the transmitter conveys the messages by simply releasing the encoded molecules into the communication media . + on contrast to the variety of information carrier , to the present days the communication media of molecular communication are concentrated on the fluid media almost exclusively . the fluid media represent the environment where the communication takes places . a typical environment is the blood , where the communication process occurs at human vessels . as is well - known , the motion of molecules in fluid media is described by the `` diffusion process '' . the molecular communication systems whose communication media are described by the diffusion process are collectively called the `` free diffusion - based molecular systems '' or simply `` diffusion - based molecular systems '' . for systems operating under the diffusion process , there is a special type of randomness which distinguishes those systems from the classical communication : the disorder of message arrivals . consider an one - dimensional molecular communication system with transmitter and receiver separated by some distance ( fig . 1 ) . for each released molecule diffusing independently in the fluid medium , the time it takes to be detected by the receiver is a random variable . if we assume that the receiver perfectly detects every molecule at its first arrival at the location of the receiver , then the detection times of the transmitted messages are the _ first passage times _ of the corresponding molecules , which are random variables depending only on the fluid medium , the corresponding transmitted molecules , and the distance from transmitter to receiver . denote the first passage time of the @xmath5th message by @xmath197 . it is possible to see that @xmath198 with positive probability even if @xmath199 , i.e. , it can happen that _ the later transmitted message arrives earlier_. more generally , if we release a set of molecules in sequence , then it can happen that the receiver receives these molecules in _ any _ order . the disorder of message arrivals is inevitable in diffusion - based molecular communication . rather unfortunately , the classical tools dealing with disorders of message arrivals are targeting at the link - layer disorders ( e.g. @xcite ) and do not suit the case of diffusion - based disorder . specifically , there are no existing mathematical tools describing the behavior of random ordering of arbitrary set of infinite random variables . to cope with it , we must develop useful channel model that describes the phenomena of the random ordering of first passage times , and by useful we mean the capability of applying the existing communication theory to the channel . the following is the strategy we take : we first find out the essential features of first passges times occurring in diffusion - based systems and define the notion of _ permutation channel induced by the random ordering of first passage times_. afterwards , we prove that the permutation channel is with adima , and apply the channel coding theorems to such channel . despite the randomness described above , the first passage times in diffusion - based molecular systems do contain some regularities . the followings are the essential properties of first passage times in diffusion - based systems : * we require that the @xmath200 is a _ delayed version _ of @xmath197 ; that is , there are two independent random variables @xmath201 and @xmath202 having the same distribution , and a random variable @xmath203 independent of @xmath204 and @xmath205 , such that @xmath206 @xmath207 + the above condition actually says that we are releasing the encoded molecules _ in sequence _ ( see fig . the intuitive meaning is that @xmath208 is the _ inter - transmission _ time of the @xmath5th and @xmath209th messages , and @xmath204 ( or @xmath205 ) is the first passage distribution of the @xmath5th message . if we release the @xmath5th message , wait for a time @xmath208 , and release the @xmath209th message , then the time it takes for the @xmath5th and @xmath209th molecules to diffuse to the receiver is of the above form . note that the above condition can be applied iteratively on @xmath5 to derive @xmath210 @xmath211 where @xmath204 and @xmath205 are independent and have the same distribution , and all @xmath208 s are independent . note also that the @xmath197 s are pairwise independent . + let us give some examples . if the system is `` synchronous '' ( which is the case in @xcite , @xcite , and @xcite ) then @xmath212 for a real number @xmath213 and @xmath214 . the work in @xcite also considers an asynchronous communication scheme for which @xmath208 s are i.i.d . random variables . + * the second restriction we impose on the first passage times is a qualitative one : we shall require that the arrival order of each @xmath197 is _ `` effected '' `` most exclusively '' _ by the @xmath215 s for those whose indices are close ( or , equivalently , @xmath216 s are small ) . the precise meaning of `` effected '' and `` most exclusively '' are made clear by the next section , where we need the notion of the _ permutation channel _ to state them properly . + our second restriction aims at capturing the nature of the diffusion processes occurring in molecular communication . the common feature of these diffusion processes is that if the release times of two molecules are @xmath217 and @xmath218 , then the probability of the later released molecule advancing the first is very small provided @xmath219 is large , _ regardless of any feature of the transmitted molecules ( e.g. , weight , radius , ... ) _ . in other words , to infer the order of @xmath215 in @xmath220 , it suffices to observe the nearby @xmath221 and @xmath222 for some large @xmath19 . + let us give a concrete example to support the belief . consider the diffusion process governed by the _ brownian motion with drift _ , which is adopted in most diffusion - based molecular communication systems ( see , e.g. , @xcite , @xcite , and @xcite ) . the first passage time of a released molecule is given by @xcite @xmath223 where @xmath67 is the diffusion coefficient , @xmath51 is the distance from the transmitter to the receiver , and @xmath224 is the drifting velocity . suppose the inter - transmission times @xmath208 s are supported on @xmath225 and @xmath226 . then , according to the first restriction , the probability of the @xmath5th released molecule advancing the first is @xmath227 by discarding all the irrelevant constants we see that @xmath228 behaves like @xmath229 for large @xmath230 . therefore , for large @xmath230 , @xmath231 in fact , the above derivation shows that if @xmath232 , then @xmath233 in @xmath5 , which means that the probability of the @xmath5th release molecule advancing the first molecule converges to @xmath40 in @xmath234 ( that is , the probability decays super - exponentially ) . similar reasonings apply to @xmath235 with all @xmath236 s . in particular , if @xmath237 , then by union bound @xmath238 that is , the probability of the event `` any molecule released after the @xmath239th message advancing the @xmath5th released molecule is vanishing in @xmath239 '' . this strong and universal property of diffusion processes is the one we shall capitalize upon when formulating the mathematical structure of diffusion - based molecular channels . + in molecular communication , the receivers are designed to match the information carrier chosen by the transmitter . for example , if the information are carried on the molecule types , then the receiver is some nanomachines capable of recognizing these types . if the transmitter emits different concentration levels of a compound , then the receiver is usually operating on a chemical reaction involving the transmitted compound . some examples for the first case are @xcite , @xcite , and @xcite . for the later case , see , e.g. , @xcite or @xcite . just as the white noise may occur in classical communication , the receiver might contain some sources of detection noise . again , in all cases of molecular communication , the detection noise of a message is governed by its nearby received messages `` most exclusively '' . the mathematical formulation of a receiver is given in the next section . for an integer @xmath19 , denote the set of symmetric group of order @xmath19 by @xmath240 ; that is , @xmath240 is the set of all bijections of integers @xmath241 onto itself . an element @xmath242 is called a _ permutation _ and can be represented by the permutation matrix : @xmath243{ccccc } 1 & 2 & 3 & \cdots & n \\ b_1 & b_2 & b_3 & \cdots & b_n \end{array } \right)\ ] ] where @xmath244 exhausts @xmath241 , indicating the position after the effect of @xmath245 ; that is , @xmath246 . we also consider the _ infinite permutation _ defined as any bijection of the set of all integers @xmath247 onto itself . the set of all infinite permutations are denoted by @xmath248 . similar to @xmath240 , any @xmath249 has its permutation matrix representation : @xmath243{ccccc } \cdots & -1 & 0 & 1 & \cdots \\ \cdots & b_{-1 } & b_0 & b_1 & \cdots \end{array } \right)\ ] ] for some doubly infinite sequence @xmath250 as the position indicators . let @xmath251 be any probability measure on @xmath248 ( with borel field being all subsets of @xmath248 in discrete topology ) . the effect of @xmath243{ccccc } \cdots & -1 & 0 & 1 & \cdots \\ \cdots & b_{-1 } & b_0 & b_1 & \cdots \end{array } \right)\ ] ] on an input sequence @xmath252 is viewed as the permutation of index : @xmath253 and @xmath254 is , of course , the probability of such index permutation . given a finite set of random variables @xmath255 , let @xmath256 denote the @xmath5th largest order statistic of @xmath255 ; that is , @xmath257 . the order statistics of @xmath258 s induce a probability measure @xmath251 on @xmath240 by considering the probability @xmath254 of @xmath243{ccccc } 1 & 2 & 3 & \cdots & n \\ b_1 & b_2 & b_3 & \cdots & b_n \end{array } \right)\ ] ] as @xmath259 . we shall say that @xmath251 is induced by the ordering of @xmath197 s . for the case of @xmath248 the ordering of an infinite sequence of random variables is ambiguous . to this end , we need another way of identifying the bijections in @xmath248 . for a finite subindex set @xmath260 ( i.e. , @xmath261 s are integers and @xmath262 ) and a @xmath263,@xmath243{ccccc } \cdots & -1 & 0 & 1 & \cdots \\ \cdots & b_{-1 } & b_0 & b_1 & \cdots \end{array } \right),\ ] ] define the _ local @xmath260 version _ of @xmath245 as @xmath264{ccccc } k_1 & k_2 & k_3 & \cdots & k_n \\ b'_1 & b'_2 & b'_3 & \cdots & b'_n \end{array } \right)\ ] ] where @xmath265 is the @xmath5th least integer of @xmath266 . for example , the local @xmath267 version of @xmath264{ccccc } 1 & 2 & 3 & 4 & 5 \\ 5 & 4 & 3 & 1 & 2 \end{array } \right)\ ] ] is @xmath264{ccc } 1 & 3 & 5 \\ 3 & 2 & 1 \end{array } \right)\ ] ] since @xmath268 . obviously , for @xmath269 , @xmath270 if and only if every local version of @xmath245 and @xmath271 coincides . therefore , if we view the local @xmath260 version as the permutation matrix induced by the order statistics on the random variables @xmath272 , then these partial orderings of the infinite sequence @xmath273 uniquely determine an infinite permutation @xmath249 . however , the explicit close - form solution of the distribution @xmath251 on @xmath248 induced by @xmath273 is a difficult and long standing problem for which all existing solutions require strong assumptions ( see , e.g. , the case for i.i.d . and exchangable random variables in @xcite ) . fortunately , we can take a different route to consider the _ qualitative _ properties of @xmath251 and the qualitative features are enough for our application , as illustrated below . we shall say that a probability measure @xmath251 on @xmath248 is induced by the orderings of @xmath273 if @xmath251 possesses the following property : for any finite subindex @xmath260 and any set of integers @xmath274 exhausting @xmath241 , we have @xmath275 where the summation is taken over all @xmath263 with local @xmath260 version @xmath264{ccccc } k_1 & k_2 & k_3 & \cdots & k_n \\ b'_1 & b'_2 & b'_3 & \cdots & b'_n \end{array } \right).\ ] ] in other words , for any finite sub - collection @xmath276 of @xmath273 , we demand that the probability of the event @xmath277 equals the sum of probability of the infinite permutations for which their local @xmath260 versions are the permutation matrix generated by the ordering of @xmath276 . for difficulty illustrated before , the explicit construction for @xmath251 satisfying the condition ( 2 ) is beyond the scope of this paper . in particular , the uniqueness of @xmath251 is not of our concern : any @xmath251 with property ( 2 ) is a good model for the channel under our consideration . if @xmath251 is not unique we can simply choose any one as the representative . we are now ready to define the notion of a _ let @xmath278 be any input sequence of a diffusion - based molecular communication system . as discussed in section iv.a.2 , @xmath51 corresponds a set of random variables @xmath279 where @xmath197 is the first passage time of the message @xmath280 . the ordering of these first passage times then induces a probability measure @xmath281 on @xmath248 . permutation channel @xmath46 _ on a same input and output alphabet , say @xmath1 , is defined by @xmath282 where @xmath249 and @xmath283 means ( 1 ) . with slight abuse of notation we also say that @xmath284 or @xmath251 is a permutation channel . + with the help of the permutation channel , we now give the following characterizations of a new class of channel , which is the mathematical channel model capturing all essential features of diffusion - based molecular channels considered in section iv.a . a diffusion - based molecular channel @xmath52 $ ] is a cascade of a permutation channel @xmath251 on @xmath1 and a stationary ergodic adima channel @xmath285 $ ] , where @xmath251 is induced by the first passage times of input messages . $ ] is called the receiver channel . the permutation channel has the following property : let @xmath51 and @xmath145 be two input sequences and induce @xmath281 and @xmath286 on @xmath248 , respectively . for @xmath132 , there exists integers @xmath18 and @xmath287 with @xmath288 such that for all integer @xmath19 , if @xmath136 for @xmath289 , then the following holds : ( i ) : : for any @xmath249 with @xmath290 and @xmath291 , we have @xmath292 ( ii ) : : the `` outlier '' probability is small : @xmath293 in both @xmath281 and @xmath286 measure . in other words , @xmath294 where the summation is taken over all @xmath245 with the property in the probability bracket . let us give the engineering interpretation to all the assumptions we made . in section iv.a we have illustrated that the messages arrive at the receiver in disorder for diffusion - based molecular communication systems . since such disorder comes from the random ordering of the first passage times , it does contain some implicit regularities . suppose we aim at an interval @xmath295 at the output of the permutation channel . then condition ( i ) says that , since @xmath136 for an interval @xmath296 much larger than @xmath297 , their corresponding first passage times are equally distributed within @xmath296 and consequently the output distributions of @xmath295 given @xmath51 and @xmath145 should be close . moreover , since the probability of @xmath298 vanishes in @xmath299 in all diffusion processes in molecular communication , any model for a diffusion - based molecular channel must be able to capture the fact that the `` crossing probability @xmath298 '' is small in @xmath300 , which is the content of condition ( ii ) . the total effect at the receiver is simply modeled as an adima channel . the notion of adima already characterizes fairly well a receiver whose detection of a message is `` mostly '' effected by its nearby received messages . in this section we shall prove the following fundamental theorem of diffusion - based molecular communication : let @xmath1 be a finite alphabet . for a diffusion - based molecular channel @xmath52 $ ] with a stationary ergodic source @xmath30 $ ] , \a ) @xmath301 + b ) if @xmath302 and @xmath132 , then for sufficiently large @xmath175 there exist @xmath186 block codes for all @xmath176 . if @xmath303 there exists an @xmath188 such that there do not exist @xmath189 block codes for any @xmath190 . + c ) if @xmath304 , then @xmath12 is admissible . if @xmath305 , then @xmath12 is not admissible . + we divide our proof into the following steps : ( i ) : : the permutation channel in a diffusion - based molecular channel is stationary . ( ii ) : : the permutation channel in a diffusion - based molecular channel is with adima . ( iii ) : : the cascade of adima channels is an adima channel . ( iv ) : : by * theorem 3 ) * , an adima channel is @xmath0-continuous , and therefore * theorem 4.a)@xmath306d ) * apply . ( v ) : : the diffusion - based molecular channel is strongly mixing ( and therefore by * theorem 2 * it is ergodic ) . ( vi ) : : the cascade of two ergodic channels is ergodic . by * theorem 4.e ) * the proof is complete . ( i ) : this is an immediate result of the fact that @xmath307 if and only if @xmath283 , where @xmath32 is the time - shift operator . ( ii ) : since by ( i ) the channel is stationary , it suffices to prove that for any @xmath141 and all integer @xmath19 , there exists an integer @xmath308 such that if @xmath136 for @xmath289 , then @xmath309 for any @xmath310 . let @xmath132 and @xmath44 be given . let @xmath18 be a large integer determined later and @xmath136 for @xmath289 . for each integer @xmath287 , divide the elements of @xmath311 into three classes @xmath312 , @xmath313 , and @xmath314 where @xmath315 @xmath316 and @xmath314 is the complement of @xmath317 . if @xmath318 , then @xmath136 for @xmath319 and therefore @xmath312 actually equals @xmath313 . now , by definition of a diffusion - based molecular channel , there is an @xmath18 such that if @xmath320 , then @xmath321 since there exists a @xmath249 with @xmath322 such that @xmath323 . on the other hand , if @xmath324 , then either @xmath325 . by definition of the diffusion - based molecular channel , @xmath326 since any set @xmath310 is a finite union of @xmath327 , the proof is complete . ( iii ) : we shall state the result formally and prove it in appendix a. ( iv ) : it is a fact . the * theorem 6 * provides the mathematical foundations for information theory in diffusion - based molecular communication . in this section , we discuss several issues concerning the assumptions of the * theorem 6 * along with some important implications . * byproducts of * theorem 6 * + the proof of * theorem 6 * contains many useful byproducts . we list them formally as below : + a diffusion - based molecular channel is stationary . + this is an easy consequence of the fact that the permutation channel and the receiver channel are both stationary , and the cascade of two stationary channels are stationary . + the permutation channel in a diffusion - based molecular channel is with adima , strongly @xmath131-continuous , and @xmath0-continuous . + a diffusion - based molecular channel is ergodic . + a diffusion - based molecular channel is @xmath0-continuous . + * why @xmath0-continuity ? + in the proof of * theorem 6 * we have shown that the diffusion - based molecular channel is an adima channel . since adima channel has its own coding theorems , it seems better to directly view the diffusion - based molecular channel as adima channel and not bothering using the @xmath0-continuity . this is , however , not appropriate due to the fact that the adima channel is very `` vulnerable '' to modeling errors . this is best illustrated by the following example . consider two i.i.d . binary sources with parameter @xmath361 and @xmath331 . recall that an adima channel demands the variational distances between distributions under consideration to be close , which is _ not _ the case for these two i.i.d . sources when @xmath362 since the variational distance @xmath363 between the two i.i.d . sources tends to @xmath41 as @xmath94 , _ no matter how small the error @xmath364 is_. the example tells us that if our model differs slightly from the reality , then the theorem might not carry over . unfortunately , the diffusion process is itself a mathematical assumption and not the physical reality . in particular , for the most widely adopted diffusion model , the brownian motion , various peculiar physical behaviors have been reported ( e.g. , almost - surely nowhere - differentiable path ) and now it is commonly accepted that the brownian motion is a good approximation of diffusion process but not the exact solution . there do exist realistic models for diffusion processes ( e.g. , the ornstein - uhlenbeck process ) , but they usually involve the `` stochastic differential equations '' which render the derivation of analytical results hard , if not impossible . + on contrast to the adima channel , the @xmath0-continuity is very `` robust '' to the modeling errors @xcite . in particular , by * theorem 5 * the @xmath0-limit of a sequence of @xmath0-continuous channels is @xmath0-continuous , while the limit of a sequence of adima channels is not necessarily an adima channel . this tells us that suppose we can approximate the `` true '' channel by a sequence of @xmath0-continuous channels , then the results we derived are still valid for the `` true '' channel . this is the reason we adopt the @xmath0-continuity in this paper . + * continuous alphabets are possible . + in molecular communication , it can happen that the input and output alphabets are continuous ( a classical example being the transmission of concentration waveforms over fluid media @xcite ) . we point out that there do exist works on the @xmath0-continuous channel with continuous input and output alphabets @xcite . the extension of our results to continuous alphabets is therefore possible . + * why ergodic sources ? + the notion of ergodicity comes from the statistical physics , where the introduce of ergodic process is used to describe a closed physical system . since a large application area of molecular communication is to monitor the physical systems occurring in biological world , we are naturally lead to consider the ergodic sources . besides , in most cases of communication a transmitter is itself a closed physical system , thus justifying the use of ergodicity . last but not least , the ergodic sources have so many strong mathematical properties that we could barely derive anything without it . we just ca nt help but using it ( and so do many great predecessors in information theory ) . + * why @xmath95 , @xmath128 , and @xmath97 ? + @xmath95 clearly has its importance in coding theory . the importance of @xmath97 and @xmath128 is not so obvious , as we illustrated below . + a major application of molecular communication is the monitoring of biological processes ( e.g. , the output signal of a human organ ) , which is usually assumed to be ergodic . for such scenarios , our communication problem at hand is `` given an ergodic source @xmath30 $ ] and a diffusion - based molecular channel @xmath52 $ ] , what is the best we can do ? '' the question is answered in a very strong sense by the definition of @xmath97 and * theorem 6 * : a good source / channel code for @xmath30 $ ] exists if @xmath365 ; if @xmath366 , no good source / channel code exist . + the reason for taking @xmath128 into consideration is of practical concern . suppose that we try to really find out the channel capacity of the permutation channel @xmath251 . to this end , the direct calculation of @xmath367 , @xmath368 , and @xmath369 is impractical , since the calculation of @xmath368 and @xmath369 involves all combinatorial terms for each @xmath370 ( i.e. , for each @xmath370 and a realization @xmath371 and @xmath372 , we must consider all @xmath249 for which @xmath373 ) and the passing of limit @xmath374 . on contrast , since @xmath128 is defined through the _ sample _ mutual information @xmath117 , it suffices to choose a large @xmath370 , calculate the expectation of @xmath375 , and the convergence of @xmath117 in @xmath376 and the shannon - mcmillan theorem will guarantee the closeness of @xmath375 and @xmath377 . as a promising paradigm for nano - communication , the molecular communication has been developed over the past decade . in this paper , we consider a major subclass of molecular communication systems called the diffusion - based molecular systems . solid mathematical foundations for information theory are laid down for diffusion - based molecular communication . in particular , we have created an abstract channel model capturing all the essential features of diffusion - based molecular systems , and for such channel , the capacity theorems and channel coding theorems are proven . other useful notions concerning stationarity and ergodicity of diffusion - based molecular channel are also established . although the generalization is possible , here we shall confine ourselves to a special case which suffices our purpose . for channels with finite input and output alphabets , the cascade of adima channels is an adima channel . it suffices to consider the cascade of two adima channels . the case for cascade of arbitrary number follows from induction . let @xmath64 $ ] and @xmath65 $ ] be two adima channels with finite input and output alphabets and let @xmath52 $ ] be their cascade . fix @xmath141 . since @xmath378 is with adima , for any @xmath19 there are integers @xmath287 and @xmath379 such that for any @xmath380 , @xmath381 if @xmath382 for @xmath383 . write @xmath384 by ( 3 ) , we have @xmath385 put @xmath386 . then we can further write @xmath387 therefore , @xmath388 if @xmath51 and @xmath145 are chosen with integers @xmath18 and @xmath389 such that @xmath136 for @xmath390 implies @xmath391 is small , where the condition can be achieved since @xmath64 $ ] is with adima . the proof is completed by noting that all @xmath153 is a finite union of @xmath380 . the authors would like to thank prof . ian f. akyildiz for his inspring advices during a meeting in taiwan . the authors would also like to thank prof . kwang - cheng chen , prof . chia - han lee , po - jen shih , and yen - chi lee for their contribution to this paper .

traffic in content delivery networks exhibits strong temporal variability , resulting in congestion during peak hours and resource underutilization during off - peak hours . it is therefore desirable to try to `` shift '' some of the traffic from peak to off - peak hours . one approach to achieve this is to exploit idle network resources to duplicate some of the content in memories distributed across the network . this duplication of content is called content placement or caching . the duplicated content can then be used during peak hours to reduce network congestion . from the above description , it is apparent that the network operates in two different phases : a content placement phase and a content delivery phase . in the placement phase , the network is not congested , and the system is constrained mainly by the size of the cache memories . in the delivery phase , the network is congested , and the system is constrained mainly by the rate required to serve the content requested by the users . the goal is thus to design the placement phase such that the rate in the delivery phase is minimized . there are two fundamentally different approaches , based on two distinct understandings of the role of caching , for how the placement and the delivery phases are performed . * _ providing content locally : _ in the first , conventional , caching approach , replication is used to make part of the requested content available close to the end users . if a user finds part of a requested file in a close - by cache memory , that part can be served locally . the central content server only sends the remaining file parts using simple orthogonal unicast transmissions . if more than one user requests the same file , then the server has the option to multicast a single stream to those users . + extensive research has been done on this conventional caching approach , mainly on how to exploit differing file popularities to maximize the caching gain @xcite . the gain of this approach is proportional to the fraction of the popular content that can be stored locally . as a result , this conventional caching approach is effective whenever the local cache memory is large enough to store a significant fraction of the total popular content . * _ creating simultaneous coded - multicasting opportunities : _ in this approach , which we recently proposed in @xcite , content is placed in order to allow the central server to satisfy the requests of several users with _ different _ demands with a _ single _ multicast stream . the multicast streams are generated by coding across the different files requested by the users . each user exploits the content stored in the local cache memory to enable decoding of its requested file from these data streams . since the content placement is performed before the actual user demands are known , it has to be designed carefully such that these coded - multicasting opportunities are available simultaneously for all possible requests . + in @xcite , we show that this simultaneous coded - multicasting gain can significantly reduce network congestion . moreover , for many situations , this approach results in a much larger caching gain than the one obtained from the conventional caching approach discussed above . unlike the conventional approach , the simultaneous coded - multicast approach is effective whenever the aggregate _ global _ cache size ( i.e. , the cumulative cache available at all users ) is large enough compared to the total amount of popular content , even though there is no cooperation among the caches . as mentioned above , the scheme proposed in @xcite , relies on a carefully designed placement phase in order to create coded - multicasting opportunities among users with different demands . a central server arranges the caches such that every subset of the cache memories shares a specific part of the content . it is this carefully arranged overlap among the cache memories that guarantees the coded - multicasting opportunities simultaneously for all possible user demands . while the assumption of a centrally coordinated placement phase was helpful to establish the new caching approach in @xcite , it limits its applicability . for example , the identity or even just the number of active users in the delivery phase may not be known several hours in advance during the placement phase . as another example , in some cases the placement phase could be performed in one network , say a wifi network , to reduce congestion in the delivery phase in another network , say a cellular network . in either case , coordination in the placement phase may not be possible . this raises the important question whether lack of coordination in the placement phase eliminates the significant rate reduction promised by the simultaneous coded - multicast approach proposed in @xcite . put differently , the question is if simultaneous coded - multicasting opportunities can still be created without a centrally coordinated placement phase . in this paper , we answer this question in the positive by developing a caching algorithm that creates simultaneous coded - multicasting opportunities without coordination in the placement phase . more precisely , the proposed algorithm is able to operate in the placement phase with an unknown number of users situated in isolated networks and acting independently from each other . thus , the placement phase of the proposed algorithm is _ decentralized_. in the delivery phase , some of these users are connected to a server through a shared bottleneck link . in this phase , the server is first informed about the set of active users , their cache contents , and their requests . the proposed algorithm efficiently exploits the multicasting opportunities created during the placement phase in order to minimize the rate over the shared bottleneck link . we show that our proposed decentralized algorithm can significantly improve upon the conventional uncoded scheme . moreover , we show that the performance of the proposed decentralized coded caching scheme is close to the one of the centralized coded scheme of @xcite . users connected to a server storing @xmath0 files through a shared bottleneck link . the horizontal axis is the size of the cache memory ( normalized by the file size ) at each user ; the vertical axis shows the peak rate ( again normalized by the file size ) over the shared link in the delivery phase . the dashed green curve depicts the rate achieved by the conventional uncoded caching scheme advocated in the prior literature . the solid black curve depicts the rate achieved by the decentralized coded caching scheme proposed in this paper . the dashed blue curve depicts the rate achieved by the centralized coded caching algorithm from the recent paper @xcite . ] these two claims are illustrated in fig . [ fig : intro ] for a system with @xmath1 users and @xmath0 pieces of content . for example , when each user is able to cache @xmath2 of the files , the peak rate of the conventional uncoded scheme is equivalent to transmitting @xmath3 files . however , in the proposed decentralized coded scheme , the peak rate is equivalent to transmitting only about @xmath4 files . by comparing this to the performance of the centralized coded scheme , we can see that the rate penalty for decentralization of the placement phase of the caching system is modest . the remainder of this paper is organized as follows . section [ sec : problem ] formally introduces the problem setting . section [ sec : main_proposed ] presents the proposed algorithm . in section [ sec : performance ] , the performance of the proposed algorithm is evaluated and compared with the uncoded and the centralized coded caching schemes . in section [ sec : extensions ] , the results are extended to other topologies of practical interest . section [ sec : discussions ] discusses various implications of the results . to gain insight into how to optimally operate content - distribution systems , we introduce here a basic model for such systems capturing the fundamental challenges , tensions , and tradeoffs in the caching problem . for the sake of clarity , we initially study the problem under some simplifying assumptions , which will be relaxed later , as is discussed in detail in sections [ sec : extensions ] and [ sec : discussions ] . we consider a content - distribution system consisting of a server connected through an error - free shared ( bottleneck ) link to @xmath5 users . the server stores @xmath6 files each of size @xmath7 bits . the users each have access to a cache able to store @xmath8 bits for @xmath9 $ ] . this scenario is illustrated in fig . [ fig : setting ] . files of size @xmath7 bits each is connected through a shared link to @xmath5 users each with a cache of size @xmath8 bits . in the figure , @xmath10 and @xmath11 . ] the system operates in two phases : a _ placement _ phase and a _ delivery _ phase . the placement phase occurs when the network load is low . during this time , the shared link can be utilized to fill the caches of the users . the main constraint in this phase is the size of the cache memory at each user . the delivery phase occurs after the placement phase when the network load is high . at this time , each user requests one file from the server , which proceeds to transmit its response over the shared link . given the output of the shared link ( observed by all users ) and its cache content , each user should be able to recover its requested file . the main constraint in this phase is the load of the shared link . the objective is to minimize the worst - case ( over all possible requests ) rate of transmission over the shared link that is needed in the delivery phase . we now formalize this problem description . in the placement phase , each user is able to fill its cache as an arbitrary function ( linear , nonlinear , ) of the @xmath6 files subject only to its memory constraint of @xmath8 bits . we emphasize that the requests of the users are not known during the placement phase , and hence the caching function is not allowed to depend on them . in the delivery phase , each of the @xmath5 users requests one of the @xmath6 files and communicates this request to the server . let @xmath12 be the request of user @xmath13 . the server replies to these requests by sending a message over the shared link , which is observed by all the @xmath5 users . let @xmath14 be the number of bits in the message sent by the server . we impose that each user is able to recover its requested file from the content of its cache and the message received over the shared link with probability arbitrary close to one for large enough file size @xmath7 . denote by @xmath15 the worst - case normalized rate for a caching scheme . our objective is to minimize @xmath16 in order to minimize the worst - case network load @xmath17 during the delivery phase . clearly , @xmath16 is a function of the cache size @xmath8 . in order to emphasize this dependence , we will usually write the rate as @xmath18 . the function @xmath18 expresses the _ memory - rate tradeoff _ of the content - distribution system . the following example illustrates the definitions and notations and introduces the uncoded caching scheme advocated in most of the prior literature . this uncoded caching scheme will be used as a benchmark throughout the paper . [ eg : conventional ] consider the caching scenario with @xmath19 files and @xmath20 users each with a cache of size @xmath11 . in the uncoded caching scheme , each of the two files @xmath21 and @xmath22 are split into two parts of equal size , namely @xmath23 and @xmath24 . in the placement phase , both users cache @xmath25 , i.e. , the first part of each file . since each of these parts has size @xmath26 , this satisfies the memory constraint of @xmath27 bits . consider now the delivery phase of the system . assume that each user requests the same file @xmath21 , i.e. , @xmath28 . the server responds by sending the file part @xmath29 . clearly , from their cache content and the message received over the shared link , each user can recover the requested file @xmath23 . the ( normalized ) rate in the delivery phase is @xmath30 . assume instead that user one requests file @xmath21 and user two requests file @xmath22 , i.e. , @xmath31 and @xmath32 . the server needs to transmit @xmath33 to satisfy these requests , resulting in a rate in the delivery phase of @xmath34 . it is easy to see that this is the worst - case request , and hence @xmath35 for this scheme . for general @xmath6 , @xmath5 , and @xmath36 , the uncoded scheme caches the first @xmath37 fraction of each of the @xmath6 files . therefore , in the delivery phase , the server has to send the remaining @xmath38 fraction of the requested files . the resulting rate in the delivery phase , denoted by @xmath39 for future reference , is @xmath40 for @xmath41 and @xmath11 , this yields @xmath42 , as before . as we will see , this conventional caching scheme can be significantly improved upon . in particular , see example [ eg : illustration ] in section [ sec : main_proposed ] . one important feature of the uncoded scheme introduced in example [ eg : conventional ] is that it has a _ decentralized _ placement phase . by that we mean that the cache of each user is filled independently of other users . in particular , the placement operation of a given user neither depends on the identity nor the number of other users in the system . as a result , the users could , in fact , contact different servers at different times for the placement phase . having a decentralized placement phase is thus an important robustness property for a caching system . this is discussed further in sections [ sec : main_proposed ] and [ sec : extensions ] . as was mentioned earlier , the system description introduced in this section makes certain simplifying assumptions . in particular , we assume a system having a single shared broadcast link , with a cache at each user , and we focus on worst - case demands , with synchronized user requests in the delivery phase . all these assumptions can be relaxed , as is discussed in sections [ sec : extensions ] and [ sec : discussions ] . we now present a new algorithm ( referred to as decentralized coded caching scheme in the following ) for the caching problem . in the statement of the algorithm , we use the notation @xmath43 to denote the bits of the file @xmath44 requested by user @xmath45 cached exclusively at users in @xmath46 . in other words , a bit of file @xmath44 is in @xmath43 if it is present in the cache of every user in @xmath46 and if it is absent from the cache of every user outside @xmath46 . we also use the notation @xmath47 \defeq \{1,2,\dots , k\}$ ] and @xmath48 \defeq \{1,2,\dots , n\}$ ] . the proposed algorithm consists of a placement procedure and two delivery procedures . in the placement phase , we always use the same placement procedure . in the delivery phase , the server chooses the delivery procedure minimizing the resulting rate over the shared link . user @xmath45 caches a random @xmath49-bit subset of file @xmath50 [ alg:1_cache ] [ alg:1_sloop ] [ alg:1_sloop ] server sends @xmath51 [ alg:1_send ] server sends enough random linear combinations of bits in file @xmath50 for all users requesting it to decode [ alg:1_send2 ] [ rem : alg ] the @xmath52 operation in line [ alg:1_send ] of algorithm [ alg:1 ] represents the bit - wise xor operation . all elements @xmath53 are assumed to be zero padded to the length of the longest element . we illustrate the algorithm [ alg:1 ] with a small example . [ eg : illustration ] consider the caching problem with @xmath19 files @xmath21 and @xmath22 , and @xmath20 users each with a cache of size @xmath36 . in the placement phase of algorithm [ alg:1 ] , each user caches a subset of @xmath54 bits of each file independently at random . as a result , each bit of a file is cached by a specific user with probability @xmath55 . let us focus on file @xmath21 . the actions of the placement procedure effectively partition file @xmath21 into @xmath56 subfiles , @xmath57 where , for @xmath58 , @xmath59 denotes the bits of file @xmath21 that are stored in the cache memories of users in @xmath46 . as shorthand for @xmath60 . similarly , we write @xmath61 for @xmath62 . ] for example , @xmath63 are the bits of @xmath21 available in the cache memories of users one and two , whereas @xmath64 are the bits of @xmath21 available exclusively in the cache memory of user one . by the law of large numbers , @xmath65 with probability approaching one for large enough file size @xmath7 . therefore , we have with high probability : * @xmath66 is approximately @xmath67 . * @xmath68 and @xmath69 are approximately @xmath70 . * @xmath71 is approximately @xmath72 . the same analysis holds for file @xmath22 . we now consider the delivery phase in algorithm [ alg:1 ] . as we will see later ( see remark [ rem : thm ] below ) , for the scenario at hand the first delivery procedure will be used . assume that user one requests file @xmath21 and user two requests file @xmath22 . the iteration in line [ alg:1_sloop ] of algorithm [ alg:1 ] starts with @xmath73 . by line [ alg:1_sloop ] , this implies that we consider the set @xmath74 . observe that : * the cache of user two contains @xmath75 , which is needed by user one . hence , @xmath76 . * the cache of users one contains @xmath77 , which is needed by user two . hence , @xmath78 . as a result , in line [ alg:1_send ] of algorithm [ alg:1 ] , the server transmits @xmath79 over the shared link . user one can solve for @xmath75 from the received message @xmath79 and the cached subfile @xmath77 . user two can solve for @xmath77 from the message @xmath79 and the cached subfile @xmath75 . therefore , @xmath79 is simultaneously useful for @xmath73 users . thus , even though the two users request different files , the server can successfully multicast useful information to both of them . we note that the normalized ( by @xmath7 ) size of @xmath80 is @xmath70 . the second iteration in line [ alg:1_sloop ] is for @xmath81 . in this iteration , the server simply sends @xmath82 and @xmath83 in line [ alg:1_sloop ] . each of these transmissions has normalized size @xmath67 . from @xmath75 computed in iteration one , @xmath84 received in iteration two , and its cache content @xmath85 , user one can recover the requested file @xmath86 . similarly , user two can recover the requested file @xmath22 . summing up the contributions for @xmath73 and @xmath81 , the aggregate size ( normalized by @xmath7 ) of the messages sent by the server is @xmath87 this can be rewritten as @xmath88 in particular , for @xmath11 , the rate of algorithm [ alg:1 ] is @xmath89 . this compares to a rate of @xmath42 achieved by the uncoded caching scheme described in example [ eg : conventional ] in section [ sec : problem ] . while the improvement in this scenario is relatively small , as we will see shortly , for larger values of @xmath6 and @xmath5 , this improvement over the uncoded scheme can be large . [ rem : knownk ] the placement procedure of algorithm [ alg:1 ] is _ decentralized _ , in the sense that the user s caches are filled independently of each other . this implies that neither the identity nor even the number of users sharing the same bottleneck link during the delivery phase need to be known during the earlier placement phase . this decentralization of the placement phase enables the content - distribution system to be much more flexible than a centralized placement phase . this flexibility is essential . for example , in wireline networks , some of the users may not request any file in the delivery phase . in wireless networks , users may move from one network or cell to another , and hence might not even be connected to the same server in the two phases . in either case , the result is that the precise number and identity of users in the delivery phase is unknown in the placement phase . one of the salient features of the decentralized algorithm proposed in this paper is that it can easily deal with these situations . this flexibility is also crucial to deal with asynchronous user requests , as is explained in detail in section [ sec : extensions_synchronization ] . it is also a key ingredient to extending the coded - caching approach to scenarios with nonuniform demands or with online cache updates , as is discussed further in section [ sec : discussions ] and in the follow - up works @xcite and @xcite . [ _ greedy coding strategy _ ] the first delivery procedure in algorithm [ alg:1 ] follows a _ greedy _ strategy . it first identifies and forms coded messages that are useful for all @xmath90 users . in the next iteration , it forms coded messages that are useful for a subset of @xmath91 users . the iteration continues until it identifies messages that are useful for only @xmath81 user . [ _ simplified decision rule _ ] [ remark : m ] algorithm [ alg:1 ] provides two delivery procedures . the general rule is to choose the procedure which minimizes the resulting rate over the shared link . a simple alternative rule to decide between these two procedures is as follows : if @xmath92 , employ the first procedure ; otherwise , employ the second procedure . the performance loss due to this simpler rule can be shown to be small . with some minor modifications . ] we now analyze the performance of the proposed decentralized coded caching scheme given by algorithm [ alg:1 ] . section [ sec : performance_rate ] provides an analytic expression for the rate of algorithm [ alg:1 ] . section [ sec : performance_decentralized ] compares the proposed decentralized coded caching scheme with the decentralized uncoded caching scheme from example [ eg : conventional ] ( the best previously known decentralized caching scheme ) . section [ sec : performance_centralized ] compares the proposed decentralized coded caching scheme with the optimal centralized caching scheme and the caching scheme from @xcite ( the best known centralized caching scheme ) . the performance of decentralized coded caching is analyzed in the next theorem , whose proof can be found in appendix [ sec : proofs_achievable1 ] . [ thm : achievable1 ] consider the caching problem with @xmath6 files each of size @xmath7 bits and with @xmath5 users each having access to a cache of size @xmath8 bits with @xmath93 $ ] . algorithm [ alg:1 ] is correct and , for @xmath7 large enough , achieves rate arbitrarily close to @xmath94 [ rem : thm ] we note that if @xmath95 or @xmath96 , then the minimum in @xmath97 is achieved by the first term so that @xmath98 this is the rate of the first delivery procedure in algorithm [ alg:1 ] . since @xmath95 or @xmath99 is the regime of most interest , the majority of the discussion in the following focuses on this case . [ rem : zero ] theorem [ thm : achievable1 ] is only stated for @xmath100 . for @xmath101 algorithm [ alg:1 ] is easily seen to achieve a rate of @xmath102 we see that @xmath103 is the continuous extension of @xmath97 for @xmath104 . to simplify the exposition , we will not treat the case @xmath101 separately in the following . we point out that the rate @xmath97 of algorithm [ alg:1 ] consists of three distinct factors . the first factor is @xmath5 ; this is the rate without caching . the second factor is @xmath105 ; this is a _ local _ caching gain that results from having part of the requested file already available in the local cache . the third factor is a _ gain that arises from using the caches to create simultaneous coded - multicasting opportunities . see example [ eg : illustration ] in section [ sec : main_proposed ] for an illustration of the operational meaning of these three factors . it is instructive to examine the performance of the proposed decentralized coded caching scheme ( algorithm [ alg:1 ] ) for large and small values of cache size @xmath36 . for simplicity , we focus here on the most relevant case @xmath106 , i.e. , the number of files is at least as large as the number of users , so that the rate @xmath97 of algorithm [ alg:1 ] is given by remark [ rem : thm ] . as a baseline , we compare the result with the uncoded caching scheme , introduced in example [ eg : conventional ] in section [ sec : problem ] . this is the best previously known algorithm with a decentralized placement phase . for @xmath95 , the uncoded scheme achieves the rate @xmath107 which is linear with slope of @xmath108 throughout the entire range of @xmath36 . for small cache size @xmath109 $ ] , the rate achieved by algorithm [ alg:1 ] behaves approximately , and by analyzing the constant in the @xmath110 expression it can be shown that this is a good approximation in the regime @xmath109 $ ] . ] as @xmath111 in this regime , @xmath97 scales approximately linearly with the memory size @xmath36 with slope @xmath112 : increasing @xmath36 by one decreases the rate by approximately @xmath113 . this is illustrated in fig . [ fig : regimes ] . comparing and , we have the following observations : * _ order-@xmath5 improvement in slope : _ the slope of @xmath97 around @xmath101 is approximately @xmath114 times steeper than the slope of @xmath39 . thus , the reduction in rate as a function of @xmath36 is approximately @xmath114 times faster for algorithm [ alg:1 ] than for the uncoded scheme . in other words , for small @xmath36 the scheme proposed here makes approximately @xmath114 times better use of the cache resources : an improvement on the order of the number of users in the system . this behavior is clearly visible in fig . [ fig : intro ] in section [ sec : intro ] . * _ virtual shared cache : _ consider a hypothetical setting in which the @xmath5 cache memories are collocated and shared among all @xmath5 users . in this hypothetical system , arising from allowing complete cooperation among the @xmath5 users , it is easy to see that the optimal rate is @xmath115 . comparing this to , we see that , up to a factor @xmath116 , the proposed decentralized coded caching scheme achieves the same order behavior . therefore , this scheme is essentially able to create a single virtually shared cache , even though the caches are isolated without any cooperation between them . on the other hand , for @xmath117 $ ] we can approximate , we have @xmath118 in the regime @xmath117 $ ] , and the pre - constant in the order notation converges to @xmath119 as @xmath120 . ] @xmath121 in this regime , the rate achieved by algorithm [ alg:1 ] scales approximately inversely with the memory size : doubling @xmath36 approximately halves the rate . this is again illustrated in fig . [ fig : regimes ] . comparing and , we have the following observation : * _ order-@xmath5 improvement in rate : _ in this regime , the rate of the proposed decentralized coded scheme can be up to a factor @xmath5 smaller than the uncoded scheme : again an improvement on the order of the number of users in the system . this behavior is again clearly visible in fig . [ fig : intro ] in section [ sec : intro ] . achieved by algorithm [ alg:1 ] for @xmath122 files and @xmath123 users ( see theorem [ thm : achievable1 ] ) . the function @xmath97 behaves approximately linearly for @xmath109 $ ] and behaves approximately as @xmath124 for @xmath117 $ ] ( both approximations are indicated by dotted curves ) . ] we have compared the performance of the proposed decentralized coded caching scheme to the uncoded caching scheme , which is the best previously known _ decentralized _ algorithm for this setting . we now compare the decentralized coded caching scheme to the rate achievable by _ centralized _ caching schemes . we start with an information - theoretic lower bound on the rate of _ any _ centralized caching scheme . we then consider the rate of the best known centralized caching scheme recently introduced in @xcite . [ thm : comparison_converse ] let @xmath97 be the rate of the decentralized coded caching scheme given in algorithm [ alg:1 ] , and let @xmath125 be the rate of the optimal centralized caching scheme . for any number of files @xmath6 and number of users @xmath5 and for any @xmath9 $ ] , we have @xmath126 the proof of theorem [ thm : comparison_converse ] , presented in appendix [ sec : proofs_comparison_converse ] , uses an information - theoretic argument to lower bound the rate of the optimal scheme @xmath125 . as a result , theorem [ thm : comparison_converse ] implies that _ no _ scheme ( centralized , decentralized , with linear caching , nonlinear caching , ) regardless of is computational complexity can improve by more than a constant factor upon the efficient decentralized caching scheme given by algorithm [ alg:1 ] presented in this paper . theorem [ thm : comparison_converse ] implies that uncoded caching in the placement phase combined with linear greedy coding in the delivery phase is sufficient to achieve a rate within a constant factor of the optimum . we now compare the rate @xmath97 of decentralized coded caching to the rate @xmath127 of the best known centralized coded caching scheme . in ( * theorem 2 ) , @xmath127 is given by @xmath128 for @xmath129 , and the lower convex envelope of these points for all remaining values of @xmath9 $ ] . [ fig : intro ] in section [ sec : intro ] compares the performance @xmath127 of this centralized coded caching scheme to the performance @xmath97 of the decentralized coded caching scheme proposed here . as can be seen from the figure , the centralized and decentralized caching algorithms are very close in performance . thus , there is only a small price to be paid for decentralization . indeed , we have the following corollary to theorem [ thm : comparison_converse ] . [ thm : comparison_centralized ] let @xmath97 be the rate of the decentralized coded caching scheme given in algorithm [ alg:1 ] , and let @xmath127 be the rate of the centralized coded caching scheme from @xcite . for any number of files @xmath6 and number of users @xmath5 and for any @xmath9 $ ] , we have @xmath130 corollary [ thm : comparison_centralized ] shows that the rate achieved by the decentralized coded caching scheme given by algorithm [ alg:1 ] is at most a factor @xmath131 worse than the one of the best known centralized algorithm from @xcite . this bound can be tightened numerically to @xmath132 for all values of @xmath5 , @xmath6 , and @xmath36 . hence , the rate of the decentralized caching scheme proposed here is indeed quite close to the rate of the best known centralized caching scheme . it is instructive to understand why the decentralized scheme performs close to the centralized one . in the centralized scheme , content is placed in the placement phase such that in the delivery phase every message is useful for exactly @xmath133 users . in the decentralized scheme , we can not control the placement phase as accurately . however , perhaps surprisingly , the number of messages that are useful for about @xmath133 users is nevertheless the dominant term in the overall rate of the decentralized scheme . more precisely , from the proof of theorem [ thm : achievable1 ] in appendix [ sec : proofs_achievable1 ] , we can write the rate @xmath97 of the decentralized coded scheme as a convex combination of the rate @xmath127 of the centralized coded scheme : @xmath134 with @xmath135 and @xmath136 . the dominant term in this sum is @xmath127 occurring at @xmath137 , as is illustrated in fig . [ fig : concentration ] . this observation explains why the centralized and the decentralized schemes have approximately the same rate . in this section , we extend the results presented so far to some important cases arising in practical systems . in particular , we show how to handle networks with tree topologies in section [ sec : extensions_tree ] , caches shared by several users in section [ sec : extensions_shared ] , and users with asynchronous requests in section [ sec : extensions_synchronization ] . the basic problem setting considered so far considers users connected to the server through a single shared bottleneck link . we showed that the rate of our proposed algorithm over the shared link is within a constant factor of the optimum . here we extend this result to a more general network with tree structure ( see fig . [ fig : tree ] ) . files of size @xmath7 bits each is connected through a tree - structured network to @xmath5 users each with a cache of size @xmath8 bits . internal nodes of the tree represent routers . in this figure , @xmath138 , and @xmath11 . the proposed placement and delivery procedures together with a routing algorithm achieves the order - optimal rate over every link @xmath139 of the network . ] consider a directed tree network , oriented from the root to the leaves . the server is located at the root of the tree , and users with their caches are located at the leaves . each internal node of the network represents a router . the router decides what to transmit over each of its outgoing links as a function of what it received over its single incoming link from its parent . we again assume that the system operates in two phases . in the placement phase , the caches are populated without knowledge of users future demands . in the delivery phase , the users reveal their requests , and the server has to satisfy these demands exploiting the cached content . for this network , we propose the following caching and routing procedures . for the placement phase , we use the same placement procedure as in algorithm [ alg:1 ] . for the delivery phase , we use the two delivery procedures detailed in algorithm [ alg:1 ] , but with the simplified decision rule explained in remark [ remark : m ] . in other words , if @xmath140 , the server creates coded packets according to the first delivery procedure . if @xmath141 , it uses the second delivery procedure and the server creates linear combinations of the bits of each file without coding across different files . it remains to describe the operations of the routers at the internal nodes of the tree . each router operates according to the following simple rule . the router at node @xmath142 forwards a coded message over link @xmath139 if and only if that coded message is directly useful to at least one of the descendant leaves of node @xmath143 . to be precise , let us assume that @xmath140 so that the server uses the first delivery procedure . thus for each subset @xmath144 $ ] of users , the server creates the coded message @xmath51 . this coded message is useful for all users in @xmath46 and is completely useless for the remaining users . the router located at node @xmath142 forwards this coded message @xmath51 over the link @xmath139 , if at least one of the descendants of @xmath143 ( including @xmath143 itself if it is a leaf ) is in the set @xmath46 . a similar routing procedure is used for @xmath141 . the performance of this scheme is analyzed in appendix [ sec : proofs_tree ] . we show there that , for @xmath145 , the rate of this scheme over the link @xmath139 is equal to @xmath146 where @xmath147 is the number of descendant leaves of node @xmath143 . for @xmath148 , it is easy to see that the rate over the link @xmath139 is equal to @xmath149 the rate over every link in the tree network can be shown to be within a constant factor of optimal . to prove approximate optimality for edge @xmath139 , we consider the subtree rooted at @xmath143 together with @xmath142 and the edge @xmath139 . we can then use the same bound used in theorem [ thm : comparison_converse ] over this subtree , treating @xmath139 as the shared bottleneck link . this result shows that , for tree - structured networks , caching and routing can be performed separately with at most a constant factor loss in performance compared to the optimal joint scheme . this means that the proposed placement and delivery procedures are universal in the sense that they do not depend on the topology of the tree network connecting the server to the caches at the leaves . [ eg : universal ] consider @xmath5 users connected to a server through @xmath5 orthogonal links ( i.e. , no shared links ) . for this topology the optimal rate over each link can be achieved without coding . however , it is easy to see that the proposed universal scheme achieves the same optimal rate . thus , depending on the network topology , we may be able to develop simpler schemes , but the performance will be the same up to a constant factor as the proposed universal scheme . [ sec : private ] consider the original scenario of users sharing a single bottleneck link . as an example , assume we have @xmath19 files and @xmath20 users as described in example [ eg : illustration ] in section [ sec : main_proposed ] . observe that a user does not need all messages sent by the server over the shared link in order to recover its requested file . for example , in order to recover file @xmath21 , user one only needs @xmath79 and @xmath84 . thus , if a router is located right where the shared link splits into the two private links , it can forward only these two messages over the private link to user one . by the analysis in this section , the resulting normalized rate over the private link to user one is then @xmath150 we note that in the uncoded scheme the rate over each private link is also @xmath38 . hence , we see that by proper routing the rate over the private links for both the coded as well as the uncoded schemes are the same . the reduction of rate over the shared link achieved by coding does therefore not result in an increase of rate over the private links . this conclusion holds also for general values of @xmath6 , @xmath5 , and @xmath36 . the problem setting considered throughout this paper assumes that each user has access to a private cache . in this example , we evaluate the gain of shared caches . this situation arises when the cache memory is located close to but not directly at the users . we consider a system with @xmath5 users partitioned into subsets , where users within the same subset share a common cache ( see fig . [ fig : shared ] ) . for simplicity , we assume that these subsets have equal size of @xmath151 users , where @xmath151 is a positive integer dividing @xmath5 . we also assume that the number of files @xmath6 is greater than the number of caches @xmath152 . to keep the total amount of cache memory in the system constant , we assume that each of the shared caches has size @xmath153 bits . files of size @xmath7 bits each is connected through a shared link to @xmath5 users each with a cache of size @xmath8 bits . users are partitioned into equal subsets each with @xmath151 users . users in each subsets share their caches . in this figure , @xmath154 , @xmath155 , and @xmath156 . ] we can operate this system as follows . define @xmath152 super users , one for each subset of @xmath151 users . run the placement procedure of algorithm [ alg:1 ] for these @xmath152 users with cache size @xmath153 . in the delivery phase , treat the ( up to ) @xmath151 files requested by the users in the same subset as a single super file of size @xmath157 . applying theorem [ thm : achievable1 ] to this setting yields an achievable rate of in theorem [ thm : achievable1 ] by replacing @xmath36 by @xmath158 ( since each cache has now size @xmath153 instead of @xmath8 ) , replacing @xmath5 by @xmath152 ( since there are @xmath152 super users ) , and multiplying the result by an extra factor of @xmath151 ( since each super file is @xmath151 times the size of a normal file ) . ] @xmath159 let us again consider the regimes of small and large values of @xmath36 of @xmath160 . for @xmath109 $ ] , we have @xmath161 comparing this to the small-@xmath36 approximation of @xmath97 ( for a system with private caches ) , we see that @xmath162 i.e. , there is only a small effect on the achievable rate from sharing a cache . this should not come as a surprise , since we have already seen in section [ sec : performance_decentralized ] that for small @xmath36 , @xmath97 behaves almost like a system in which all @xmath5 caches are combined . hence , there is no sizable gain to be achieved by having collaboration among caches in this regime . consider then the regime @xmath163 . here , we have @xmath164 and from @xmath165 the difference between the two approximations is only in the second factor . we recall that this second factor represents the caching gain due to making part of the files available locally . quite naturally , this part of the caching gain improves through cache sharing , as a larger fraction of each file can be stored locally . up to this point , we have assumed that in the delivery phase all users reveal their requests simultaneously , i.e. , that the users are perfectly synchronized . in practice , however , users reveal their requests at different times . in this example , we show that the proposed algorithm can be modified to handle such asynchronous user requests . we explain the main idea with an example . consider a system with @xmath166 files @xmath167 , and @xmath168 users . we split each file into @xmath169 consecutive segments , e.g. , @xmath170 and similarly for @xmath22 and @xmath171 . here @xmath169 is a positive integer selected depending on the maximum tolerable delay , as will be explained later . to be specific , we choose @xmath172 in this example . and @xmath173 , respectively . ] in the placement phase , we simply treat each segment as a file . we apply the placement procedure of algorithm [ alg:1 ] . for the delivery phase , consider an initial request @xmath174 from user one , say for file @xmath21 . the server responds by starting delivery of the first segment @xmath175 of file @xmath21 . meanwhile , assume that user two requests file @xmath176 , say @xmath22 , as shown in fig . [ fig : segments ] . the server puts the request of user two on hold , and completes the delivery of @xmath175 for user one . it then starts to deliver the second segment @xmath177 of @xmath21 and the first segment @xmath178 of @xmath22 using the delivery procedure of algorithm [ alg:1 ] for two users . delivery of the next segments @xmath179 and @xmath180 is handled similarly . assume that at this point user three requests file @xmath181 , say @xmath171 , as shown in fig . [ fig : segments ] . after completing the current delivery phase , the server reacts to this request by delivering @xmath182 , @xmath183 , and @xmath184 to users one , two , and three , respectively , using the delivery procedure of algorithm [ alg:1 ] for three users . the process continues in the same manner as depicted in fig . [ fig : segments ] . we note that users two and three experience delays of @xmath185 and @xmath173 as shown in the figure . the maximum delay depends on the size of the segments . therefore , segment size , or equivalently the value of @xmath169 , can be adjusted to ensure that this delay is tolerable . we point out that the number of effective users in the system varies throughout the delivery phase . due to its decentralized nature , the proposed caching algorithm is close to optimal for any value of users as discussed in remark [ rem : knownk ] . this is instrumental for the segmentation approach just discussed to be efficient . caching random linear combinations of file segments is a popular prefetching scheme . in this example , we argue that in some scenarios this form of caching can be quite inefficient . to be precise , let us focus on a specific scenario with @xmath5 users and @xmath186 files , where each user has sufficient cache memory to store half of the files , i.e. @xmath187 . according to theorem [ thm : achievable1 ] , algorithm [ alg:1 ] achieves a rate of less than one , i.e. , @xmath188 . on the other hand , the rate achieved by caching of random linear combinations can be shown to be at most @xmath189 , which is significantly larger than @xmath97 for large number of users @xmath5 . indeed , assume that user one requests file @xmath21 . recall that each user has cached @xmath26 random linear combinations of file @xmath21 . with high probability , these random linear combinations span a @xmath26-dimensional space at each user and the subspaces of different users do not overlap . for example , consider users two and three . as a consequence of this lack of overlap , these two users do not have access to a shared part of the file @xmath21 . this implies that , in the delivery phase , the server can not form a linear combination that is simultaneously useful for three users . in other words , the server can form messages that are at most useful simultaneously for up to two users . a short calculation reveals that then the server has to send at least @xmath190 bits over the shared link . this inefficiency of caching random linear combinations can be interpreted as follows . the placement phase follows two contradicting objectives : the first objective is to spread the available content as much as possible over the different caches . the second objective , is to ensure maximum overlap among different caches . the system performance is optimized if the right balance between these two objectives is struck . caching random linear combinations maximizes the spreading of content over the available caches , but provides minimal overlap among them . at the other extreme , the conventional scheme maximizes the overlap , but provides only minimal spreading of the content . as a consequence , both of these schemes can be highly suboptimal . our problem formulation focuses on worst - case requests . in some situation , this is the correct figure of merit . for example , in a wireless scenario , whenever the delivery rate required for a request exceeds the available link bandwidth , the system will be in outage , degrading user experience . in other situations , for example a wireline scenario , excess rates might only incur a small additional cost and hence might be acceptable . in such cases , the average rate is the right metric , especially when files have different popularities . this is discussed further in @xcite . in practical scenarios , the set of popular files is time varying . to keep the caching algorithm efficient , the cache contents have to be dynamically updated to track this variation . a popular scheme to update the caches is to evict the last - recently used ( lru ) file from the caches and replace it with a newly requested one . this lru eviction scheme is known to be approximately optimal for systems with a single cache @xcite . however , it is not efficient for networks with multiple caches as considered here . this problem of online caching with several caches is investigated in @xcite . the decentralized algorithm [ alg:1 ] presented in this paper turns out to be a crucial ingredient of the suggested online coded caching algorithm in @xcite . we first prove correctness . note that , since there are a total of @xmath6 files , the operations in line [ alg:1_cache ] of algorithm [ alg:1 ] satisfies the memory constraint of @xmath8 bits at each user . hence the placement phase of algorithm [ alg:1 ] is correct . for the delivery phase , assume the server uses the first delivery procedure , and consider a bit in the file requested by user @xmath45 . if this bit is already cached at user @xmath45 , it does not need to be sent by the server . assume then that it is cached at some ( possibly empty ) set @xmath191 of users with @xmath192 . consider the set @xmath193 in line [ alg:1_sloop ] . by definition , the bit under consideration is contained in @xmath53 , and as a consequence , it is included in the sum sent by the server in line [ alg:1_send ] . since @xmath194 for every other @xmath195 , user @xmath45 has access to all bits in @xmath196 from its own cache . hence , user @xmath45 is able to recover the requested bit from this sum . this shows that the first delivery procedure is correct . the second delivery procedure is correct as well since the server sends in line [ alg:1_send2 ] enough linear combinations of every file for all users to successfully decode . this shows that the delivery phase of algorithm [ alg:1 ] is correct . it remains to compute the rate . we start with the analysis of the second delivery procedure . if @xmath197 , then in the worst case there is at least one user requesting every file . consider then all users requesting file @xmath50 . recall that each user requesting this file already has @xmath198 of its bits cached locally by the operation of the placement phase . an elementary analysis reveals that with high probability for @xmath7 large enough at most @xmath199 random linear combinations need to be sent in line [ alg:1_send ] for all those users to be able to decode . we will assume that the file size @xmath7 is large and ignore the @xmath200 term in the following . since this needs to be done for all @xmath6 files , the normalized rate in the delivery phase is @xmath201 if @xmath202 , then there are at most @xmath5 different files that are requested . the same analysis yields a normalized rate of @xmath203 thus , the second procedure has a normalized rate of @xmath204 for @xmath93 $ ] . we continue with the analysis of the first delivery procedure . consider a particular bit in one of the files , say file @xmath50 . since the choice of subsets is uniform , by symmetry this bit has probability @xmath205\ ] ] of being in the cache of any fixed user . consider now a fixed subset of @xmath206 out of the @xmath5 users . the probability that this bit is cached at exactly those @xmath206 users is @xmath207 hence the expected number of bits of file @xmath50 that are cached at exactly those @xmath206 users is @xmath208 in particular , the expected size of @xmath53 with @xmath209 is @xmath210 moreover , for @xmath7 large enough the actual realization of the random number of bits in @xmath53 is in the interval @xmath211 with high probability . for ease of exposition , we will again ignore the @xmath200 term in the following . consider a fixed value of @xmath212 in line [ alg:1_sloop ] and a fixed subset @xmath46 of cardinality @xmath212 in line [ alg:1_sloop ] . in line [ alg:1_send ] , the server sends @xmath213 bits . since there are @xmath214 subsets @xmath46 of cardinality @xmath212 , the loop starting in line [ alg:1_sloop ] generates @xmath215 bits . summing over all values of @xmath212 yields a total of @xmath216 bits being sent over the shared link . substituting the definition of @xmath217 yields a rate of the first delivery procedure of @xmath218 for @xmath93 $ ] . since the server uses the better of the two delivery procedures , and show that algorithm [ alg:1 ] achieves a rate of @xmath219 using that @xmath220 this can be simplified to @xmath221 concluding the proof . recall from theorem [ thm : achievable1 ] that @xmath222 using that @xmath223 this can be upper bounded as @xmath224 for all @xmath93 $ ] . moreover , we have from ( * ? ? ? * theorem 2 ) @xmath225 we will treat the cases @xmath226 and @xmath227 separately . assume first that @xmath228 . by , @xmath229 and by with @xmath81 , @xmath230 hence @xmath231 for @xmath228 . assume in the following that @xmath227 . we consider the cases @xmath232 , \\ ( \max \{1 , n / k \ } , n/12 ] , \\ ( n/12 , n ] , \end{cases}\ ] ] separately . assume first that @xmath233 . by , @xmath234 on the other hand , by with @xmath235 , @xmath236 where in @xmath237 we have used that @xmath238 and @xmath239 . hence @xmath240 for @xmath241 and @xmath242 . assume then that @xmath243 . by , @xmath244 on the other hand , by with @xmath245 , @xmath246 where in @xmath237 we have used that @xmath247 and that @xmath248 . hence @xmath249 for @xmath250 and @xmath251 . finally , assume that @xmath252 . by , @xmath253 on the other hand , by with @xmath81 , @xmath254 hence , @xmath255 for @xmath250 and @xmath252 . combining , , , and yields that @xmath256 for all @xmath6 , @xmath5 , and @xmath257 . as is shown in appendix [ sec : proofs_achievable1 ] , for @xmath7 large enough the actual realization of the random number of bits in @xmath53 is in the interval @xmath211 with high probability , and where @xmath217 . as before , we will again ignore the @xmath200 term in the following . recall that only a subset of coded messages generated in line [ alg:1_sloop ] in algorithm [ alg:1 ] pass through link @xmath139 , namely only those @xmath258 for which the subset @xmath46 has at least one element among leave descendants of node @xmath143 . we split @xmath46 into @xmath259 and @xmath260 where @xmath259 is the subset of descendant leaves of node @xmath143 , and @xmath261 . denote the cardinalities of @xmath259 and @xmath260 by @xmath262 and @xmath263 , respectively , so that @xmath264 . with this , only coded messages with @xmath265 are forwarded over link @xmath139 . the number of bits sent over this link is then equal to @xmath266 substituting @xmath217 yields the desired result .

gdel s completeness theorem ( for first - order classical logic ) is one of the most important theorems in logic . it is concerned with a duality ( in a naive sense ) between proofs and models : for every proposition @xmath0 , _ either@xmath4or@xmath5 . _ here @xmath1 ranges over the set of proofs , @xmath2 over the class of models , and @xmath6 reads `` @xmath1 is a proof of @xmath0 . '' one can imagine a debate on a general proposition @xmath0 , where player tries to justify @xmath0 by giving a proof and opponent tries to refute it by giving a countermodel . the completeness theorem states that exactly one of them wins . actually , the theorem gives us far more insights than stated . -12 pt**finite proofs vs infinite models:**a very crucial point is that proofs are always finite , while models can be of arbitrary cardinality . completeness thus implies compactness and lwenheim - skolem theorems , leading to constructions of various nonstandard models . 2 . -12 pt**nondeterministic principles : * * proof of gdel s completeness theorem relies on a strong nondeterministic principle such as knig s or zorn s lemma , in contrast to the trivial completeness theorem with respect to the class of boolean algebras . -12 pt**matching of two inductions:**__provability _ _ is defined by induction on proofs , while _ truth _ by induction on formulas . the two inductions are somehow ascribed to the essence of syntax and semantics , respectively , and the completeness theorem states that they do match . unlike the real debate , however , there is no interaction between proofs and models in gdel s theorem . a more interactive account of completeness is given by girard s _ ludics _ ( @xcite ; see @xcite for good expositions ) . ludics is a variant of game semantics , which has the following prominent features . 1 . -12 pt**monism : * * proofs and models are not distinguished by their ontological status , but by their structural properties . the common objects are called _ designs_. 2 . -12 pt**existentialism : * * _ behaviours _ ( semantic types ) are built from designs , in contrast to the ordinary game semantics ( hyland - ong @xcite ) where one begins with the definition of arenas ( types ) and then proceeds to strategies ( proofs ) . -12 pt**normalization as interaction : * * designs ( hence proofs and models ) interact together via normalization . it induces an _ orthogonality relation _ between designs in such a way that @xmath7 holds if the normalization of @xmath1 applied to @xmath2 converges . a behaviour @xmath0 is defined to be a set of designs which is equivalent to its biorthogonal ( @xmath8 ) . in this setting , girard shows a completeness theorem for proofs @xcite , which roughly claims that any `` winning '' design in a behaviour is a proof of it . in view of the interactive definition of behaviour , it can be rephrased as follows : for every ( logical ) behaviour @xmath0 and every ( proof - like ) design @xmath1 , _ either@xmath6or@xmath9 and @xmath2 defeats @xmath1 ) . _ here , `` @xmath10 '' means @xmath11 , and `` @xmath2 defeats @xmath1 '' means @xmath12 . hence the right disjunct is equivalent to @xmath13 . namely , @xmath14 if and only if @xmath6 , that is a typical full completeness statement . notice that @xmath10 no more entails absolute unprovability of @xmath0 ( it is rather relativized to each @xmath1 ) , and there is a real interaction between proofs and models . actually , girard s original ludics is so limited that it corresponds to a polarized fragment of multiplicative additive linear logic , which is too weak to be a stand - alone logical system . as a consequence , one does not really observe an opposition between finite proofs and infinite models , since one can always assume that the countermodel @xmath2 is finite ( related to the finite model property for @xmath15 @xcite ) . indeed , proving the above completeness is easy once _ internal completeness _ ( a form of completeness which does not refer to any proof system @xcite ) for each logical connective has been established . in this paper , we employ a term syntax for designs introduced in @xcite , and extend girard s ludics with duplication ( contraction ) and its dual : universal nondeterminism ( see @xcite and references therein ) . although our term approach disregards some interesting locativity - related phenomena ( normalization as merging of orders and different sorts of tensors @xcite ) , our calculus is easier to manipulate and closer to the tradition of @xmath16 , @xmath17 , @xmath18 , @xmath19-calculi and other more recent syntaxes for focalized classical logic ( @xcite ) . our resulting framework is as strong as a polarized fragment of linear logic with exponentials ( @xcite ; see also @xcite ) , which is in turn as strong as a constructive version of classical propositional logic . we then prove the completeness theorem above in this extended setting . here , universal nondeterminism is needed on the model side to well interact with duplicative designs on the proof side . this is comparable to the need of noninnocent " ( and sometimes even nondeterministic ) opponents to have full completeness with respect to deterministic , but nonlinear player s strategies . unlike before , we can not anymore assume the finiteness of models , since they are not sufficient to refute infinite proof attempts . as a result , our proof is nontrivial , even after the internal completeness theorem has been proved . indeed , our proof exhibits a striking similarity with schtte s proof of gdel s completeness theorem @xcite . given a ( proof - like ) design @xmath1 which is not a proof of @xmath0 , we explicitly construct a countermodel @xmath2 in @xmath20 which defeats @xmath1 , essentially using knig s lemma . soundness is proved by induction on proofs , while completeness is by induction on types . thus our theorem gives matching of two inductions . finally , it implies an analogue of lwenheim - skolem theorem , ( and also the finite model property for the linear fragment ) , which well illustrates the opposition between finite proofs and infinite models with arbitrary cardinality . in game semantics , one finds a number of similar full completeness results . however , the connection with gdel s completeness seems less conspicuous than ours . typically , innocent strategies in hyland - ong games most naturally correspond to _ bhm trees _ , which can be _ infinite _ ( @xcite ) . thus , in contrast to our result , one has to impose finiteness / compactness on strategies in an external way , in order to have a correspondence with _ finite _ @xmath16-terms . although this is also the case in @xcite , we show that such a finiteness assumption is not needed in ludics : infinitary proof attempts are always defeated by infinitary models . the paper is organized as follows . in section [ s - prelim ] we describe the syntax of ( untyped ) designs ; in section [ s - behaviour ] we move to a typed setting and introduce behaviours ( semantic types ) . in section [ s - full ] we introduce our proof system and prove completeness for proofs . finally , section [ s - conclusion ] concludes the paper . in this paper , we employ a process calculus notation for designs , inspired by the close relationship between ludics and linear @xmath19-calculus @xcite . precisely , we extend the syntax introduced by the second author @xcite adding a ( universal ) nondeterministic choice operator @xmath21 . although @xcite mainly deals with linear designs , its syntax is designed to deal with nonlinear ones without any difficulty . however , in order to obtain _ completeness _ , we also need to incorporate the dual of nonlinearity , that is _ universal nondeterminism _ @xcite . it is reminiscent of differential linear logic @xcite , which has nondeterministic sum as the dual of contraction ; the duality is essential for the separation property @xcite ( see also @xcite for separation of bhm trees ) . a similar situation also arises in hyland - ong game semantics @xcite , where _ nonlinear _ strategies for player may contain a play in which opponent behaves _ noninnocently _ ; opponent s noninnocence is again essential for full completeness . designs are built over a given _ signature _ @xmath22 , where @xmath23 is a set of _ names _ @xmath24 and @xmath25 is a function which assigns to each name @xmath26 its _ arity _ @xmath27 . let @xmath28 be a countable set of variables @xmath29 . over a fixed signature @xmath30 , a _ positive action _ is @xmath31 with @xmath32 , and a _ negative action _ is @xmath33 where variables @xmath34 are distinct and @xmath35 . we often abbreviate a sequence of variables @xmath36 by @xmath37 . in the sequel , we always assume that an expression of the form @xmath38 stands for a negative action , @xmath39 and @xmath37 is a sequence consisting of @xmath40 distinct variables . if @xmath26 is a nullary name we simply write @xmath26 for the negative action on name @xmath26 . [ designs ] for a fixed signature @xmath30 , the class of * positive designs * @xmath41 , that of * predesigns * @xmath42 , and that of * negative designs * @xmath43 are _ coinductively _ defined as follows : @xmath44 where : 1 . @xmath45 ; 2 . @xmath46 and the formal sum @xmath47 has @xmath48-many components @xmath49 ; 3 . @xmath50 is built from a set @xmath51 of predesigns with @xmath52 an arbitrary index set . we denote arbitrary designs by @xmath53 . the set of * designs * , consisting of all positive , negative and predesigns , is denoted by @xmath54 . any subterm @xmath55 of @xmath56 is equivalently called a _ subdesign _ of @xmath56 . notice that designs are _ coinductively _ defined objects . in particular , infinitary designs are included in our syntax , just as in the original ludics @xcite . it is strictly necessary , since we want to express both proof attempts and countermodels as designs , both of which tend to be infinite . informally , designs may be regarded as infinitary @xmath16-terms with _ named _ applications , _ named _ and _ superimposed _ abstractions and a _ universal nondeterministic _ choice operator @xmath21 . more specifically , a predesign @xmath57 can be thought of as iterated application @xmath58 of an @xmath40-ary name @xmath32 . in the sequel , we may abbreviate @xmath59 by @xmath60 . if @xmath26 is a nullary name we simply write @xmath61 . on the other hand , a negative design of the form @xmath62 can be thought of as iterated abstraction @xmath63 of an @xmath40-ary name @xmath32 . a family @xmath64 of abstractions indexed by @xmath23 is then superimposed to form a negative design @xmath47 . since @xmath47 is built from a family indexed by @xmath23 , there can not be any overlapping of name in the sum . each @xmath62 is called an ( additive ) * component*. a predesign is called a * cut * if it is of the form @xmath65 . otherwise , it is of the form @xmath66 and called a * head normal form*. as we shall see in detail in subsection [ normalization subsection ] , cuts have substantial computational significance in our setting : in fact a cut @xmath67 can be _ reduced _ to another design @xmath68 $ ] . namely , when the application is of name @xmath69 , one picks up the component @xmath70 from the family @xmath64 . notice that the arities of @xmath71 and @xmath72 always agree . then , one applies a simultaneous @xmath73-reduction " @xmath74 $ ] . the _ head variable _ @xmath75 in an head normal form @xmath66 plays the same role as a pointer in a strategy does in hyland - ong games and an address ( or locus ) in girard s ludics . on the other hand , a variable @xmath75 occurring in a bracket as in @xmath76 @xmath77 , @xmath75 , @xmath78 does not correspond to a pointer nor address . rather , it corresponds to an identity axiom ( initial sequent ) in sequent calculus , and for this reason is called an * identity*. if a negative design @xmath79 simply consists of a variable @xmath75 , then @xmath79 is itself an identity . the positive design @xmath80 denotes _ divergence _ ( or _ partiality _ ) of the computation , in the sense we will make more precise in the next subsection . we also use @xmath80 to encode partial sums . given a set @xmath81 of negative actions with distinct names @xmath82 , we write @xmath83 to denote the negative design @xmath84 , where @xmath85 if @xmath86 , and @xmath87 otherwise . we also use an informal notation @xmath88 to denote @xmath83 . so far , the syntax we are describing is essentially the same as the one introduced in @xcite . a novelty of this paper is the nondeterministic conjunction operator @xmath21 , which allows us to build a positive design @xmath50 from a set @xmath51 of predesigns with @xmath52 an arbitrary index set . each @xmath89 is called a * conjunct * of @xmath1 . we write @xmath90 ( * daimon * ) for the empty conjunction @xmath91 . this design plays an essential role in ludics , since it is used to define the concept of orthogonality . although @xmath90 is usually given as a primitive ( see @xcite ) , we have found it convenient and natural to identify ( or rather encode ) @xmath90 with the empty conjunction . as we shall see , its computational meaning exactly corresponds to the usual one : @xmath90 marks the _ termination _ of a computation . put in another way , our nondeterministic conjunction can be seen as a generalization of the termination mark . a design @xmath56 may contain free and bound variables . an occurrence of subterm @xmath62 _ binds _ the free variables @xmath37 in @xmath92 . variables which are not under the scope of the binder @xmath93 are _ free_. we denote by @xmath94 the set of free variables occurring in @xmath56 . as in @xmath16-calculus , we would like to identify two designs which are _ @xmath95-equivalent _ up to renaming of bound variables . but it is more subtle than usual , since we also would like to identify , @xmath96 with @xmath97 whenever @xmath98 and @xmath99 are @xmath95-equivalent . to enforce these requirements simultaneously and hereditarily , we define an equivalence relation by coinduction . by _ renaming _ we mean a function @xmath100 . we write @xmath101 for the identity renaming , and @xmath102 $ ] for the renaming that agrees with @xmath103 except that @xmath102(x)=z$ ] . the set of renamings is denoted by @xmath104 . [ d - alpha ] a binary relation @xmath105 is called a * design equivalence * if for any @xmath106 such that @xmath107 , one of the following holds : 1 . @xmath108 ; 2 . @xmath109 , @xmath110 and we have : a. for any @xmath111 there is @xmath112 such that @xmath113 , b. for any @xmath112 there is @xmath111 such that @xmath113 ; 3 . @xmath114 , @xmath115 and @xmath116 for every @xmath117 ; 4 . @xmath118 , @xmath119 and @xmath120 ; 5 . @xmath121 , @xmath122 and @xmath123 ) { \ ; \cr\ ; } ( q_a , \rho[\vec{z}_a/\vec{y}_a])$ ] for every @xmath32 and some vector @xmath124 of fresh variables . we say that two designs @xmath56 and @xmath55 are _ equivalent _ if there is a design equivalence @xmath125 such that @xmath126 . see @xcite for further details . henceforth we _ always _ identify two designs @xmath56 and @xmath55 , and write @xmath127 by abuse of notation , if they are equivalent in the above sense . the following lemma is a straightforward extension of lemma 2.6 of @xcite . it makes it easier to prove equivalence of two designs ( just as the `` bisimulation up - to '' technique in concurrency theory makes it easier to prove bisimilarity of two processes ) . [ l - equiv ] let @xmath125 be a binary relation on designs such that if @xmath128 then one of the following holds : 1 . @xmath108 ; 2 . @xmath109 , @xmath110 , and we have : 1 . for any @xmath111 there is @xmath112 such that @xmath129 , 2 . for any @xmath112 there is @xmath111 such that @xmath129 ; 3 . @xmath114 , @xmath115 and @xmath130 for every @xmath117 ; 4 . @xmath131 ; 5 . @xmath132 , @xmath133 and @xmath134 for every @xmath32 . if @xmath135 , then @xmath56 and @xmath55 are equivalent . as a notational convention , a unary conjunction @xmath136 is simply written as @xmath98 this allows us to treat a predesign as a positive design . we also write : 1 . @xmath137 if @xmath1 is a conjunction and @xmath98 is a conjunct of @xmath1 ; 2 . @xmath138 if _ either _ @xmath139 , _ or _ both @xmath1 and @xmath140 are conjunctions and for all @xmath141 , @xmath137 . thus @xmath138 indicates that @xmath1 has more conjuncts than @xmath140 unless @xmath139 . we also extend the conjunction operator to positive designs and abstractions as follows . [ conj binary ] 1 . as for positive designs , we set @xmath142 2 . as for abstractions , negative designs of the form @xmath47 , observe that since we are working up to renaming of bound variables , it is no loss of generality to assume that in any pair @xmath47 , @xmath143 , one has @xmath144 for every @xmath145 . we set : @xmath146 observe the following : 1 . the set of positive designs forms a semilattice with respect to @xmath147 and @xmath148 . 2 . @xmath149 for any positive design @xmath1 . the previous definition can be naturally generalized to arbitrary sets as follows : [ conj extension ] 1 . given a set @xmath150 of positive designs , we define the positive design @xmath151 as follows : if @xmath152 , we set @xmath153 . 2 . if @xmath154 , we set @xmath155 . otherwise , @xmath150 is a nonempty set of conjunctions and we set : @xmath156 2 . given a set @xmath150 of abstractions , we define the abstraction @xmath151 as : @xmath157 in particular , if @xmath152 then @xmath158 . notice that @xmath159 , as long as @xmath56 ranges over positive designs or abstractions . a design @xmath56 is said : 1 . * total * , if @xmath160 ; 2 . * closed * , if @xmath56 has no occurrence of free variable ; 3 . * linear * ( or * affine * , more precisely ) , if for any subdesign of the form @xmath161 , the sets @xmath162 , , @xmath163 are pairwise disjoint ; 4 . * deterministic * , if in any occurrence of subdesign @xmath164 , @xmath52 is either empty ( we have @xmath90 ) or a singleton ( we have a predesign ) . * cut - free * , if it does not contain a cut as a subdesign ; 6 . * identity - free * , if it does not contain an identity as subdesign . we remark that the notion of design introduced in @xcite exactly corresponds in our terminology to that of deterministic design . furthermore , considering the specific signature @xmath165 given below , we can also express in our setting girard s original notion of design : let us consider the signature @xmath166 where : 1 . @xmath167 consists of finite subsets of @xmath168 ; 2 . @xmath169 is the function that maps a finite subset @xmath170 to its cardinality @xmath171 . girard s designs correspond to total , linear , deterministic , cut - free and identity - free designs which have a finite number of free variables over the signature @xmath165 . see @xcite for more details . ludics is an interactive theory . this means that designs , which subsume both proofs and models , interact together via normalization , and types ( behaviours ) are defined by the induced orthogonality relation ( section [ s - behaviour ] ) . several ways to normalize designs have been considered in the literature : abstract machines @xcite , abstract merging of orders @xcite , and terms reduction @xcite . here we extend the last solution @xcite . as in untyped @xmath16-calculus , normalization is not necessarily terminating , but in our setting a new difficulty arises through the presence of the operator @xmath21 . we define the normal forms in two steps , first giving a _ nondeterministic _ reduction rule which finds head normal forms whenever possible , and then expanding it corecursively . as usual , let @xmath172 $ ] denote the design obtained by the simultaneous and capture - free substitution of negative designs @xmath173 for @xmath174 in @xmath56 . given positive designs @xmath175 , we write @xmath176 if @xmath177 and @xmath178 $ ] . we denote by @xmath179 the transitive closure , and by @xmath180 the reflexive transitive closure of @xmath181 . given two binary relations @xmath182 on designs , we write @xmath183 to denote the relation given by their composition @xmath184 for instance , we write @xmath185 if there exists @xmath140 such that @xmath186 and @xmath187 . we now give examples of reductions and some remarks . 1 . @xmath188 . @xmath189 ( recall that @xmath190 stands for @xmath191 by our conventions on partial sums ) . 3 . let @xmath192 . we have @xmath193 and @xmath194 . 4 . for @xmath195 , let us consider @xmath196 . we then have an infinite reduction sequence @xmath197 since @xmath198 and the latter design reduces to @xmath199 = n \;|\ ; { { \overline{a}}}\langle n \rangle = p$ ] . let @xmath200 . we have the following reduction : @xmath201 \ = \ b(t).q \;|\ ; { { \overline{b}}}\langle w \rangle \ , \wedge \ , z|\overline{c } \langle m[b(t).q / y ] \rangle.\ ] ] we therefore have @xmath202 and @xmath203 \rangle$ ] . since @xmath204 is a cut , we have : @xmath205 \rangle \ { -\!\!\!\!\!\rightharpoonup}\ q[w / t].\ ] ] 6 . the special designs @xmath90 and @xmath80 do not reduce to anything ( as we will see , they are normal forms ) . 7 . by its definition , our reduction is not closed under context " if @xmath176 and @xmath1 ( resp . @xmath140 ) occurs as a subdesign of @xmath56 ( resp . @xmath55 ) , nothing ensures that @xmath206 . for instance a negative design ( or an head normal form ) having an occurrence of cut as subdesign does not reduces to anything . to expand the reduction under context " we will use definition [ normal form ] . notice that any _ closed _ positive design @xmath1 has one of the following forms : @xmath90 , @xmath80 and @xmath207 , where @xmath89 are cuts . the conjunction then reduces to another closed positive design . hence any sequence of reductions starting from @xmath1 either terminates with @xmath90 or @xmath80 or it diverges . by _ stipulating _ that the normal form of @xmath1 in case of divergence is @xmath80 , we obtain a dichotomy between @xmath90 and @xmath80 : the normal form of a closed positive design is either @xmath90 or @xmath80 . this leads us to the following definition of normal form : [ normal form ] the * normal form function * @xmath208 is defined by corecursion as follows : @xmath209 we observe that when @xmath1 is a closed positive design , we have @xmath210 precisely when _ all _ reduction sequences from @xmath1 are finite and terminate with @xmath90 ; thus our nondeterminism is _ universal _ rather than existential . this , however , does not mean that the set @xmath211 is finite ; even when it is infinite , it may happen that @xmath210 . the following facts are easily observed : [ l - facts ] 1 . if @xmath212 , @xmath213 and @xmath214 , then @xmath215 . 2 . @xmath216 implies @xmath217 . furthermore , if @xmath1 is a predesign , then @xmath176 implies @xmath218 . 3 . @xmath219 , for any set @xmath150 of positive designs . notice that the first statement means that the composed relation @xmath220 is equivalent to @xmath181 as far as total designs are concerned . in @xcite , it is illustrated how words and deterministic finite automata are represented by ( deterministic ) designs in ludics . we may extend the idea to trees and finite tree automata in presence of nondeterminism . rather than describing it in full detail , we will only give an example which illustrates the power of nondeterminism to express ( topdown ) finite tree automata . we consider the set of _ finite trees _ labelled with @xmath221 which are at most binary branching . it is defined by the following grammar : @xmath222 here , @xmath223 represents a tree with the root labelled by @xmath26 and with two subtrees @xmath224 . in particular , @xmath225 represents a leaf labelled by @xmath26 . we simply write @xmath26 in this case . suppose that the signature @xmath226 contains a unary name @xmath227 , binary names @xmath221 and a nullary name @xmath228 . we write @xmath229 for the positive action @xmath230 . we abbreviate @xmath231 by @xmath232 , so that we have @xmath233.\ ] ] each tree is then represented by a deterministic linear negative design as follows : @xmath234 now consider the positive design @xmath235 $ ] defined by the following equations : @xmath236 & : = & x|{\mathord{\downarrow}}\langle a(x , y ) . q_1[x]\wedge q_0[y]\ ; + \ ; b(x , y ) . q_2[x]\wedge q_2[y]\rangle , \\ q_1[x ] & : = & x|{\mathord{\downarrow}}\langle b(x , y ) . q_2[x]\wedge q_2[y]\rangle , \\ q_2[x ] & : = & x|{\mathord{\downarrow}}\langle \epsilon.\maltese\rangle.\end{aligned}\ ] ] this design @xmath140 works as an automata accepting all trees of the form @xmath237 . indeed , given @xmath238 , it works nondeterministically as follows : @xmath239 \\ & \rotatebox[origin = c]{200}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } & & \rotatebox[origin = c]{340}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } \\ q_1[b^\star ] & & & & q_0[a(b , b)^\star]\\ \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } & & & \rotatebox[origin = c]{200}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } & & \rotatebox[origin = c]{340}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } \\ q_2[\epsilon^\star ] & & \qquad\qquad q_1[b^\star ] & & & & q_0[b^\star ] \\ \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*$ } & & \qquad\qquad \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } & & & & \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*\ni$ } \\ \maltese & & \qquad\qquad q_2[\epsilon^\star ] & & & & q_2[\epsilon^\star ] \\ & & \qquad\qquad \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*$ } & & & & \rotatebox[origin = c]{270}{${-\!\!\!\!\!\rightharpoonup}^*$ } \\ & & \qquad\qquad \maltese & & & & \maltese \end{array}\ ] ] hence we conclude @xmath240 ] \ ! ] } = \maltese$ ] , @xmath140 `` accepts '' the tree @xmath238 . in this subsection , we prove one of the fundamental properties of designs which we will need later : [ t - assoc ] let @xmath56 be a design and @xmath241 be negative designs . we have : @xmath242 { \rrbracket}= { \llbracket } { \llbracket}d { \rrbracket } [ { \llbracket}n_1 { \rrbracket}/y_1 , \ldots , { \llbracket}n_n { \rrbracket}/y_n ] { \rrbracket}.\ ] ] associativity corresponds to a weak form of the church - rosser property : the normal form is the same even if we do not follow the head reduction strategy . in this paper we are not concerned with the full church - rosser property , and leave it as an open question . the proof consists of several stages and it can be skipped at first reading . to prove associativity , first notice that a simultaneous substitution @xmath243 $ ] can be turned into a sequential one of the form @xmath244\cdots[n_n / z_n]$ ] by renaming @xmath245 by fresh variables @xmath246 as follows : @xmath247 = d[z_1/y_1 , \dots , z_n / y_n][n_1/z_1]\cdots[n_n / z_n].\ ] ] moreover , we have : @xmath248\!]}[\ ; { [ \ ! [ n_1 ] \!]}/y_1 , \dots , { [ \ ! [ n_n ] \!]}/y_n ] = { [ \ ! [ \ ; d[z_1/y_1 , \dots , z_n / y_n]\ ; ] \!]}[\ ; { [ \ ! [ n_1 ] \!]}/z_1]\cdots [ \ ; { [ \ ! [ n_n ] \!]}/z_n].\ ] ] this allows us to work with sequential substitutions rather than simultaneous ones . we define a binary relation @xmath249 on designs by : 1 . @xmath250 if @xmath251 \cdots [ n_n / y_n]$ ] and @xmath252\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] for some @xmath253 such that @xmath254 for @xmath255 . [ l - a1 ] suppose that @xmath256\cdots[n_n / y_n]$ ] and @xmath257\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] so that @xmath258 . when @xmath259 , two cases can be distinguished . 1 . if @xmath260 and @xmath261\cdots[n_n / y_n]$ ] , then there exists @xmath262 such that @xmath263 and @xmath264 . otherwise , there exists @xmath262 such that @xmath265 and @xmath264 . 1 . by lemma [ l - facts ] ( 2 ) , we have @xmath266\!]}\leq{[\ ! [ p_1 ] \!]}$ ] that implies @xmath267\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] . hence by letting @xmath268\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] , we have @xmath263 and @xmath264 . if ( 1 ) is not the case , a cut must be created by substitution of some @xmath269 for a head variable of @xmath270 . hence @xmath270 must contain a head normal form @xmath271 as conjunct for some @xmath272 and @xmath273 , so that @xmath1 contains a cut @xmath274\cdots[n_n / y_n ] = n_j |{\overline{a}}{\langle \vec{m } \rangle}[n_1/y_1]\cdots[n_n / y_n]\ ] ] ( the equality due to @xmath254 for @xmath275 ) and @xmath276[n_1/y_1]\cdots[n_n / y_n ] $ ] @xmath277 . in this case , @xmath266\!]}$ ] contains @xmath278\ ! ] } \rangle}$ ] so that @xmath140 contains @xmath279\ ! ] } \rangle } [ { [ \ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n ] = { [ \ ! [ n_j ] \ ! ] } |{\overline{a}}{\langle { [ \ ! [ \vec{m } ] \ ! ] } \rangle } [ { [ \ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n].\ ] ] since @xmath280\ ! ] } = { \textstyle\sum}a(\vec{x}).{[\ ! [ r_a ] \!]}$ ] , we have @xmath281\!]}[\vec{{[\ ! [ m ] \!]}}/\vec{x } ] [ { [ \ ! [ n_1 ] \!]}/y_1 ] \cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] . let @xmath262 be the latter design . since the simultaneous substitutions @xmath282 $ ] and @xmath283\!]}}/\vec{x}]$ ] can be made sequential , we have @xmath284 . [ l - a2 ] if @xmath258 and @xmath265 , then there exists some @xmath285 such that @xmath284 and @xmath286 . suppose that @xmath256\cdots[n_n / y_n]$ ] , @xmath257\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] and @xmath265 . then @xmath266\!]}$ ] must contain @xmath278\ ! ] } \rangle}$ ] for some @xmath287 and @xmath288 . thus , @xmath289 . suppose also @xmath273 so that @xmath280\ ! ] } = { \textstyle\sum}a(\vec{x}).{[\ ! [ r_a ] \!]}$ ] . now the situation is as follows : @xmath140 contains @xmath279\ ! ] } \rangle } [ { [ \ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n ] = { [ \ ! [ n_j ] \ ! ] } |{\overline{a}}{\langle { [ \ ! [ \vec{m } ] \ ! ] } \rangle } [ { [ \ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n],\ ] ] so we have @xmath290\!]}[\vec{{[\ ! [ m ] \!]}}/\vec{x } ] [ { [ \ ! [ n_1 ] \!]}/y_1 ] \cdots [ { [ \ ! [ n_n ] \!]}/y_n ] = q'.\ ] ] on the other hand , @xmath291 \cdots [ n_n / y_n ] & \\ & = & n_j |{\overline{a}}{\langle \vec{m } \rangle}[n_1/y_1 ] \cdots [ n_n / y_n ] & \\ & { -\!\!\!\!\!\rightharpoonup } & r_a[\vec{m}/\vec{x}_a ] [ n_1/y_1 ] \cdots [ n_n / y_n ] & = p ' , \end{array}\ ] ] which implies @xmath292 . since the simultaneous substitutions @xmath282 $ ] and @xmath283\!]}}/\vec{x}]$ ] can be made sequential , we have @xmath284 . [ l - a3 ] suppose that @xmath258 . then @xmath293\!]}=\omega$ ] if and only if @xmath294\ ! ] } = \omega$ ] . 1 . for the ` if ' direction , we distinguish two cases . 1 . if there is an infinite reduction sequence from @xmath140 , then there is also an infinite sequence from @xmath1 by lemma [ l - a2 ] . 2 . if @xmath295 , then there is @xmath285 such that @xmath296 and @xmath297 . namely , @xmath285 can be written as @xmath298\cdots[n_n / y_n]$ ] and @xmath299\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] . the latter means that @xmath266\ ! ] } = \omega$ ] , which implies @xmath300\ ! ] } = \omega$ ] . from this and @xmath296 , we conclude @xmath293\ ! ] } = \omega$ ] . 2 . for the ` only - if ' direction , if @xmath301 , we easily obtain @xmath294\ ! ] } = \omega$ ] . otherwise , there is an infinite reduction sequence @xmath302 . suppose that @xmath303\cdots[n_n / y_n]$ ] and @xmath304\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots[{[\ ! [ n_n ] \!]}/y_n]$ ] . our purpose is to build either a finite reduction sequence @xmath305 or an infinite reduction sequence @xmath306 . two cases arise : 1 . the reductions take place inside @xmath270 and independently of @xmath307 . namely , there is an infinite reduction sequence @xmath308 such that @xmath309\cdots[n_n / y_n]$ ] for every @xmath310 . then @xmath266\!]}= \omega$ ] , which implies @xmath311 . so we have @xmath294\!]}=\omega$ ] . otherwise , there is at most a finite sequence @xmath312 such that @xmath309\cdots[n_n / y_n]$ ] for @xmath313 and @xmath314 contains a head normal form that is responsible for the reduction @xmath315 . by repeatedly applying lemma [ l - a1 ] ( 1 ) , we obtain @xmath262 such that @xmath263 and @xmath316 . since @xmath315 , there exists @xmath317 such that @xmath318 and @xmath319 by lemma [ l - a1 ] ( 2 ) . hence by lemma [ l - facts ] ( 1 ) , we obtain @xmath320 . + in the former case , we are already done . in the latter case , we still have an infinite reduction sequence @xmath321 and @xmath319 . hence we may repeat the same argument to prolong the reduction sequence @xmath322 . hence we eventually obtain @xmath294\!]}=\omega$ ] . [ l - a4 ] suppose that @xmath258 . 1 . if @xmath323 , then there exist @xmath324 such that @xmath325 and @xmath326 , , @xmath327 . conversely , if @xmath328 , then there exist @xmath329 such that @xmath330 and @xmath326 , , @xmath327 . suppose that @xmath331 . by lemmas [ l - a1 ] and [ l - facts ] ( 1 ) ( which states that the composed relation @xmath332 is identical with @xmath181 ) , there is @xmath262 such that @xmath333 and @xmath264 . since @xmath334 , we may write @xmath335\cdots [ n_n / y_n ] = x|{\overline{a}}{\langle \vec{k}[n_1/y_1]\cdots [ n_n / y_n ] \rangle}\ ] ] for some @xmath336 , where @xmath337 , and @xmath262 contains @xmath338\ ! ] } [ { [ \ ! [ n_1 ] \!]}/y_1]\cdots [ { [ \ ! [ n_n ] \!]}/y_n ] = x|{\overline{a}}{\langle \vec{{[\ ! [ k ] \!]}}[{[\ ! [ n_1 ] \!]}/y_1]\cdots [ { [ \ ! [ n_n ] \!]}/y_n ] \rangle}.\ ] ] hence by letting @xmath339\!]}[{[\ ! [ n_1 ] \!]}/y_1]\cdots [ { [ \ ! [ n_n ] \!]}/y_n]$ ] we obtain @xmath340 for every @xmath341 . since @xmath342 , namely @xmath343 , the claim holds . conversely , suppose that @xmath344 . by lemma [ l - a2 ] , there is @xmath285 such that @xmath345 and @xmath264 . the rest is similar to the above . [ l - a5 ] if @xmath346 , then either @xmath347 for a variable @xmath348 , or @xmath349 , @xmath350 and @xmath351 for every @xmath32 . immediate . the following lemma completes the proof of theorem [ t - assoc ] . if @xmath352 , then @xmath353\ ! ] } = { [ \ ! [ e_0 ] \!]}$ ] . define a binary relation @xmath125 on designs as follows : 1 . for positive ( resp . negative ) designs @xmath354 , we have @xmath128 if @xmath355\!]}$ ] , @xmath356\!]}$ ] , and @xmath357 for some @xmath358 and @xmath359 . 2 . for predesigns @xmath360 , we have @xmath361 if @xmath362\ ! ] } , \dots , { [ \ ! [ m_m ] \ ! ] } \rangle}$ ] , @xmath363\ ! ] } , \dots , $ $ { [ \ ! [ l_m ] \ ! ] } \rangle}$ ] , and @xmath340 for every @xmath364 . we now verify that this @xmath125 satisfies the conditions of lemma [ l - equiv ] . first , let @xmath175 be positive designs such that @xmath365 , @xmath366\!]}$ ] , @xmath367\!]}$ ] , and @xmath368 for some @xmath270 and @xmath369 . 1 . if @xmath266\!]}=\omega$ ] , then @xmath370\!]}= \omega$ ] too by lemma [ l - a3 ] . hence ( 1 ) holds . 2 . if @xmath266\!]}$ ] is a conjunction , then @xmath370\!]}$ ] is not @xmath80 by lemma [ l - a3 ] , so is a conjunction . if @xmath266\!]}$ ] contains @xmath371\ ! ] } , \dots , { [ \ ! [ m_m ] \ ! ] } \rangle}$ ] , then @xmath372 . since @xmath368 , lemma [ l - a4 ] yields @xmath373 for some @xmath374 . namely , @xmath370\!]}$ ] contains @xmath375\ ! ] } \rangle}$ ] . moreover , we have @xmath340 for every @xmath364 . similarly , one can show that if @xmath370\!]}$ ] contains @xmath375\ ! ] } \rangle}$ ] , then @xmath266\!]}$ ] contains @xmath376\ ! ] } \rangle}$ ] and @xmath340 for every @xmath364 . hence ( 2 ) holds . let @xmath360 be predesigns such that @xmath377 , i.e. , @xmath362\ ! ] } , \dots , { [ \ ! [ m_m ] \ ! ] } \rangle}$ ] , @xmath363\ ! ] } , \dots , $ $ { [ \ ! [ l_m ] \ ! ] } \rangle}$ ] , and @xmath340 for every @xmath364 . it immediately follows that @xmath378\ ! ] } { \ ; \cr\ ; } { [ \ ! [ l_i ] \!]}$ ] for every @xmath364 . also , @xmath379 . hence ( 3 ) holds . finally , let @xmath380 be negative designs such that @xmath381\!]}$ ] , @xmath382\!]}$ ] , and @xmath383 for some @xmath384 and @xmath385 . 1 . if @xmath386 , then @xmath387 . hence ( 4 ) holds . 2 . otherwise , @xmath79 must be of the form @xmath388\!]}$ ] and @xmath389 . since @xmath383 , @xmath385 is of the form @xmath390 and @xmath391 for every @xmath32 by lemma [ l - a5 ] . so @xmath392\ ! ] } = { \textstyle\sum}a(\vec{x}_a).{[\ ! [ q_a ] \!]}$ ] and @xmath393\ ! ] } { \ ; \cr\ ; } { [ \ ! [ q_a ] \!]}$ ] . hence ( 5 ) holds . therefore , if @xmath357 , we have @xmath353\ ! ] } = { [ \ ! [ e_0 ] \!]}$ ] by lemma [ l - equiv ] . this section is concerned with the type structure of ludics . we describe orthogonality and behaviours in [ ortho - s ] , logical connectives in [ ss - logical ] and finally explain ( the failure of ) internal completeness of logical connectives in [ ss - internal ] . in the rest of this paper , we mainly restrict ourselves to a special subclass of designs : we only consider designs which are _ total _ , _ cut - free _ , and _ identity - free_. generalizing the terminology in @xcite , we call them _ standard designs_. in other words : [ d - standard ] a design @xmath56 is said * standard * if it satisfies the following two conditions : a. _ cut - freeness and identity - freeness _ : @xmath56 can be coinductively generated by the following restricted version of the grammar given in definition [ designs ] : @xmath394 b. _ totality _ : @xmath160 . the totality condition is due to the original work @xcite . it has a pleasant consequence that behaviours ( see below ) are never empty . we also remark that the lack of identities can be somehow compensated by considering their infinitary @xmath395 expansions , called _ faxes _ in @xcite . in our setting , the infinitary @xmath395 expansion of an identity @xmath75 is expressed by the negative standard design @xmath396 defined by the equation : @xmath397 we refer to @xcite for more details . we are now ready to define orthogonality and behaviours . a positive design @xmath1 is said * atomic * if it is standard and @xmath398 for a certain fixed variable @xmath399 . here plays the same role as the empty address @xmath400 " does in @xcite : @xmath399 may be thought of as a fixed and predetermined location . " ] a negative design @xmath79 is said * atomic * if it is standard and @xmath401 . two atomic designs @xmath402 of opposite polarities are said * orthogonal * and written @xmath403 ( or equivalently @xmath404 ) when @xmath405 { \rrbracket}= \maltese$ ] . if @xmath150 is a set of atomic designs of the same polarity , then its * orthogonal set * , denoted by @xmath406 , is defined by @xmath407 . the meaning of @xmath21 and the associated partial order @xmath147 can be clarified in terms of orthogonality . for atomic designs @xmath354 of the same polarity , define @xmath408 if and only if @xmath409 . @xmath408 means that @xmath55 has more chances of convergence than @xmath56 when interacting with other atomic designs . the following is easy to observe . [ conj and ort ] 1 . @xmath410 is a preorder . @xmath138 implies @xmath411 for any pair of atomic positive designs @xmath175 . 3 . let @xmath150 and @xmath412 be sets of atomic designs of the same polarity . then @xmath413 implies @xmath414 . in particular , @xmath415 for any atomic positive design @xmath1 . , which does not officially belong to the universe of atomic designs . ] this justifies our identification of @xmath90 with the empty conjunction @xmath416 . [ r - separation ] designs in @xcite satisfy the _ separation property _ : for any designs @xmath354 of the same polarity , we have @xmath417 if and only if @xmath418 . but when the constraint of linearity is removed , this property no more holds , as observed in @xcite ( see also @xcite ) . in our setting , separation does not hold , even when @xmath56 and @xmath55 are deterministic ( atomic ) designs . for instance , consider the following two designs @xcite : @xmath419 it is easy to see that in our setting @xmath403 holds if and only if @xmath79 has an additive component of the form @xmath420 for arbitrary index set @xmath52 and arbitrary standard negative designs @xmath421 with @xmath422 . the same holds for @xmath140 , as can be observed from the following reduction sequence ( for readability , we only consider the case in which @xmath79 has a component of the form @xmath423 , the general case easily follows ) : @xmath424 & = & n \;|\ ; { \mathord{\downarrow}}\langle{\mathord{\uparrow}}(y ) . p [ n / x_0]\rangle \\ & { -\!\!\!\!\!\rightharpoonup } & \left ( { \mathord{\uparrow}}(y ) . p [ n / x_0]\right ) \;|\ ; { \mathord{\downarrow}}\langle m[{\mathord{\uparrow}}(y ) . p [ n / x_0]/z]\rangle \\ & { -\!\!\!\!\!\rightharpoonup } & p[n / x_0]\ { -\!\!\!\!\!\rightharpoonup}^\ast \ \maltese.\end{aligned}\ ] ] ( if @xmath79 does not have a component of the form discussed above , we have @xmath425 { -\!\!\!\!\!\rightharpoonup}^ * \omega$ ] . ) we therefore conclude @xmath426 , even though @xmath427 . although possible , we do not define orthogonality for nonatomic designs . accordingly , we only consider _ atomic _ behaviours which consist of atomic designs . [ d - beh ] a * behaviour * @xmath150 is a set of atomic standard designs of the same polarity such that @xmath428 . a behaviour is positive or negative according to the polarity of its designs . we denote positive behaviours by @xmath429 and negative behaviours by @xmath430 . orthogonality satisfies the following standard properties : [ closure operator ] let @xmath431 be sets of atomic designs of the same polarity . we have : 1 . @xmath432 . 2 . @xmath433 . 3 . @xmath434 . 4 . @xmath435 . in particular , any orthogonal set is a behaviour . 5 . @xmath436 . in particular , the intersection of two behaviours is a behaviour . we also observe that @xmath437 and @xmath438 implies @xmath439 when @xmath150 is a behaviour . among all positive ( resp . negative ) behaviours , there exist the least and the greatest behaviours with respect to set inclusion:@xmath440 where @xmath441 plays the role of the design called _ negative daimon _ in @xcite . notice that behaviours are always nonempty due to the totality condition : any positive ( resp . negative ) behaviour contains @xmath90 ( resp . @xmath442 ) . now that we have given behaviours , we can define _ contexts of behaviours _ and then the _ semantical entailment _ @xmath443 in order to relate designs to contexts of behaviours . these constructs play the role of typing environments in type systems . they correspond to _ sequents of behaviours _ , in the terminology of @xcite . a. a * positive context * @xmath444 is of the form @xmath445 , where @xmath34 are distinct variables and @xmath446 are ( atomic ) positive behaviours . we denote by @xmath447 the set @xmath448 . + a * negative context * @xmath449 is a positive context @xmath444 enriched with an ( atomic ) negative behaviour @xmath450 , to which no variable is associated . b. the * semantical entailment * is the binary relation @xmath443 between designs and contexts of behaviours of the same polarity defined as follows : + @xmath451 if and only if : 1 . @xmath1 is standard ; 2 . @xmath452 ; 3 . @xmath453 { \rrbracket}= \maltese$ ] for any @xmath454 , , @xmath455 . + @xmath456 if and only if : 1 . @xmath79 is standard ; 2 . @xmath457 ; 3 . @xmath458/x_0 ] { \rrbracket}= \maltese$ ] for any @xmath454 , , @xmath455 , @xmath459 . clearly , @xmath460 if and only if @xmath461 , and @xmath462 if and only if @xmath463\in \bp$ ] . furthermore , associativity ( theorem [ t - assoc ] ) implies the following quite useful principle : [ l - closure ] 1 . @xmath464 if and only if @xmath465 { \rrbracket}\models \bgamma$ ] for any @xmath466 ; 2 . @xmath467 if and only if @xmath468 { \rrbracket}\models \bgamma$ ] for any @xmath469 ; 3 . @xmath470 if and only if @xmath471 { \rrbracket}\models \bgamma,\bn$ ] for any @xmath466 . 1 . let @xmath1 be a standard design with @xmath472 and @xmath444 a context @xmath473 . first , we claim that @xmath465 { \rrbracket}$ ] is a standard design when @xmath464 and @xmath466 . indeed , it is obviously cut - free . it is also identity - free because so are @xmath474 and neither substitution @xmath475 $ ] nor normalization @xmath465 { \rrbracket}$ ] introduces identities . totality will be shown below . we also note that @xmath476 { \rrbracket } ) \subseteq \{x_1 , \dots , x_n\}$ ] , since @xmath2 is an atomic negative design that is always closed . + next , we observe that @xmath477 { \rrbracket}= { \llbracket}\ ; { \llbracket}p[m / z ] { \rrbracket}\ ; [ \vec{k}/\vx ] { \rrbracket}$ ] for any list @xmath478 of standard negative designs . indeed , notice that @xmath479 = p[m / z ] [ \vec{k}/\vx]$ ] since @xmath2 is closed , and @xmath480\ ! ] } = k_i$ ] since @xmath481 is cut - free . hence by associativity , we obtain : @xmath482 { \rrbracket}= { \llbracket}p[m / z ] [ \vec{k}/\vx ] { \rrbracket}= { \llbracket}\ ; { \llbracket}p[m / z ] { \rrbracket}\ ; [ { \llbracket}\vec{k } { \rrbracket}/\vx ] { \rrbracket}= { \llbracket}\ ; { \llbracket}p[m / z ] { \rrbracket}\ ; [ \vec{k } /\vx ] { \rrbracket}. \ ] ] in particular , @xmath477 { \rrbracket}= \maltese$ ] implies the totality of @xmath465 { \rrbracket}$ ] . + we are now ready to prove the first claim . writing @xmath483 for @xmath484 , , @xmath485 , we have : @xmath486 { \rrbracket}= \maltese \mbox { for every $ \vec{k}\in \bgamma{{}^{\bot}}$ and $ m\in \bp^\bot$ } , \\ & \longleftrightarrow & { \llbracket}\ ; { \llbracket}p[m / z ] { \rrbracket}\ ; [ \vec{k } /\vx ] { \rrbracket}= \maltese \mbox { for every $ \vec{k}\in \bgamma{{}^{\bot}}$ and $ m\in \bp^\bot$ } , \\ & \longleftrightarrow & { \llbracket}p[m / z ] { \rrbracket}\models \bgamma \mbox { for every $ m\in \bp^\bot$.}\end{aligned}\ ] ] 2 . and ( 3 ) are proven in a similar way . we just mention that the crucial equalities @xmath487/x_0 ] { \rrbracket } & = & { \llbracket}\ ; { \llbracket}q[n / x_0 ] { \rrbracket}\ ; [ \vec { k } /\vx ] { \rrbracket},\\ { \llbracket}q[n[\vec{k}/\vx , m / z]/x_0 ] { \rrbracket } & = & { \llbracket}q [ \ ; { \llbracket}n[m / z ] { \rrbracket}\ ; [ \vec{k}/\vx]/x_0 ] { \rrbracket},\end{aligned}\ ] ] which are needed to show ( 2 ) and ( 3 ) respectively , can be straightforwardly derived from associativity . we next describe how to build behaviours by means of _ logical connectives _ in ludics . [ d - lc ] an * @xmath40-ary logical connective * @xmath95 is a pair @xmath488 where : 1 . @xmath489 is a sequence of distinct variables ; 2 . @xmath490 is a finite set of negative actions such that : 1 . the names @xmath491 are distinct ; 2 . @xmath492 for each @xmath493 . two logical connectives are identified if one is obtained from another by renaming of variables . we can intuitively explain the structure of logical connectives in terms of standard connectives of linear logic as follows . the variables @xmath494 play the role of _ placeholders _ for ( immediate ) subformulas , while @xmath495 determines the _ logical structure _ of @xmath95 . an action @xmath496 can be seen as a kind of @xmath497-ary tensor product " @xmath498 indexed by the name @xmath26 . the whole set @xmath495 can be thought of as @xmath499-ary additive sum " of its elements : @xmath500 in appendix [ pol ll ] we give a more precise correspondence between logical connective in our sense and connectives of polarized linear logic @xcite . [ example ] consider the logical connective @xmath501 . by the previous discussion , we can intuitively think of it as @xmath502 when @xmath95 is applied to @xmath503 , it gives the formula @xmath504 we now define behaviours built by logical connectives . [ d - log conn beh ] given an @xmath497-ary name @xmath26 , an @xmath40-ary logical connective @xmath505 with @xmath506 and behaviours @xmath507 we define : 1 . @xmath508 2 . @xmath509 3 . @xmath510 where the indices @xmath511 vary for each @xmath512 and are determined by the variables @xmath513 . we call the set @xmath514 the _ ethics _ of @xmath515 . [ ethics ] an ethics is a set of atomic predesigns which are by construction linear in @xmath399 . it can be seen as a generator " of a behaviour defined by logical connectives in the following sense . for positives , we have by definition @xmath516 . for negatives , we have by proposition [ closure operator ] ( 3 ) : @xmath517 let @xmath95 be the logical connective as given in example [ example ] and @xmath518 negative behaviours . we have @xmath519 . [ linear ] logical connectives @xmath520{180}{\&}\ , } , \with , { \mathord{\uparrow}},\bot , \top$ ] can be defined if the signature @xmath226 contains a nullary name @xmath521 , unary names @xmath522 and a binary name @xmath523 . we also give notations to their duals for readability . @xmath524 where @xmath228 denotes the empty sequence . we do not have exponentials here , because we are working in a nonlinear setting so that they are already incorporated into the connectives . with these logical connectives we can build behaviours corresponding to usual linear logic types ( we use infix notations such as @xmath525 rather than the prefix ones @xmath526 ) . @xmath527 the next theorem illustrates a special feature of behaviours defined by logical connectives . it also suggests that nonlinearity and universal nondeterminism play dual roles . [ p - duphalf ] [ fundi ] [ p - log ] let @xmath528 be an arbitrary positive behaviour . 1 . @xmath529 \in \bp$ ] . 2 . @xmath530 . moreover , if @xmath528 is _ obtained by applying a logical connective _ , that is @xmath531 for some @xmath95 , @xmath532 , then : 1 . the converse of _ ( 1 ) _ ( _ duplicability _ ) and 2 . the converse of _ ( 2 ) _ ( _ closure under @xmath21 _ ) hold . 1 . for any @xmath533 , we have @xmath534 { \rrbracket}=\maltese$ ] . hence , @xmath535[n / x_0 ] { \rrbracket}= \maltese$ ] , and so @xmath536 \in \bp^{\bot\bot } = \bp$ ] . 2 . by proposition [ conj and ort ] ( 3 ) , we have @xmath537 for any @xmath538 . since @xmath539 is a behaviour , it is upward closed with respect to @xmath410 . hence the claim holds . 3 . for the sake of readability , we consider the binary case and show that @xmath540 implies @xmath541 . the general case can be proven using the same argument . + let @xmath542 . to prove @xmath543 , by remark [ ethics ] , it is sufficient to show that @xmath544 is orthogonal to any @xmath545 . since by construction @xmath399 occurs only once at the head position of @xmath546 , we only have to show that @xmath547 . + let @xmath548 and @xmath549 so that @xmath550 . since @xmath551 is a predesign , we have by lemma [ l - facts ] ( 2 ) , ( 3 ) : @xmath552 { \rrbracket}= { \llbracket}p_a[\vec{k}/\vx ] { \rrbracket}\wedge { \llbracket}q_a[\vec{k}/\vx ] { \rrbracket}= { \llbracket}n \;|\ ; { { \overline{a}}}\langle \vec{k } \rangle { \rrbracket}\wedge { \llbracket}m \;|\ ; { { \overline{a}}}\langle \vec{k } \rangle { \rrbracket}.\ ] ] since @xmath553 , we have @xmath554 and @xmath555 . our claim then immediately follows . 4 . let @xmath556\in \bp = \coal\langle \bn_1,\ldots,\bn_n \rangle$ ] . it suffices to show that @xmath557 { \rrbracket}=\maltese$ ] holds for any @xmath558 . but we have just proven that @xmath559 , and so @xmath560[n\wedge m / x_0 ] { \rrbracket}= { \llbracket}p[n\wedge m / x_1 , n\wedge m / x_2 ] { \rrbracket}= \maltese$ ] . since @xmath561 by proposition [ conj and ort ] ( 3 ) , we have @xmath562 { \rrbracket}= \maltese$ ] . theorem [ p - log ] can be considered as an internal , monistic form of soundness and completeness for the contraction rule : soundness corresponds to point ( 1 ) while completeness to its converse ( 3 ) , duplicability . however , in the sequel we only use point ( 1 ) ( in theorem [ soundness ] ) and point ( 4 ) ( in lemma [ l - comp ] ) of theorem [ p - log ] . in @xcite , girard proposes a purely monistic , local notion of completeness , called _ internal completeness_. it means that we can give a precise and direct description to the elements of behaviours ( built by logical connectives ) without using the orthogonality and without referring to any proof system . it is easy to see that negative logical connectives enjoy internal completeness : [ t - negative ] let @xmath563 be a logical connective with @xmath564 and @xmath565 an atomic negative design . we have : @xmath566 where the indices @xmath567 are determined by the variables @xmath568 . let @xmath548 be an atomic negative design and @xmath569 . since @xmath570 $ ] is a predesign and @xmath399 occurs only at the head position of @xmath1 , we have by lemma [ l - facts ] ( 2 ) : @xmath571 { \rrbracket}= { \llbracket}{\textstyle\sum}a(\vx).p_a \;|\ ; { { \overline{a}}}\langle n_1,\ldots , n_m \rangle { \rrbracket}= { \llbracket}p_a[n_1/z_{i_1},\ldots , n_m / z_{i_m } ] { \rrbracket}.\ ] ] this means that @xmath572 ( see remark [ ethics ] ) if and only if for every @xmath573 and for every @xmath574 , @xmath575 { \rrbracket}=\maltese $ ] if and only if for every @xmath573 , @xmath576 ( see definition [ d - context ] ( b ) ) . notice that in the above , @xmath577 can be arbitrary when @xmath578 . thus our approach is `` immaterial '' in that we do not consider material designs ( see @xcite for the definition of material design ) . the original `` material '' version of internal completeness @xcite can be easily derived from our immaterial one . a remarkable example of internal completeness for negative behaviours is provided for the logical connective @xmath579 : @xmath580 above , the irrelevant components of the sum are suppressed by `` @xmath581 . '' up to materiality ( removal of irrelevant additive components ) , @xmath582 , which has been defined by _ intersection _ , is isomorphic to the _ cartesian product _ of @xmath528 and @xmath583 . this isomorphism is called _ the mystery of incarnation _ " in @xcite . as to positive connectives , @xcite proves internal completeness theorems for additive and multiplicative ones separately in the linear and deterministic setting . they are integrated in @xcite as follows : when the universe of standard designs is restricted to linear and deterministic ones , we have @xmath584 however , this is no more true with nonlinear designs . a counterexample is given below . [ example completeness ] let us consider the behaviour @xmath585 and the designs @xmath586 and @xmath587 of remark [ r - separation ] . by construction , @xmath1 belongs to @xmath528 . since @xmath588 , @xmath140 also belongs to @xmath528 . however , @xmath589 , since @xmath590 is not atomic and so can not belong to @xmath591 . this motivates us to directly prove completeness for proofs , rather than deriving it from internal completeness as in the original work @xcite . in @xcite a weaker form of internal completeness is proved , which is enough to derive a weaker form of full completeness : all _ finite _ `` winning '' designs are interpretations of proofs . while such a finiteness assumption is quite common in game semantics , we will show that it can be avoided in ludics . we end this section with the following remark . the main linear logic isomorphism , namely the exponential one @xmath592 can be expressed in our notation as @xmath593 . in our setting it is possible to prove that those behaviours are morally " isomorphic , in the sense that they are isomorphic if we consider designs equal up to materiality and @xmath55 are equal up to materiality in a behaviour @xmath594 if they only differ in occurrences of positive subdesign which are irrelevant for the normalization against designs of @xmath595 . ] . we can in fact define a pair of maps @xmath596 on designs such that : 1 . @xmath597 and @xmath598 ; 2 . if @xmath1 and @xmath140 are equal up to materiality in @xmath599 , then @xmath600 and @xmath601 are equal up to materiality in @xmath602 , and similarly for @xmath603 ; 3 . for any @xmath604 , we have that @xmath605 and @xmath1 are equal up to materiality in @xmath599 , and similarly for the other direction . we postpone a detailed study of isomorphisms of types and related issues to a subsequent work . having set up the framework , we now address the main problem : an interactive form of gdel completeness . we first introduce the proof system in [ proof system ] , then examine its soundness in [ ss - soundness ] , and finally prove completeness in [ completeness ] , in a way quite analogous to the proof of gdel s theorem based on proof search ( often attributed to schtte @xcite ) . we will now introduce a proof system . in our system , logical rules are automatically generated by logical connectives . since the names which constitute the logical connectives are chosen among the names of a signature @xmath30 , the set of logical connectives vary for each signature @xmath226 . thus , our proof system is parameterized by @xmath226 . if one chooses @xmath226 rich enough , the constant - only fragment of polarized linear logic ( @xcite ; see also @xcite ) can be embedded , as we will show in appendix [ pol ll ] . in the sequel , we focus on _ logical _ behaviours , which are composed by using logical connectives only . [ logical behaviours ] a behaviour is * logical * if it is inductively built as follows ( @xmath95 denotes an arbitrary logical connective ) : @xmath606 notice that the orthogonal of a logical behaviour is again logical . as advocated in the introduction , our monistic framework renders both proofs and models as homogeneous objects : designs . [ proof , models ] a * proof * is a standard design ( definition [ d - standard ] ) in which all the conjunctions are unary . in other words , a proof is a total , deterministic and @xmath90-free design without cuts and identities . a * model * is a linear standard design ( in which conjunctions of arbitrary cardinality may occur ) . we will use proofs as proof - terms for syntactic derivations in the proof system to be introduced below . in that perspective , it is reasonable to exclude designs with non - unary conjunctions from proofs , because they do not have natural counterparts in logical reasoning . for instance , the nullary conjunction ( daimon ) and the binary one would correspond to the following `` inference rules '' respectively : @xmath607 with @xmath608 an arbitrary sequent . notice that we have not specified yet what a proof actually proves . hence it might be better called `` proof attempt '' or `` untyped proof '' or `` para - proof . '' on the other hand , we restrict models to linear designs just to emphasize the remarkable fact that _ linear _ designs do suffice for defeating any failed proof attempt that is possibly _ nonlinear_. given a design @xmath56 , let @xmath609 be the set of occurrences of positive actions @xmath31 in @xmath56 . the _ cardinality _ of @xmath56 is defined to be the cardinality of @xmath609 . for instance , the fax @xmath610 ( see section [ ortho - s ] ) is an infinite design in this sense . also , both proofs and models can be infinite . a _ positive _ ( resp . _ negative _ ) _ sequent _ is a pair of the form @xmath611 ( resp . @xmath612 ) where @xmath1 is a positive proof ( resp . @xmath79 is a negative proof ) and @xmath444 is a positive context of logical behaviours ( definition [ d - context ] ( a ) ) such that @xmath613 ( resp . @xmath614 ) . we write @xmath615 for a generic sequent . intuitively , a sequent @xmath615 should be understood as a claim that `` @xmath56 is a proof of @xmath616 '' or `` @xmath56 is of type @xmath616 . '' our proof system consists of two sorts of inference rules : 1 . a _ positive _ rule @xmath617 : + @xmath618{z|\coa\langle m_1 , \dots , m_m \rangle\vdash \bgamma } { m_1 \vdash \bgamma , \bn_{i_1 } & \dots & m_m \vdash \bgamma , \bn_{i_m } & ( z:\coal\langle\bn_1 , \dots , \bn_n\rangle \in \bgamma)}\ ] ] where @xmath488 , @xmath489 and @xmath619 so that the indices @xmath620 are determined by the variables @xmath621 . 2 . a _ negative _ rule @xmath622 : @xmath623{{\textstyle\sum}a(\vx).p_a \vdash \bgamma , \alpha(\bp_1 , \dots , \bp_n ) } { \{p_a \vdash \bgamma , z_{i_1 } : \bp_{i_1 } , \dots , z_{i_m } : \bp_{i_m}\}_{a(\vx)\in \alpha_0 } } \ ] ] where , as in the positive rule , the indices @xmath624 are determined by the variables @xmath621 for each @xmath512 . + we assume that @xmath37 are fresh , do not occur in @xmath444 . this does not cause a loss of generality since variables in @xmath95 can be renamed ( see definition [ d - lc ] ) . + notice that a component @xmath70 of @xmath47 can be arbitrary when @xmath625 . hence we again take an `` immaterial '' approach ( theorem [ t - negative ] ) . observe that the positive rule @xmath617 involves implicit uses of the contraction rule on positive behaviours . the weakening rule for positive behaviours is implicit too ; in the bottom up reading of a proof derivation , unused formulas are always propagated to the premises of any instance of rule . it should also be noted that proof search in our system is deterministic . in particular , given a positive sequent @xmath626 , the head variable @xmath627 and the first positive action @xmath31 completely determine the next positive rule to be applied bottom - up ( if there is any ) . it is also possible to adopt a `` material '' approach in the proof system by simply requiring @xmath628 when @xmath625 in the rule @xmath622 . then a proof @xmath56 is finite ( @xmath609 is a finite set ) whenever @xmath629 is derivable for some @xmath630 . thus , as in ordinary sequent calculi , our proof system accepts only _ essentially finite _ proofs for derivable sequents ( finite up to removal of irrelevant parts ) . [ example top fax ] to clarify the last point , we observe that for any ( _ possibly infinite _ ) negative proof @xmath79 with @xmath631 , the sequent @xmath632 is derivable by the instance of the negative rule with @xmath633 . in fact , this corresponds to the usual top - rule of linear logic ( see also example [ linear rules ] ) : this means that for a ( possibly infinite ) negative proof @xmath79 there is a finite derivation of @xmath632 . by contrast , in the `` material '' approach we only have where @xmath634 is the unique negative proof which has cardinality @xmath635 . [ linear rules ] for linear logic connectives ( example [ linear ] ) , the positive and negative rules specialize to the following ( taking here the `` material '' approach ) : @xmath636{z|\bullet\langle m_1 , m_2\rangle \vdash \bgamma } { m_1 \vdash \bgamma , \bn_1 & m_2 \vdash \bgamma , \bn_2 & ( z:\bn_1 \pmb{\otimes } \bn_2 \in \bgamma ) } \quad \infer[\pmb{{\,\rotatebox[origin = c]{180}{\&}\,}}]{\wp(x_1 , x_2 ) . p \vdash \bgamma , \bp_1 \pmb{{\,\rotatebox[origin = c]{180}{\&}\ , } } \bp_2}{p \vdash \bgamma , x_1 : \bp_1 , x_2 : \bp_2}\\[1em ] \infer[(\pmb{\llplus } , \iota_i)]{z|\iota_i\langle m\rangle \vdash \bgamma } { m \vdash \bgamma , \bn_i & ( z:\bn_1 \pmb{\llplus } \bn_2 \in \bgamma ) } \qquad \infer[\pmb{\llwith}]{\pi_1(x_1 ) . p_1 + \pi_2(x_2 ) . p_2 \vdash \bgamma , \bp_1 \pmb{\llwith } \bp_2}{p_1 \vdash \bgamma , x_1 : \bp_1 & p_2 \vdash \bgamma , x_2 : \bp_2}\\[1em ] \infer[({\pmb{\mathord{\downarrow } } } , { \mathord{\downarrow}})]{z|{\mathord{\downarrow}}\langle n\rangle \vdash { { \mathbf{\gamma } } } } { n \vdash \bgamma , { { \mathbf{n } } } & ( z : { \pmb{\mathord{\downarrow}}}{{\mathbf{n}}}\in \bgamma ) } \qquad \infer[{\pmb{\mathord{\uparrow}}}]{{\mathord{\uparrow}}(x).p \vdash \bgamma , { \pmb{\mathord{\uparrow}}}\bp}{p \vdash \bgamma , x : \bp}\\[1em ] \infer[(\llone,\overline{*})]{z|\overline{*}\vdash\bgamma } { ( z:\llone \in \bgamma ) } \qquad \infer[({\pmb{\bot}})]{*.p\vdash\bgamma,{\pmb{\bot}}}{p\vdash\bgamma } \qquad \infer[({\pmb{\top}})]{{\textstyle\sum}a(\vx).\omega \vdash\bgamma,{\pmb{\top } } } { } \end{array}\ ] ] the inference rules given above are all sound . namely we have : [ soundness ] if @xmath615 is derivable in the proof system , then @xmath637 . by induction on the length of the derivation of @xmath615 . we have two cases , one for each sort of rule . 1 . suppose that the last inference rule is + @xmath618{z|\coa\langle m_1 , \dots , m_m \rangle\vdash \bgamma } { m_1 \vdash \bgamma , \bn_{i_1 } & \dots & m_m \vdash \bgamma , \bn_{i_m } & ( z:\coal\langle\bn_1 , \dots , \bn_n\rangle \in \bgamma)}\ ] ] where @xmath638 and @xmath639 for some @xmath640 . + the induction hypothesis gives us @xmath641 for every @xmath642 . by lemma [ l - closure ] ( 3 ) , @xmath643 { \rrbracket}\in \bn_{i_j}$ ] for every @xmath644 and by definition [ d - lc ] , we have that @xmath645 , that is @xmath646 . + applying lemma [ l - closure ] ( 1 ) , we get @xmath647 and by theorem [ p - duphalf ] ( 1 ) we conclude @xmath648 . 2 . suppose now that the last inference rule is + @xmath623{{\textstyle\sum}a(\vx).p_a \vdash \bgamma , \alpha(\bp_1 , \dots , \bp_n ) } { \{p_a \vdash \bgamma , \vx : \vec{\bp}_a\}_{a(\vx)\in \alpha_0 } } \ ] ] where @xmath649 and @xmath650 stands for @xmath651 . we assume that the variables @xmath652 and @xmath37 are disjoint in any premise . + the induction hypothesis gives us @xmath653 for every @xmath654 . by lemma [ l - closure ] ( 1 ) , for every @xmath655 , @xmath656 { \rrbracket}\models \vx : \vec{\bp}_a$ ] . + then , we can apply theorem [ t - negative ] to obtain @xmath657 , that is @xmath658 . notice that in @xmath659 the components @xmath70 for @xmath660 can be arbitrary . + we finally apply lemma [ l - closure ] ( 3 ) and conclude @xmath661 . although our proof system does not include a cut rule officially , the semantics validates it as follows . 1 . if @xmath662 and @xmath663 , then @xmath465 { \rrbracket}\models \bgamma$ ] . 2 . if @xmath664 and @xmath663 , then @xmath471 { \rrbracket}\models \bn,\bgamma$ ] . let @xmath444 be @xmath473 , let @xmath665 and write @xmath666 for @xmath667 . 1 . by lemma [ l - closure ] , we have @xmath668 { \rrbracket}\models z : \bp$ ] and @xmath669 { \rrbracket}\models \bp{{}^{\bot}}$ ] , so that @xmath670 \in \bp$ ] and and @xmath671 . hence , @xmath672[m'/x_0 ] { \rrbracket}= { \llbracket}p'[m'/z ] { \rrbracket}= \maltese$ ] . from this fact and associativity ( theorem [ t - assoc ] ) , we can derive @xmath673 { \rrbracket}[\vec { k } /\vx ] { \rrbracket}=\maltese$ ] , which proves @xmath465 { \rrbracket}\models \bgamma$ ] . 2 . let @xmath140 be an arbitrary design in @xmath674 . by lemma [ l - closure ] , we obtain @xmath468 { \rrbracket}\models \bgamma , z : \bp$ ] and @xmath675 { \rrbracket}[\vec{k}/\vx ] { \rrbracket}\models z : \bp$ ] . on the other side , we have @xmath669 { \rrbracket}\models \bp{{}^{\bot}}$ ] . from @xmath676 \in \bp$ ] and and @xmath671 , we obtain @xmath677[m'/x_0 ] { \rrbracket}= { \llbracket}q'[m'/z ] { \rrbracket}= \maltese$ ] . from this fact and associativity , we can derive @xmath678 { \rrbracket}[\vec { k } /\vx ] \ ; /x_0 ] { \rrbracket}=\maltese$ ] , which proves @xmath471 { \rrbracket}\models \bn,\bgamma$ ] . thanks to the previous proposition , we can naturally strengthen our proof system as follows . first , we consider sequents of the form @xmath615 where @xmath56 is a proof with cuts " ( a proof in the sense of definition [ proof , models ] except that the cut - freeness condition is not imposed ) . second , we add the following cut rule : @xmath679{d[n / z]\vdash \mathbf{\xi},\bgamma } { d \vdash \mathbf{\xi } , \bgamma , z:\bp & n \vdash \bgamma , \bp^\bot}\ ] ] where @xmath680 is either empty or it consists of a negative logical behaviour @xmath450 . the soundness theorem can be naturally generalized as follows : if @xmath615 is derivable in the proof system with the cut rule above , then @xmath681 . let us finally establish the other direction of theorem [ soundness ] , namely : [ full completeness ] a sequent @xmath682 is derivable in the proof system if and only if @xmath683 . in particular , for any positive logical behaviour @xmath528 and a proof @xmath1 , @xmath684 is derivable if and only if @xmath685 . similarly for the negative case . before proving the theorem , let us recall a well - established method for proving gdel completeness based on proof search ( often attributed to schtte @xcite ) . it proceeds as follows : 1 . given an unprovable sequent @xmath608 , find an open branch in the cut - free proof search tree . 2 . from the open branch , build a countermodel @xmath2 in which @xmath608 is false . the proof below follows the same line of argument . we can naturally adapt ( 1 ) to our setting , since the bottom - up cut - free proof search in our proof system is deterministic in the sense that at most one rule applies at each step . moreover , it never gets stuck at the negative sequent , since a negative rule is always applicable bottom - up . adapting ( 2 ) is more delicate . for simplicity , we assume that the sequent @xmath686 is positive ; the argument below can be easily adapted to the negative case . so , suppose that a positive sequent @xmath687 with @xmath688 does not have a derivation . by knig s lemma , there exists a branch @xmath689 in the cut - free proof search tree , @xmath690 which is _ either _ finite and has the topmost sequent @xmath691 with @xmath692 to which no rule applies anymore , _ or _ infinite . in the latter case , we set @xmath693 . our goal is to build models @xmath694 such that @xmath695 ] \!]}= \omega.\ ] ] more generally , we define negative designs 1 . @xmath696 for every @xmath310 ( @xmath697 if @xmath698 ) ; 2 . @xmath699 for every variable @xmath75 occurring in the branch . below , @xmath95 and @xmath73 stand for logical connectives : @xmath488 , @xmath700 . to define @xmath696 we distinguish three cases : ( i ) when @xmath701 and @xmath702 , let @xmath703 . \(ii ) when @xmath704 and @xmath705 , suppose that @xmath691 is of the form @xmath706 but @xmath707 so that the proof search gets stuck . then let @xmath708 . recall that the partial sum @xmath709 has @xmath710 as component by our convention . \(iii ) for @xmath711 , suppose that the relevant part of the branch @xmath689 is of the form : @xmath712{z|\coa\langle m_1,\ldots , m_{{k-1 } } , { \textstyle\sum}b(\vy).p_{i+1 } , m_{k+1},\ldots , m_{{m}}\rangle\vdash \btheta_i } { \infer[(\beta)]{{\textstyle\sum}b(\vy ) . \vdash \btheta_i , \bn_{i_k } } { \infer*{p_{i+1 } \vdash \btheta_i , y_1 : \bq_1 , \dots , y_l : \bq_l}{}}}},\ ] ] where @xmath713 contains @xmath714 , @xmath715 with @xmath716 , and @xmath717 @xmath718 , @xmath719 with @xmath720 . namely , the situation is as follows ( to be read bottom - up ) : 1 . the head variable of @xmath721 is @xmath627 , so @xmath714 is chosen from the context @xmath713 and the rule @xmath617 is applied . among @xmath497 upper sequents , the @xmath499th one is taken in the branch . @xmath722 is negative , and the unique negative behaviour in @xmath723 is @xmath717 @xmath718 , so the rule @xmath724 is applied . among the upper sequents ( recall that there is one sequent for each action in @xmath725 ) , the one corresponding to @xmath726 is taken in the branch . in this case , we define @xmath727 here , the main additive component of @xmath696 begins with @xmath728 because ( 1 ) @xmath721 begins with the positive action @xmath31 , ( 2 ) the @xmath499th upper sequent is taken in the branch , and ( 3 ) the upper sequent corresponding to @xmath729 is taken . the other additive components @xmath730 are needed to ensure that our countermodel belongs to the behaviour @xmath731 ( see lemma [ l - comp ] ( 1 ) ) . the subdesigns @xmath732 are given by @xmath733 notice that each @xmath734 is a negative design , so the above conjunction is a defined operation ( in the sense of definition [ conj extension ] ( 2 ) ) . we claim that @xmath696 is well - defined , because variables @xmath71 are chosen fresh , so do not appear freely below @xmath735 . hence subdesigns @xmath732 do not have @xmath736 with @xmath737 as conjunct . namely , @xmath696 depends only on @xmath734 with @xmath738 . this gives rise to a recursive procedure and @xmath696 arises in the limit of the procedure . notice also that the set @xmath739 can be empty and in such a case , we have that @xmath740 . the above is an instance of corecursive definition . it is possible to formally justify it by employing _ design generators _ developed in @xcite ( see in particular theorem 2.12 of @xcite ) . an alternative way is to define @xmath696 ( and @xmath741 ) as the limit of its finite approximations . here we briefly outline this latter approach . we assume that @xmath693 . the idea is to chop off the branch @xmath689 at height @xmath742 , where @xmath742 is an arbitrary natural number , and define finite approximations @xmath743 and @xmath744 . then @xmath696 and @xmath741 arise as the limit when @xmath745 . more concretely , given a natural number @xmath742 , we define @xmath743 by downward induction from @xmath746 to @xmath747 as follows : 1 . when @xmath748 , the sequent @xmath749 is of the form @xmath750 . we let @xmath751 . when @xmath752 , we proceed as in the case ( iii ) above . namely , @xmath753 where actions @xmath93 , @xmath754 and the index @xmath755 are determined as before . now observe that the sequence @xmath756 is `` monotone increasing '' in the sense that @xmath757 has more conjuncts than @xmath758 whenever @xmath759 . the same for @xmath760 with @xmath761 . hence we can naturally obtain the `` limits '' @xmath762 this construction ends up with the same as the previous recursive one . observe that each @xmath696 and @xmath699 thus constructed are surely models , atomic linear designs . theorem [ full completeness ] is a direct consequence of the following two lemmas . the first lemma crucially rests on _ induction on logical behaviours _ , that is an analogue of _ induction on formulas _ , which lies at the core of logical completeness in many cases . [ l - comp ] for @xmath763 appearing in the branch @xmath689 above , suppose that @xmath721 has a head variable @xmath627 and @xmath764 . then : 1 . @xmath765 ; 2 . @xmath766 . by induction on the construction of @xmath767 . 1 . suppose that @xmath704 . since @xmath80 does not have a head variable , the case ( i ) does not apply . hence we are in the case ( ii ) , namely @xmath768 , for some logical connective @xmath95 and logical behaviours @xmath532 . thus , @xmath769 , and @xmath708 . + by internal completeness for negative connectives ( theorem [ t - negative ] ) , we have @xmath770 where @xmath771 and the expression @xmath772 abbreviates the positive context @xmath773 . since @xmath774 trivially holds for every @xmath512 , we have @xmath775 . + when @xmath776 , the case ( iii ) applies . in the same notation , we have that @xmath777 , @xmath717 @xmath778 , and @xmath779 where actions @xmath93 , @xmath754 , the index @xmath755 and the variables @xmath780 are determined by the relevant part of the branch @xmath689 as described above . + by induction hypothesis on ( 2 ) , we have that @xmath781 . hence , @xmath782 . since @xmath783 are atomic ( closed ) , we may derive @xmath784 . we also have @xmath785 for every @xmath786 . hence , by internal completeness again , @xmath787 . 2 . it follows from ( 1 ) since @xmath788 is a negative logical behaviour and so closed under @xmath21 ( theorem [ p - log ] ( 4 ) ) . the proof of the next lemma suggests a similarity between the construction of our countermodels and the _ bhm - out _ technique ( see , @xcite ) , that constructs a suitable term context in order to visit a specific position in the bhm tree of a given @xmath16-term . recall that the initial sequent of our open branch @xmath689 is @xmath687 with @xmath688 , so that @xmath789 . we have : [ l - bohm ] @xmath695 ] \ ! ] } = \omega.\ ] ] we first prove that there is a reduction sequence @xmath790 { -\!\!\!\!\!\rightharpoonup}^ * p_{i+1 } [ \cm(w_1)/w_1 , \dots , \cm(w_t)/w_t]\ ] ] for any @xmath776 , where @xmath791 and @xmath792 are the free variables of @xmath721 and @xmath793 , respectively . suppose that @xmath721 is as in the case ( iii ) above , so has the head variable @xmath794 . by writing @xmath795 $ ] for @xmath796 $ ] and noting that @xmath797 is a ( defined ) conjunction that contains @xmath798 as conjunct , we have : @xmath799 & = & \cm(z ) \ | \ \coa\langle m_1[\theta],\ldots , m_{{k-1}}[\theta ] , { \textstyle\sum}b(\vy).p_{i+1}[\theta ] , m_{k+1}[\theta],\ldots , m_{{m}}[\theta]\rangle \\ & { -\!\!\!\!\!\rightharpoonup } & \left({\textstyle\sum}b(\vy).p_{i+1}[\theta]\right ) \ ; & { -\!\!\!\!\!\rightharpoonup } & p_{i+1}[\theta , \cm(y_{1})/ y_{1 } , \ldots , \cm(y_{l})/y_{l}],\\ \end{array}\ ] ] as desired . when @xmath693 , we have obtained an infinite reduction sequence from @xmath800 $ ] . otherwise , @xmath800 { -\!\!\!\!\!\rightharpoonup}^ * p_{max}[\theta]$ ] , for some substitution @xmath795 $ ] . in case ( i ) , we have @xmath801 = \omega$ ] , while in case ( ii ) , we have @xmath802 . so , @xmath803 = z|\coc\langle \vec{m}\rangle [ \theta ] = \cm(z)|\coc\langle \vec{m}[\theta]\rangle { -\!\!\!\!\!\rightharpoonup}\omega,\ ] ] because @xmath797 contains @xmath709 as conjunct , and @xmath804 has @xmath710 as component . theorem [ full completeness ] now follows easily . suppose that @xmath805 is not derivable . then we obtain models @xmath806 , , @xmath807 by lemma [ l - comp ] and @xmath808 ] \!]}=\omega$ ] by lemma [ l - bohm ] . this means that @xmath809 . our explicit construction of the countermodels yields a by - product : [ lowen ] 1 . let @xmath1 be a proof and @xmath528 a logical behaviour . if @xmath810 , then there is a countable model @xmath811 ( @xmath812 is a countable set ) such that @xmath813 . 2 . furthermore , when @xmath1 is linear , there is a finite and deterministic model @xmath811 such that @xmath813 . the second statement is due to the observation that when @xmath1 is linear the positive rule @xmath617 can be replaced with a linear variant : @xmath814{z|\coa\langle m_1 , \dots , m_m \rangle\vdash \bgamma , z:\coal\langle\bn_1 , \dots , \bn_n\rangle } { m_1 \vdash \bgamma_1 , \bn_{i_1 } & \dots & m_m \vdash \bgamma_m , \bn_{i_m}},\ ] ] where @xmath815 are disjoint subsets of @xmath444 . we then immediately see that the proof search tree is always finite , and so is the model @xmath699 . it is deterministic , since each variable occurs at most once as head variable in a branch so that all conjunctions are at most unary . we have presented a gdel - like completeness theorem for proofs in the framework of ludics , aiming at linking completeness theorems for provability with those for proofs . we have explicitly constructed a countermodel against any failed proof attempt , following schtte s idea based on cut - free proof search . our proof employs knig s lemma and reveals a sharp opposition between finite proofs and infinite models , leading to a clear analogy with lwenhein - skolem theorem . our proof also employs an analogue of the bhm - out technique @xcite ( see the proof of lemma [ l - bohm ] ) , though it does not lead to the separation property ( remark [ r - separation ] ) . in hyland - ong game semantics , player s innocent strategies most naturally correspond to possibly infinite bhm trees ( see , @xcite ) . one could of course impose finiteness ( or compactness ) on them to have correspondence with finite proofs . but it would not lead to an explicit construction of opponent s strategies defeating infinite proof attempts . although finiteness is imposed in @xcite too , our current work shows that it is not necessary in ludics . our work also highlights the duality : @xmath816 the principle is that _ when proofs admit contraction , models have to be nondeterministic _ ( whereas they do not have to be nonlinear ) . a similar situation arises in some variants of @xmath16-calculus and linear logic , when one proves the separation property . we mention @xcite , where the authors add a nondeterministic choice operator and a numeral system to the pure @xmath16-calculus in order to internally ( interactively ) discriminate two pure @xmath16-terms that have different bhm trees . however , in contrast to our work , the nondeterminism needed for their purpose is of existential nature : a term converges if at least one of the possible reduction sequences starting from it terminates . in @xcite , the separation property for differential interaction nets @xcite is proven . a key point is that the exponential modalities in differential interaction nets are more `` symmetrical '' than in linear logic . in our setting , the symmetry shows up between nonlinearity and nondeterministic conjunctions ( nonuniform elements ) . it is typically found in theorem [ p - duphalf ] , which reveals a tight connection between duplicability of positive logical behaviours and closure under nondeterministic conjunctions of negative logical behaviours . similar nonuniform structures naturally arise in various semantical models based on coherence spaces and games , such as finiteness spaces @xcite , indexed linear logic and nonuniform coherence spaces @xcite , nonuniform hypercoherences @xcite , and asynchronous games @xcite ( see also @xcite ) . for future work , we plan to extend our setting by enriching the proof system with propositional variables , second order quantifiers and nonlogical axioms . by moving to the second order setting , we hope to give an _ interactive _ account to gdel s incompleteness theorems as well . we are deeply indebted to pierre - louis curien , who gave us a lot of useful comments . our thanks are also due to the anonymous referees . 10 andreoli , j .- : logic programming with focusing proofs in linear logic . * 2*(3 ) ( 1992 ) 297347 . barendregt , h. p. : the lambda calculus : its syntax and semantics . north - holland ( 1981 ) . basaldella , m. , faggian , c. : ludics with repetition ( exponentials , interactive types and completeness ) . in : lics . ( 2009 ) 375384 . bhm , c. : alcune propriet delle forme @xmath817-normali nel @xmath818-calcolo . publicazioni dellistituto per le applicazioni del calcolo * 696 * ( 1968 ) . boudes , p. : non - uniform hypercoherences . notes theor . * 69 * ( 2002 ) 6282 . bucciarelli , a. , ehrhard , t. : on phase semantics and denotational semantics : the exponentials . pure appl . 109*(3 ) ( 2001 ) : 205241 . curien , p .- l . : abstract bhm trees . struct . in comp . * 8*(6 ) ( 1998 ) 559591 . curien , p .- l . : introduction to linear logic and ludics , part ii . advances in mathematics ( china ) * 35*(1 ) ( 2006 ) 144 . curien , p .- l . : notes on game semantics . manuscript ( 2006 ) . curien , p .- , herbelin , h. : abstract machines for dialogue games . panoramas et synthses * 27 * ( 2009 ) 231275 . curien , p .- l . , munch - maccagnoni , g. : the duality of computation under focus . in : proc . of ifip tcs . dezani - ciancaglini , m. , intrigila , b. , venturini - zilli , m. : . in : ictcs98 . ( 1998 ) 123 . ehrhard , t. : finiteness spaces . struct . in comp . * 15*(4 ) ( 2005 ) 615646 . ehrhard , t. , regnier , l. : differential interaction nets . sci . * 364*(2 ) ( 2006 ) 166195 . faggian , c. : travelling on designs . in : csl . ( 2002 ) 427441 . faggian , c. : interactive observability in ludics : the geometry of tests . sci . * 350*(2 ) ( 2006 ) 213233 . faggian , c. , piccolo , m. : ludics is a model for the finitary linear pi - calculus . in : tlca . ( 2007 ) 148162 . faggian , c. , piccolo , m. : partial orders , event structures , and linear strategies . in : tlca . ( 2009 ) 95111 . girard , j .- y . : on the meaning of logical rules i : syntax vs. semantics . in berger , u. , schwichtenberg , h. , eds . : computational logic . heidelberg springer - verlag ( 1999 ) 215272 . girard , j .- y . : on the meaning of logical rules ii : multiplicatives and additives . foundation of secure computation , berger and schwichtenberg edts ( 2000 ) 183212 . girard , j .- y . : locus solum : from the rules of logic to the logic of rules . struct . in comp . sci . * 11*(3 ) ( 2001 ) 301506 . hyland , j.m.e . , ong , c.h.l . : on full abstraction for pcf : i , ii , and iii . * 163*(2 ) ( 2000 ) 285408 . lafont , y. : the finite model property for various fragments of linear logic . j. symb . log . * 62*(4 ) ( 1997 ) 12021208 . laurent , o. : tude de la polarization en logique . phd thesis , univ . aix - marseille ii ( 2002 ) . laurent , o. : polarized games . . logic * 130*(1 - 3 ) ( 2004 ) 79123 . maurel , f. : un cadre quantitatif pour la ludique . phd thesis , univ . paris vii ( 2004 ) . mazza , d. , pagani , m. : the separation theorem for differential interaction nets . in : lpar . ( 2007 ) 393407 . mellis , p .- a . : asynchronous games 2 : the true concurrency of innocence . * 358*(2 - 3 ) ( 2006 ) 200228 . schtte , k. : ein system des verknpfenden schliessens . logic grundlagenf . * 2 * ( 1956 ) 5567 . terui , k. : ludics . ( 2008 ) to appear in theor . in this appendix , we show a correspondence between the proof system for ludics introduced in [ proof system ] and the constant - only propositional fragment of _ polarized linear logic _ @xmath819 @xcite . this will ensure that our proof system is rich enough to capture a constructive variant of constant - only propositional classical logic . we recall the syntax of the constant - only propositional fragment of @xmath819 . the _ formulas _ are split into positive and negative ones and generated by the following grammar : @xmath820 the linear negation is defined in the usual way . sequent _ is of the form @xmath821 with @xmath822 a multiset of formulas . the inference rules of @xmath819 are given below : strict sequents will play a crucial role for the correspondence between @xmath819 and the proof system for ludics ( theorem [ t - llp ] ) . the intuition , which we will formalize later , is that a strict sequent @xmath824 can be thought of as a sequent of the proof system of ludics ( omitting the information about designs ) of the form @xmath825 . a. positive subderivation : @xmath827 that consists of positive inference rules only , where @xmath828 is a positive formula obtained from @xmath829 by applying positive connectives , and @xmath511 . for instance , if @xmath828 is of the form @xmath830 , there are two positive subderivations with conclusion @xmath831 : + @xmath832 b. negative subderivation : @xmath833 that consists of negative inference rules only , where @xmath834 is a negative formula obtained from @xmath835 by applying negative connectives , and @xmath836 consist of formulas in @xmath837 . for instance , if @xmath834 is of the form @xmath838 , then there is ( essentially ) one negative subderivation with conclusion @xmath839 : + @xmath840 c. @xmath841-dereliction : @xmath842 the above decomposition motivates us to cluster the logical connectives of the same polarity into _ synthetic connectives _ ( @xcite ) . consider the expressions finitely generated by : @xmath843 where @xmath75 ranges over the set of variables . we write @xmath844 ( resp.@xmath845 ) to denote the set of variables occurring in @xmath846 ( resp . @xmath847 ) . @xmath846 is a _ positive synthetic connective _ if for every subexpression of @xmath846 of the form @xmath848 , @xmath849 and @xmath850 are disjoint . for instance , @xmath851 is a positive synthetic connective while @xmath852 is not . likewise , @xmath847 is a _ negative synthetic connective _ if for every subexpression of @xmath847 of the form @xmath853 , @xmath854 and @xmath855 are disjoint . this condition is needed when we translate synthetic connectives to logical connectives of ludics . we indicate the variables occurring in @xmath846 by writing @xmath856 , and similarly for @xmath847 . given a negative synthetic connective @xmath847 , its _ dual _ @xmath857 is obtained by replacing @xmath858 with @xmath859 , @xmath860 with @xmath861 , @xmath862 with @xmath863 , @xmath864 with @xmath865 , and @xmath841 with @xmath866 respectively , in each occurrence of symbol . @xmath867 is similarly defined . + the _ formulas _ of @xmath819 are then redefined inductively as follows : @xmath868 where @xmath869 is obtained from @xmath856 by substituting @xmath870 for @xmath871 ( @xmath872 ) . notice that when @xmath873 , @xmath874 can be any combination of @xmath859 and @xmath861 using @xmath863 and @xmath865 . to each positive synthetic connective @xmath875 , we can naturally associate a set of inference rules as follows . consider all possible positive subderivations with conclusion @xmath876 in the sense of ( i ) above . to each such derivation @xmath827 we associate an inference rule : @xmath877 for instance , to @xmath878 , we associate two inference rules : likewise , each negative synthetic connective @xmath880 comes equipped with a unique inference rule derived from the negative subderivation with conclusion @xmath881 ( see ( ii ) above ) . for instance , @xmath882 is equipped with : observe the asymmetry between the positive and negative cases here ; in the negative case , we leave the @xmath841-formulas @xmath884 in the premises . these formulas are to be dealt with by the @xmath841-dereliction rule . 1 . @xmath226 contains a nullary name @xmath521 and a unary name @xmath227 . if @xmath226 contains an @xmath40-ary name @xmath26 and an @xmath497-ary name @xmath69 , it also contains @xmath40-ary names @xmath886 , @xmath887 and an @xmath888-ary name @xmath889 ( example [ linear ] ) . given a negative synthetic connective @xmath847 , we inductively associate a set @xmath890 of negative actions of ludics as follows : @xmath891 notice that when @xmath892 , the variables @xmath37 occur in @xmath847 . hence @xmath893 above is certainly a negative action , since @xmath37 and @xmath71 are disjoint sequences due to the definition of negative synthetic connective . we finally let @xmath894 , where @xmath895 lists the variables occurring in @xmath847 . a positive synthetic connective @xmath846 is interpreted by @xmath896 . for instance , when @xmath878 and @xmath882 , we have @xmath897 with @xmath898 this induces a polarity - preserving translation from the formulas of @xmath819 to the logical behaviours of ludics : @xmath899 to establish a connection with @xmath819 , we simplify the proof system of [ proof system ] by taking its _ skeleton _ , namely by omitting all information about designs . the resulting proof system , which we call @xmath900 , consists of two sorts of inference rules : @xmath618{\vdash \bgamma } { \vdash \bgamma , \bn_{i_1 } & \dots & \vdash \bgamma , \bn_{i_m } & ( \coal\langle\bn_1 , \dots , \bn_n\rangle \in \bgamma ) } \qquad \infer[(\alpha ) ] { \vdash \bgamma , \alpha(\bp_1 , \dots , \bp_n ) } { \{\vdash \bgamma , \bp_{i_1 } , \dots , \bp_{i_m}\}_{a(\vx)\in \alpha_0}}\ ] ] where @xmath488 , @xmath489 , @xmath573 and the indices @xmath620 are determined by the variables @xmath621 . one can annotate derivations in @xmath900 with designs as in section [ proof system ] . therefore the above theorem means that ludics designs can be used as term syntax for @xmath819 , as far as strict sequents and derivations are concerned ( although we have to verify carefully that the translation preserves the reduction relation ) . 1 . to each action @xmath907 , we associate the synthetic connective @xmath908 ( @xmath909 , if @xmath26 is nullary ) ; 2 . to each logical connective @xmath910 , we associate the synthetic connective @xmath911 ( @xmath912 , if @xmath913 ) ; 3 . to each logical behaviour , we associate the formula of @xmath819 @xmath914 4 . to each positive context @xmath915 of system @xmath900 , we associate the multiset @xmath916 of formulas of @xmath819 ; 5 . to each negative context @xmath449 of system @xmath900 , we associate the multiset of formulas @xmath917 . it is routine to define an isomorphism between @xmath918 and @xmath919 ( resp . between @xmath920 and @xmath921 ) in some natural sense . moreover , the translation of a sequent of system @xmath900 always results in a strict sequent of @xmath819 . we therefore conclude by theorem [ t - llp ] :

in correlation analysis of binned data the quantity of interest is the covariance of observable @xmath0 between arbitrary bins @xmath1 and @xmath2 , given by @xmath3 where @xmath0 is the bin content and over - lines indicate averages over independent measurements . for example , in high - energy collisions between atomic nuclei @xmath0 could represent the number of sub - atomic particles produced and detected for each collision , or _ event _ , within bins defined by particle 3-momentum . for example , such bins could be constructed using the transverse momentum ( @xmath4 ) ( component of 3-momentum perpendicular to the direction of the colliding beams ) , the azimuth angle ( @xmath5 ) in the plane transverse to the beam , the pseudorapidity ( @xmath6 ) where @xmath7 and @xmath8 is the polar angle relative to the beam direction , the azimuthal angle difference ( @xmath9 ) for arbitrary particles 1 and 2 , and the pseudorapidity difference ( @xmath10 ) . quantity @xmath11 is the number of particle - pairs in 2d bin @xmath12 and @xmath13 is calculated by averaging product @xmath11 over all collision events in the event collection . pairs of particles from the same event are referred to as _ sibling _ pairs . averages @xmath14 and @xmath15 are calculated using all events in the collection where quantity @xmath16 is the _ reference_. for a given set of measured events the covariance will have a unique numerical value . the goal of this paper is to calculate the statistical error in this quantity by following and extending the error propagation and error reduction method of reid and trainor @xcite . in practical applications measured quantities @xmath17 , @xmath18 and @xmath11 are affected by experimental inefficiency , acceptance and contamination . inefficiency and acceptance losses in single particle counts and some contamination effects are readily corrected via ratio @xmath19 . however , two - particle inefficiencies , which occur when signals in the detectors from two particles are unresolved , can not be corrected this way . such inefficiencies are due to finite detector resolution and , if uncorrected , produce significant artifacts in the correlations for heavy - ion collisions @xcite . the conventional correction method @xcite involves removing particle pairs whose signals ( e.g. induced ionization , secondary particle showers , etc . ) fall within the resolution limits of the detector , and then removing pairs of particles from mixed - events which would have the same , relative locations in the detectors . corrected results are obtained by increasing the minimum required separation between the detected signals of two nearby particles from zero until ratio @xmath19 stabilizes . the practical consequence of this correction procedure is that the reference @xmath20 must be calculated by constructing uncorrelated pair counts in 2d bin @xmath12 by averaging over pairs of particles where each particle in a pair is selected from different collision events , referred to as _ mixed - events_. the statistical error of the covariance in eq . ( [ eq1 ] ) equals the standard deviation of the distribution of covariance values corresponding to independent , statistically equivalent event samples ( event collections ) of underlying parent distributions for quantities @xmath17 , @xmath18 and @xmath11 . analytical calculations of this error therefore represent @xmath17 , @xmath18 as random , event - wise fluctuating quantities relative to the parent distribution . the sibling and mixed - event pair - numbers are similarly represented . in ref . @xcite reid and trainor derived a practical mixed - event method for calculating the reference in which random , event - wise fluctuations ( noise ) in @xmath17 and @xmath18 , which are common to both sibling and mixed - event pair numbers , cancel in the covariance , significantly reducing the errors . for large data volumes summing the total number of mixed - event pairs can be computationally demanding . various event - mixing algorithms have been developed by the heavy - ion community to reduce the necessary computation time while retaining sufficient statistical accuracy . one such method was discussed in ref . the choice of the reference and the event - mixing method strongly affects the statistical errors in the final correlation measurement . in this paper the statistical noise reduction method of reid - trainor will be extended and applied to other , practical event - mixing algorithms . the consequences of these event - mixing choices , or references , for the statistical uncertainties in the correlations will be quantified . the present application is for ultra - relativistic heavy - ion collisions such as those measured by the star experiment @xcite and by the experiments at the large hadron collider ( lhc ) @xcite . the methods presented here are also directly applicable to correlation analysis of multi - particle production from any type of particle collision . the present event - mixing technique for constructing an uncorrelated reference distribution has analogues in , for example , cosmology and acoustics . measurements of the relative distance correlation between galaxies within an angular patch of the sky require an uncorrelated reference distribution . the latter can be constructed from cross - correlated pairs of galaxies observed in different sky patches or from randomly generated distributions of galaxies @xcite . in acoustical analysis of multiple , independent time series the autocorrelation , or time - lag dependence , for each time series must be referenced to a cross correlation between two independent time series having the same lag time @xcite . in these two examples angular patches of sky or individual time series correspond to collision events and binned numbers of galaxies or acoustical amplitudes correspond to binned number of particles in the present analysis . this paper is organized as follows . in sec . [ sec : general ] the reid - trainor procedure is derived and extended . in sec . [ sec : other ] their method is applied to other event - mixing algorithms . monte carlo studies are discussed in sec . [ sec : montecarlo ] . conclusions are given in sec . [ sec : conclusions ] . in ref . @xcite the event collection was separated into pairs of events with similar , total number of detected particles , or _ multiplicity_. event - mixing was only applied between the two events in each pair . for arbitrary 2d bin @xmath12 ( @xmath21 ) the total number of sibling pairs of particles in the collection is given by the sum _ mn & = & _ g=1^n_g _ j=1 ^ 2 ( |m+_j)(|n + _ j ) [ eq2 ] and the mixed - event pair sum is given by _ mn & = & _ g=1^n_g _ j^ > j=1 ^ 2 [ ( |m+_j)(|n + _ j^ ) + ( |m+_j^)(|n + _ j ) ] [ eq3 ] where indices @xmath22 and @xmath23 denote events and event groups , respectively . variables @xmath24 and @xmath25 are the parent distribution number of particles in bins @xmath26 ( bins are denoted with subscripts ) and fluctuations are represented with random variables @xmath27 and @xmath28 as in ref.@xcite . event group index @xmath23 is suppressed in the notation for fluctuations @xmath27 and @xmath28 . the number of pairs of events in the collection is @xmath29 where @xmath30 is the number of events in the collection . note that averages of random variables @xmath27 and @xmath28 within an event collection only vanish in the @xmath31 limit . for large event numbers the summations in eqs . ( [ eq2 ] ) and ( [ eq3 ] ) are approximately given by _ mn & & n_events ( |m|n + ) + m_mn & & n_events ( |m|n ) , [ eq4 ] where @xmath32 is non - zero if the fluctuations in bins @xmath1 and @xmath2 are correlated . for relativistic heavy - ion collisions @xmath33 @xcite and for the purpose of calculating statistical errors the small contributions of @xmath34 can be neglected . the large event number limits are defined as @xmath35 . from the above discussion the correlation quantity of interest is @xmath36 and we calculate the number of correlated pairs per reference pair given by @xcite & = & - 1 _ mn - 1 . [ eq5 ] the statistical error in @xmath37 equals the statistical error in @xmath38 , denoted by @xmath39 and is given by ( ) ^2 & = & ( ) ^2 + ( ) ^2 - [ eq6 ] where @xmath40 , etc . are variances , @xmath41 is a covariance , and @xmath42 . simplifying eq . ( [ eq6 ] ) and averaging over event collections yields ( _ mn)^2 & = & |r_mn^2 + & = & + & = & [ eq7 ] where angle - brackets represent the average over independent event collections , @xmath43 ^ 2 $ ] is the variance of difference @xmath44 , @xmath45 , and @xmath46 . the key result of ref . @xcite was to show that event - pair - wise mixing eliminates contributions of single - particle fluctuations , leaving only those contributions from fluctuations in the number of pairs . using eqs . ( [ eq2 ] ) and ( [ eq3 ] ) the variance in the numerator of eq . ( [ eq7 ] ) simplifies to ( s_mn - m_mn)^2 & = & ^2 + & = & _ j=1^n_events _ j^2 _ j^2 + _ j=1,odd^n_events ( _ j^2 _ j+1 ^ 2 + _ j+1 ^ 2 _ j^2 ) [ eq8 ] where averages over products of bin - wise fluctuations from different events vanish ( see appendix a ) . carrying out the above event averaging gives ( s_mn - m_mn)^2 & = & n_events _ ^2 _ ^2 + n_g _ ^2 _ ^2 + _ ^2 _ ^2 + & = & 2 n_events _ ^2 _ ^2 + & & 2 n_events |m|n 2 n_events |m|n , [ eq9 ] where @xmath47 and @xmath48 are the variances in bins @xmath1 and @xmath2 , respectively . these variances result from the following averages : _ j=1^n_events _ j^2 _ j^2 & = & _ ^2 _ ^2 , + _ j=1^n_events _ j^2 _ j+1 ^ 2 & = & _ ^2 _ ^2 . these factorized results are valid if the single - particle fluctuations in bins @xmath1 and @xmath2 are uncorrelated , consistent with the above assumption that correlation corrections to the statistical errors are negligible . in the last two steps in eq . ( [ eq9 ] ) the limit of poisson distributions is assumed , denoted here and in the remainder of this paper with a limit symbol ( arrow ) . in the last line in eq . ( [ eq9 ] ) event collection average @xmath49 is approximated with @xmath50 from a single event - collection . correlated single - particle fluctuations in the event - wise occupancies of different bins ( other than that caused by collision dynamics ) could also be significant if the range of multiplicities in the event collection is large or if the bin size is large relative to the detector acceptance . in such cases the event - wise frequency distributions of @xmath51 and @xmath52 would be inconsistent with a poisson distribution . in sec . [ sec : montecarlo ] the frequency distributions for realistic bin sizes are shown to be accurately represented with poisson distributions . finally , the statistical error in @xmath38 for the event - mixing algorithm in ref . @xcite is & = & = . [ eq10 ] note that this error equals @xmath53 , or the square - root of the sum of the relative variances of the sibling and mixed - event pair numbers and is not determined by the errors in the bin - wise single - particle number @xcite . for 2d diagonal bins ( @xmath54 ) only unique particle pairs are counted ( no self - pairs ) which requires a separate derivation . in this case @xmath55 and @xmath56 are given by _ mm & = & _ j=1,odd^n_events [ eq11 ] + m_mm & = & _ j=1,odd^n_events ( |m+_j ) ( |m+_j+1 ) [ eq12 ] and as before the large event number limits are |s_mm & = & n_events ( |m^2 - |m + _ ^2 ) n_events |m^2 [ eq13 ] + variance quantity yields ( s_mm - m_mm)^2 & = & ^2 + & = & _ j=1,odd^n_events ( _ j^2 + _ j+1 ^ 2 -2 _ j _ j+1 - 2|m - _ j - _ j+1)^2 + & = & _ j=1^n_events ( _ j^4 + _ j^2 + 2|m^2 - 2_j^3 - 4|m_j^2 ) + & & + 6 _ j=1,odd^n_events _ j^2 _ j+1 ^ 2 [ eq15 ] where in going from the first to the second line in eq . ( [ eq15 ] ) cross terms between different events cancel in the poisson limit and in the last line we use the fact that event - collection averages of terms linear in the fluctuation vanish . inserting the moments of poisson distributions ( see appendix b ) , the resulting variance is ( s_mm - m_mm)^2 & & n_events |m^2 [ eq16 ] and the statistical error in @xmath57 , assuming poisson distributions , is & = & = . [ eq17 ] the multiplicative factor in the numerator for this diagonal bin error is @xmath58 larger than that for off - diagonal bins given in eq . ( [ eq10 ] ) due to the reduced number of unique sibling and mixed - event pairs in diagonal bins . in general the event collection could be sub - divided into an arbitrary number of groups of events with @xmath59 events per group where @xmath60 and with @xmath61 combinations of mixing event - pairs within each group . for this general case the sibling and mixed - event particle - pair sums and their limits are given by _ mn & = & _ g=1^n_g _ j=1^ ( |m+_j)(|n+_j ) + m_mn & = & _ g=1^n_g _ j^ > j=1^ [ ( |m+_j)(|n+_j^ ) + ( |m+_j^)(|n+_j ) ] + there are more mixed - event particle pairs than sibling and the ratio @xmath38 must be normalized such that the correlation signal , @xmath44 , equals zero in the absence of correlations . the ratio @xmath38 for the general case is therefore defined as _ mn & = & = ( - 1 ) [ eq20 ] and _ mn - 1 & = & . [ eq21 ] the variance is evaluated as before where ^2 & = & \ { _ g=1^n_g [ ( - 1 ) _ j=1^ _ j _ j - _ j^ > j = 1^ ( _ j _ j^ + _ j^ _ j ) ] } ^2 + & = & _ g=1^n_g [ ( - 1 ) _ j=1^ _ j _ j - _ j^ > j = 1^ ( _ j _ j^ + _ j^ _ j ) ] ^2 + & = & _ g=1^n_g [ ( - 1 ) _ j=1^ _ j _ j - _ j^ j _ j _ j^ ] ^2 [ eq22 ] where cross terms among different event groups vanish . this last expression can be further simplified using the fact that the event - collection average of @xmath62 for different events @xmath63 vanishes , resulting in ^2 & = & _ g=1^n_g [ ( - 1)^2 _ j=1^ _ j^2 _ j^2 + _ j^ j=1^ _ j^2 _ j^^2 ] + & & = _ g=1^n_g [ ( ^2 - 2 + 1 -1 ) _ j=1^ _ j^2 _ j^2 + _ j^ , j=1^ _ j^2 _ j^^2 ] + & & = n_g _ g=1^n_g _ j=1^ _ j^2 _ j^2 + _ g=1^n_g _ j=1^ _ j^2 _ j^=1^ _ [ eq23 ] the first term is an average over all events in the collection and results in the product of variances @xmath64 in the absence of correlated fluctuations between different bins . the second term is the average over all groups of the product of variances for each group . ( [ eq23 ] ) therefore reduces to ^2 & = & n_g ( ^2 - 2 ) _ ^2 _ ^2 + _ g=1^n_g ( _ ^2 + _ g ) ( _ ^2 + _ g ) + & & = n_g ( - 1 ) _ ^2 _ ^2 ( - 1 ) n_events ( |m|n ) , [ eq24 ] where the variances within each group are expressed as the average variance for all event collections plus random fluctuations @xmath65 and @xmath66 , where averages @xmath67 , @xmath68 , and @xmath69 equal zero . the error in an off - diagonal bin is therefore & = & = . [ eq25 ] this error equals that in eq . ( [ eq10 ] ) when @xmath70 and approaches @xmath71 as the number of events in each mixing group increases and approaches the entire number of events in the collection . again , note that this error equals @xmath53 . the improvement in statistical accuracy achieved by increasing @xmath59 can be weighed against the increased computational requirements in order to optimize the number of event groups to be used in the analysis of data . in typical analyses of heavy - ion collision data @xcite correlations are calculated by reading an unordered list of event data and sorting the particle kinematic information from those events into subsets corresponding to a global event - property such as overall multiplicity , a proxy for the overlap , or _ centrality _ , between the colliding nuclei . for each centrality all sibling particle - pairs are processed and to reduce computational requirements mixed - event particle - pairs are only processed for the first and second events in the collection , then for the second and third events , and so on until all events in the centrality bin have been processed . this event - mixing algorithm can be extended such that each event is mixed with the next two events in the list , or the next three events , etc . in this section statistical errors are derived for the above single , double and multiple - event mixing algorithms . it will be shown that the noise reduction obtained in ref . @xcite also occurs for these other types of event - mixing methods . in this event mixing algorithm particle pairs are constructed for all events by mixing particles in event @xmath22 with those in event @xmath72 . the number of sibling and mixed - event particle pairs and their limits are given by _ mn & = & _ j=1^n_events ( |m+_j)(|n+_j ) _ j=1^n_events s_mn^j + m_mn & = & _ j=1^n_events - 1 [ ( |m+_j)(|n+_j+1 ) + ( |m+_j+1)(|n+_j ) ] + & & _ j=1^n_events - 1 ( m_mn^j , j+1 + m_mn^j+1,j ) + definitions @xmath73 and @xmath74 are introduced for brevity . the approximation @xmath75 occurs because the last event in the event list has no other event to mix with . for @xmath76 those missing pairs can be ignored in the above summations . in this single - event mixing algorithm there are approximately twice as many mixed - event pairs as sibling pairs . therefore the variance to be calculated is given by ( 2s_mn - m_mn)^2 & = & ^2 . [ eq28 ] twice the sum over sibling pairs can be written as two , identical summations . the resulting sibling and mixed - event sums can be separated into sums over even and odd values of event number @xmath22 . as a result eq . ( [ eq28 ] ) can be rearranged into sums with the identical structure as that in ref . @xcite , thus achieving the same level of noise reduction for this alternate event - mixing algorithm . the result is ( 2s_mn - m_mn)^2 & = & ^2 . [ eq29 ] from ref . @xcite we find that _ mn^j + s_mn^j+1 -m_mn^j , j+1 - m_mn^j+1,j & = & _ j _ j + _ j+1 _ j+1 - _ j _ j+1 - _ j+1 _ j . + [ eq30 ] substituting the above result into eq . ( [ eq29 ] ) and combining the even and odd numbered summations gives ( 2s_mn - m_mn)^2 & = & ^2 + & = & _ j=1^n_events ( 2 _ j _ j - _ j _ j+1 - _ j+1 _ j)^2 [ eq31 ] where cross terms for different events vanish as before . continuing , we find that the above equation in the poisson limit reduces to _ j=1^n_events ( 4 _ j^2 _ j^2 + _ j^2 _ j+1 ^ 2 + _ j+1 ^ 2 _ j^2 ) & = & 6 n_events _ ^2 _ ^2 + & & 6 n_events ( |m|n ) . [ eq32 ] the statistical error for off - diagonal bins is & = & = . [ eq33 ] this error is a factor of @xmath77 smaller than that for the reid - trainor method due to the larger number of mixed - event pairs ( twice as many pairs ) generated by this algorithm . it is important to note that the noise reduction achieved via the mixing algorithm in ref . @xcite is also achieved with the single - event mixing method and the error is determined by pair number where the error in eq . ( [ eq33 ] ) equals @xmath53 . for diagonal bins _ mm & = & _ j=1^n_events ( |m+_j)(|m+_j - 1 ) + & = & _ j=1^n_events ( s_mm^j -|m -_j ) + m_mm & = & _ j=1^n_events - 1 ( |m+_j)(|m+_j+1 ) = _ j=1^n_events - 1 m_mm^j , j+1 + particle pairs are included in the summations . the variance normalization factor , @xmath78 , for the diagonal bins is also equal to 2 . the variance quantity is then given by ( 2s_mm - m_mm)^2 & = & ^2 + & & = ^2 [ eq36 ] where @xmath79 . separating the above into summations over even and odd event numbers , doubling the sibling - pair summation , assuming that @xmath76 ( in order that the contribution of missing mixed - events at the end of the event list is negligible ) , and using the noise reduction result from ref . @xcite result in ( 2s_mm - m_mm)^2 & = & ^2 + & & = _ j=1^n_events ( _ j^4 + _ j^2 _ j+1 ^ 2 + |m^2 + _ j^2 - 2|m_j^2 - 2_j^3 ) + & & + _ j j^ ( _ j^2 _ j^^2 - |m_j^2 -|m_j^^2 + |m^2 ) . [ eq37 ] the second term vanishes in the @xmath76 limit where n_events^2 ( _ ^4 - |m _ ^2 - |m _ ^2 + |m^2 ) & & n_events^2 ( |m^2 - |m^2 - & = & 0 . [ eq38 ] the first term in eq . ( [ eq37 ] ) , using the moments of a poisson distribution in appendix b , simplifies to ( 2s_mm - m_mm)^2 & = & + & & n_events _ j=1^n_events ( _ j^4 + _ j^2 _ j+1 ^ 2 + |m^2 + _ j^2 - 2|m_j^2 - 2_j^3 ) + & & n_events ( 3|m^2 + |m ) + |m^2 + |m^2 + |m -2|m^2 - 2|m = 3n_events [ eq39 ] the statistical error in diagonal bins for the single - event mixing algorithm is & = & = [ eq40 ] where the multiplicative factor in the numerator is @xmath58 larger than that for the off - diagonal errors in eq . ( [ eq33 ] ) . in this method particles in each event @xmath22 are mixed with particles in events @xmath72 and @xmath80 . the sibling and mixed - event pair summations , using the symbols @xmath73 and @xmath74 introduced above , and their limits are given by _ mn & = & _ j=1^n_events s_mn^j + m_mn & = & _ j=1^n_events - 1 ( m_mn^j , j+1 + m_mn^j+1,j ) + _ j=1^n_events - 2 ( m_mn^j , j+2 + m_mn^j+2,j ) + & & 4n_events ( |m|n ) . [ eq42 ] the normalization factor for @xmath38 is approximately 4 and the variance to be calculated is ( 4s_mn - m_mn)^2 & = & ^2 . [ eq43 ] by writing out the sum over sibling pairs four times it is found that the resulting summations in eq . ( [ eq43 ] ) can be separated into sums over pairs of events where arbitrary event @xmath22 is mixed with event @xmath81 , and in this example @xmath82 and 2 . for further brevity the symbol @xmath83 is introduced which is defined by _ mn^j , k & & s_mn^j + s_mn^j+k - m_mn^j , j+k - m_mn^j+k , j . [ eq43b ] after rearranging the sums the above variance in the @xmath76 limit can be expressed as ( 4s_mn - m_mn)^2 & = & ^2 [ eq44 ] where , as in ref . mn^j , k & = & _ j _ j + _ j+k - _ j _ j+k - _ j+k _ j . [ eq45 ] substituting @xmath83 into eq . ( [ eq44 ] ) and recombining the summations gives ( 4s_mn - m_mn)^2 & = & ^2 . [ eq46 ] eliminating cross terms involving products of fluctuations between different events results in ( 4s_mn - m_mn)^2 & = & _ j=1^n_events ( 16 _ j^2 _ j^2 + _ j^2 _ j+1 ^ 2 + _ j+1 ^ 2 _ j^2 + _ j^2 _ j+2 ^ 2 + _ j+2 ^ 2 _ j^2 ) + & = & 20 n_events _ ^2 _ ^2 20 n_events ( |m|n ) . [ eq47 ] the statistical error in off - diagonal bins for the double - event mixing algorithm is given by & = & = [ eq48 ] which is smaller than the above , single - event mixing error and is @xmath84 smaller than the error resulting from the reid - trainor method . again , this error equals @xmath53 . the error for diagonal bins is obtained as in the above sections and is given by & = & . [ eq49 ] the preceding event - mixing methods can be readily extended to an arbitrary number of mixing events @xmath85 . in the following derivation however we continue to assume that @xmath86 . the sums of sibling and mixed - event particle pairs in off - diagonal bins and their limits are _ mn & = & _ j=1^n_events s_mn^j + m_mn & = & _ j=1^n_events -1(m_mn^j , j+1 + m_mn^j+1,j ) + _ j=1^n_events -2(m_mn^j , j+2 + m_mn^j+2,j ) + & + & + _ j=1^n_events -k(m_mn^j , j+k + m_mn^j+k , j ) + eq . ( [ eq43b ] ) the variance is given by ( 2ks_mn - m_mn)^2 & = & ^2 [ eq52 ] where event index @xmath87 , @xmath88 . recombining the summations over sibling pair quantities @xmath89 results in @xmath90 . the summations over the last two terms , @xmath91 , include each mixed - event only once and may be simplified to @xmath92 . the variance can then be expressed as ( 2ks_mn - m_mn)^2 & = & ^2 . [ eq53 ] as before the average over cross terms between different events vanishes resulting in ( 2ks_mn - m_mn)^2 & = & _ j=1^n_events ( 4k^2 _ j^2 _ j^2 + _ k=1^k ( _ j^2 _ j+k^2 + _ j+k^2 _ j^2 ) ) + & & = n_events ( 4k^2 + 2k)_^2 _ ^2 ( 4k^2 + 2k ) n_events(|m|n ) [ eq54 ] and the error in off - diagonal bins is & = & = . [ eq55 ] evaluating this result for @xmath93 and 2 gives the above errors for single - event and double - event mixing in eqs . ( [ eq33 ] ) and ( [ eq48 ] ) , respectively . the above error equals @xmath53 . to summarize , for each event - mixing algorithm presented here the sibling minus mixed - event pair summations can be accurately rearranged into sums over event - pair - wise differences between sibling and mixed - event pairs as in ref . @xcite , resulting in significant statistical noise reduction . the statistical errors for the correlation quantity @xmath94 are determined by the total numbers of sibling and mixed - event pairs of particles in 2d bin @xmath12 , given by @xmath53 . the correlation quantity @xmath94 in eq . ( [ eq5 ] ) represents the event - averaged number of correlated particle pairs per final - state particle pair produced in the collisions . in some analyses @xcite the authors reported the number of correlated particle pairs per final - state particle in which case @xmath94 was scaled by a prefactor @xcite proportional to total event - multiplicity . for example , in the analysis of au + au collisions at @xmath95 = 200 gev @xcite the prefactor was computed using efficiency corrected , charged - particle yields within the detector acceptance defined by pf_ci^ ( , ) & = & ^1/2 = [ eq56 ] for correlations on 2d angular space @xmath96 for arbitrary particle pair ( 1,2 ) and at mid - rapidity where @xmath97 is approximately constant . subscript `` ci '' ( charge independent ) means that all charge - sign pairs are included , i.e. @xmath98 , @xmath99 , @xmath100 , and @xmath101 . other kinematic projections are possible . in ref . @xcite correlations on 2d transverse momentum space were presented with corresponding prefactor pf_ci^(p_t , p_t ) & = & ^1/2 . [ eq57 ] statistical errors in the prefactor could be included in the final , reported correlation quantities but this factor is usually treated as a simple scaling of the correlation measurement . insight into the dynamics underlying the observed correlations can be gained by making further selections on the particle pairs . for example , correlations for positive versus negative charged particle pairs , or like - charge - sign pairs ( @xmath98 and @xmath99 ) versus unlike - sign pairs ( @xmath100 and @xmath101 ) , can help differentiate charge - ordering effects @xcite in hadronization , resonance decay contributions , quantum correlations for identical particles @xcite , and photon pair - production contamination . selection of low-@xmath4 or high-@xmath4 particles can help differentiate soft and hard - scattering processes @xcite . similarly , for correlations on 2d transverse momentum - space particle pairs with smaller or larger relative azimuthal angles , e.g. @xmath102 ( near - side ) or @xmath103 ( away - side ) , in conjunction with charge - pair selections , can help differentiate the @xmath4 structures of near - side , jet - like fragmentation from that of away - side , or back - to - back dijet fragmentation . if the prefactor , which is the square - root of a pair density , is computed using the same charge - sign and/or kinematic selections used in constructing the correlation @xmath94 , then the statistical error in the final quantity , @xmath104 , is unaffected by those selections . this is a result of the noise reduction inherent in the event - mixing methods presented here and in ref . @xcite where the statistical error is inversely proportional to the square - root of the number of particle pairs in the bin , @xmath105 . for example , if only like - sign pairs are selected the total number of pairs is reduced by 1/2 if the numbers of positive and negative particles are the same ( approximately true for relativistic heavy - ion collisions ) . the prefactor is therefore reduced by @xmath106 while the error in each bin is increased by @xmath58 . similarly , selecting only near - side or away - side particle pairs reduces the number of pairs by 1/2 which again decreases the prefactor by @xmath106 and increases the bin errors by @xmath58 . in general , such pair selections do not affect the prefactor scaled statistical errors studied here . this result is summarized as follows : & = & pf_ci [ eq58 ] where the left - hand - side represents the root - mean - square statistical error of quantity @xmath104 for arbitrary charge - sign and/or azimuth angle pair selections . ( [ eq58 ] ) applies to each event - mixing method discussed here . throughout the preceding derivations it was assumed in the final steps that the frequency distributions for event - wise particle number in bins @xmath1 and @xmath2 were consistent with poisson distributions , i.e. variance @xmath47 equaled mean @xmath24 . in practical data analysis , collision events are usually grouped into subsets based on global event properties such as total multiplicity in the detector acceptance @xcite . clearly the multiplicity range chosen for the event collections can affect the frequency distributions in the @xmath1 and @xmath2 bins . in addition , if the @xmath1,@xmath2 bin sizes are comparable to the acceptance , then the multiplicity distribution of the event collection will also affect that in the bins . in this section monte carlo simulations are used to determine the differences between actual event - wise particle number frequency distributions in the bins and a poisson distribution . if the distributions differ significantly from a poisson , then the actual variances should be used in the above derivations . in au + au collisions at energies of 200 gev per colliding nucleon + nucleon pair over a thousand charged particles can be produced at mid - rapidity within the tracking acceptance of the star detector @xcite ( @xmath107 gev/@xmath108 , @xmath109 and @xmath110 in azimuth ) . for events selected with a minimum - bias trigger @xcite the multiplicity frequency distribution for total , observed multiplicity , @xmath111 , is approximately proportional to @xmath112 except near the lower and upper multiplicity end - points @xcite . the events are then subdivided into multiplicity classes or centrality . for example , the multiplicity ranges for peripheral , intermediate ( mid - central ) , and head - on ( most - central ) collisions from ref . @xcite are [ 15,35 ] , [ 152,187 ] and [ 952,1002 ] , respectively . for the present example , correlations on 2d transverse momentum space were considered where the kinematic variable used is transverse rapidity , defined by @xmath113 $ ] , @xmath114 and for non - identified charged particles @xmath115 , which regulates the singularity in @xmath116 at @xmath117 , was assumed to be the pion mass , 0.14 gev/@xmath118 . the transverse rapidity spectrum for each multiplicity bin was obtained from the efficiency and contamination corrected @xmath4 spectrum data @xcite and is accurately represented by a levy distribution given by & = & . [ eq59 ] parameters @xmath119 and @xmath2 from data fitting for peripheral , mid - central and most - central collisions are ( 0.154 gev , 10.42 ) , ( 0.199 gev , 13.32 ) , and ( 0.226 gev , 17.55 ) , respectively . amplitude @xmath120 is determined by event - wise multiplicity . in the correlation analysis particle pairs were binned on 2d @xmath121 space where @xmath122 $ ] with @xmath123 2d @xmath12 bins of equal size . in the simulations , the power - law distribution @xmath112 was randomly sampled within each of the above multiplicity ranges , the single - particle distribution in eq . ( [ eq59 ] ) was sampled for the event - wise multiplicity , the 25 single - particle @xmath116 bins were filled , and the number of particles in each @xmath116 bin for each event was recorded . the particle number frequency distributions for the 25 @xmath116 bins for the simulated events were compared with poisson distributions . this was done by calculating the ratio of the computed variance , @xmath124 , to the variance for a poisson distribution , @xmath125 , where @xmath126 is the mean particle number in the @xmath116 bin . the resulting ratios were compared with unity . the uncertainty in the ratio of variances for each @xmath116 bin was estimated by repeating the above simulation procedure many times , corresponding to statistically independent collections of events . the mean and width of the resulting distribution of variance ratios were computed . results for @xmath127 for six of the twenty - five @xmath116 bins are listed in table [ table1 ] and shown in fig . [ figure1 ] for the three event - multiplicity ranges listed above . for the peripheral , mid - central and most - central collisions @xmath128 , @xmath129 and @xmath129 events were generated , respectively . one - hundred event collections were simulated for each centrality . the calculations provided sufficient statistical accuracy to determine a significant difference between @xmath127 and unity for most @xmath116 bins . in all cases @xmath130 agreed with the poisson limit @xmath131 to within 3% or less , thus confirming the poisson assumption for the present application to au + au collisions at 200 gev . [ cols="^,^,^,^ " , ] ( color online ) ratios @xmath127 as described in the text and listed in table [ table1 ] for three event - multiplicity ranges where the solid , dashed - dotted and dashed lines connecting the points correspond to peripheral , mid - central and most - central collisions , respectively.,width=384 ] monte carlo simulations were also used to verify the noise reduction predicted in ref . @xcite and for the derivations in sections [ sec : general ] and [ sec : other ] . the simulations were also used to check the error scaling factors @xmath58 , @xmath132 and @xmath133 for the reid - trainor , single- and double - event mixing algorithms . in ref . @xcite and in appendix c the statistical error for correlations obtained with mixed - event particle pairs taken from a different set of events than those used for the sibling particle pairs is shown to be proportional to @xmath134^{1/2}$ ] . this error depends on particle number and therefore differs significantly from the errors derived in ref.@xcite and in sections [ sec : general ] and [ sec : other ] which are proportional to @xmath135 . finally , the simulations were used to verify that the errors in 2d diagonal bins are larger than those in off - diagonal bins by a factor of @xmath58 . as in the preceding simulation the @xmath112 frequency distribution was randomly sampled to obtain an event multiplicity , the @xmath116 distribution in eq . ( [ eq59 ] ) was sampled , and the 25 single - particle @xmath116 bins were filled . sibling and mixed - event pairs were counted for each 2d @xmath12 bin on @xmath121 space using the event - mixing algorithms in ref . @xcite ( see sec . [ sec : general ] ) , the single - event mixing described in sec . [ sec : single ] , or the double - event mixing described in sec . [ sec : double ] . the correlation quantity @xmath94 was computed for all @xmath12 bins using all simulated events in the collection . this entire process was repeated many times thus generating independent , statistically equivalent event collections and corresponding values for @xmath94 . from these distributions of @xmath94 the means and widths in each @xmath12 bin were obtained , including those for on - diagonal and off - diagonal bins . the resulting monte carlo uncertainties are directly comparable to the analytic results for statistical errors in eqs . ( [ eq10 ] ) , ( [ eq17 ] ) , ( [ eq33 ] ) , ( [ eq40 ] ) , ( [ eq48 ] ) , and ( [ eq49 ] ) . results are reported below for the mid - central multiplicity bin with @xmath136 $ ] and for two diagonal @xmath137 bins on @xmath121 defined by ranges ( [ 1.56,1.70],[1.56,1.70 ] ) and ( [ 2.96,3.10],[2.96,3.10 ] ) , and for the corresponding off - diagonal @xmath12 bin defined by ( [ 1.56,1.70],[2.96,3.10 ] ) . the mean , single - particle numbers in these two @xmath116 bins are 14.3 and 3.39 , respectively . the scaling of the error with particle pair number , rather than with single particle number can be checked by comparing the simulated errors for two diagonal bins . for the above diagonal bins the ratios of the analytical errors in the lower to upper @xmath116 bins are 0.237 for particle - pair scaling ( i.e. with noise reduction ) and 0.462 for single - particle scaling ( statistically independent events for sibling and mixed - event particle - pairs , see appendix c ) . the simulated ratios for the reid - trainor , single - event mixing , and double event - mixing algorithms are 0.241 , 0.226 and 0.236 , respectively , which agree with the noise reduced , particle - pair scaling . the scaling factors for the errors can be checked by comparing the analytic ratio for single - event mixing to reid - trainor mixing , equal to @xmath138 , and the ratio for double - event mixing to reid - trainor mixing , equal to @xmath139 , to the corresponding ratios from the simulations . the monte carlo error results for the lower and upper @xmath116 diagonal bins are 0.851 and 0.911 ( ratio of single - event mixing to reid - trainor mixing ) and 0.785 and 0.803 ( ratio of double - event mixing to reid - trainor mixing ) , respectively . both pairs of monte carlo results are in good agreement with the corresponding analytic expressions . finally , the @xmath58 increase in the diagonal bin errors can be checked via the ratio @xmath140 where @xmath141 is the error in bin @xmath12 . the analytical value of this ratio equals 1/2 for the three event - mixing algorithms considered in this section . the simulation results for the reid - trainor , single - event mixing , and double - event mixing algorithms using the above bins are 0.52 , 0.52 and 0.50 , respectively , in good agreement with the analytic errors . the simulated and analytic absolute errors agree within 2%-6% , where the few percent effects of non - poisson distributions discussed above contribute here as well . the scaling factors discussed above and the differences between the errors obtained with particle - pair number scaling versus single - particle number scaling are much larger than these few percent differences . overall , the simulations confirm the analytic expressions . the poisson distribution assumption should be checked for each application . the event - mixing and error propagation method of reid - trainor @xcite for a multiplicity - ordered event collection , grouped into pairs of events for particle - pair mixing , was generalized to include event - mixing groups with arbitrary number of events . the method was then applied to other event - mixing algorithms which have been used in correlation analysis of relativistic heavy - ion collision data . the noise reduction arising from the partial cancellation of event - wise single - particle multiplicity fluctuations was shown to occur for those other methods in which particle - pair mixing is done between each event in the collection and the next 1 , 2 or @xmath142 ( where @xmath143 ) events in the list . the statistical errors for analyses which use the event - mixing group method of ref . @xcite are given in eqs . ( [ eq10 ] ) and ( [ eq17 ] ) for two events per mixing group and in eq . ( [ eq25 ] ) for @xmath144 events in each event - mixing group . errors for correlation analysis based on single - event mixing and double - event mixing are given in eqs . ( [ eq33 ] ) , ( [ eq40 ] ) and eqs . ( [ eq48 ] ) , ( [ eq49 ] ) , respectively . correlation errors for all event - mixing algorithms in which each event is mixed with the next @xmath142 events in the list are given in eq . ( [ eq55 ] ) . statistical errors for correlations reported as the number of correlated pairs per final - state particle , including charge and/or kinematic dependent pair selections @xcite , are given in eq . ( [ eq58 ] ) . the statistical errors in correlations for each of the event - mixing algorithms studied here using the reid - trainor method are inversely proportional to the square - root of the number of pairs of particles in a bin . this result is in contrast to that obtained when the two - particle reference distribution is constructed from a set of events independent from those used for sibling pairs . in the latter case the errors are inversely proportional to the square - root of the number of particles in the bin and can be much larger than the errors which follow from the event - mixing methods discussed here and in ref . @xcite . as long as the sum over sibling particle pairs minus mixed - event particle pairs can be accurately approximated by sibling minus mixed - event particle pair differences summed over pairs of events as in ref . @xcite and in this paper , then the reid - trainor error reduction will follow and the correlation errors will be inversely proportional to the square - root of the total , bin - wise number of particle pairs . it should also be noted that error propagation algorithms based on bin - wise single - particle counts which neglect covariance @xmath41 in eq . ( [ eq6 ] ) will produce erroneous results . monte carlo simulations were employed to test the accuracy of the poisson distribution approximations used in the error calculations . simulations were also used to verify the predicted reduction in statistical errors for the reid - trainor , single - event and double - event mixing algorithms . finally , the analytic error results presented here may be helpful in optimizing correlation data analysis when more aggressive event - mixing algorithms are considered and the benefit from reduced statistical errors needs to be weighed against the additional computational cost . the authors would like to thank professor tom trainor of the univ . of washington for many informative discussions relevant to this work . this research was supported in part by the office of science of the u. s. department of energy under grants de - fg02 - 94er40845 and de - sc0013391 . in deriving eq . ( [ eq8 ] ) event averages and cross terms involving averages over pairs of different events must be calculated . the necessary steps are given here . similar calculations are required for the other event - mixing algorithms described in this paper . starting with the first line in eq . ( [ eq8 ] ) , replacing the group summation in the first part of eq . ( [ eq8 ] ) with event summations , expanding the quadratic term , and retaining the non - vanishing averages yield the following : ( s_mn - m_mn)^2 & = & ^2 + & = & ^2 + & = & _ j=1,odd^n_events ( _ j _ j - _ j _ j+1 ) ^2 + _ j=2,even^n_events ( _ j _ j - _ j _ j-1 ) ^2 + & + & _ j j^,odd ( _ j _ j - _ j _ j+1 ) ( _ j^ _ j^ - _ j^ _ j^+1 ) + & + & _ j j^,even ( _ j _ j - _ j _ j-1 ) ( _ j^ _ j^ - _ j^ _ j^-1 ) + & + & 2 _ j=1,odd^n_events _ j^=2,even^n_events ( _ j _ j - _ j _ j+1 ) ( _ j^ _ j^ - _ j^ _ j^-1 ) + & = & _ j=1^n_events _ j^2 _ j^2 + _ j=1,odd^n_events _ j^2 _ j+1 ^ 2 + _ j=2,even^n_events _ j^2 _ j-1 ^ 2 . the last line is equivalent to eq . ( [ eq8 ] ) . event collection averages over cross terms involving pairs of different events , for example & & _ j=1,odd^n_events _ j^2 _ j _ j+1 , _ j j^,odd _ j _ j _ j^ _ j^ , _ j j^,odd _ j _ j _ j^ _ j^+1 + & & _ j=1,odd^n_events _ j^=2,even^n_events _ j _ j _ j^ _ j^ , vanish because bin - wise multiplicity fluctuations are uncorrelated from event - to - event . cross term _ j=1,odd^n_events _ j^=2,even^n_events _ j _ j _ j^ _ j^-1 & & + & & = _ j=1,odd^n_events _ ^n_events _ j _ j _ j^ _ j^-1 + _ j=1,odd^n_events _ j _ j+1 _ j^2 also vanishes for the same reason where the @xmath145 term is separated out . moments of the multiplicity frequency distribution in arbitrary bin @xmath1 involve averages of the event - wise fluctuations _ j=1^n_events _ j^p where exponent @xmath146 and @xmath147 , where @xmath148 is the multiplicity in bin @xmath1 for event @xmath22 . this average can be expressed as _ j=1^n_events ( m_j - |m)^p & = & _ m _ j=1^n(m ) [ ( m_j - |m)^p]|_m_j = m + & = & _ m ( m-|m)^p where @xmath149 is the number of events with @xmath150 particles in bin @xmath1 and @xmath151 sums over all events with @xmath152 . for poisson distributions @xmath153 where @xmath154 . the moments are calculated via the summations given by _ j=1^n_events ( m_j - the resulting poisson weighted sums can be related to bell polynomials @xcite , defined by b_p(|m ) & = & _ m=0^ m^p p(m,|m ) where b_0(|m ) & = & 1 + b_1(|m ) & = & |m + b_2(|m ) & = & |m^2 + |m + b_3(|m ) & = & |m^3 + 3|m^2 + |m + b_4(|m ) & = & |m^4 + 6|m^3 + 7|m^2 + |m and so on . the first few values of @xmath155 for @xmath156 $ ] are 1 , 0 , @xmath24 , @xmath24 , @xmath157 , respectively . the statistical error for correlation quantity @xmath94 when independent sets of events are used to calculate the sibling and mixed - event pair sums is derived here for the event - mixing algorithm in ref . @xcite . as before _ mn & = & _ j=1^n_events ( |m+_j)(|n+_j ) + @xmath158 and @xmath159 are statistically equivalent . the error of @xmath94 squared equals @xmath160 , however , the fluctuations in single - particle number no longer cancel as in ref . @xcite and eq . ( [ eq8 ] ) . substituting the above summations for @xmath55 and @xmath56 into @xmath161 and noting that all cross terms between different events vanish , result in ( s_mn - m_mn)^2 & = & _ j=1^n_events ( |m^2_j^2 + |n^2_j^2 + _ j^2 _ j^2 ) + & & + _ j^=1^n_events ( |m^2_j^^2 + |n^2_j^^2 ) + _ j^=1,odd^n_events ( _ j^^2 _ j^+1 ^ 2 + _ j^+1 ^ 2 _ j^^2 ) + & & = 2n_events |m^2 _ ^2 + |n^2 _ ^2 + _ ^2 _ ^2 2n_events ( |m^2|n + |n^2|m + |m|n ) and the error in @xmath38 is & = & ^1/2 . comparing this error to the corresponding statistical error when the same event collection is used for the sibling and mixed - event pairs in eq . ( [ eq10 ] ) yields the ratio & = & . for 200 gev au + au collisions with events selected in the mid - central range @xmath136 $ ] and for the off - diagonal @xmath12 bin on @xmath121 , defined by the ranges ( [ 1.56,1.70],[2.96,3.10 ] ) where @xmath24 = 14.3 and @xmath25 = 3.39 , the above ratio of statistical errors is 0.23 . the noise reduction of the reid - trainor method is more than a factor of 4 in this instance . m. s. daugherity , `` two - particle correlations in ultra relativistic heavy ion collisions , '' ph.d . thesis , the university of texas at austin , ( 2008 ) , https://drupal.star.bnl.gov/star/files/daugherity_dissertation.pdf . e. w. oldag , `` two - particle correlations in angular and momentum space in relativistic heavy ion collisions at star , '' ph.d . thesis , the university of texas at austin , ( 2013 ) , https://drupal.star.bnl.gov/star/files/oldag_dissertation_20132.pdf .

nuclei are quantum many - body systems with both bound and unbound states . a realistic _ ab initio _ description of light nuclei with predictive power must have the capability to describe both classes of states within a unified framework . over the past decade , significant progress has been made in our understanding of the properties of the bound states of light nuclei starting from realistic nucleon - nucleon ( @xmath11 ) interactions , see e.g. ref . @xcite and references therein , and more recently also from @xmath11 plus three - nucleon ( @xmath12 ) interactions @xcite . the solution of the nuclear many - body problem becomes more complex when scattering or nuclear reactions are considered . for @xmath13 and 4 nucleon systems , the faddeev @xcite and faddeev - yakubovsky @xcite as well as the hyperspherical harmonics ( hh ) @xcite or the alt , grassberger and sandhas ( ags ) @xcite methods are applicable and successful . however , _ ab initio _ calculations for unbound states and scattering processes involving more than four nucleons in total are quite challenging . the first _ ab initio _ many - body neutron-@xmath0he scattering calculations were performed within the green s function monte carlo method using the argonne @xmath11 potential and the illinois @xmath12 interaction @xcite . also , resonances in he isotopes were investigated within the coupled - cluster method using the gamow basis @xcite . in a new development , we have recently combined the _ ab initio _ no - core shell model ( ncsm ) @xcite and the resonating - group method ( rgm ) @xcite , into a new many - body approach @xcite ( _ ab initio _ ncsm / rgm ) capable of treating bound and scattering states of light nuclei in a unified formalism , starting from fundamental inter - nucleon interactions . the ncsm is an _ ab initio _ approach to the microscopic calculation of ground and low - lying excited states of light nuclei with realistic two- and , in general , three - nucleon forces . the rgm is a microscopic cluster technique based on the use of @xmath14-nucleon hamiltonians , with fully anti - symmetric many - body wave functions built assuming that the nucleons are grouped into clusters . although most of its applications are based on the use of binary - cluster wave functions , the rgm can be formulated for three ( and , in principle , even more ) clusters in relative motion @xcite . the use of the harmonic oscillator ( ho ) basis in the ncsm results in an incorrect description of the wave - function asymptotic and a lack of coupling to the continuum . by combining the ncsm with the rgm , we complement the ability of the rgm to deal with scattering and reactions with the use of realistic interactions , and a consistent _ ab initio _ description of the nucleon clusters , achieved via the ncsm . presently the ncsm / rgm approach has been formulated for processes involving binary - cluster systems only . however , extensions of the approach to include three - body cluster channels are feasible , also in view of recent developments on the treatment of both three - body bound and continuum states ( see , e.g. , refs . @xcite ) . as described in detail in refs . @xcite , the _ ab initio _ ncsm / rgm approach has been already applied to study the @xmath15-@xmath16h , @xmath15-@xmath17he , @xmath15-@xmath18be , and @xmath19-@xmath20he scattering processes , and address the parity inversion of the @xmath21be ground state , using realistic @xmath11 potentials . in that work , we demonstrated convergence of the approach with increasing basis size in the case of the @xmath22 and @xmath23 scattering . the @xmath15-@xmath18be calculations were , on the other hand , perfomed only in a limited basis due to the computational complexity of the ncsm calculations of the @xmath24be eigenstates . it is the purpose of the present paper to expand the applicability of the ncsm / rgm beyond the lightest nuclei by using sufficiently large @xmath25 ho excitations to guarantee convergence of the calculation with the ho basis expansion of both the cluster wave functions and the localized rgm integration kernels . the use of large @xmath25 values is now feasible due to the recent introduction of the importance truncated ( it ) ncsm scheme @xcite . it turns out that many of the basis states used in the ncsm calculations are irrelevant for the description of any particular eigenstate , e.g. , the ground state or a set of low - lying states . therefore , if one were able to identify the important basis states beforehand , one could reduce the dimension of the matrix eigenvalue problem without losing predictive power . this can be done using an importance truncation scheme based on many - body perturbation theory @xcite . we make use of the it ncsm wave functions for the cluster eigenstates , in particular the eigenstates of the target nucleus of the binary nucleon - nucleus system , and calculate the one- and two - body densities that are then used to obtain the ncsm / rgm integration kernels . we benchmark the it approach in basis sizes accessible by the full calculation and apply it within still larger basis sizes until convergence is reached for target nuclei as heavy as @xmath2c or @xmath3o . in this study , we employ a similarity renormalization group ( srg ) @xcite evolved chiral n@xmath26lo @xmath11 potential @xcite ( srg - n@xmath26lo ) that is soft enough to allow us reach convergence within about @xmath27 ho excitations in the basis expansion . in sect . [ formalism ] , we briefly overview the ncsm / rgm formalism and present for the first time the it ncsm scheme that includes both ground and low - lying excited states in the set of reference states . next , we present scattering calculation results for the @xmath28-@xmath0he and @xmath29-@xmath0he systems in sect . in particular , we compare the calculated phase shifts to an r - matrix analysis of experimental data and , further , calculated differential cross sections and analyzing powers in the energy range 6 - 19 mev to the corresponding experimental data . neutron elastic and inelastic scattering on @xmath1li and proton elastic and inelastic scattering on @xmath1be are investigated in sect . we present phase shifts , cross sections and scattering lengths . we predict resonances in @xmath7li and @xmath7be that have not been clearly identified in experiments yet . in sect . [ n12c ] , we discuss nucleon-@xmath2c results for both bound and unbound states of @xmath30c and @xmath30n , obtained including two @xmath2c bound states , the ground and the first @xmath31 state , in the ncsm / rgm coupled - channel calculations . in sect . [ n16o ] , we present results for the nucleon-@xmath3o system . in the ncsm / rgm coupled - channel calculations , we take into account the @xmath3o ground state and up to the lowest three @xmath3o negative - parity states . conclusions are given in sect . [ conclusions ] . the _ ab inito _ ncsm / rgm approach was introduced in ref . @xcite with details of the formalism given in ref . @xcite . here we give a brief overview of the main points . in the present paper , we limit ourselves to a two - cluster rgm , which is based on binary - cluster channel states of total angular momentum @xmath32 , parity @xmath33 , and isospin @xmath34 , @xmath35^{(j^\pi t)}\,\frac{\delta(r - r_{a - a , a})}{rr_{a - a , a}}\,.\label{basis}\end{aligned}\ ] ] in the above expression , @xmath36 and @xmath37 are the internal ( antisymmetric ) wave functions of the first and second cluster , containing @xmath38 and @xmath39 nucleons ( @xmath40 ) , respectively . they are characterized by angular momentum quantum numbers @xmath41 and @xmath42 coupled together to form channel spin @xmath43 . for their parity , isospin and additional quantum numbers we use , respectively , the notations @xmath44 , and @xmath45 , with @xmath46 . the cluster centers of mass are separated by the relative coordinate @xmath47 where @xmath48 are the @xmath14 single - particle coordinates . the channel states ( [ basis ] ) have relative angular momentum @xmath49 . it is convenient to group all relevant quantum numbers into a cumulative index @xmath50 . the former basis states can be used to expand the many - body wave function according to @xmath51 as the basis states ( [ basis ] ) are not anti - symmetric under exchange of nucleons belonging to different clusters , in order to preserve the pauli principle one has to introduce the appropriate inter - cluster anti - symmetrizer , schematically @xmath52 where the sum runs over all possible permutations @xmath53 that can be carried out among nucleons pertaining to different clusters , and @xmath29 is the number of interchanges characterizing them . the coefficients of the expansion ( [ trial ] ) are the relative - motion wave functions @xmath54 , which represent the only unknowns of the problem . to determine them one has to solve the non - local integro - differential coupled - channel equations @xmath55 \frac{g^{j^\pi t}_\nu(r)}{r } = 0\,,\label{rgmeq}\ ] ] where the two integration kernels , the hamiltonian kernel , @xmath56 and the norm kernel , @xmath57 contain all the nuclear structure and anti - symmetrization properties of the problem . in particular , the non - locality of the kernels is a direct consequence of the exchanges of nucleons between the clusters . we have used the notation @xmath58 and @xmath59 to denote the total energy in the center - of - mass frame , and the intrinsic @xmath14-nucleon microscopic hamiltonian , respectively . the formalism presented above is combined with the _ ab initio _ ncsm in two steps : first , we note that the hamiltonian can be written as @xmath60 where @xmath61 and @xmath62 are the ( @xmath38)- and @xmath39-nucleon intrinsic hamiltonians , respectively , @xmath63 is the relative kinetic energy and @xmath64 is the sum of all interactions between nucleons belonging to different clusters after subtraction of the average coulomb interaction between them , explicitly singled out in the term @xmath65 ( @xmath66 and @xmath67 being the charge numbers of the clusters in channel @xmath68 ) . we use identical realistic potentials in both the cluster s hamiltonians and inter - cluster interaction @xmath64 . accordingly , @xmath36 and @xmath37 are obtained by diagonalizing @xmath61 and @xmath62 , respectively , in the model space spanned by the ncsm @xmath69 ho basis . note that in the present paper we use soft srg evolved @xmath11 potentials . therefore , there is no need to derive any further effective interaction tailored to the model space truncation as with these soft interactions our calculations converge in the model spaces we are able to reach . second , we replace the delta functions in the localized parts of the hamiltonian ( [ h - kernel ] ) and the norm ( [ n - kernel ] ) kernels with their representation in the ho model space . we use identical ho frequency as for the cluster eigenstate wave functions and a consistent model space size ( @xmath70 ) . we emphasize that this replacement is performed only for the localized parts of the kernels . the diagonal parts coming from the identity operator in the antisymmetrizers , the kinetic term and the average coulomb potential are treated exactly . in this paper , we apply the ncsm / rgm formalism in the single - nucleon projectile basis , i.e. , for binary - cluster channel states ( [ basis ] ) with @xmath71 ( with channel index @xmath72 ) . as an illustration , let s discuss in more detail the norm kernel that is rather simple in this basis : @xmath73 we can easily recognize a direct term , in which initial and final state are identical ( corresponding to diagram @xmath74 of fig . [ diagram - norm - pot ] ) , and a many - body correction due to the exchange part of the inter - cluster anti - symmetrizer ( corresponding to diagram @xmath75 of fig . [ diagram - norm - pot ] ) . we note that in calculating the matrix elements of the exchange operator @xmath76 we replaced the delta function of eq . ( [ basis ] ) with its representation in the ho model space as discussed above . this is appropriate as the transposition @xmath76 operator acting on the traget wave function is short - to - medium range . on the contrary , the @xmath77-function coming from the identity is treated exactly . the presence of the inter - cluster anti - symmetrizer affects also the hamiltonian kernel , and , in particular , the matrix elements of the interaction . for a @xmath11 potential one obtains a direct term involving interaction and exchange of two nucleons only ( see diagrams ( @xmath78 ) and ( @xmath79 ) of fig . [ diagram - norm - pot ] ) , and an exchange term involving three - nucleons . diagram ( @xmath80 ) of fig . [ diagram - norm - pot ] describes this latter term , in which the last nucleon is exchanged with one of the nucleons of the first cluster , and interacts with yet another nucleon . for more details on the integration kernels in the single - nucleon projectile basis we refer the readers to ref . @xcite . being translationally - invariant quantities , the norm and hamiltonian kernels can be naturally " derived working within the ncsm jacobi - coordinate basis . however , by introducing slater - determinant channel states of the type @xmath81^{(j^\pi t ) } r_{n\ell}(r^{(a)}_{\rm c.m.})\ , , \label{sd - basis}\end{aligned}\ ] ] in which the eigenstates of the @xmath82-nucleon fragment are obtained in the sd basis ( while the second cluster is still a ncsm jacobi - coordinate eigenstate ) , it can be easily demonstrated that translationally invariant matrix elements can be extracted from those calculated in the sd basis of eq . ( [ sd - basis ] ) by inverting the following expression : @xmath83 here @xmath84 represents any scalar and parity - conserving translational - invariant operator ( @xmath85 , @xmath86 , etc . ) . we exploited both jacobi - coordinate and sd channel states to verify our results . the use of the sd basis is computationally advantageous and allows us to explore reactions involving @xmath29-shell nuclei , as done in the present work . in order to calculate the parts of the integration kernels depicted in fig . [ diagram - norm - pot ] ( b ) , ( c ) and ( d ) , all information that we need from the sd basis calculation are one - body densities of the target eigenstates . for the ( e ) part of the integration kernel in fig . [ diagram - norm - pot ] , we need two - body densities of the target eigenstates obtained in the sd basis . due to the presence of the norm kernel @xmath87 , eq . ( [ rgmeq ] ) does not represent a system of multichannel schrdinger equations , and @xmath54 do not represent schrdinger wave functions . the short - range non - orthogonality , induced by the non - identical permutations in the inter - cluster anti - symmetrizers , can be removed by introducing normalized schrdinger wave functions @xmath88 where @xmath89 is the square root of the norm kernel , and applying the inverse - square root of the norm kernel , @xmath90 , to both left and right - hand side of the square brackets in eq . ( [ rgmeq ] ) . this procedure , explained in more detail in ref . @xcite , leads to a system of multichannel schrdinger equations @xmath91\frac{\chi^{j^\pi t}_{\nu } ( r)}{r } \nonumber\\[2 mm ] & & + \sum_{\nu^\prime}\int dr^\prime\,r^{\prime\,2 } \,w^{j^\pi t}_{\nu \nu^\prime}(r , r^\prime)\,\frac{\chi^{j^\pi t}_{\nu^\prime}(r^\prime)}{r^\prime } = 0,\label{r - matrix - eq}\end{aligned}\ ] ] where @xmath92 is the energy eigenvalue of the @xmath93-th cluster ( @xmath46 ) , and @xmath94 is the overall non - local potential between the two clusters , which depends on the channel of relative motion , while it does not depend on the energy . these are the equations that we finally solve to obtain both our scattering and bound - state results . the primary limitation for the range of applicability of the ncsm in terms of particle number @xmath14 and model spaces size @xmath95 results from the factorial growth of the dimension of the @xmath96 space . except for light isotopes , it is hardly possible to obtain a converged result using a bare hamiltonian within the @xmath96 spaces that are computationally tractable . at this point the importance truncation offers a solution . the importance truncation in connection with the ncsm was introduced in ref . @xcite and discussed in detail in ref . @xcite . in the following we summarize a few key features of the it - ncsm and generalize the approach to the simultaneous description of excited states . the motivation for the importance truncation results from the observation that the expansion of any particular eigenstate of the hamiltonian in a full @xmath97-scheme ncsm space typically contains a large number of basis states with extremely small or vanishing amplitudes . the amplitudes define an adaptive truncation criterion , which takes into account the properties of the hamiltonian and the structure of the eigenstate under consideration . if those amplitudes were known at least approximately before actually solving the eigenvalue problem , one could reduce the model space to the most relevant basis states by imposing a threshold on the amplitude . the amplitude of a particular basis state @xmath98 in the expansion of a specific eigenstate can be estimated using first - order multiconfigurational perturbation theory . in order to set up a perturbation series we need an initial approximation of the target state , the so - called reference state @xmath99 . in practice this reference state will be a superposition of basis states @xmath100 from a reference space @xmath101 : @xmath102 the reference state and the amplitudes @xmath103 are typically extracted from a previous ncsm calculation . based on @xmath99 as unperturbed state , we can evaluate the first - order perturbative correction to the target state resulting from basis states @xmath104 . their first - order amplitude defines the so - called importance measure @xmath105 the energy denominator @xmath106 in a mller - plesset - type partitioning is simply given by the unperturbed harmonic - oscillator excitation energy of the basis state @xmath98 ( see ref . @xcite for details ) . imposing an importance threshold @xmath107 , we construct an importance truncated model space including all basis states with importance measure @xmath108 . since the importance measure is zero for all basis states that differ from all of the states in @xmath101 by more than a two - particle - two - hole excitation , we have to embed the construction of the importance truncated space into an iterative update cycle . after constructing the importance truncated space and solving the eigenvalue problem in that space , we obtain an improved approximation for the target state that defines a reference state for the next iteration . in order to accelerate the evaluation of the importance measure , we typically do not use the complete eigenstate as new reference state , but project it onto a reference space spanned by the basis states with @xmath109 , where @xmath110 are the coefficients resulting from the solution of the eigenvalue problem . the second threshold @xmath111 will be chosen sufficiently small so as not to affect the results for a given @xmath107 threshold . simple iterative update schemes can be devised for any type of full model spaces , as discussed in refs . specifically for the @xmath96 space of the ncsm , however , there is an efficient sequential update scheme leading to the it - ncsm(seq ) approach . it is based on the fact that all states of an @xmath112 space can be generated from the basis states of an @xmath96 space using two - particle - two - hole excitations at most . thus a single importance update starting from a reference state in an @xmath96 space gives access to all relevant states in an @xmath112 space . making use of this property , in the it - ncsm(seq ) we start with a full ncsm calculation in , e.g. , @xmath113 and use this eigenstate after applying the @xmath111 threshold as reference state for constructing the importance truncated @xmath114 space . after solving the eigenvalue problem for this importance truncated @xmath114 space we use the resulting eigenstate as reference state to construct the @xmath115 space , and so on . thus only one importance update is required for each value of @xmath95 , which makes this scheme very efficient computationally . moreover , in the limit of vanishing thresholds , @xmath116 , this scheme recovers the full @xmath96 space at each step of the sequence , i.e. , the it - ncsm(seq ) would recover the full ncsm result . based on this limiting property , we can obtain a numerical approximation to the full ncsm result by extrapolating the it - ncsm(seq ) observables obtained for a set of different importance thresholds @xmath107 ( and in principle also @xmath111 ) to @xmath117 . through this extrapolation , the contribution of discarded basis states , i.e. those with importance measures @xmath118 below the smallest threshold considered , is effectively recovered . because the control parameter @xmath107 is tied to the physical structure of the eigenstate , we observe a smooth threshold dependence for all observables , which allows for a robust threshold extrapolation . in the case of the energy we can improve the quality of the extrapolation further by considering a perturbative second - order estimate for the energy of the excluded basis states . while setting up the importance truncated space , all second - order energy contributions @xmath119 for the discarded states with @xmath120 are summed up to provide a correction @xmath121 to the energy eigenvalue . by construction this correction goes to zero in the limit @xmath117 . we use this additional information for a constrained simultaneous extrapolation of the energy to vanishing threshold with and without perturbative correction for the excluded states as described in detail in ref . @xcite . the whole concept can be generalized to the description of excited states . for the present application in connection with the ncsm / rgm , we aim at an importance truncated model space that is equally well suited for the description of the lowest @xmath122 eigenstates of the hamiltonian for given parity and total angular momentum projection . instead of using a single reference state , we employ different reference states @xmath123 , with @xmath124 , for each of the @xmath122 target states . for each reference state we define a separate importance measure @xmath125 following eq . . a basis state @xmath126 is included in the importance truncated space if at least one of the importance measures @xmath127 is above the threshold @xmath107 , i.e. , if it is relevant for the description of at least one of the @xmath122 target states it will be included . because the different eigenstates are typically dominated by different basis states , the dimension of the importance truncated space grows linearly with @xmath122 . in the it - ncsm(seq ) scheme we start with a full ncsm calculation in @xmath113 and use the lowest @xmath122 eigenstates as initial reference states @xmath123 . based on the corresponding importance measures @xmath125 the importance truncated @xmath114 space is constructed and the lowest @xmath122 eigenvectors in this space serve as new reference states ( after application of the @xmath111 threshold ) for the construction of the @xmath115 space , and so on . from a sequence of it - ncsm(seq ) calculations we obtain a set of @xmath122 eigenvectors for each value of @xmath95 which can be used to evaluate other observables . by default we compute the expectation values of @xmath128 and @xmath129 as well as the expectation values of @xmath130 and @xmath131 . indeed , since we use an importance truncated space in the @xmath97-scheme without explicit angular momentum projection , the eigenstates are not guaranteed to have good angular momentum and isospin . we therefore monitor the expectation values of @xmath128 and @xmath129 and find values which typically differ by less then @xmath132 from the exact quantum numbers . as in the full ncsm we separate spurious center - of - mass ( cm ) excitations from the physical spectrum by adding a lawson term @xmath133 to the translationally invariant intrinsic hamiltonian @xmath130 ( with the typical choice @xmath134 ) . the use of this modified hamiltonian provides at the same time a diagnostic for potential cm contaminations of the intrinsic states induced by the importance truncation . as discussed in refs . @xcite , the independence of the intrinsic energies @xmath135 on @xmath136 and the smallness of @xmath137 demonstate that the it - ncsm(seq ) solutions are free of cm contaminations . eventually , the wave functions obtained in the it - ncsm(seq ) together with the threshold extrapolated intrinsic energies form the input for the ncsm / rgm calculations discussed in the following . the purpose of the nucleon-@xmath0he calculations presented in this paper is two - fold . first , we want to check the predictive power of the srg evolved chiral interaction in the @xmath23 system , where a lot of experimental scattering data exist and where our calculations can be easily converged with respect to the size of the basis expansion . second , we want to benchmark the importance truncation scheme with the full - space calculations all the way up to very large @xmath69 spaces . the first _ ab initio _ @xmath23 scattering calculations was reported in ref . @xcite . the @xmath28-@xmath10 low - lying @xmath138 and @xmath139 @xmath53-wave resonances as well as the @xmath140 @xmath141-wave non - resonant scattering below 5 mev c.m . energy were obtained using the av18 @xmath11 potential with and without the three - nucleon force , chosen to be either the urbana ix or the illinois-2 model . the results of these green s function monte carlo ( gfmc ) calculations revealed sensitivity to the inter - nucleon interaction , and in particular to the strength of the spin - orbit force . soon after , the development of the _ ab initio _ ncsm / rgm approach allowed us to calculate both @xmath28- and ( for the first time ) @xmath29-@xmath10 scattering phase shifts for energies up to the inelastic threshold @xcite , using several realistic @xmath11 potentials , including the chiral n@xmath26lo @xcite , the @xmath142 @xcite and the cd - bonn @xcite @xmath11 potentials . nucleon-@xmath10 scattering provides one of the best - case scenarios for the application of the ncsm / rgm approach . this process is characterized by a single open channel up to the @xmath143h threshold , which is fairly high in energy . in addition , the low - lying resonances of the @xmath0he nucleus are narrow enough to be reasonably reproduced diagonalizing the four - body hamiltonian in the ncsm model space . in the present work we include the first excited state of @xmath0he , the @xmath144 state , as a closed channel in our ncsm / rgm basis space . we performed extensive nucleon-@xmath0he calculations with the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 to check convergence of our ncsm / rgm calculations . in fig . [ fig : nhe4_phaseconv ] , we present @xmath28-@xmath0he phase shift results for the @xmath141- and @xmath53-waves obtained using an ho basis expansion up to @xmath146 for for the localized parts of the ncsm / rgm integration kernels and for the @xmath0he ground- and the first - excited @xmath144 wave functions ( since these states have positive parity , the @xmath147 expansion is in fact used for the @xmath0he eigenstates ) . as seen in the figure , the phase - shift convergence is excellent . in particular , the @xmath146 and the @xmath148 curves lie on top of each other . the convergence rate demonstrated here is quite similar to that obtained using the @xmath149 @xmath11 potentatial in our erlier study ( compare the present fig . [ fig : nhe4_phaseconv ] to the left panel of fig . 13 in ref . @xcite ) . -@xmath0he phase shifts on the size of the ho basis expansion of the @xmath0he wave functions and the localized parts of the integration kernels . the @xmath0he ground state and the first @xmath144 excited states were included . the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 and the ho frequency @xmath150 mev were used . ] as shown in the previous subsection , for the @xmath23 system we are able to reach complete convergence with @xmath0he wave functions obtained within full , non - truncated , ncsm calculations . we can , therefore , test the performace of the it - ncsm scheme in this system all the way up to very large @xmath70 values and see how well the it - ncsm scheme reproduces the completetly converged results . it should be noted that for heavier @xmath151 and @xmath9 systems investigated later , full , non - truncated ncsm calculations for the @xmath152 ( @xmath153 ) target nuclei are feasible only up to @xmath154 ( @xmath155 ) . it is , therefore , desirable and important to benchmark the it - ncsm calculations in a lighter system like @xmath23 in @xmath156 calculations . in fig . [ fig : nhe4_phase_full_it ] , we compare @xmath28-@xmath0he phase shifts calculated within the ncsm / rgm with @xmath0he wave functions obtained in a full @xmath157 ncsm calculation and those obtained using @xmath0he wave functions obtained within an @xmath157 it - ncsm calculation . the agreement of the two sets of phase shifts is excellent . it should be noted that the dimension of the full @xmath157 @xmath0he ncsm basis is 6344119 . the dimension of the it - ncsm basis used here to calculate the @xmath0he wave functions was just 992578 , more than a factor of six smaller . truncation parameters @xmath158 and @xmath159 were used . the ground state energy from the full ncsm calculation is @xmath160 mev . the @xmath161 extrapolated ground state energy from the it - ncsm calculation is @xmath162 mev with a difference from the full result less than 10 kev . the excited @xmath144 energy obtained in the full ncsm calculation was 21.58 mev . the corresponding extrapolated it - ncsm result was 21.4(1 ) mev . the slightly lower accuracy of the excited state reproduction in the it - ncsm calculation is manifested in a very small deviation of the @xmath141-wave phase shift at energies above 12 mev ( less than 1 degree at 16 mev ) . it should be noted that the excited @xmath144 state is not bound . consequently , it is challenging to reproduce the excited state as well as the ground state in a importance - truncated calculation . it should be also pointed out that unlike for the energies , no phase shift extrapolation was performed . the needed one- and two - body densities were calculated from the wave functions obtained in the it - ncsm calculation with the truncation parameters described above . the excellent agreement of the full and the it - ncsm phase shifts demonstrates that no extrapolation was actually necessary . obviously , we can check the dependence of observables like phase shifts on the @xmath107 and @xmath111 and perform an extrapolation to vanishing values of these parameters if needed . -@xmath0he @xmath141- and @xmath53-wave phase shifts . results obtained with @xmath0he wave functions from full ncsm ( solid lines ) and it - ncsm ( dashed lines ) calculations are compared . the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 , the @xmath146 basis space and the ho frequency @xmath150 mev were used . see text for details on the it - ncsm calculation . ] he ( left panels ) and @xmath163he ( right panels ) compared to the r - matrix analysis of experimental data @xcite . the ncsm / rgm calculations that included the @xmath0he ground state and the @xmath144 excited state were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . the ho frequency @xmath150 mev and @xmath146 basis space were employed . ] our calculated @xmath28-@xmath0he and @xmath29-@xmath0he phase shifts are compared to those obtained from an @xmath164-matrix analysis of @xmath165he experimental data @xcite in fig . [ fig : n4he_phase ] . the agreement is quite reasonable for the @xmath141-wave , @xmath166-wave and @xmath167-wave . the @xmath168 resonance is positioned at higher energy in the calculation and the corresponding phase shifts are underestimated with respect to the @xmath164-matrix results , although the disagreement becomes less and less pronounced starting at about 8 mev . while the inclusion of negative - parity excited states of the @xmath169particle would likely increase somewhat the @xmath168 phase shifts @xcite , the observed difference is largely due to a reduction in spin - orbit strength caused by the neglect of the three - nucleon interaction in our calculations . the importance of the three - nucleon force in reproducing the @xmath164-matrix @xmath168 phase shifts was demonstrated in the gfmc @xmath28-@xmath0he calculations of ref . overall , the present results obtained with the srg - n@xmath26lo @xmath11 interaction agree better with experiment than our earlier calculations @xcite with the @xmath149 , n@xmath26lo and cd - bonn @xmath11 potentials . the only exception is the @xmath141-wave phase shift which is best described using the cd - bonn @xmath11 potential . the larger spin - orbit strength of the employed srg - n@xmath26lo potential with respect to n@xmath26lo itself is the likely responsible for the improved agreement . as our calculated phase shifts agree with the experimental ones reasonably well above the center - of - mass energy of 8 mev , we expect a similar behavior for cross section and analysing power in that energy range . this is indeed the case as shown in fig . [ fig : n4he_17mevxsay ] , where the calculated differential cross section and analyzing power are compared to experimental data from karlsruhe @xcite with polarized neutrons of @xmath17017 mev laboratory energy . for the cross section experimental data see also references in @xcite . the cross section is reproduced remarkably well at all angles and the analysing power is in reasonable agreement with the data , particularly at backward angles . the same quality of agreement can be found for all energies far from the low - lying resonances , as shown in the right panel of fig . [ fig : n4he_17mevxsay ] for the analysing power at @xmath171 mev and 19 mev . a better display of the dependence of our calculated cross section and analysing power upon the incident nucleon energy is provided by fig . [ fig : p4he_1 ] , where the @xmath163he results for these observables are compared to the data of ref . @xcite at the proton laboratory energies of @xmath172 , 7.89 , 9.89 , and 11.99 mev . as expected from the behavior of the phase shifts described earlier , for energies relatively close to the resonance region we find a rather poor agreement with experiment , particularly noticeable in the analysing power overall and in the cross section at backward angles . however , starting at about 10 mev , the agreement improves substantially and data are once again reproduced in a quite satisfactory way at higher energies , as shown in fig . [ fig : p4he_2 ] , where the ncsm / rgm @xmath163he results are compared to various experimental data sets @xcite in the energy range @xmath173 mev . the @xmath1be(@xmath29,@xmath175)@xmath7b capture reaction plays a very important role in nuclear astrophysics as it serves as an input for understanding the solar neutrino flux @xcite . while the experimental determination of the neutrino flux from @xmath7b has an accuracy of about 9% @xcite , the theoretical predictions have uncertainties of the order of 20% @xcite . the theoretical neutrino flux depends on the @xmath1be(@xmath29,@xmath175)@xmath7b s - factor . significant experimental and theoretical effort has been devoted to studying this reaction . the s - factor extrapolation to astrophysically relevant energies depends among other things on the scattering lengths of the proton scattering on @xmath1be . experimental determination of these lengths was performed recently @xcite with precision of the order of 30% . the proton-@xmath1be elastic scattering was also investigated in ref . @xcite . to benchmark the theoretical calculations used for s - factor extrapolations , an investigation of the mirror capture reaction , @xmath1li(@xmath28,@xmath175)@xmath7li , as well as the @xmath28+@xmath1li scattering is important . for example , the @xmath28+@xmath1li scattering lengths are known with a higher accuracy @xcite . li ground - state and the @xmath139 and @xmath176 excited state energy dependence on the model - space size @xmath70 , obtained within the importance - truncated ncsm ( solid lines ) , using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . the ho frequency @xmath150 mev was employed . the full - space ncsm results are shown by dashed lines . ] the first applications of the ncsm approach to the description of the @xmath1be(@xmath29,@xmath175)@xmath7b capture reaction @xcite required a phenomenological correction of the asympotic behavior of the overlap functions and , further , the scattering @xmath29+@xmath1be wave function was calculated from a phenomenological potential model . the present investigation within the _ ab initio _ ncsm / rgm approach paves the way for a complete first principles calculation of this capture reaction . here , we limit ourselves to scattering calculations and postpone the capture reaction calculations to a forthcoming paper . our current limit on the unrestricted ncsm calculations for @xmath1li and @xmath1be is @xmath154 . to improve the convergence of our scattering calculations , we utilize wave functions obtained within the it - ncsm . in that scheme , we are able to reach @xmath177 model spaces and calculate both ground as well as low - lying excited states . this is demonstrated in fig . [ fig : li7_itncsm ] . with the srg - n@xmath26lo @xmath11 potential with @xmath178 @xmath145 employed in the present study we reach convergence already around @xmath179 . also , as seen in the figure , the aggreement between the unrestricted ncsm and the it - ncsm is perfect up to the highest accessible unrestricted space , @xmath154 . -wave diagonal phase shifts of the @xmath28-@xmath1li elastic scattering ( top panel ) , elastic @xmath1li(@xmath28,@xmath28)@xmath1li cross section ( middle panel ) , and inelastic @xmath1li(@xmath28,@xmath180)@xmath1li(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1li ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . experimental data are from ref . @xcite . ] -wave diagonal phase shifts of the @xmath28-@xmath1li elastic scattering ( top panel ) , elastic @xmath1li(@xmath28,@xmath28)@xmath1li cross section ( middle panel ) , and inelastic @xmath1li(@xmath28,@xmath180)@xmath1li(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1li ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . experimental data are from ref . @xcite . ] -wave diagonal phase shifts of the @xmath28-@xmath1li elastic scattering ( top panel ) , elastic @xmath1li(@xmath28,@xmath28)@xmath1li cross section ( middle panel ) , and inelastic @xmath1li(@xmath28,@xmath180)@xmath1li(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1li ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . experimental data are from ref . @xcite . ] the ncsm / rgm coupled - channel calculations performed for the @xmath4 system include the @xmath1li ( @xmath1be ) ground state , the first excited @xmath139 state as well as the second excited @xmath176 state . it is essential to include the @xmath176 state in order to reproduce the low - lying @xmath183 resonance in @xmath7li and @xmath7b . using these three states , we are able to reach model spaces up to @xmath182 , which is sufficient concerning the ho basis expansion convergence as can be judged from fig . [ fig : li7_itncsm ] . the coupled channel calculation described above gives two bound states for the @xmath28-@xmath1li system , a @xmath31 corresponding to the experimentally observed @xmath7li ground state , bound by 2.03 mev @xcite , and a @xmath184 corresponding to the @xmath7li first excited state at @xmath185 mev , bound by 1.05 mev @xcite . the calculated states are bound by 1.16 mev and 0.17 mev , respectively , i.e. less than in experiment . this is in part due to the fact that higher excited states of @xmath1li were omitted . in fig . [ fig : li7_it_317_pwaves ] , we present our results for the diagonal @xmath53-wave phase shifts of the @xmath28+@xmath1li elastic scattering as well as the elastic @xmath1li(@xmath28,@xmath28)@xmath1li and inelastic @xmath1li(@xmath28,@xmath180)@xmath1li(1/2@xmath181 ) cross sections . at low energies , we can identify four resonances two of which can be associated with the experimentally known @xmath7li states : @xmath183 at @xmath186 mev and @xmath184 at @xmath187 mev @xcite . the other two resonances , @xmath188 and @xmath31 are not present in the @xmath7li evaluation of ref . @xcite . they do appear in many theoretical calculations including the gfmc @xcite , ncsm @xcite and recoil - corrected continuum shell model ( rccsm ) @xcite . the @xmath188 resonance also appears in the gcm calculations of ref . contributions of different resonances to the cross sections can be deduced from fig . [ fig : li7_it_317_pwaves ] . the elastic cross section is dominated by the @xmath183 resonance with some contributions from the @xmath31 resonace at higher energy . the inelastic cross section shows a peak just above the threshold due to the @xmath188 resonance and also a contribution from the @xmath184 resonance . the appearance of a @xmath188 peak just above threshold of the @xmath1li(@xmath28,@xmath180)@xmath1li(1/2@xmath181 ) reaction was also discussed in ref . @xcite ( see fig . 10 in that paper ) . the data of ref . @xcite seem to rule out a @xmath188 state so close to the threshold . it is known , however , that the position of the @xmath188 state is sensitive to the strength of the spin - orbit interaction @xcite . the three - nucleon interaction , that would increase the strenght of the spin - orbit force , was not included in our present calculations . consequently , our predicted @xmath188 state energy is likely underestimated . we note that no fit to the experimental treshold was done in the present ncsm / rgm calculations . still , as seen in the bottom panel of fig . [ fig : li7_it_317_pwaves ] , the calculated inelastic cross section is very close to the experimental data just above the threshold . -wave diagonal phase shifts of the @xmath29+@xmath1be elastic scattering ( top and middle panel ) and inelastic @xmath1be(@xmath29,@xmath189)@xmath1be(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1be ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . in the middle panel , the full - space ncsm ( solid lines ) and the it - ncsm ( dashed lines ) results in the @xmath154 basis are compared . ] -wave diagonal phase shifts of the @xmath29+@xmath1be elastic scattering ( top and middle panel ) and inelastic @xmath1be(@xmath29,@xmath189)@xmath1be(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1be ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . in the middle panel , the full - space ncsm ( solid lines ) and the it - ncsm ( dashed lines ) results in the @xmath154 basis are compared . ] -wave diagonal phase shifts of the @xmath29+@xmath1be elastic scattering ( top and middle panel ) and inelastic @xmath1be(@xmath29,@xmath189)@xmath1be(1/2@xmath181 ) cross section ( bottom panel ) . the ncsm / rgm calculation that included the @xmath1be ground state and the @xmath139 and @xmath176 excited states were done using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.02 @xmath145 . wave functions from it - ncsm calculations in the @xmath182 basis and the ho frequency of @xmath150 mev were employed . in the middle panel , the full - space ncsm ( solid lines ) and the it - ncsm ( dashed lines ) results in the @xmath154 basis are compared . ] in the mirror system , @xmath29-@xmath1be , we do not find a bound state in the same type of coupled - channel ncsm / rgm calculation as described above for @xmath28-@xmath1li . as seen in the top and the middle parts of fig . [ fig : be7_317_pwaves ] , the lowest @xmath31 resonance corresponding to the @xmath7b ground state lies at about 200 kev above the threshold . in experiment , @xmath7b is bound by 137 kev @xcite . our calculated lowest @xmath184 resonance appears at about 1 mev . it corresponds to the experimental @xmath7b @xmath184 state at @xmath190 mev ( 0.63 mev above the @xmath29-@xmath1be threshold ) . this resonance dominates the inelastic cross section as seen in the bottom part of fig . [ fig : be7_317_pwaves ] . the higher lying resonances follow similar patterns as those found in @xmath28-@xmath1li ( fig . [ fig : li7_it_317_pwaves ] ) . again , we find @xmath188 and @xmath31 resonances not included in the recent @xmath7b evaluation @xcite . we note that experimental efforts are now under way to find these resonances @xcite . we further note that our calculated @xmath191 states in @xmath7li and @xmath7b appear at a significantly higher energies than the corresponding @xmath191 states obtained within the microscopic cluster model in ref . @xcite . the middle panel of fig . [ fig : be7_317_pwaves ] demonstrates once again the good accuracy of the importance truncated calculations for a high @xmath25 , @xmath154 , model space . the it calculation reduced the @xmath1be basis from 43.6 million to 11.9 million in the present case . be(@xmath29,@xmath29)@xmath1be ( top panel ) and inelastic @xmath1be(@xmath29,@xmath189)@xmath1be(1/2@xmath181 ) ( bottom panel ) differential cross section at @xmath192 calculated within the ncsm / rgm with srg - n@xmath26lo @xmath11 potential with @xmath178 @xmath145 . ] be(@xmath29,@xmath29)@xmath1be ( top panel ) and inelastic @xmath1be(@xmath29,@xmath189)@xmath1be(1/2@xmath181 ) ( bottom panel ) differential cross section at @xmath192 calculated within the ncsm / rgm with srg - n@xmath26lo @xmath11 potential with @xmath178 @xmath145 . ] the elastic @xmath29-@xmath1be scattering was measured at @xmath193 and analyzed by the r - matrix approach @xcite . cross section calculations within the rccsm at that angle were then published in ref . @xcite and also in ref . further , elastic and inelastic cross sections at this angle were analyzed within the time - dependent approach to the continuum shell model ( tdcsm ) @xcite . our elastic and inelastic differential cross section results at @xmath193 are presented in fig . [ fig : p_be7_148 ] . in the elastic cross section , the first @xmath184 state is visible and beyond the minimum of the cross section , we can see the dominant peak due to the @xmath183 state . at higher energies , the @xmath31 state contributes as well . the inelastic cross section at @xmath193 has a similar shape as the reaction cross section shown in fig . [ fig : be7_317_pwaves ] . the first @xmath184 state peak dominates at low energy with contributions from the @xmath188 and the second @xmath184 at higher energies . our findings are in line with the rccsm results . however , we remind the reader that there is no fitting in our calculations , all results being predictions based on the srg - n@xmath26lo @xmath11 potential . because of this , the positions of our calculated resonances , e.g. , @xmath184 , @xmath183 do not exactly reproduce experiment . we do not include the experimental data in the figure as they would be shifted compared to the calculated peaks . there are at least two reasons why our predictions do not match the experimental resonances accurately . first , our nuclear hamiltonian is incomplete , e.g. no three - nucleon interaction is included . second , we omitted higher resonances of @xmath1li and @xmath1be due to numerical reasons . most likely , the omitted resonances would produce some shifts in the calculated peaks . both these points can and will be addressed in the future . still , our current results contain the bulk of the physics behind the investigated scattering processes . -wave phase shifts of the @xmath28+@xmath1li ( solid lines ) and the @xmath29+@xmath1be ( dashed lines ) elastic scattering . the calculations as described in figs . [ fig : li7_it_317_pwaves ] and [ fig : be7_317_pwaves ] . ] . the @xmath28-@xmath1li and the @xmath29-@xmath1be @xmath141-wave scattering lengths . theoretical values correspond to calculations as described in figs . [ fig : li7_it_317_pwaves ] and [ fig : be7_317_pwaves ] . experimental values are from refs . @xcite . [ cols="^,^,^,^,^ " , ] in fig . [ fig : be7_li7_317_swaves ] , we present our calculated @xmath28-@xmath1li and the @xmath29-@xmath1be @xmath141-wave phase shifts . we do not find any evidence for a @xmath194 resonance advocated in ref . @xcite and discussed in ref . the corresponding scattering lengths together with the experimental values are given in table [ tab : s - wave_scatl ] . with the exception of the @xmath29-@xmath1be @xmath195 , which has a large experimental uncertainty , our calculated scattering lengths do agree with experimental data as to their signs , there are however differences in the absolute values . again , as dicussed above , the results presented here serve only as a first step towards the _ ab initio _ investigation of the @xmath28-@xmath1li and the @xmath29-@xmath1be reactions . prospects for a realistic calculation of the @xmath1be(@xmath29,@xmath175)@xmath7b capture are excellent . here we found the @xmath7b unbound by only 200 kev . it is quite possible that @xmath7b will become bound ( with the @xmath11 potential employed here : srg - n@xmath26lo with @xmath178 @xmath145 ) by including more excited states of @xmath1be in the coupled - channel ncsm / rgm calculations . the effect of higher excited states of @xmath1be can be , in fact , most efficiently included by coupling the presently used ncsm / rgm basis with the @xmath7b ncsm eigenstates as outlined in ref . even if the @xmath7b would not be bound or , most likely , the theshold energy will not agree with the experiment , we have the possibilty to explore a variation of the srg @xmath11 potential evolution parameter @xmath196 and tune this parameter to fit the experimental threshold . we note that for any @xmath196 the srg - evolved @xmath11 potential will describe all two - nucleon properties as accurately as the original starting @xmath11 potential , here the chiral n@xmath26lo potential of ref . it should be noted that by adding the three - nucleon interaction , omitted in the present calculations due to computational reasons , the need for a fine - tuning should be significantly reduced , i.e. the results should become @xmath196 independent . for nucleon scattering calculations on @xmath2c or heavier targets within the ncsm / rgm , the use of the importance truncation becomes essential . for @xmath2c , the full - space ncsm calculations are currently limited to @xmath155 ( although successful runs were already performed for @xmath154 on the biggest supercomputers with the latest version of the code mfd @xcite ) . this is insufficient for reaching or approaching convergence of the @xmath2c ncsm calculations as seen from fig . [ fig : c12_itncsm ] and even more so of the ncsm / rgm scattering calculations . the importance - truncated calculations , on the other hand , are feasible up to @xmath177 , where convergence is reached for both the ground state as well excited states . our @xmath2c calculations are performed with the srg - n@xmath26lo @xmath11 potential with the evolution parameter @xmath197 @xmath145 , a higher value ( i.e. shorter evolution , less soft ) than that used for the lighter nuclei . the use of a small @xmath196 results in large overbinding of heavier nuclei and a significant underestimation of their radii . as seen in fig . [ fig : c12_itncsm ] , our converged @xmath2c binding energy is about 84.5(8 ) mev , smaller than the experimental value of 92 mev and , further , the agreement of the full - space and importance - truncated results is perfect all the way up to @xmath155 . state energy dependence on the model - space size @xmath70 for @xmath2c , obtained within the importance - truncated ncsm , using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 . the ho frequency @xmath198 mev was employed . the calculation is variational . no ncsm effective interaction was used . the full ncsm results were obtained with the code antoine @xcite . ] our low - energy @xmath29-@xmath2c phase shift results are shown in fig . [ fig : pc12 ] . the comparison of the @xmath157 and @xmath199 results demonstrates good convergence with respect to the ho basis expansion . the @xmath2c ground state and the first @xmath31 state were included in the coupled - channels ncsm / rgm equations . we note that we also performed a phase shift comparison of the full - space and the importance - truncated calculations up to @xmath200 and found a similarly perfect agreement as presented in fig . [ fig : nhe4_phase_full_it ] for @xmath28-@xmath0he . -@xmath2c eigenphase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 and the ho frequency @xmath198 mev . full lines ( dotted lines ) corespond to results obtained in the @xmath157 ( @xmath199 ) model space . the ground state and the first excited @xmath31 state of @xmath2c was included . the @xmath2c wave functions were obtained within the it ncsm . ] in the present @xmath29-@xmath2c calculations , we found a single bound state , @xmath139 at -2.98 mev , corresponding to the @xmath30n ground state , bound experimentally by 1.94 mev @xcite . the lowest resonance in our calculation is @xmath201 , barely visible at 0.25 mev above threshold . in experiment , this resonance is at 1.56 mev . our calculated @xmath140 resonance appears at about 1.5 mev above treshold ( in experiment at 0.42 mev above threshold ) and the @xmath202 resonance at about 4.9 mev ( in experiment at 2.61 mev ) . in the mirror system , @xmath28-@xmath2c , our ncsm / rgm calculations produce three bound states : @xmath139 at -5.34 mev corresponding to the @xmath30c ground state experimentally bound by 4.95 mev with respect to the @xmath28-@xmath2c threshold , @xmath201 , bound by 2.23 mev ( experimentally bound by 1.26 mev ) , and @xmath140 bound by just 0.03 mev ( experimentally bound by 1.86 mev ) . in experiment , there is also a @xmath202 state bound by 1.09 mev . our present ncsm / rgm calculations including the lowest @xmath188 and and the lowest @xmath31 @xmath2c states do not produce any bound @xmath202 state . -@xmath2c phase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 . the ho frequency @xmath198 mev and the model - spaces size of @xmath157 were used . the ground state and the first excited @xmath31 state of @xmath2c was included . the @xmath2c wave functions were obtained within the it ncsm . ] our low - energy @xmath28-@xmath2c diagonal phase shifts are shown in fig . [ fig : nc12 ] . the @xmath202 resonance is found at 2.8 mev ( experimenally at 1.92 mev with respect to the @xmath28-@xmath2c threshold ) . the steep drop of the @xmath140 phase shift is due to the presence of the very weakly bound @xmath140 state . we note that similarly as in the case of @xmath21be , discussed in ref . @xcite , we observe a significant decrease of the @xmath140 state energy in the @xmath28-@xmath2c ncsm / rgm calculation when compared to the standard ncsm calculation for @xmath30c . we were able to make these comparisons in model spaces up to @xmath200 where we found this drop to be about 3 mev . -@xmath2c elastic scattering below and above the calculated @xmath202 resonance . energies are in the center of mass . the calculation as described in fig . [ fig : nc12 ] . ] analysing powers were measured for proton and neutron scattering on @xmath2c @xcite and scattering experiments on polarized proton target are under way @xcite . in fig . [ fig : nc12_ay ] , we present our calculated analyzing power below and above the energy of the @xmath202 resonance . we note that our calculated @xmath202 resonance appears at 2.8 mev in the center of mass ( experimentally at 1.92 mev ) . below the resonance , the analyzing power is positive at @xmath203 and negative at @xmath204 . at energies above the resonance , the analyzing power reverses its sign . similar observations were made in calculations performed within the multichannel algebraic scattering ( mcas ) theory @xcite . see in particular fig . 5 of ref . @xcite . our calculated @xmath30n and @xmath30c bound - state levels and resonances are more spread than the experimental ones . this is a consequence of an underestimation of the @xmath2c radius found to be 2.05 fm with the srg - n@xmath26lo @xmath11 potential . to remedy this , one would have to calculate three - nucleon interaction terms induced due to the srg evolution . this can be done as described in ref . however , we still need to further develop the ncsm / rgm formalism in order to handle three - nucleon interactions in the scattering calculations . the calculation of nucleon scattering on @xmath3o is the most challenging among the systems we investigate in this paper . the @xmath10 clustering plays an important role in the structure of @xmath3o , in particular for the first excited @xmath188 state that is known to be almost impossible to reproduce in ncsm or coupled - cluster calculations . our present calculations do not include the @xmath10 clustering yet . as in the case of @xmath2c , we rely on the importance - truncated ncsm calculations for obtaining the @xmath3o wave functions as the full-@xmath70 ncsm calculations are possible only up to @xmath155 . in fig . [ fig : o16_itncsm ] , we show the ground - state convergence within the it - ncsm and a comparison to the full - space results . again , up to the largest accessible model space , the agreement between the importance - truncated and the full - space calculations is perfect . for @xmath3o , obtained within the importance - truncated ncsm , using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 . the ho frequency @xmath198 mev was employed . the calculation is variational . no ncsm effective interaction was used . the full ncsm results were obtained with the code antoine @xcite . ] o phase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 and the ho frequency @xmath198 mev in the @xmath177 model space . the ground state and of @xmath3o was included . the @xmath3o wave functions were obtained within the it ncsm . ] o phase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 . the ho frequency of @xmath198 mev was used . the @xmath205 channel is shown in the top ( bottom ) panel . model space sizes up to @xmath177 were considered . the ground state and of @xmath3o was included . the @xmath3o wave functions were obtained within the it ncsm . ] o phase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 . the ho frequency of @xmath198 mev was used . the @xmath205 channel is shown in the top ( bottom ) panel . model space sizes up to @xmath177 were considered . the ground state and of @xmath3o was included . the @xmath3o wave functions were obtained within the it ncsm . ] it is straightforward to converge nucleon-@xmath3o scattering calculations within the ncsm / rgm using the ho expansion up to @xmath177 . our calculated @xmath28-@xmath3o phase shifts are shown in fig . [ fig : no16gs_18 ] and the ho - basis expansion convegence is checked for the @xmath141- and the @xmath166-wave in fig . [ fig : no16gs_conv ] . these calculations included the @xmath3o ground state only . we find two bound states , @xmath140 at -0.88 mev and @xmath202 at -0.41 mev with respect to the @xmath28-@xmath3o threshold . in experiment , the @xmath206o ground state is @xmath202 , bound by 4.14 mev , and the @xmath140 state is the first excited state bound by 3.27 mev . there are also two additional bound states : @xmath139 and @xmath201 . those are unbound in our calculations . -@xmath3o phase shifts calculated within the ncsm / rgm using the srg - n@xmath26lo @xmath11 potential with a cutoff of 2.66 @xmath145 and the ho frequency @xmath198 mev in the @xmath207 model space . the ground state and and the lowest @xmath208 , @xmath209 and @xmath194 excited states of @xmath3o were included . the @xmath3o wave functions were obtained within the it ncsm . ] clearly , it is insufficient to consider only the ground state of @xmath3o in the coupled - channel ncsm / rgm scattering calculations . we , therefore , include in addition the three lowest @xmath3o negative parity states : @xmath208 , @xmath209 , and @xmath194 . due to computational limitations , in this case we used ho basis expansion up to @xmath70=13 . comparing fig . [ fig : no16gs312_15 ] to fig . [ fig : no16gs_18 ] , the @xmath210 negative - parity excited states of @xmath3o generate negative - parity resonances in @xmath206o . these resonances do appear , however , at much higher energy than in experiment . the reason for this is the fact that our calculated @xmath3o @xmath210 states have too large excitation energy . in particular , our calculated @xmath208 excited state has an excitation energy of 15.99 mev while experimentally it lies at just 6.13 mev . one reason for the discrepancy is the softness of the srg - n@xmath26lo @xmath11 potential we use that results in an overall overbinding of the @xmath3o ground state and in an underestimation of its radius . another aspect is the challenging problem of the it - ncsm extrapolations of the independent positive and negative - parity state calculations . the uncertainties of the relative excitation energies are higher than in same - parity calculations . on the positive side our calculation with the negative - parity states , even though with overestimated excitation energies , results in the proper ordering of the @xmath206o bound states . the ground state is @xmath202 at -1.32 mev and the @xmath140 state gains binding as well , appearing at -1.03 mev . we also investigated the @xmath29-@xmath3o scattering and @xmath206f states . when the ncsm / rgm calculations are restricted to the channels involving only the @xmath3o ground state , we find a @xmath140 resonance at 1.0 mev and a @xmath202 resonance at 2.2 mev . these resonances correspond to the @xmath206f @xmath140 first excited state , bound by 0.105 mev , and the @xmath206f @xmath202 ground state bound by 0.6 mev with respect to the @xmath29+@xmath3o threshold . by coupling channels involving the @xmath210 @xmath3o @xmath208 , @xmath209 and @xmath194 excited states , the calculated @xmath140 and @xmath202 states are still unbound resonances but their energy moves significantly closer to the threshold : the @xmath140 appears at + 0.7 mev and the @xmath202 at + 1.2 mev . the @xmath206f low - lying states were recently investigated within the coupled - cluster approach with the gamow - hartree - fock basis @xcite . in those calculations with the n@xmath26lo @xmath11 potential , the @xmath140 state is weakly bound while the @xmath202 state remains unbound by about 0.1 mev . using the srg evolved interaction , the @xmath202 state became bound with the decrease of the cutoff @xmath196 . we note that our calculated @xmath3o ground state energy , -139.0(8 ) mev ( fig . [ fig : o16_itncsm ] ) obtained with the srg - n@xmath26lo @xmath11 potential with @xmath197 @xmath145 , compares well with the ccsd coupled - cluster @xmath3o calculations : -137.6 mev with the srg - n@xmath26lo @xmath11 potential with @xmath211 @xmath145 @xcite . the differences in the positions of the @xmath140 and the @xmath202 are due to deficiencies in our description of the negative parity @xmath210 states , which could be related to the two - body hamiltonian used here as well as the uncertainties of the threshold extrapolations for the excitation energies . the inclusion of additional @xmath3o excited states would increase the absolute energy of our calculated @xmath206f states . the most efficient way to do this is by coupling the presently used ncsm / rgm basis with the @xmath206f ncsm eigenstates in as outlined in ref . by combining the importance truncation scheme for the cluster eigenstate basis with the _ ab initio _ ncsm / rgm approach , we were able to perform many - body calculations for nucleon scattering on nuclei with mass number as high as @xmath212 . with the soft srg - evolved chiral @xmath11 potentials , convergence of the calculations with respect to the ho basis expansion of the target eigenstates and the localized parts of the ncsm / rgm integration kernels can be reached using @xmath213 . we first benchmarked the it - ncsm results with the full - space ncsm results for the @xmath23 system . our neutron-@xmath0he and proton-@xmath0he calculations compare well with an r - matrix analysis of the data in particular at energies above 8 mev , and describe well measured cross sections and analysing powers for those energies . our calculations of @xmath28-@xmath1li and @xmath29-@xmath1be scattering predict low - lying @xmath188 and @xmath31 resonances in @xmath7li and @xmath7b that have not been experimentally clearly identified yet . we found that the prospects of a realistic _ ab initio _ calculation of the @xmath1be(@xmath29,@xmath175)@xmath7b capture within our approach are very good . in the present calculations we found the @xmath7b unbound by only 200 kev . it is quite possible that @xmath7b will become bound ( with the @xmath11 potential employed here : srg - n@xmath26lo with @xmath178 @xmath145 ) by including more excited states of @xmath1be in the coupled - channel ncsm / rgm calculations . even if the @xmath7b will still not be bound or , most likely , the threshold energy will not agree with the experiment , we have the possibility to explore a variation of the srg @xmath11 potential evolution parameter @xmath196 and tune this parameter to fit the experimental threshold . the use of the importance - trunacted basis becomes essential in calculations with @xmath2c or @xmath3o targets as the full - space ncsm calculations are limited to @xmath155 . our @xmath28-@xmath2c and @xmath29-@xmath2c investigations included @xmath2c ground and the first excited @xmath31 states . we found a single bound state , @xmath140 in @xmath30n as in experiment . in @xmath30c , we found three bound states with the @xmath202 state still unbound contrary to experiment . our calculated spectrum of @xmath8 states is more spread than in experiment due to the underestimation of the @xmath2c radius , a consequence of the softness of the srg - evolved @xmath11 interaction . the description of nucleon scattering on @xmath3o within our formalism was the most challenging . the @xmath10 clustering that plays an important role in the structure of @xmath3o is not yet included in our present calculations . further , the @xmath210 @xmath3o excited states are more difficult to treat in the it - ncsm approach , as the extrapolations of excitation energies are done from the independent ground state and the negative - parity state calculations . we found a strong impact of the @xmath210 @xmath3o states on the positions of the lowest @xmath9 states . for example , correct ordering of the @xmath202 and the @xmath140 states in @xmath206o was obtained only when the @xmath210 states were included . overall , we find that the inclusion of additional excited states of the target nuclei would be beneficial in all studied systems and more significant with the increase of @xmath14 . coupled - channel ncsm / rgm calculations with many excited states of the traget are computationally challenging . the most efficient way of including the effects of such states is by coupling the presently used ncsm / rgm basis , consisting of just a few lowest excited states , with the ncsm eigenstates of the composite system as outlined in ref . @xcite . work on this coupling is under way . the use of the srg - evolved @xmath11 interaction facilitates convergence of the ncsm / rgm calculations with respect to the ho basis expansion . on the other hand , due to the softness of these interactions , radii of heavier nuclei become underestimated . to remedy this , one would have to calculate three - nucleon interaction terms induced due to the srg evolution . this can be done as described in ref . it is essential to further develop the ncsm / rgm formalism in order to handle three - nucleon interactions , both genuine and the srg - evolution induced , in the scattering calculations . numerical calculations have been performed at the llnl lc facilities and at the nic , jlich . prepared in part by llnl under contract de - ac52 - 07na27344 . support from the u. s. doe / sc / np ( work proposal no . scw0498 ) , llnl ldrd grant pls-09-erd-020 , and from the u. s. department of energy grant de - fc02 - 07er41457 is acknowledged . this work is supported in part by the deutsche forschungsgemeinschaft through contract sfb 634 and by the helmholtz international center for fair within the framework of the loewe program launched by the state of hesse . + r. b. wiringa , s. c. pieper , j. carlson , v. r. pandharipande , phys . c * 62 * , 014001 ( 2000 ) ; s. c. pieper and r. b. wiringa , ann . nucl . part . sci . * 51 * , 53 ( 2001 ) ; s. c. pieper , k. varga and r. b. wiringa , phys . c * 66 * , 044310 ( 2002 ) .

in detection statistics , the relation between the average number of counts during a detection time @xmath0 , @xmath1 , and the fluctuations around this value , expressed by the variance @xmath2-@xmath3 , i.e. the variance to mean ratio q^2 ~ is often used to characterise the statistics of the particle field detected . for instance in the case of neutrons emitted from a radioactive source following a simple poisson statistics , this ratio is obviously equal to unity . however , for a neutron chain in a multiplying medium , such as a subcritical reactor with a source or a fissile sample with an inherent neutron source due to spontaneous fission , the branching character represented by the fission process has the consequence that the individual detections will not be independent , rather positive correlations exist between them . hence the variance to mean ratio is larger than unity , and the deviation from unity carries information on the medium in which the branching process ( neutron multiplication ) took place . this fact was used by feynman and de hoffmann in 1944 - 1956 @xcite for the derivation of a formula for a branching process where the variance to mean was above unity , @xmath4 . the @xmath5-function became called the feynman y - function , characterising the deviation of the relative variance from unity . both its time dependence , expressed by the prompt neutron decay constant @xmath6 , as well as its asymptotic value , carry information on the sought parameters of the system . the original application of these studies was related to the theoretical description of statistical fluctuations of the number of neutrons in multiplying medium or in other words , to the determination of the level of subcriticality . therefore , the above - mentioned research remained classified for several years . the fundamental principles of the feynman - alpha theory have been extensively described in a number of publications , e.g. @xcite . about a decade ago the interest to this subject returned again in connection to the on - line measurement of subcritical reactivity of accelerator - driven systems ( ads ) . whereas the original feynman - alpha formulas referred to a homogeneous system in an monoenergetic ( `` one - group '' ) description @xcite , dealing only with one exponent or decay constant , the further experiments , e.g. the yalina @xcite and muse @xcite , showed the appearance of more than one decay constant and , therefore , the possible need of extension of the one - group one - region ( also referred to as `` one - point '' ) feynman - alpha formulas to more energy groups and spatial regions . several attempts were made towards the explanation of multiple exponential modes by the spatial effects @xcite . by that time it was decided that the future ads - systems will be driven by pulsed neutron and spallation sources which lead to the extension of the theory of variance - to - mean formulas for a continuous source with poisson statistics to the cases of pulsed and spallation neutron sources with different definition of the pulse shapes and pulsing manner @xcite . latter analysis @xcite showed the close link between the application of feynman - alpha formulas to subcriticality measurements and safeguards . in line with the above , the suggestion of new safeguards technique for mox / spent fuel assay @xcite , the differential die - away self - interrogation ( ddsi ) technique , displayed the interest towards the energy - dependent aspects of neutron counting . in connection with this , the two - group feynman - alpha theory was elaborated in @xcite where delayed neutrons were neglected , and in @xcite with inclusion of delayed neutron precursors . however , fast fission and thermal detections were neglected in both papers . the results of further considerations of the importance of the energy - aspect in evaluation of the real systems shows that `` a measured variance - to - mean ratio in fast systems may be contaminated by the energy - higher order mode effect except when the system is near - critical @xcite '' . in the light of recent advances in detector technologies in safeguards towards the development of fast neutron detection systems with scintillators , the knowledge of the energy - dependent behavior of neutron counting became a very important issue to be taken into account in feynman - alpha theory . the authors of @xcite showed that the short and long time behavior of the @xmath7-function can be used to assay the amount of @xmath8pu and the absolute amount of @xmath9pu+@xmath10pu in the reprocessed fuel . therefore , one part of this paper is devoted to the derivation of the general case of one - point two - group feynman - alpha formulas , when fast fission and thermal detections and delayed neutrons are included . however in some cases , for example , when the fission chambers are used as detectors , the energy importance makes way for the region - dependent aspect . this issue has not well been studied previously , although some expressions for the one - group two - region feynman - alpha formulas can be found in @xcite . however , even these investigations are limited to the case of delayed neutron precursors having been neglected and detections accounted for only in one region . thus , the second part of this paper is devoted to the derivation of the general case of the two - point one - group feynman - alpha formulas , when detections and delayed neutrons are accounted for in both regions . it has to be noted that the present paper does not carry out fully an analysis of the diagnostic value of the obtained formulas the same way as it was made in the traditional works based on a one - group treatment in a single ( infinite ) homogeneous medium . in the traditional case the time dependence of the feynman @xmath5 function is characterised essentially with one decay constant which can clearly be related to the subcriticality of the system . in the case of using two energy groups and two spatial regions , the number of decay constants increases and each of them becomes a much more involved function of the increased number of material properties ( reaction intensities ) that the treatment of different regions and energy intervals incurs . the sought system parameters become very involved functions of these decay constants , and no attempt is made in this paper on the investigation of how these parameters can be extracted from the measurements . this is deferred to later work . the objective of the present work is to give a clear and transparent derivation of the various variance - to - mean formulas as functions of the reaction and transition intensities , and to compare the solutions for the different cases . in this paper , the two - point two - group , the two - group one - point ( with delay neutrons ) and the one - group two - point ( with delayed neutrons ) feynman - alpha formulas were derived by using the kolmogorov forward approach @xcite . in the general model used for derivations we assume that the neutron population consists of two groups of neutrons : fast ( denoted as 1 ) and thermal ( denoted as 2 ) . fast and thermal neutrons can undergo different reactions ( i ) listed below : * absorption ( @xmath11 ) , * fission ( @xmath12 ) , * detection ( @xmath13 ) , * removal from the fast group to the thermal ( @xmath14 ) . unlike in the terminology , used in the traditional one - group treatments , absorbtion here stands only for capture . the decay constant of the delayed neutron precursors is given as @xmath15 . in addition , both the fast and thermal neutrons can transit from one region to the others , in both directions . in all models the source is considered as releasing n particles with probability _ @xmath16 _ at an emission event . in this paper a term _ `` two - point '' _ has the same meaning as _ `` two - region''_. for the two - group two - point model it was assumed that two adjacent infinite and homogeneous half - space regions ( denoted as a and b ) with different independent reaction intensities for absorption of fast and thermal neutrons ( @xmath17 , @xmath18 , @xmath19 , @xmath20 ) , fission induced by fast and thermal neutrons ( @xmath21 , @xmath22 , @xmath23 , @xmath24 ) and detection of fast and thermal neutrons ( @xmath25 , @xmath26 , @xmath27 , @xmath28 ) . the two regions are coupled by two passage intensities ( @xmath29 , @xmath30 , @xmath31 , @xmath32 ) in two different directions describes the intensity of particles ( group i ) leaving region a for region b and @xmath33 is the intensity of particles ( group i ) transferring to region a from region b. ] . thus , each of the reactions for the different groups of neutrons can be described by transition intensities , as shown in figure [ fig:1 ] . total intensities including both the reactions and transitions between the regions for the fast and the thermal neutrons are denoted as @xmath34 and @xmath35 , @xmath36 and @xmath37 for regions a and b , respectively : _ a1 = _ a1a + _ a1f + a2 = _ a2a + _ a2f + _ a2 t + _ a2d _ b1 = _ b1a+ _ b1f + _ b1 t + _ b1r + _ b1d _ b2=_b2a + _ b2f + _ b2 t + _ b2d the slowing down process , i.e. the removal of neutrons from the fast group to the thermal group is described by the removal reaction intensity @xmath38 . in the two - point two - group model we also include two extraneous compound poisson sources of fast neutrons placed in different regions , a and/or b , with intensities @xmath39 and @xmath40 . in the following , two special cases of the above general form will be described briefly . because in the lower dimensionality of the special cases , inclusion of delayed neutrons is possible . in the two - group one - point feynman - alpha model ( figure [ fig:25 ] ) , we assume that the medium is infinite and homogeneous . the neutron population consists of two groups of neutrons , fast and thermal . a compound poisson source of fast neutrons with emission intensity @xmath41 is included in the model . thus , the total transition intensities for the fast and thermal neutrons , denoted as @xmath42 and @xmath43 , are given as : _ 1 = _ 1a + _ 1f + _ r + _ 1d _ 2 = _ 2a + _ 2f + _ 2d the assumption behind the one - group two - point model is that the two adjacent homogeneous half - space regions ( denoted as a and b ) with independent reaction intensities for detection ( @xmath44 , @xmath45 ) , absorption @xmath46 and @xmath47 , and fission @xmath48 and @xmath49 are coupled by two passage intensities @xmath50 and @xmath51 in two different directions . the decay constants of delayed neutron precursors are given as @xmath52 and @xmath53 for regions a and b , as shown in figure [ fig:3 ] . thus , total transition intensities for region a and region b are denoted as @xmath54 and @xmath55 : _ a = _ aa + _ af + _ at + _ ad _ b = _ ba + _ bf + _ bt + _ bd in the model we include two compound poisson sources of fast neutrons in regions a and b with emission intensities @xmath56 and @xmath57 , respectively . the sources are considered as releasing @xmath58 particles in one emission with the probability distributions of @xmath59 and @xmath60 , respectively . for the induced fission reaction , we consider that @xmath61 neutrons and @xmath62 delayed neutron precursors are emitted with the probability distributions @xmath63 and @xmath64 for the fission reaction in region a and region b , respectively . based on the main concept and assumptions used the two - group two - region , the two - group one - region and the two - region one - group feynman - alpha formulas are elaborated as below . in order to derive the two - point two - group feynman - alpha theory let us assume that the source @xmath65 is switched on in the region a / b at the time @xmath66 , while the detection process is started at the fixed time instant @xmath67 , where @xmath68 and @xmath69 . let the random processes @xmath70 , @xmath71 , @xmath72 and @xmath73 represent the number of fast neutrons in region a , fast neutrons in region b , thermal neutrons in region a and thermal neutrons in region b at the time @xmath74 and @xmath75 , @xmath76 , @xmath77 , @xmath78 - the number of fast and thermal particle detections in the regions a and b in the time interval [ @xmath67 , @xmath79 , respectively . for convenience , we consider _ @xmath67=0_. thus , the joint probability of having @xmath80 fast neutrons in region a , @xmath81 fast neutrons in region b , @xmath82 thermal neutrons in region a , @xmath83 thermal neutrons in region b at time @xmath0 , @xmath84 fast neutrons have been detected in region a , and @xmath85 fast neutrons have been detected in region b , @xmath86 thermal neutrons have been detected in region a , and @xmath87 thermal neutrons have been detected in region b during the period of time @xmath88 can be defined as : @xmath89 . by summing up the probabilities of all mutually exclusive events of the particle not having or having a specific reaction within the infinitesimally small time interval d@xmath0 , one can write : = -(_a1n_a1+_a2n_a2+_b1n_b1+_b2n_b2+s_a+s_b)p(n_a1,n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ a1a(n_a1 + 1)p(n_a1 + 1,n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ a2a(n_a2 + 1)p(n_a1,n_a2 + 1,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ b1a(n_b1 + 1)p(n_a1,n_a2,n_b1 + 1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ b2a(n_b2 + 1)p(n_a1,n_a2,n_b1,n_b2 + 1,z_a1,z_b1,z_a2,z_b2,t ) + _ a1f_k^n_a1 + 1(n_a1 + 1-k)f_a1(k)p(n_a1 + 1-k , n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ b1f_k^n_b1 + 1(n_b1 + 1-k)f_b1(k)p(n_a1,n_a2,n_b1 + 1-k , n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ a2f_k^n_a1(n_a2 + 1)f_a2(k)p(n_a1-k , n_a2 + 1,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ b2f_k^n_b1(n_b2 + 1)f_b2(k)p(n_a1,n_a2,n_b1-k , n_b2 + 1,z_a1,z_b1,z_a2,z_b2,t ) + _ a1t(n_a1 + 1)p(n_a1 + 1,n_a2,n_b1 - 1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ b1t(n_b1 + 1)p(n_a1 - 1,n_a2,n_b1 + 1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ a2t(n_a2 + 1)p(n_a1,n_a2 + 1,n_b1,n_b2 - 1,z_a1,z_b1,z_a2,z_b2,t ) + _ b2t(n_b2 + 1)p(n_a1,n_a2 - 1,n_b1,n_b2 + 1,z_a1,z_b1,z_a2,z_b2,t ) + _ ar(n_a1 + 1)p(n_a1 + 1,n_a2 - 1,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + _ br(n_b1 + 1)p(n_a1,n_a2,n_b1 + 1,n_b2 - 1,z_a1,z_b1,z_a2,z_b2,t ) + _ a1d(n_a1 + 1)p(n_a1 + 1,n_a2,n_b1,n_b2,z_a1 - 1,z_b1,z_a2,z_b2,t ) + _ b1d(n_b1 + 1)p(n_a1,n_a2,n_b1 + 1,n_b2,z_a1,z_b1 - 1,z_a2,z_b2,t ) + _ a2d(n_a2 + 1)p(n_a1,n_a2 + 1,n_b1,n_b2,z_a1,z_b1,z_a2 - 1,z_b2,t ) + _ b2d(n_b2 + 1)p(n_a1,n_a2,n_b1,n_b2 + 1,z_a1,z_b1,z_a2,z_b2 - 1,t ) + s_a_n^n_a1p_a(n)p(n_a1-n , n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) + s_b_n^n_b1p_b(n)p(n_a1,n_a2,n_b1-n , n_b2,z_a1,z_b1,z_a2,z_b2,t ) with initial conditions p(n_a1,n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t = t_0t_0 ) = _ n_a1,0 _ n_a2,0 _ n_b1,0 _ n_b2,0 _ z_a1,0 _ z_a2,0 _ z_b1,0 _ z_b2,0 and _ n_a1 _ n_a2 _ n_b1 _ n_b2 p(n_a1,n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t = t_dt_0 ) = _ z_a1,0 _ z_a2,0 _ z_b1,0 _ z_b2,0 and @xmath90 is the number distribution of neutrons in a fission of type _ i_. this equation can be solved by using the generating function technique in the way similar to as described in @xcite . by defining the following generating function for the probability distribution @xmath91 : g(x_a , y_a , x_b , y_b , m , n , o , p , t)= _ n_a1 _ n_a2 _ n_b1 _ n_b2 _ z_a1 _ z_a2 _ z_b1 _ z_b2 x_a^n_a1 y_a^n_a2 x_b^n_b1 y_b^n_b2 m^z_a1 n^z_a2 o^z_b1 p^z_b2 * p(n_a1,n_a2,n_b1,n_b2,z_a1,z_b1,z_a2,z_b2,t ) with initial condition for @xmath92 g(x_a , y_a , x_b , y_b , m , n , o , p , t = t_0t_0 ) = 1 and g(1,1,1,1,m , n , o , p , t = t_dt_0 ) = 1 the following partial differential equation is obtained : = [ _ a1a + _ ary_a + _ a1tx_b + _ a1 dm - _ a1x_a + q_a1(x_a ) _ a1f ] + [ _ a2a + _ a2ty_b + _ a2dn - _ a2y_a + q_a2(x_a ) _ a2f ] + [ _ b1a + _ bry_b + _ b1tx_a + _ b1do - _ b1x_b + q_b1(x_b ) _ b1f ] + [ _ b2a + _ b2ty_a + _ b2dp - _ b2y_b + q_b2(x_b ) _ b2f ] + s_a [ r_a(x_a ) - 1 ] g + s_b [ r_b(x_b ) - 1 ] g , where q_i ( x ) = _ k x^k f_if(k ) n p_q(n ) x^n for the sake of simplicity , some identities are used in the solution as below ( @xmath93 ) : .r ( x)|_x=1= _ n n p_q(n ) = r .r ( x)|_x=1= _ n n(n-1 ) p_q(n ) = r thus , @xmath94 ( @xmath95 ) , @xmath96 ( @xmath97 ) , @xmath98 ( @xmath99 ) , @xmath100 ( @xmath101 ) and @xmath102 , @xmath103 , @xmath104 , @xmath105 stand for the first and second factorial moments of the number of neutrons emitted in a fission process and in a source event , respectively . the index @xmath106 denotes fission induced by fast or thermal neutrons , respectively . in a steady subcritical medium with a steady source , when @xmath107 , the following stationary solutions for the neutron populations @xmath108 , @xmath109 , @xmath110 , @xmath111 , @xmath112 , @xmath113 , @xmath114 , @xmath115 are obtained as below : r_a ( _ a2 t _ ( -_b1+_b1f _ b1 ) + _ a2 ( _ b1 _ b2-_b1f _ b2 _ b1-_b2f _ br _ b2 ) ) ) r_b ( _ a1 ( _ a2 _ b2-_a2 t _ b2t)+_a1f ( -_a2 _ b2+_a2 t _ b2t)_a1-_a2f _ ar _ b2 _ a2 ) ) ) ) + s_b r_b ( _ ar _ b1 t _ b2+_b2 t _ br ( _ a1-_a1f_a1 ) ) ) _ ar _ b1t+_br ( _ a1 _ a2-_a1f _ a2_a1-_a2f _ ar _ a2 ) ) ) by introducing the modified second factorial moment of the random variables a and b as follows @xmath116=@xmath117 - @xmath118,@xmath119 and then taking cross- and auto - derivatives , the following system of differential equations of modified second factorial moments ( @xmath120 , @xmath121 , @xmath122 , @xmath123 , @xmath124 , @xmath125 , @xmath126 ) for the neutron population are obtained as below : _ x_a x_a = 2 _ b1 t _ x_a x_b+ 2 _ a2f _ a1-_a1)_x_a x_a + s_a r_a+_a2f _ a2 |n_a2 + _ a1f _ a1|n_a1 _ x_a y_a = _ b1 t _ y_a x_b+_a2f_a2 _ y_a y_a+_b2 t _ x_a y_b+(_a1f _ a1 -_a1-_a2 ) _ x_a y_a+_ar _ x_a x_a _ x_a x_b = _ b1 t _ x_b x_b+_a2f_a2 _ y_a x_b+_b2f _ b2 _ x_a y_b + ( _ a1f _ a1 -_a1 ) _ x_a x_b+(_b1f _ b1-_b1 ) _ x_a x_b+_a1 t _ x_a y_b = _ b1 t _ x_b y_b+_a2f_a2 _ y_a y_b+(_a1f _ a1 -_a1-_b2 ) _ x_a y_b+_br _ x_a x_b+_a2 t _ x_a y_a _ x_b x_b = 2_b2f _ b2_x_b y_b+2(_b1f _ b1-_b1 ) _ x_b x_b+2 _ a1 t _ x_a x_b + s_b r_b+_b2f _ b2|n_b2+_b1f _ b1 |n_b1 _ y_a x_b = _ _ x_b y_b+_b2f _ b2 _ y_a y_b+(_b1f _ b1-_b1-_a2 ) _ y_a x_b+_ar _ x_a x_b+_a1 t _ x_a y_a _ x_b y_b = _ b2f _ b2 _ y_b y_b+(_b1f _ b1-_b1-_b2 ) _ x_b y_b+_br _ x_b x_b+_a2 t _ y_a x_b+_a1 t _ x_a y_b _ y_a y_a = 2 _ b2 t _ y_a y_b-2 _ a2 _ y_a y_a+2 _ ar _ x_a y_a _ y_a y_b = _ b2t_y_b y_b-_a2 _ y_a y_b-_b2 _ y_a y_b+_br _ y_a x_b+_a2 t _ y_a y_a+_ar _ x_a y_b _ y_b y_b = -2 _ b2_y_b y_b+2 _ br _ x_b y_b+2 _ a2 t _ y_a y_b this system can be solved in the stationary state ( when left hand sides are equal to 0 ) . the final expression of two - point two - group feynman - alpha formulas for fast detections is given as below : = 1+y(t)=1 + _ i=1 ^ 4 y_i ( 1 - ) the four roots , namely @xmath127 , @xmath128 , @xmath129 and @xmath130 can be obtained by solving the forth order characteristic equation in @xmath131 with known coefficients a , b , c , d , obtained from the temporal laplace transform of the time - dependent equations for @xmath132 etc . : ^4 + a^3+b^2+c+d = 0 where a= _ a1+_a2+_b1+_b2-_a1f _ a1-_b1f _ b1 b= -_a1 t _ b1t+_a2 _ b2-_a2 t _ b2t-_a2 ( _ a1f _ a1-_a1)-_b2 ( _ a1f _ a1-_a1)-_a2 ( _ b1f _ b1-_b1)-_b2 ( _ b1f _ b1-_b1)+(_a1f _ a1-_a1 ) ( _ b1f _ b1-_b1)-_a2f _ ar _ -_b2f _ br _ b2 c= -_a1 t _ b1t-_a1 t _ b2-_a2 _ b2 ( _ a1f _ a1-_a1)+_a2 t _ b2 t ( _ a1f _ a1-_a1)-_a2 _ b2 ( _ b1f _ b1-_b1)+_a2 t _ b2 t ( _ b1f _ b1-_b1)+_a2 ( _ a1f _ a1-_a1 ) ( _ b1f _ b1-_b1)+_b2 ( _ a1f _ a1-_a1 ) ( _ b1f _ b1-_b1)-_a2f _ ar a2 + _ a2f _ ar ( _ b1f _ b1-_b1 ) _ a2-_a2 _ b2f _ br _ b2 + _ b2f _ br ( _ a1f _ a1-_a1 ) _ b2 d= -_a1 t _ a2 _ b1 t _ b2+_a1 t _ a2 t _ b2t+_a2 _ b2 ( _ a1f _ a1-_a1 ) ( _ b1f _ b1-_b1)-_a2 t _ b2 t ( _ a1f _ a1-_a1 ) ( _ b1f _ b1-_b1)-_a1 t _ a2f _ b2 t _ br _ b2 ( _ b1f _ b1-_b1 ) _ a2 -_a2 t _ ar _ b1 t _ b2f _ b2+_a2 _ b2f _ br ( _ a1f _ a1-_a1 ) _ b2 + _ a2f _ ar _ b2f _ br _ a2 _ b2=_1 _ 2 _ 3 _ 4 if detection of fast neutrons is performed in region a , then the functions @xmath133 , @xmath134 , @xmath135 and @xmath136 should be used in the form : -y_1 = -y_2 = - y_3 = - y_4 = where , k_3 = _ x_a x_a k_2=_a2f _ x_a y_a q_a2(x_a)+_a2 _ x_a x_a-_b1f _ x_a x_a q_b1(x_b)+_b1 _ x_a x_a+_b1 t _ x_a x_b+_b2 _ x_a x_a k_1 = _ a2f _ b1f _ x_a y_a q_a2(x_a ) q_b1(x_b)+_a2f _ b1 _ x_a y_a q_a2(x_a)+_a2f _ b2 _ x_a y_a q_a2(x_a)+_a2f _ b2 t _ x_a y_b q_a2(x_a)-_a2 _ b1f _ x_a x_a q_b1(x_b)+_a2 _ b1 _ x_a x_a+_a2 _ b1 t _ x_a x_b+_a2 _ b2 _ x_a x_a-_a2 t _ b2 t _ x_a x_a-_b1f _ b2 _ x_a x_a q_b1(x_b)+_b1 t y_b q_b2(x_b)-_b2f _ br _ x_a x_a q_b2(x_b)+_b1 _ b2 _ x_a x_a+_b1 t _ b2 _ x_a x_b k_0= _ a2f _ b1f _ b2 _ x_a y_a q_a2(x_a ) q_b1(x_b)-_a2f _ b1f _ b2 t _ x_a y_b q_a2(x_a ) q_b1(x_b)-_a2f _ b2f _ br _ x_a y_a q_a2(x_a ) q_b2(x_b)+_a2f _ b1 _ b2 _ x_a y_a q_a2(x_a)+_a2f _ b1 _ b2 t _ x_a y_b q_a2(x_a)+_a2f _ b2 t _ br _ x_a x_b q_a2(x_a)-_a2 _ b1f _ b2 _ x_a x_a q_b1(x_b)+_a2 _ b1 t _ b2f _ x_a y_b q_b2(x_b)-_a2 _ b2f _ br _ x_a x_a q_b2(x_b)+_a2 _ b1 _ b2 _ x_a x_a+_a2 _ b1 t _ b2 _ x_a x_b+_a2 t _ b1f _ b2 t _ x_a q_b1(x_b)+_a2 t _ x_a y_a q_b2(x_b)-_a2 t _ b1 _ b2 t _ x_a x_a-_a2 t _ x_b it can be shown that : y_0 = y_1+y_2 + y_3 + y_4= if a thermal neutron detector is placed in region a , then the following @xmath133 , @xmath134 , @xmath135 and @xmath136 functions are to be used : -y_1= -y_2 = - y_3 = - y_4 = where , l_3 = _ y_a y_a l_2=-_a1f _ y_a y_a q_a1(x_a)+_a1 _ y_a y_a+_ar _ x_a y_a-_b1f _ y_a y_a q_b1(x_b)+_b1 _ y_a y_a+_b2 _ y_a y_a+_b2 t _ y_b y_a l_1 = _ a1f _ b1f _ y_a y_a q_a1(x_a ) q_b1(x_b)-_a1f _ b1 _ y_a y_a q_a1(x_a)-_a1f _ b2 _ y_a y_a q_a1(x_a)-_a1f y_b y_a q_a1(x_a)-_a1 _ b1f _ y_a y_a q_b1(x_b)+_a1 _ b1 _ y_a y_a+_a1 _ b2 _ y_a y_a+_a1 _ b2 t _ y_b y_a-_a1 t _ x_a y_a q_b1(x_b)+_ar _ b1 _ x_a y_a+_ar _ b1 t _ x_b y_a+_ar _ b2 _ x_a y_a-_b1f _ b2 _ y_a y_a q_b1(x_b)-_b1f _ b2 t _ y_b y_a q_b1(x_b)-_b2f _ br _ y_a y_a q_b2(x_b)+_b1 _ b2 _ y_a y_a+_b1 _ b2 t _ y_b y_a+_b2 t _ br _ x_b y_a l_0= _ a1f _ b1f _ b2 _ y_a y_a q_a1(x_a ) q_b1(x_b)+_a1f _ b1f _ b2 t _ y_b y_a q_a1(x_a ) q_b1(x_b)+_a1f _ b2f _ br _ y_a y_a q_a1(x_a ) q_b2(x_b)-_a1f _ b1 _ b2 _ y_a y_a q_a1(x_a)-_a1f _ b1 _ b2 t _ y_b y_a q_a1(x_a)-_a1f _ b2 t _ br _ x_b y_a q_a1(x_a)-_a1 _ b1f _ b2 _ y_a y_a q_b1(x_b)-_a1 _ b1f _ b2 t _ y_b y_a q_b1(x_b)-_a1 y_a y_a q_b2(x_b)+_a1 _ b1 _ b2 _ y_a y_a+_a1 _ b1 _ b2 t _ y_b y_a+_a1 _ b2 t _ br _ y_a-_a1 t _ b1 t _ b2 _ y_a y_a-_a1 t _ b1 t _ b2 t _ y_b y_a+_a1 t _ _ x_a y_a-_ar _ b1f _ b2 _ x_a y_a q_b1(x_b)+_ar _ y_b y_a q_b2(x_b)-_ar x_a y_a q_b2(x_b)+_ar _ b1 _ b2 _ x_a y_a+_ar _ b1 t _ b2 _ x_b y_a it can be shown that : y_0=y_1+y_2+y_3+y_4= for the case when a fast neutron detector is placed in region b , the following @xmath133 , @xmath134 , @xmath135 and @xmath136 functions should be used : -y_1= -y_2 = - y_3 = - y_4 = where , m_3 = _ x_b x_b m_2=-_a1f _ x_b x_b q_a1(x_a)+_a1 _ x_b x_b+_a1 t _ x_a x_b+_a2 _ x_b x_b+_b2f _ x_b y_b q_b2(x_b)+_b2 _ x_b x_b m_1 = -_a1f _ a2 _ x_b x_b q_a1(x_a)-_a1f _ b2f _ x_b y_b q_a1(x_a ) q_b2(x_b)-_a1f _ b2 _ x_b x_b q_a1(x_a)+_a1 t _ a2f _ x_b y_a q_a2(x_a)-_a2f _ ar _ x_b x_b q_a2(x_a)+_a1 _ a2 _ x_b x_b+_a1 _ b2f _ x_b y_b q_b2(x_b)+_a1 _ b2 _ x_b x_b+_a1 t _ a2 _ x_a x_b+_a1 t _ b2 _ x_a x_b+_a2 _ b2f _ x_b y_b q_b2(x_b)+_a2 _ b2 _ x_b x_b+_a2 t _ b2f _ x_b y_a q_b2(x_b)-_a2 t _ x_b m_0=-_a1f _ a2 _ b2f _ x_b y_b q_a1(x_a ) q_b2(x_b)-_a1f _ a2 _ b2 _ x_b x_b q_a1(x_a)-_a1f _ a2 t x_b y_a q_a1(x_a ) q_b2(x_b)+_a1f _ a2 t _ x_b q_a1(x_a)+_a1 t _ a2f _ x_b y_a q_a2(x_a)+_a1 t _ y_b q_a2(x_a)-_a2f _ ar _ b2f _ x_b y_b q_a2(x_a ) q_b2(x_b)-_a2f _ ar _ b2 _ x_b x_b q_a2(x_a)+_a1 _ a2 _ x_b y_b q_b2(x_b)+_a1 _ a2 _ b2 _ x_b x_b+_a1 _ x_b y_a q_b2(x_b)-_a1 _ a2 t _ b2 t x_b+_a1 t _ x_a x_b-_a1 t _ x_b+_a2 t _ ar _ b2f _ x_a x_b q_b2(x_b ) it can be shown that : y_0=y_1+y_2+y_3+y_4= if a thermal neutron detector is placed in region b , the following @xmath133 , @xmath134 , @xmath135 and @xmath136 functions should be used : -y_1= -y_2 = - y_3 = - y_4 = where , p_3 = _ y_b y_b p_2=-_a1f _ y_b y_b q_a1(x_a)+_a1 _ y_b y_b+_a2 _ y_b y_b+_a2 t _ y_b y_a-_b1f _ y_b y_b q_b1(x_b)+_b1 _ y_b y_b+_br _ x_b y_b p_1 = -_a1f _ a2 _ y_b y_b q_a1(x_a)-_a1f _ a2 t _ y_b y_a q_a1(x_a)+_a1f _ b1f _ y_b y_b q_a1(x_a ) q_b1(x_b)-_a1f _ b1 _ y_b y_b q_a1(x_a)-_a1f _ br _ x_b y_b q_a1(x_a)-_a2f _ ar _ y_b y_b q_a2(x_a)+_a1 _ a2 _ y_b y_b+_a1 _ a2 t _ y_b y_a-_a1 _ b1f _ y_b y_b q_b1(x_b)+_a1 _ b1 _ y_b y_b+_a1 _ y_b-_a1 t _ y_b y_b+_a1 t _ br _ x_a y_b-_a2 _ b1f _ y_b y_b q_b1(x_b)+_a2 _ b1 _ y_b y_b+_a2 _ y_b+_a2 t _ ar _ y_b-_a2 t _ y_b y_a q_b1(x_b)+_a2 t _ b1 _ y_b y_a p_0=_a1f _ a2 _ b1f _ y_b y_b q_a1(x_a ) q_b1(x_b)-_a1f _ a2 _ b1 _ y_b y_b q_a1(x_a)-_a1f _ a2 _ br _ x_b y_b q_a1(x_a)+_a1f _ a2 t _ b1f _ y_b y_a q_a1(x_a ) q_b1(x_b)-_a1f _ a2 t _ b1 _ y_b y_a q_a1(x_a)+_a1 t _ a2f _ y_b y_a q_a2(x_a)+_a2f _ ar _ b1f _ y_b y_b q_a2(x_a ) q_b1(x_b)-_a2f _ ar _ b1 _ y_b y_b q_a2(x_a)-_a2f _ ar _ br _ x_b y_b q_a2(x_a)-_a1 _ a2 _ b1f _ y_b y_b q_b1(x_b)+_a1 _ a2 _ b1 _ y_b y_b+_a1 _ a2 _ x_b y_b-_a1 _ y_b y_a q_b1(x_b)+_a1 _ a2 t _ b1 _ y_b y_a-_a1 t _ a2 _ b1 t _ y_b y_b+_a1 t _ a2 _ y_b-_a1 t _ y_b y_a-_a2 t _ ar _ b1f _ x_a y_b q_b1(x_b)+_a2 t _ ar _ b1 _ x_a y_b+_a2 t _ ar _ b1 t _ y_b it can be shown that : y_0=y_1+y_2+y_3+y_4=. quantitative examples of the feynman y(t ) function will be given shortly . in order to derive the two - group one - point feynman - alpha theory let us assume that the source s is switched on at the time @xmath66 , while the detection process is started at the fixed time instant @xmath67 , where @xmath68 and @xmath69 . for convenience , we consider _ @xmath67=0_. let the random processes @xmath137 , @xmath138 and @xmath139 represent the number of fast neutrons , thermal neutrons and delayed neutron precursors at the time @xmath74 , and @xmath140 , @xmath141 - the number of fast and thermal particle detections in the time interval _ [ @xmath67 , t ] _ , respectively . thus , the joint probability of having @xmath142 fast neutrons , @xmath143 thermal neutrons and c delayed neutron precursors present in system at time @xmath0 , and that @xmath144 fast neutrons and @xmath145 thermal neutrons have been detected during the period @xmath88 can be defined as @xmath146 . by summing up the probabilities of the mutually exclusive events of the particle not having or having a specific reaction or that there is a source emission within the infinitesimally small time interval d@xmath0 , one can write : = -(_1n_1+_2n_2+c+s_1)p(n_1,n_2,c , z_1,z_2,t ) + _ 1a(n_1 + 1)p(n_1 + 1,n_2,c , z_1,z_2,t ) + _ 2a(n_2 + 1)p(n_1,n_2 + 1,c , z_1,z_2,t ) + _ 1f_k^n_1 + 1_l^c(n_1 + 1-k)f_1f(k , l)p(n_1 + 1-k , n_2,c - l , z_1,z_2,t ) + _ 2f_k^n_1_l^c(n_2 + 1)f_2f(k , l)p(n_1-k , n_2 + 1,c - l , z_1,z_2,t ) + _ r(n_1 + 1)p(n_1 + 1,n_2 - 1,c , z_1,z_2,t ) + _ 1d(n_1 + 1)p(n_1 + 1,n_2,c , z_1 - 1,z_2,t ) + _ 2d(n_2 + 1)p(n_1,n_2 + 1,c , z_1,z_2 - 1,t ) + ( c+1)p(n_1 - 1,n_2,c+1,z_1,z_2,t ) + s_1_n^n_1p_q(n)p(n_1-n , n_2,c , z_1,z_2,t ) with initial condition p(n_1,n_2,c , z_1,z_2,t = t_0t_0 ) = _ n_1,0 _ n_2,0 _ c,0 _ z_1,0 _ z_2,0 and _ n_1 _ n_2 _ c p(n_1,n_2,c , z_1,z_2,t = t_dt_0 ) = _ z_1,0 _ z_2,0 by defining the following generating function for the probability distribution @xmath147 : g(x , y , v , m , n , t)= _ n_1 _ n_2 _ c _ z_1_z_2 x^n_1 y^n_2 v^c m^z_1 n^z_2 p(n_1,n_2,c , z_1,z_2,t ) with initial condition for @xmath92 g(x , y , v , m , n , t = t_0t_0 ) = 1 and g(1,1,1,m , n , t = t_dt_0 ) = 1 the following partial differential equation is obtained : = [ _ 1a+ _ ry + q_1(x , v ) _ 1f + _ 1d m - _ 1 x ] + [ _ 2a + q_2(x , v)_2f + _ 2d n - _ 2 y ] + ( x - v)+ s_1 [ r(x ) - 1 ] g , where q_1 ( x , v ) = _ k _ v^l f_1f(k , l ) q_2 ( x , v ) = _ k _ v^l f_2f(k , l ) r(x ) = _ n p_q(n ) x^n here , @xmath148__(k , l ) _ _ is the probability of having _ k _ prompt neutrons and _ l _ delay neutron precursors produced in a fission event induced by a fast neutron , @xmath149__(k , l ) _ _ is the probability of having _ k _ prompt neutrons and _ l _ delay neutron precursors produced in a fission event induced by a thermal neutron . the effective delayed neutron fraction is @xmath150 , @xmath151 and @xmath152 are the average total number of neutrons per fast and thermal induced fission , respectively . for the sake of simplicity , some identities are used in the solution as below ( @xmath93 ) : .q_i ( x , v)|_x=1,v=1= _ k _ l k f_if(k , l ) = ( 1-)_i .q_i ( x , v)|_x=1,v=1= _ k _ l l f_if(k , l ) = _ i and .r ( x)|_x=1= _ n n p_q(n ) = r .r ( x)|_x=1= _ n n(n-1 ) p_q(n ) = r in a steady subcritical medium with a steady source , when @xmath107 , the following stationary solutions for the neutron populations @xmath153 , @xmath154 and @xmath155 , and detection counts @xmath144 and @xmath145 are obtained as below : by introducing the modified second factorial moment of the random variables a and b as follows @xmath116= @xmath117 - @xmath118 , @xmath119 and then taking cross- and auto - derivatives , the following system of differential equations of modified second factorial moments for the neutron population are obtained as below : _ x x = s_1 r + _ 2f _ 2pp |n_2 + _ 1f _ 1pp |n_1 + 2 _ x v+ 2 _ x x + 2 ( 1- ) _ 2f _ 2 _ x y _ x y = _ y v+(1- ) _ 2f _ 2 _ y y+_x y+_r_x x _ y y=-2 _ 2_yy + 2 _ r _ xy _ x v=_vv+_2f _ 2pd |n_2 + ( 1- ) _ 2f _ 2 _ yv+_1f_1pd |n_1 + _ xv+_2f _ 2 _ xy+_1f _ 1 _ xx _ y v = ( --_2 ) _ yv + _ 2f _ 2 _ yy+_r _ xv+_1f _ 1 _ xy _ vv= -2 _ vv + _ 2f _ 2dd|n_2 + 2 _ 2f _ 2 _ yv+_1f _ 1dd |n_1 + 2 _ 1f _ 1 _ xv the three coefficients @xmath127 , @xmath128 and @xmath129 can be obtained by solving the third order equation in @xmath131 with known constant coefficients a , b , c : ^3 + a^2+b+c = 0 where a= _ 1 _ 1f-_1 _ 1f++_1+_2=-(_1 + _ 2+_3 ) b= _ 2 _ 1 _ 1f-_1 _ 1f-_2 _ 1 _ 1f+_2 _ 2f _ r-_2 _ 2f _ r+_1+_2 _ 1+_2 c=-_2 _ 1 _ 1f-_2 _ 2f _ r+_1 _ 2=-_1 _ 2_3 the stationary modified variance of the fast particle detections can be obtained from the coupled equation system by using the laplace transform technique : _ x m = _ vm+(1- ) _ 2f _ 2 _ ym+(-_1+(1- ) _ 1f _ 1 ) _ xm+_1d _ xx _ y m= -_2 _ ym+_r _ xm+_1d _ xy _ v m = -_vm+_2f _ 2 _ ym+_1d _ xv+_1f _ 1 _ xm _ mm = 2 _ 1d_xm the same can be done to define the stationary modified variance of the thermal particle detections via solving the following coupled equation system : _ x n = _ vn+(1- ) _ 2f _ 2 _ yn+(-_1+(1- ) _ 1f _ 1 ) _ xn+_2d _ n = -_2 _ yn+_r _ xn+_2d _ yy _ v n = -_vn+_2f _ 2 _ yn+_2d _ yv+_1f _ 1 _ xn _ nn = 2 _ 2d_yn some second moment notations were introduced as follows : .q_i ( x , v)|_x=1,v=1= _ k _ l k(k-1 ) f_if(k , l ) = _ ipp .q_i ( x , v)|_x=1,v=1= _ k _ l l(l-1 ) f_if(k , l ) = _ idd .q_i ( x , v)|_x=1,v=1= _ k _ l kl f_if(k , l ) = _ ipd in which @xmath93 . thus , the solution for the two - group one - point feynman - alpha formula for fast and thermal detection particles can be written as below : = 1 + y(t)=1 + _ i=1 ^ 3 y_i ( 1 - ) . for fast particle detections the following expressions should be used : -y_1=- -y_2 = -y_3 = with k_2=_xx k_1=-_2f _ 2 _ xy+_2f _ 2 _ xy+_xv+_xx+_2 _ xx k_0=_2f _ 2 _ xy+_2 _ xv+_2 _ xx it can be shown that : y_0 = y_1+y_2 + y_3= if a thermal neutron detector is used , then the following expressions should be considered : -y_1=- -y_2 = -y_3 = with l_2=_yy l_1=_1f _ 1 _ yy-_1f _ 1 _ yy+_r _ xy+_yy+_1 _ yy l_0=-_1f _ 1 _ yy+_r _ xy+_r _ yv+_1 _ yy it can be shown that : y_0 = y_1+y_2 + y_3= similarly as in the derivation of two - group one - point version of feynman - alpha formula , in the one - group two - point feynman - alpha theory the joint probability of having @xmath156 neutrons in region a , @xmath157 neutrons in region b , @xmath158 delayed neutron precursors presented in region a , @xmath159 delayed neutron precursors presented in region b at time @xmath0 , @xmath160 neutrons have been detected in region a , and @xmath161 neutrons have been detected in region b in the system during the period of time @xmath88 can be defined as @xmath162.by summing up all mutually exclusive events of the particle not having or having a specific reaction within the infinitesimally small time interval d@xmath0 , it can be written : = -(_an_a+_bn_b+_acc_a+_bcc_b+s_a+s_b)p(n_a , n_b , c_a , c_b , z_a , z_b , t ) + _ aa(n_a+1)p(n_a+1,n_b , c_a , c_b , z_a , z_b , t ) + _ ba(n_b+1)p(n_a , n_b+1,c_a , c_b , z_a , z_b , t ) + _ af_k^n_a+1_l^c_a(n_a+1-k)f_a(k , l)p(n_a+1-k , n_b , c_a - l , c_b , z_a , z_b , t ) + _ bf_k^n_b+1_l^c_b(n_b+1-k)f_b(k , l)p(n_a , n_b+1-k , c_a , c_b - l , z_a , z_b , t ) + _ at(n_a+1)p(n_a+1,n_b-1,c_a , c_b , z_a , z_b , t ) + _ bt(n_b+1)p(n_a-1,n_b+1,c_a , c_b , z_a , z_b , t ) + _ ac(c_a+1)p(n_a-1,n_b , c_a+1,c_b , z_a , z_b , t ) + _ bc(c_b+1)p(n_a , n_b-1,c_a , c_b+1,z_a , z_b , t ) + _ ad(n_a+1)p(n_a+1,n_b , c_a , c_b , z_a-1,z_b , t ) + _ bd(n_b+1)p(n_a , n_b+1,c_a , c_b , z_a , z_b-1,t ) + s_a_n^n_ap_a(n)p(n_a - n , n_b , c_a , c_b , z_a , z_b , t ) + s_b_n^n_bp_b(n)p(n_a , n_b - n , c_a , c_b , z_a , z_b , t ) with initial condition p(n_a , n_b , c_a , c_b , z_a , z_b , t = t_0t_0 ) = _ n_a,0 _ n_b,0 _ c_a,0 _ c_b,0 _ z_a,0 _ z_b,0 and _ n_a _ n_b _ c_a_c_b p(n_a , n_b , c_a , c_b , z_a , z_b , t = t_dt_0 ) = _ z_a,0 _ z_b,0 by defining the following generating function for the probability distribution @xmath163 : g(x , y , v , w , m , n , t ) = _ n_a _ n_b _ c_a_c_b _ z_a_z_b x^n_a y^n_b v^c_aw^c_b m^z_a n^z_bp(n_a , n_b , c_a , c_b , z_a , z_b , t ) with initial condition for @xmath92 g(x , y , v , w , m , n , t = t_0t_0 ) = 1 and g(1,1,1,1,m , n , t = t_dt_0 ) = 1 a partial differential equation in the variables @xmath164 in terms of generating function can be obtained : = [ _ aa+_aty + q_a(x , v ) _ af + _ ad m - _ a x ] + [ _ ba + _ btx + q_b(y , w)_bf + _ bd n - _ b y ] + _ ac ( x - v ) + _ bc ( y - w ) + s_a [ r_a(x ) - 1 ] g + s_b [ r_b(y ) - 1]g where q_a ( x , v ) = _ v^l f_a(k , l ) q_b ( y , w ) = _ k _ w^l f_b(k , l ) r_a(x)=_n p_a(n ) x^n r_b(y ) = _ n p_b(n ) y^n here , @xmath165 and @xmath166 are the effective delayed neutron fractions in region a and region b , respectively . for the sake of simplicity , some identities are used in the solution as below ( @xmath167 ) : .q_a ( x , v)|_x=1,v=1= _ k _ l k f_af(k , l ) = ( 1-_a)_a .q_a ( x , v)|_x=1,v=1= _ k _ l l f_af(k , l ) = _ a_a .q_b ( y , w)|_y=1,w=1= _ k _ l k f_bf(k , l ) = ( 1-_b)_b .q_b ( y , w)|_y=1,w=1= _ k _ l l f_bf(k , l ) = _ b_b and .r_a ( x)|_x=1= _ n n p_a(n ) = r_a .r_a ( x)|_x=1= _ n n(n-1 ) p_a(n ) = r_a .r_b ( y)|_y=1= _ n n p_b(n ) = r_b .r_b ( y)|_y=1=_n n(n-1 ) p_b(n ) = r_b in a steady subcritical medium with a steady source , when @xmath107 , the following stationary solutions for the neutron population and detection counts are obtained as follows : by introducing the modified second factorial moments and then taking cross- and auto - derivatives , the following system of differential equations of modified second factorial moments for the neutron population are obtained as below : _ x x = 2 _ xx + 2 _ bt _ xy + 2 _ ac_xv+ s_a r_a+ _ af _ app|n_a _ x y = _ ac_yv+ _ bc_xw+ _ _ y y = 2 _ yy + 2 _ at _ xy + 2 _ bc_yw+ s_b r_b+ _ bf _ bpp |n_b _ xv= ( ( 1-_a ) _ a _ af - _ a - _ ac ) _ xv+_a _ a _ af _ xx+ _ ac _ vv + _ bt _ yv+ _ af _ apd |n_a _ yv = _ bc _ vw+_yv + _ at _ xv+_a _ af _ a _ xy _ vv = -2 _ ac_vv+2 _ a _ af _ a _ xv + _ af _ add |n_a _ xw= _ ac _ vw+ _ bt _ yw+_xw + _ b _ bf _ b _ xy _ yw= _ bc_ww + _ yw + _ b _ bf _ b _ yy + _ at_xw + _ bf _ bpd |n_b _ vw= ( - _ ac- _ bc ) _ vw + _ b _ bf _ b _ yv+_a _ af _ a _ xw _ ww= -2 _ bc _ ww + 2 _ b _ bf _ b _ yw + _ bf _ bdd |n_b where .q_a ( x , v)|_x=1,v=1= _ k _ l k(k-1 ) f_af(k , l ) = _ app .q_b ( y , w)|_y=1,w=1= _ k _ m k(k-1 ) f_bf(k , m ) = _ bpp .q_a ( x , v)|_x=1,v=1= _ k _ l l(l-1 ) f_af(k , l ) = _ add .q_b ( y , w)|_y=1,w=1= _ k _ l l(l-1 ) f_bf(k , l ) = _ bdd .q_a ( x , v)|_x=1,v=1= _ k _ l kl f_af(k , l ) = _ apd .q_b ( y , w)|_y=1,w=1= _ k _ l kl f_bf(k , l ) = _ bpd the system above is solved for stationary case when @xmath168 . four roots @xmath127 , @xmath128 , @xmath129 and @xmath130 can be obtained by solving the forth order equation with coefficients a , b , c , d specified as below : ^4 + a^3+b^2+c+d = 0 a=_a _ af _ a-_af _ a+_a+_ac+_b _ bf _ b-_bf _ b+_b+_bc b=-_ac _ af _ a+_a _ ac+_a _ af _ b _ a-_a _ af _ bf _ a _ b+_a _ af _ b _ bf _ a _ b-_af _ b _ bf _ a _ b+_af _ bf _ a _ b-_af _ b _ a+_a _ af _ bc bc _ a+_a _ b _ bf _ b-_a _ bf _ b+_a _ b+_a _ bc+_ac _ b _ bf _ b-_ac _ bf _ b+_ac _ b+_ac _ bc-_at _ bt-_bc _ bf _ b+_b _ bc c=-_ac _ af _ b _ bf _ a _ b+_ac _ af _ bf _ a _ b-_ac _ af _ b _ a-_ac _ af _ bc _ a+_a _ ac _ b _ bf _ b-_a _ ac _ bf _ b+_a _ ac _ b+_a _ ac _ bc+_a _ af _ b _ bc _ a-_a _ af _ bc _ bf _ a _ b+_af _ bc _ bf _ a _ b-_af _ b _ bc _ a-_a _ bf _ b+_a _ b _ bc-_ac _ at _ bt-_ac _ bc _ bf _ b+_ac _ b _ bc-_at _ bc _ bt d=_ac _ af _ bc _ bf _ a _ b-_ac _ af _ b _ bc _ a-_a _ ac _ bc _ bf _ b+_a _ ac _ b _ bc-_ac _ at _ bc _ bt the stationary modified variance of the particle detections in region a can be obtained from the coupled equation system by using the laplace transform technique : _ xm = _ ac_vm + _ bt _ ym + _ xm + _ ad _ xx _ ym= _ bc _ wm+_ym+ _ at _ xm+ _ ad _ xy _ vm = - _ ac_vm + _ a _ af _ a _ xm + _ ad _ xv _ wm = - _ bc _ ym+ _ ad _ xw _ mm = 2 _ ad_xm a similar coupled equation system can be derived for the particle detections in region b : _ xn= _ ac _ vn+ _ bt _ yn+_xn+ _ bd_xy _ yn= _ bc_wn + _ yn+ _ at _ xn + _ bd _ yy _ vn= - _ ac_vn + _ a _ af _ a _ xn + _ bd _ yv _ wn= - _ bc _ wn + _ b _ bf _ b _ yn + _ bd _ yw _ nn = 2 _ bd_yn thus , a final expression for the two - point one - group feynman - alpha formula for region a and b is written below : = 1+y(t)+ 1 + _ i=1 ^ 4 y_i ( 1 - ) if the detector is placed in region a , the following expressions for the functions @xmath169 should be used : -y_1 = -y_2 = - y_3 = - y_4 = and it can be proved that : y_0 = y_1+y_2 + y_3 + y_4= where k_3=_xx k_2=_ac _ xv+_ac _ xx+_b _ bf _ xx _ b-_bf _ xx _ b+_b _ xx+_bc _ xx+_bt _ xy k_1=_ac _ b _ bf _ xv _ b-_ac _ bf _ xv _ b+_ac _ b _ bf _ xx _ b-_ac _ bf _ xx _ b+_ac _ b _ xv+_ac _ b _ xx+_ac _ bc _ xv+_ac _ bc _ xx+_ac _ bt _ xy-_bc _ bf _ b+_b _ bc _ xx+_bc _ bt _ xw+_bc _ bt _ xy k_0=-_ac _ bc _ bf _ xv _ b-_ac _ bc _ bf _ xx _ b+_ac _ b _ bc _ xx+_ac _ bc _ bt _ xw+_ac _ bc _ bt _ xy if the detector is placed in region b , the following expressions for the functions @xmath169 should be used : -y_1 = -y_2 = - y_3 = - y_4 = and it can be proved that : y_0 = y_1+y_2 + y_3 + y_4= where l_3=_yy l_2=_a _ af _ yy _ a-_af _ yy _ a+_a _ yy+_ac _ yy+_at _ xy+_bc _ yw+_bc _ yy l_1=-_ac _ af _ yy _ a+_a _ ac _ yy+_a _ af _ bc _ yw _ a-_af _ bc _ yw _ a+_a _ af _ bc _ yy _ a-_af _ bc _ yy _ a+_a _ bc _ yw+_a _ bc _ yy+_ac _ at _ xy+_ac _ at _ yv+_ac _ bc _ yw+_ac _ bc _ yy+_at _ bc _ xy l_0=-_ac _ af _ bc _ yw _ a-_ac _ af _ bc _ yy _ a+_a _ ac _ bc _ yw+_a _ ac _ bc _ yy+_ac _ at _ bc _ xy+_ac _ at _ bc _ yv in the following , we shall perform a comparison of the two - point two - group version of the feynman - alpha theoretical formula to the two - point one - group , the one - point two - group and the one - point one - group ( i.e. traditional ) versions . [ sec:9 ] in order to compare the four different versions of the feynman - alpha theory , quantitative values of the transition probabilities and reaction intensities were obtained by using monte - carlo simulations in a way similar to that described in @xcite . the simulation setup consists of two regions , region a and region b , as shown in figure [ fig:2 ] . region a represents nuclear material ( radius 4.46 cm ) , in particular a mixture of 2.5% @xmath170u and 97.5% @xmath171u . region b consists of a moderating material with a thickness of 21 cm . the neutron source emits the neutrons with an energy of 2.5 mev . two cases are considered in the simulations , one when the neutron source is in the center of region a and another when the neutron source is at a distance of 15 cm from the center of the nuclear material , in region b. two point detectors , in region a and in region b , are included in the simulation setup . delay neutron precursors are not included in the simulations . initially , the transition probabilities and reaction intensities are obtained in simulations for the two - point two - group case . then , the values of reaction intensities of the two - point two - group case are condensed in order to get the reaction intensities which correspond to the two - point one - group , the one - point two - group and one - point one - group cases . afterwards , these values are used in order to obtain the values of the @xmath7 and @xmath172 coefficients in the feynman - alpha formulas for the cases when the source is in region a , and in region b for fast neutron detections and thermal neutron detections , as shown in table [ table : tab1]-[table : tab4 ] , respectively . .the values of the @xmath173 and @xmath174 calculated for four various versions of feynman - alpha formulas ( the source is in region a , fast neutron detector is used either in region a or b ) . [ cols= " > , > , > , > , > " , ] [ table : tab4 ] since there is only one region considered in the two - group one - point and the one - point one - group feynman - alpha formulas , the coefficients are the same for the detection in the different regions of the initial system used for the simulations . the same is true for the energy - dependent factor in the two - point one - group and the one - point one - group feynman - alpha formulas , the coefficients are the same for the fast and thermal neutron detection . * attention ! * in the studies described below we assume that the two - point two - group version of feynman - alpha formulas gives the most accurate predictions as the most involved one among the four various versions , i.e. the two - point two - group , the two - point one - group , the two - group one - point and one - point one - group theories . figures [ fig:4]-[fig:5 ] show a quantitative illustration of the dependence of the variance to mean of the number of fast neutron detections on the detection time for four versions of feynman - alpha theories when the source is in region a. different curves in figures [ fig:4]-[fig:5 ] are created based on the parameter values from table [ table : tab1]-[table : tab2 ] . as it is shown in figure [ fig:4 ] , when fast neutrons are detected in region a , the two - point two - group , the two - point one - group and one - point one - group versions of the feynman - alpha theoretical formulas give very similar results . however , the one - point two - group version of the formulas overestimates the asymptotic ratio of the variance to mean . thus , we can conclude that the region dependence of the model plays a more important role than the energy dependence for the case when the source and the fast neutron detector are both placed in the region of the nuclear material . therefore , in this situation all three versions of the feynman - alpha theory , the two - point two - group , the two - point one - group and one - point one - group , can be used , although it is more time - efficient to use the one - group one - point version of the feynman - alpha theory . as an example , in reality this case may be related to the measurements performed in the spent fuel pool when the detector is placed in the control tube of fuel assembly . in the case when the fast neutron detector is in region b ( figure [ fig:5 ] ) , a slight difference is observed between the two - point two - group and the two - point one - group versions of feynman - alpha theories . at the same time , the one - point two - group and one - point one - group versions of feynman - alpha theory significantly overestimate the values of variance to mean ratio obtained with the two - point two - group version of the formulas . thus , in this case two versions of feynman - alpha theory , the two - point two - group and the two - point one - group can be used , although it is more time - efficient to use the two - point one - group version for quantitative estimates . the differences between the various versions of feynman - alpha theory are significantly higher when the neutron source is placed in region b and the fast neutron detector is in either region a or b , see figures [ fig:6]-[fig:7 ] . in all cases when the fast neutron detector is used , the two - point two - group version of feynman - alpha formulas produce high values of the asymptotic variance - to - mean ratio compared to results obtained with other versions , i.e. the two - point one - group , one - point two - group and one - point one - group versions of the theory . regarding thermal neutron detection , when the source and the detector are in region a ( figure [ fig:8 ] ) , the three special versions of the feynman - alpha theory , the two - point one - group , the one - point two - group and the one - point one - group , all deviate significantly from the two - point two - group version . however , the one - point one - group theory gives very similar predictions of the ratio of the variance to mean as the two - point one - group theory . at the same time , the two - group one - point theory provides somewhat more accurate results . thus , the impact of the energy - dependence appears to be somewhat higher than the impact of the space - dependence . when the thermal neutron detection is performed in region b ( figure [ fig:9 ] ) , we may conclude that both the space - dependent and energy - dependent aspects play important role for this case . if the source is in region b and detection is performed in region a ( figure [ fig:10 ] ) , the two - point one - group version of the feynman - alpha theory gives results which are closer to the one obtained with two - point two - group theory . thus , the impact of the space - dependence to the final results is higher than the impact of energy - dependence . although , for a case of detector and source being placed in region b ( figures [ fig:11 ] ) , the two - point one - group , two - group one - point and one - group one - point versions provide results of the variance - to - mean ratio that are significantly deviating from the ratio obtained by using the two - point two - group theory . thus , the space - dependent and energy - dependent aspects , both play the important role in this situation . in general , we can say that for the thermal neutron detections when the detection is done in region a , the energy - dependence has a higher impact to the ratio of the variance to mean than the space - dependent factor . on the other hand , for detection in region b both factors should be equally taken into account . the two - group two - point version of feynman - alpha theory was derived with a use of the forward master equation technique . the two - group one - point feynman - alpha theory ( with delay neutrons ) is extended by including fast neutron detections and and fast fissions . the two - point one - group variance - to - mean formula ( with delay neutrons ) is enhanced as well , by including detection and source terms in both regions . thus , this gives the possibility of treating fast reflected systems in a more accurate way , by treating the counts separately in the fast and the thermal groups ( or in the nuclear material ( fissile region ) and reflector regions ) . the comparative study of the two - group two - point , the two - group one - point , the one - group two - point and the one - group one - point feynman - alpha models is made by using the specific reaction intensities obtained in monte - carlo simulations . it is shown that for all cases when the fast neutron detector is used in measurements , the space - dependent aspect has a higher impact on the final results than the energy - dependent aspect . in particular , when the source and the fast neutron detector , both placed in the region of nuclear material three versions of feynman - alpha theory provides similar accuracy in the determination of the variance to mean ratio . namely , the two - point two - group , the two - point one - group and one - point one - group can be used , although it is more time - efficient to use the one - group one - point version of the feynman - alpha theory . the situation is not so optimistic for a case , when the fast neutron detector is in region b , because a slight difference is observed between all versions of the theories . the one - point two - group and one - point one - group versions of feynman - alpha theory significantly overestimate the values of variance to mean ratio obtained with the two - point two - group version of the formulas . therefore , in this case two versions of feynman - alpha theory , the two - point two - group , the two - point one - group can be considered as the accurate qualitative estimates . regarding the use of the thermal neutron detections , both energy- and space - dependent factors are important to take into account . when the detection is done in region a , the space - dependence has a higher impact to the ratio of the variance to mean than the energy - dependence , while , for detection in region b both factors should be equally considered .

one aspect of space weather forecasting involves the prediction of coronal mass ejection ( cme ) arrival times at earth , which may or may not lead to a geomagnetic storm at that time . assuming it is reasonably well established that a solar eruption is indeed headed towards earth , an accurate assessment of its arrival time depends on both an accurate measurement of the cme s initial velocity , and some estimate of how that velocity will change during its passage through the interplanetary medium ( ipm ) . the stereo mission @xcite provides substantial improvements in our ability to study both these aspects of cme kinematics . the lateral views that the two stereo spacecraft have of the sun - earth line provide ideal vantage points for measuring the radial velocities of earth - directed cmes . in contrast , from the perspective of a near - earth instrument such as the large angle spectrometric coronagraph ( lasco ) instrument @xcite on board the _ solar and heliospheric observatory _ ( soho ) , an observer will only be able to perceive the lateral expansion of the cme . as for ipm propagation , the heliospheric imagers on board the two stereo spacecraft allow for continuous cme tracking all the way to 1 au . we will be studying here the effects of the ambient solar wind on a shock generated by a fast cme originating on 2011 march 7 . the source region of the cme is just to the east of a coronal hole , which has a dramatic effect on the shape of the shock both close to the sun and in the ipm . the western half of the shock propagates outwards through high speed wind from the coronal hole , ultimately hitting stereo - a on march 9 , while the eastern half propagates outwards through slow speed wind , ultimately hitting earth a day later on march 10 . in addition to the indications of shock deformation provided by the in situ data , heliospheric images of the event also show signatures of the uneven shock geometry caused by the high speed wind from the coronal hole . this is therefore an ideal event for studying the effects of an inhomogeneous solar wind on shock morphology . we will empirically reconstruct the shock shape from the data , and we will also examine whether mhd models of the inner heliosphere can successfully reproduce the inferred shape . figure 1 shows a carrington map of euv emission from the solar corona , for carrington rotation number 2107 , created from 195 bandpass euvi images from stereo - a . the positions of earth , stereo - a , and stereo - b are indicated for 2011 march 7 . at this time , stereo - a was located @xmath0 ahead of earth in its orbit around the sun , at a distance of 0.96 au from sun - center , and stereo - b was @xmath1 behind earth , at a distance of 1.02 au . figure 2 explicitly displays the spacecraft geometry in the ecliptic plane , in heliocentric aries ecliptic coordinates . there are two cmes on 2011 march 7 that we will be modeling , one from active region ar1166 , and one from ar1164 . the locations of these two active regions are identified in figure 1 . the first cme , which we will call cme1 , begins with an m1.9 flare from ar1166 at 13:45 , while the second cme , cme2 , begins with an m3.7 flare from ar1164 at 19:43 . it is the second event that we are primarily interested in , but the two cmes end up overlapping in the inner heliosphere , so it is necessary to consider observations of both . each stereo spacecraft carries a package of imagers called the sun - earth connection coronal and heliospheric investigation ( secchi ) , which includes two coronagraphs ( cor1 and cor2 ) and two heliospheric imagers ( hi1 and hi2 ) that observe cmes both close to the sun and in the inner heliosphere @xcite . the fields of view of hi1 and hi2 are illustrated in figure 2 . the top two panels of figure 3 show images of the two march 7 cmes as seen from stereo - a s perspective , specifically by the cor2-a telescope . these images , and all other images shown in this paper , are displayed in running difference format , where the previous image is subtracted from each image to effectively remove static coronal structures and emphasize the dynamic cme . the second cme consists of two distinct components : the ejecta , which is directed towards the northeast in the image , and a bright front out ahead of the ejecta , which is the shock wave created by this fast , supersonic cme . curiously , the ejecta is not at all well centered within the shock , in contrast to other cmes with visible shocks that we have studied recently @xcite . although both the ejecta and shock have a northward component to their trajectory , the shock appears centered more to the west than the ejecta ( i.e. , more towards stereo - a ) . we attribute this asymmetry to the presence of a coronal hole just to the west of the cme s source region , as shown in figure 1 . studying the effects of the coronal hole on this shock s morphology is the focal point of this paper . in contrast to cme2 , there is no bright front out ahead of cme1 indicative of a shock , presumably because cme1 is a much slower eruption and is not able to create a bright shock front . in cor2-a , the ejecta of the two cmes are both directed towards the northeast , both possessing an angular extent of about @xmath2 . the more easterly longitude of cme1 is indicated both by the location of its source region relative to that of cme2 ( see figure 1 ) and by lasco images of the two cmes . the bottom panel of figure 3 is a lasco / c3 image in which both events are seen , with cme1 directed somewhat to the northeast of the sun - earth line , and the brighter cme2 directed to the northwest . both 2011 march 7 cmes can be tracked all the way to 1 au using stereo s heliospheric imagers . for cme1 , stereo - b provides a better vantage point for tracking the cme into the inner heliosphere , while for cme2 stereo - a is better . figure 4 shows hi1-a and hi2-a images of cme2 . in hi1-a , cme2 quickly overtakes cme1 , such that by the time of the image in figure 4 the two cmes are superimposed on each other , complicating interpretation of the data . although the expansion rate of cme2 slows dramatically in hi1-a , its leading edge still moves well in front of cme1 , and in the hi2-a image in figure 4 , the cme front seen is entirely that of cme2 . as cme2 approaches the left edge of the hi1-a field , the shock at the front of the cme appears to split in two , as a foreground part of the shock appears to move ahead of a background portion . this `` doubling '' of the shock front becomes even more apparent early in the hi2-a field of view . the two fronts labeled f1 and f2 in the hi2-a image in figure 4 are both part of the cme2 shock . our interpretation of this visage is illustrated in figure 2 . the part of the cme2 shock propagating towards stereo - a in high speed wind emanating from the coronal hole seen in figure 1 becomes more radially extended than the part of the shock above the ejecta , which is propagating more towards earth through slow speed wind . this creates a discontinuity in the shock shape in between the longitudes of earth and stereo - a , thereby creating two tangent points with the shock as viewed from stereo - a . the outermost tangent point creates the outermost front , f1 , which is the foreground part of the shock from stereo - a s vantage point , and the inner tangent point yields the f2 front , corresponding to the slower part of the shock directed more towards earth . this kind of asymmetry is to be expected on the basis of models of cme propagation into inhomogeneous solar wind @xcite . to summarize , images from stereo demonstrate that the coronal hole s presence near the cme2 source region significantly affects the shape of the cme2 shock . the two main observational signatures of this are the shock not being centered on the ejecta in cor2-a , and the doubling of the shock front in hi2-a . the shock asymmetry can be studied further using in situ data from stereo - a , and from _ wind _ at the l1 lagrangian point near earth . the cme2 shock is broad enough to have hit both spacecraft , despite a nearly @xmath2 separation in longitude . figure 5 shows plasma parameters measured at stereo - a and _ wind_. focusing first on stereo - a , where the meaurements are made by the plastic and impact instruments on board the spacecraft @xcite , on march 8 there is a density peak accompanied by a dramatic increase in wind speed and temperature , indicating that this is a corotating interaction region ( cir ) associated with the high speed wind emanating from the coronal hole just to the west of the cme2 source region ( ar1164 ; see figure 1 ) . the shock of cme2 hits stereo - a at about 6:48 ut on march 9 , indicated by substantial jumps in density , velocity , temperature , and magnetic field . after a cme shock , a signature of the cme ejecta driving the shock is often observed . that is not the case at stereo - a , where nothing but normal high speed wind follows the shock . this is an excellent example of the `` driverless shocks '' studied by @xcite . the lack of a driver at stereo - a is not a surprise based on the apparent trajectory of the ejecta away from the sun - spacecraft line in cor2-a ( see figure 3 ) . the cme2 shock also hits _ wind _ , but it does so over a day later at about 7:44 ut on march 10 ( see figure 5 ) . this delayed arrival time is consistent with the shock shape displayed in figure 2 . the shock arrival time discrepancy between stereo - a and _ wind _ provides another valuable diagnostic for the degree of asymmetry in the shock induced by the coronal hole and its high speed wind . unlike at stereo - a , there is ejecta observed following the cme2 shock at earth . not only is there a large density peak associated with this ejecta , but a lengthy period of negative @xmath3 on march 1011 ( see @xmath4 panel in figure 5 ) , which produces a modest but lengthy geomagnetic storm at this time , with the planetary k index reaching as high as @xmath5 the question is whether the ejecta is from cme1 or cme2 . for that matter , how sure are we that the shock seen by _ wind _ is associated with cme2 instead of cme1 ? we will return to these issues after describing in detail how we have tried to empirically reconstruct the morphology and kinematics of the two 2011 march 7 cmes . kinematic modeling of cme1 and cme2 is a necessary aspect of their 3-d reconstruction . for both cmes we track the leading edge of the ejecta ( as opposed to the shock in the case of cme2 ) . for cme1 , stereo - b provides the best vantage point to view the cme s ipm propagation , so stereo - b images are used to measure its elongation angle , @xmath6 , from sun - center as a function of time . for cme2 , stereo - a images are used instead . these elongation angles are converted to actual distances from sun - center , @xmath7 , using the equation first championed by @xcite , @xmath8 where @xmath9 is the observer s distance to the sun , and @xmath10 is the angle between the cme trajectory and the observer s line of sight to the sun . this equation is derived assuming that cme fronts can be approximated as spheres centered halfway in between their leading edges and the sun . by at least crudely taking into account a cme s lateral extent , this equation will be applicable to more viewing geometries than the alternative `` fixed-@xmath10 '' approxmiation @xcite , @xmath11 which assumes an infinitely narrow cme . however , for the advantageous near - lateral viewing geometries that we have here for cme1 and cme2 , both equations will yield similar results @xcite . measurements of the cme trajectory angle @xmath10 are ultimately established by the morphological modeling described in section 4 . for cme1 , @xmath12 ( from stereo - b ) , and for cme2 , @xmath13 ( from stereo - a ) . based on these values and equation ( 1 ) , figure 6 shows plots of cme distance versus time for cme1 and cme2 . in order to extract velocity profiles from these measurements , we use a simple three - phase kinematic model that we have used before to fit the kinematic profiles of fast cmes @xcite . this approach assumes the cme s motion can be approximated by an initial phase of constant acceleration as the cme ramps up to its maximum speed , followed by a phase of constant deceleration as the cme is slowed by its interaction with the ipm , and finally by a phase of constant velocity . solid lines in figure 6 indicate the best fits to the data with this model . for cme1 , the leading edge accelerates quickly to a maximum speed of 853 km s@xmath14 in the cor1 field of view , followed by a deceleration of @xmath15 m s@xmath16 , until a final velocity of 676 km s@xmath14 is reached at a distance of 39.1 r@xmath17 from the sun . the faster cme2 accelerates to 1648 km s@xmath14 , followed by a deceleration of @xmath18 m s@xmath16 , reaching a final velocity of 713 km s@xmath14 at a distance of 55.6 r@xmath17 . it is interesting that both cmes end up at about the same speed , despite starting out so differently . the nearly identical asymptotic velocities indicates that the cmes are propagating into solar wind with similar properties . this is not surprising considering that the cmes are propagating in similar directions , but just far apart not to overlap too much . in recent papers , we have developed tools for empirically reconstructing the 3-d structure of cmes and their shocks from stereo imagery @xcite . for cme ejecta we generally assume a flux rope ( fr ) geometry , as there exists an extensive literature providing observational support for magnetic frs lying at the heart of many cmes . this support comes from both in situ data ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and imaging data @xcite . in the empirical reconstruction process , 3-d frs are constructed using a parametrized functional form , which can produce frs of many shapes and orientations , including ones with elliptical rather than circular cross sections . a similar parametrized prescription is used to generate lobular fronts to estimate the shapes of shock fronts sometimes observed ahead of cmes , as in the case of cme2 @xcite . using these parametrized descriptions of cmes and their shocks , the idea is to create 3-d density distributions that can then be used to compute synthetic images for comparison with actual images . this comparison is done in a very comprehensive fashion , considering observations from stereo - a , stereo - b , and soho / lasco ; and considering both images of the cme close to the sun in coronagraphic images , and far from the sun in stereo s heliospheric images . parameters are adjusted to maximize agreement between the synthetic and actual images . perfect agreement is not expected , since the model places mass only on the surface of the fr and nothing in its interior , but the goal is for the model to reproduce the basic outline of the cme structure in the images as well as possible . we use these procedures to reconstruct the 3-d morphology of the two 2011 march 7 cmes , with figure 7 showing the result . figure 7a shows the cmes at 21:40 ut on march 7 , before the cme2 fr has decelerated very much . by the time of figure 7b ( at 5:10 ut on march 8) cme2 has caught up with cme1 , but it has also decelerated to a speed not much higher than that of cme1 ( see figure 6 ) , so further propagation outwards does not yield much additional motion of the cmes relative to each other . there is only a modest amount of overlap between the two cmes in the reconstruction , but this region of overlap happens to be directed towards earth ( see figure 2 ) . the fr of cme1 is a very fat one , with a highly elliptical cross section . the trajectory of the center of the flux rope is directed @xmath19 north and @xmath20 east ( e.g. , n22e27 ) of the sun - earth line . the west leg of the fr is tilted @xmath21 above an east - west orientation . for cme2 , the fr is directed towards n35w28 , with a west leg tilted @xmath22 below an east - west orientation . simple , self - similar expansion is assumed for the fr components of the two cmes . such is not the case for the cme2 shock , however . the asymmetric and time - dependent behavior of the shock morphology has already been described qualitatively in section 2 . within the framework of our model , the shock is initially assumed to be a symmetric , lobular front , as shown in figure 7a , with a trajectory @xmath23 degrees from that of the cme2 fr , at n45w58 relative to the sun - earth line . however , about 5 hours after the start of the cme , while it is in the hi1-a field of view and the cme is decelerating , we allow a discontinuity to develop in the shock front . the western part of the shock headed towards stereo - a , associated with front f1 in figure 4 , is allowed to expand outwards farther than the eastern part of the shock headed towards earth , associated with front f2 . we experiment with different degrees of asymmetry , and different longitudinal locations for the discontinuity . we settle on a shock extent 1.35 times greater for the western part of the shock than for its eastern part , with the discontinuity at a longitude of w48 relative to earth . the resulting shock shape is shown in figures 2 and 7b . synthetic images computed from the 3-d reconstruction are shown in the left panels of figures 3 and 4 . these are computed from the model density cubes using full 3-d thomson scattering calculations @xcite , and are displayed in the figures in running difference format , consistent with the real images . some random noise has been added to the synthetic images for aesthetic purposes , to better match the appearance of the images . it is worth emphasizing that when converging on the best possible model parameters , _ all _ stereo and lasco images are considered , not just the select few we can show in figures 3 and 4 . the synthetic hi2-a image in figure 4 shows how the shock discontinuity in the model does yield two distinct fronts ( f1 and f2 ) in the synthetic image . this resembles the shock s appearance in the real image , although the agreement between the synthetic and real images is far from perfect . for one thing , the model shock clearly extends too far to the south . improving matters would presumably require making the shock even more asymmetric than it already is , but there is a limit to the shapes that can be made with the parametrized functional forms we are using for fr and shock shape modeling . the reconstructed shock shape is not only designed to approximate the shape of the shock in the images , but also to reproduce the arrival time of the shock observed at _ wind _ and stereo - a . this means we need to know precisely the shock kinematics specifically towards earth and stereo - a , and figure 6 explicitly shows the kinematic profiles of the shock in those directions . towards earth , the shock distance is smaller than the top of the fr because earth and _ wind _ are being hit by the flank of the shock ( see figure 2 ) . since self - similar expansion applies to this part of cme2 , this means that the shock velocity towards earth is proportionally slower as well . towards stereo - a , the kinematics are somewhat more complicated because of the shock discontinuity that is allowed to develop while cme2 is decelerating . figure 6 shows the kinematic profile that results from this discontinuity . mostly because of the factor of 1.35 increase in shock radial extent , the shock speed at stereo - a is almost 400 km s@xmath14 faster than at earth . thus , the shock arrives much earlier at stereo - a than at _ wind _ , as observed . the top panel of figure 5 shows the density profiles predicted by the model at _ wind _ and stereo - a . the density peaks associated with the shock agree with the observed arrival time of the shock to within a couple hours . as mentioned in section 2 , at earth there is geoeffective cme ejecta observed after the shock . our reconstruction shows a broad density peak after the cme2 shock at _ wind _ , consistent with this . inspection of figure 2 reveals that this material is not from cme2 but from cme1 . the reconstruction suggests that after the cme2 shock strikes earth , there is then a glancing blow from the fr component of cme1 . since it is such a grazing incidence , and because the cme2 fr is also not too far away from the sun - earth line , confidence in this conclusion is not high , but we do find it more likely that the cme1 fr accounts for the geoeffective ejecta than the cme2 fr . in this interpretation , the density peak seen by _ wind _ at the end of march 10 would be cme1 material that had been shocked by the cme2 shock when cme2 overtakes cme1 shortly after cme2 enters the hi1-a field of view . figure 1 shows the inferred trajectories of the various components of cme1 and cme2 relative to their source regions . the cme1 fr is directed somewhat northeast of ar1166 , which is where the flare associated with the cme occurs . this trajectory is consistent with the euv observations of the event from the solar dynamics observatory ( sdo ) , which show that the euv dimming associated with the eruption is mostly north of ar1166 . the cme2 fr ends up directed about @xmath24 to the east of its source region , ar1164 . this eastward deflection is probably caused by the coronal hole to the west of ar1164 . @xcite provide many examples of cme ejecta that are deflected away from coronal holes , demonstrating that this is a common effect . in contrast , the shock expands readily into the fast outflow from the coronal hole . a faster lateral propagation speed through the coronal hole is possibly indicative of higher alfvn speeds in and above the coronal hole , compared to elsewhere . we have associated the march 10 shock observed by _ wind _ with the cme2 shock , but this interpretation is not definite considering the close proximity of cme1 , and considering that earth is near the eastern edge of the cme2 shock . we can search for support for the cme2 association by using the in situ data and the hydromagnetic rankine - hugoniot ( rh ) jump conditions ( e.g. , * ? ? ? * ) to infer the shock normal at _ wind_. the cme2 shock as reconstructed here is centered north of the ecliptic plane , and well to the west of the earth . thus , the shock normal at earth is expected to be in a southeasterly direction . this is the prediction that we intend to test . it is possible to estimate a shock normal solely from the magnetic field and/or velocity measurements using the coplanarity properties of the rh equations @xcite . however , these techniques do not work well for all possible shock geometries . they also do not consider all possible constraints on the problem , involving all relevant plasma measurements and the full set of rh equations . increasingly sophisticated computation techniques have since been developed to more precisely determine shock normal characteristics from single spacecraft measurements @xcite . we have developed our own computational tools to evaluate the normal of the cme2 shock , which closely resembles the prescription of ( * ? ? ? * hereafter ks08 ) . we refer the reader to that paper for a full list of the relevant rh equations . the jump conditions are generally expressed as six separate equations , but in practice there are eight , since the equations expressing conservation of tangential momentum flux and tangential electric field are vector equations that can each be decomposed into two scalar equations , for vector components parallel and perpendicular to the shock front ( see , e.g. , * ? ? ? utilizing these eight equations requires pre- and post - shock measurements of eight quantities : density , temperature ( or pressure ) , vector velocity , and vector magnetic field . figure 8 displays these eight quantities for the cme2 shock , both at stereo - a and at _ wind_. the data are shown with 1 minute time resolution . time intervals of about 20 minutes duration are used to assess the pre- and post - shock plasma state . for each plasma quantity an initial mean and standard deviation are computed within the interval . we then throw out points more than 1.5 standard deviations from the mean to exclude particularly anomalous excursions . we then recompute the mean and standard deviation , and the resulting values are illustrated with horizontal lines in figure 8 . the rh equations apply only in the rest frame of the shock , so the velocity and field components measured in figure 8 must be rotated into the shock frame before numbers can be plugged into the rh equations . this requires the assumption of a shock normal ( @xmath10,@xmath4 ) , and a shock velocity normal to the shock ( @xmath25 ) . the goal as described by ks08 is to systematically explore the three - dimensional parameter space described by ( @xmath10,@xmath4,@xmath25 ) to assess where the rh conditions are most precisely met . our approach is similar , but rather than keep @xmath25 as a free parameter , we instead compute it from the radial shock speed measured in the stereo images , @xmath26 . figure 6b indicates that @xmath27 km s@xmath14 at stereo - a , and @xmath28 km s@xmath14 at _ wind_. if ( @xmath10,@xmath4 ) are defined in a spacecraft - centered rtn coordinate system , then @xmath29 thus , in practice our parameter space is just two - dimensional , defined only by the shock normal ( @xmath10,@xmath4 ) . each of the eight jump conditions , @xmath30 ( i=1 - 8 ) , can be written as @xmath31}-\mbox{[post - shock state ] } = 0.\ ] ] so for a given ( @xmath10,@xmath4 ) normal we can plug the numbers into each equation and see how close each equation is to zero . however , assessing how well @xmath30 are approximating zero requires uncertainties to be computed for each @xmath30 . we do this using monte carlo simulations , where the 8 pre - shock and 8 post - shock measurements illustrated in figure 8 are varied in a manner consistent with the displayed mean and standard deviation . we also vary @xmath26 in the simulations , assuming 5% uncertainties in our measurements of this parameter . for each trial , we compute a value of @xmath30 and after all trials are complete we then compute the standard deviation of these values , @xmath32 . we can then compute a @xmath33 number @xcite to quantify how well each assumed shock normal is collectively fitting the rh conditions : @xmath34 ^ 2.\ ] ] the best fit normal is simply where the @xmath33 array has its minimum value of @xmath35 . if we define @xmath36 , then contour plots of @xmath37 can be used to define confidence intervals for @xmath10 and @xmath4 . we validate our shock normal computation code using the same synthetic shocks created by ks08 to validate their routine . the one significant difference between our approach and that of ks08 , besides the treatment of @xmath25 as a measured rather than a free parameter , is that ks08 do not explicitly measure pre- and post - shock plasma parameters ( and uncertainties ) but instead plug sets of measurements for specific pre- and post - shock times into the rh equations , considering all pre- and post - shock times within some time interval to do their version of our monte carlo simulations . we experimented with a version of our code that performs the calculation in this manner , and we did not find any dramatic difference in results for the cme2 shock studied here . in figure 9 , we show the @xmath38 contours for the cme2 shock at both stereo - a and _ wind_. following common practice , we draw contours corresponding to probabilities associated with the 1- , 2- , and 3-@xmath39 levels of a normal distribution , which are 68.3% , 95.4% , and 99.7% confidence levels . for stereo - a , the best fit has @xmath40 . interpreting this number requires knowledge of the degrees of freedom , @xmath41 , which in this case is @xmath42 , i.e. , eight jump conditions minus two free parameters . thus , the reduced chi - squared , @xmath43 , somewhat below the expectation value of 1 @xcite . consultation with a chi - squared probability table ( or direct computation of the distribution ) reveals that for @xmath42 , @xmath40 corresponds to a probability value of @xmath44 , meaning there is a 97% chance that the measured parameter uncertainties illustrated in figure 8 can explain the magnitude of @xmath35 . one consequence of the low @xmath35 value is that the confidence intervals are rather large for stereo - a in figure 9 . in contrast , for the shock at _ wind _ , @xmath45 , corresponding to @xmath46 and @xmath47 . a 24% chance that the assumed uncertainties can account for the magnitude of @xmath35 is still high enough to consider this a reasonably good fit , although the low value might indicate that the uncertainties in figure 8 may be a little too low . the confidence contours are naturally much tighter for _ wind _ in figure 9 than for stereo - a . for stereo - a , we find a shock normal of ( @xmath10,@xmath4)=(@xmath48 , @xmath49 ) , while for _ wind _ we find ( @xmath10,@xmath4)=(@xmath50 , @xmath51 ) . the quoted 1-@xmath39 uncertainties are estimated not from the @xmath38 contours , but from monte carlo simulations where we vary the measurements illustrated in figure 8 in a manner consistent with the displayed mean and standard deviation , perform the analysis just described , and after many trials the standard deviations of the resulting best fit ( @xmath10,@xmath4 ) values provides the 1-@xmath39 uncertainties in ( @xmath10,@xmath4 ) . we find that uncertainties estimated in this direct manner are larger than those estimated from the @xmath38 contours , possibly due to nonnormal characteristics of the uncertainties in this particular problem @xcite . both normals are south - directed ( i.e. , negative @xmath4 ) , which is what we would expect given the northward direction of the cme2 shock . at stereo - a , the shock may be oriented somewhat towards the east ( i.e. , negative @xmath10 ) , which is not necessarily expected , but uncertainties in @xmath10 are fairly high for stereo - a . most importantly , the _ wind _ analysis indicates that the shock there is highly oriented towards the east ( i.e. , negative @xmath10 ) , consistent with expectations for the cme2 shock , and not consistent with a cme1 association . this provides strong support for the mar . 10 shock at _ wind _ indeed being the cme2 shock . in passing , we note that about 10 hours before the cme2 shock hits stereo - a , it encounters the _ venus express _ spacecraft at venus . the location of venus is shown in figure 2 . a quick coplanarity analysis @xcite based only on the magnetic field data yields ( @xmath10,@xmath4)=(@xmath52 , @xmath53 ) , reasonably consistent with our measurements at stereo - a , but with an even more eastward orientation , which as mentioned above would not be easy to explain . sophisticated 3-d mhd modeling codes are becoming increasingly useful for modeling the solar wind and transients within it . models of this sort include the enlil code currently being used as an operational space weather modeling tool at noaa s space weather prediction center @xcite , and the space weather modeling framework ( swmf ) package @xcite . we here test whether this sort of modeling can reproduce the general asymmetric shape of the cme2 shock inferred from the empirical reconstruction . the model of @xcite already demonstrates an ability to produce shock asymmetries of this sort due to solar wind inhomogeneities . the code we use here is a well established model described most extensively by @xcite , which has been used to confront stereo data before @xcite . this model combines the hakamada - akasofu - fry ( haf ) code ( version 2 ; * ? ? ? * ) , which computes the solar wind s evolution out to 18 r@xmath17 , and a fully 3-d mhd code that then carries the simulation out to 285 r@xmath17 @xcite . the inner boundary conditions for the haf part of the code are derived from solar magnetograms and resulting source surface maps using the wang - sheeley - arge model @xcite . this establishes the ambient solar wind into which cme2 is launched . in the model a cme front is produced by introducing a velocity pulse at the inner boundary . for various reasons , the haf code can not properly model the lateral propagation of a cme disturbance into an adjacent coronal hole . for one thing , the haf code is essentially a 1-d radial propagation model , rather than a 3-d code . for another , the piston used to initiate a fast cme like cme2 is a very large one , which means that part of the velocity pulse extends into the coronal hole right from the start , even if its center is outside the hole . truly modeling the propagation of a cme front into a coronal hole would probably require a full 3-d mhd model with a lower inner boundary than ours , a smaller piston , and with a higher spatial resolution than we are using . however , the existing model described above should still be able to study shock asymmetries induced by radial propagation into an inhomogeneous medium . since the early lateral propagation effects can not be modeled properly , we place the piston at the empirically inferred center of the shock ( n45w58 ) right from the start rather than at the fr location ( n35w28 ) , where the driver is really located . as in past models ( e.g. , * ? ? ? * ; * ? ? ? * ) , the velocity pulse is gaussian - shaped , with the velocity of the piston decreasing with angular distance from piston - center . the width of the piston is described by the gaussian @xmath39 parameter @xcite , which in this case is @xmath54 . temporally , the velocity pulse consists of a 140 minute exponential rise in velocity up to a maximum speed of 1550 km s@xmath14 at piston - center , followed by a 140 minute fall back to the original ambient solar wind speed . this leads to a compression wave that arrives at both stereo - a and earth near cme2 s actual observed time of arrival at those locations . figure 10 shows the shape of the modeled disturbance as it approaches 1 au . although the model does not precisely reproduce the empirically inferred shock shape in figure 2 , at least qualitatively , the cme front exhibits the same sort of asymmetry inferred for the cme2 shock , being much farther from the sun west of the piston longitude , where fast solar wind predominates , compared to east of the piston location , where slow wind predominates . this is encouraging for ongoing attempts to explore how these kinds of mhd models can be used to improve cme arrival time forecasts at 1 au . we have empirically reconstructed the morphology of two cmes from 2011 march 7 , denoted cme1 and cme2 , with a particular focus on the shape of the shock produced by the faster cme2 . the location of cme2 in between earth and stereo - a , with a shock broad enough to hit both , allows us to bring an extensive collection of observations to bear on assessing the characteristics of the event , involving both imaging and in situ data . our findings are summarized as follows : 1 . : : a coronal hole to the west of cme2 s source location has noticeable effects on both the flux rope driver of the cme , which is deflected to the east away from the hole , and the shock front , which in contrast to the flux rope expands more readily over and above the coronal hole . visual evidence for the shock asymmetry induced by the presence of the coronal hole is apparent in both coronagraphic and heliospheric images of the cme shock . : in situ data also provide evidence for cme2 shock asymmetries induced by the coronal hole , as the shock is observed to arrive at stereo - a over a day before being observed by _ wind _ near earth . we use the in situ measurements and the rankine - hugoniot shock jump conditions to measure the shock normals at stereo - a and _ wind_. the orientations of the shock normals in both locations are in reasonably good agreement with what is expected on the basis of the shock morphology inferred from the stereo images . 3 . : : the asymmetries inferred for the cme2 shock are qualitatively reproduced by an mhd model of the cme , which clearly shows the shock expanding radially more rapidly in regions of high speed wind . 4 . : : after the cme2 shock hits _ wind _ , cme ejecta is also observed that produces a modest geomagnetic storm at earth . based on our empirical reconstruction , we tentatively attribute this material not to cme2 but to cme1 , which would imply that this is cme1 ejecta that was shocked by the cme2 shock close to the sun . this work has been supported by nasa award nnh10an83i to the naval research laboratory . the stereo / secchi data are produced by a consortium of nrl ( us ) , lmsal ( us ) , nasa / gsfc ( us ) , ral ( uk ) , ubham ( uk ) , mps ( germany ) , csl ( belgium ) , iota ( france ) , and ias ( france ) . in addition to funding by nasa , nrl also received support from the usaf space test program and onr . this work has also made use of data provided by the stereo plastic and impact teams , supported by nasa contracts nas5 - 00132 and nas5 - 00133 . the hakamada - akasofu - fry solar wind model version 2 ( hafv2 ) was provided to nrl / ssd by a software license from exploration physics international , inc . ( expi ) . abraham - schrauner , b. 1972 , j. geophys . , 77 , 736 acua , m. h. , curtis , d. , scheifele , j. l. , russell , c. t. , schroeder , p. , szabo , a. , & luhmann , j. g. 2008 , space sci . rev . , 136 , 203 arge , c. n. , & pizzo , v. 2000 , j. geophys . res . , 105 , 10465 billings , d. e. 1966 . a guide to the solar corona ( new york : academic press ) bevington , p. r. , & robinson , d. k. 1992 , data reduction and error analysis for the physical sciences ( new york : mcgraw - hill ) bothmer , v. , & schwenn , r. 1998 , ann . geophys . , 16 , 1 brueckner , g. e , et al . 1995 , sol . , 162 , 357 burlaga , l. f. 1988 , j. geophys . res . , 93 , 7217 chen , j. , et al . 1997 , apj , 490 , l191 colburn , d. s. , & sonett , c. p. 1966 , space sci . , 5 , 439 eyles , c. j. , et al . 2009 , sol . , 254 , 387 fry , c. d. , sun , w. , deehr , c. s. , dryer , m. , smith , z. , akasofu , s. -i . , tokumaru , m. , & kojima , m. 2001 , j. geophys . , 106 , 20985 galvin , a. b. , et al . 2008 , space sci . , 136 , 437 gibson , s. e. , & low , b. c. 1998 , apj , 493 , 460 gopalswamy , n. , mkel , p. , xie , h. , akiyama , s. , & yashiro , s. 2009 , j. geophys . res . , 114 , a00a22 hakamada , k. , & akasofu , s. -i . 1982 , space sci . , 31 , 3 han , s. m. , wu , s. t. , & dryer , m. 1988 , comput . fluids , 16 , 81 howard , r. a , et al . 2008 , space sci . rev . , 136 , 67 kahler , s. w. , & webb , d. f. 2007 , j. geophys . res . , 112 , a09103 kaiser , m. l. , kucera , t. a. , davila , j. m. , st . cyr , o. c. , guhathakurta , m. , & christian , e. 2008 , space sci . , 136 , 5 koval , a. , & szabo , a. 2008 , j. geophys . res . , 113 , a10110 ( ks08 ) krall , j. 2007 , apj , 657 , 559 lepping , r. p. , & argentiero , p. d. 1971 , j. geophys . , 76 , 4349 lepping , r. p. , jones , j. a. , & burlaga , l. f. 1990 , j. geophys . , 95 , 11957 lugaz , n. , roussev , i. i. , & gombosi , t. i. 2011 , adv . space res . , 48 , 292 lugaz , n. , vourlidas , a. , & roussev , i. i. 2009 , ann . geophys . , 27 , 3479 luhmann , j. g. , et al . 2008 , space sci . , 136 , 117 manchester , w. b. , gombosi , t. i. , roussev , i. , de zeeuw , d. l. , sokolov , i. v. , powell , k. g. , tth , g , & opher , m. 2004 , j. geophys . res . , 109 , a01102 marubashi , k. 1986 , adv . space res . , 6 , 335 odstrcil , d. , & pizzo , v. j. 1999 , j. geophys . res . , 104 , 28225 odstrcil , d. , & pizzo , v. j. 2009 , sol . phys . , 259 , 297 press , w. h. , flannery , b. p. , teukolsky , s. a. , & vetterling , w. t. 1989 , numerical recipes ( cambridge : cambridge university press ) riley , p. , gosling , j. t. , & pizzo , v. j. 1997 , j. geophys . res . , 102 , 14677 sheeley , n. r. , jr . , et al . 2008 , apj , 675 , 853 shu , f. s. 1992 , the physics of astrophysics , vol . 2 : gas dynamics ( mill valley : university science books ) szabo , a. 1994 , j. geophys . res . , 99 , 14737 thernisien , a. f. r. , howard , r. a. , & vourlidas , a. 2006 , apj , 652 , 763 tth , g. , et al . 2005 , j. geophys . res . , 110 , a011126 vias , a. f. , & scudder , j. d. 1986 , j. geophys . res . , 91 , 39 wang , y. -m . , & sheeley , n. r. , jr . 1990 , apj , 355 , 726 wood , b. e. , & howard , r. a. 2009 , apj , 702 , 901 wood , b. e. , howard , r. a. , & socker , d. g. 2010 , apj , 715 , 1524 wood , b. e. , wu , c. -c . , howard , r. a. , socker , d. g. , & rouillard , a. p. 2011 , apj , 729 , 70 wu , c. -c . , dryer , m. , wu , s. t. , wood , b. e. , fry , c. d. , liou , k. , & plunkett , s. 2011 , j. geophys . res . , 116 , a12103 wu , c. -c . , fry , c. d. , dryer , m. , wu , s. t. , thompson , b. , liou , k. , & feng , x. s. 2007a , adv . space res . , 40 , 1827 wu , c. -c . , fry , c. d. , wu , s. t. , dryer , m. , & liou , k. 2007b , j. geophys . res . , 112 , a09104 wu , s. t. , andrews , m. d. , & plunkett , s. p. 2001 , space sci . rev . , 95 , 191

the tight correlation between the mass of a supermassive black hole and the properties of its host galaxies spheroid component is now well established in the local universe ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , indicating a coeval growth history of galaxies and their central black holes . however , the detailed role of this correlation for our understanding of galaxy evolution and black hole growth is still an open question . the standard theoretical scenario explains this relation by joint triggering of star formation and black hole activity via major mergers , while the black hole growth is self - regulated via active galactic nuclei ( agn ) feedback that shuts off star formation and quenches the accretion on to the black hole ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? alternatively , it may be possible to reproduce the observed black hole - galaxy correlations without the need for self - regulation via agn feedback , either as a natural consequence of a common merger history @xcite , or in a black hole growth scenario via gravitational torques @xcite . an observational constraint to distinguish between these alternative scenarios and to constrain individual theoretical models ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) is provided by the redshift evolution of the @xmath1bulge relations . however , direct dynamical black hole mass measurements are not feasible at higher redshifts . while integrated constraints can be obtained from the study of the black hole mass function ( bhmf ) and the agn luminosity function @xcite , the only route for observational studies on individual objects is to use broad line agn samples . the underlying assumption for this approach is that broad line agn obey the same relationship as non - active galaxies ( e.g. * ? ? ? * ; * ? ? ? * ) . for broad line agn the black hole mass can be estimated employing the virial method ( e.g. * ? ? ? * ; * ? ? ? the main challenge for these studies is to determine the properties of the agn host galaxy , targeting either the velocity dispersion @xcite , the galaxy luminosity @xcite or the stellar mass @xcite . while early studies suggested a clear trend of positive evolution in the @xmath0-bulge relation , they preferentially focused on luminous quasars and their hosts , sampling from the bright end of the agn luminosity function . more recent studies , using fainter agn , tend to find mild or no evolution @xcite . a compilation of current literature results is shown in fig . [ fig : bhmstevo ] . indeed , sample selection is a particular important issue for these studies , as selection effects are almost inevitable @xcite . in particular , @xcite argued that intrinsic scatter in the @xmath0-bulge relation and a steep exponential cutoff in the galaxy distribution function will lead to biased apparent relations . this bias is exacerbated by the effects of a bright flux limit in the employed agn samples . in ( * hereafter sw11 ) we presented a common framework , based on the bivariate distribution function of black hole mass and spheroid property , to investigate and model the luminosity bias and other selection biases on the @xmath0-bulge relations . we additionally discussed an active fraction bias , which is introduced if only _ active _ black holes are selected and if the probability to be in an active stage depends on black hole mass . it is possible to account for these effects if ( 1 ) the selection function of the respective sample is well known , and ( 2 ) the underlying distribution functions , such as the galaxy distribution function , the active bhmf and the eddington ratio distribution function ( erdf ) , are known . however , in particular at high redshift this is currently not the case . therefore , we are facing the problem that even if we are able to obtain reasonably large agn samples with individual black hole mass estimates and host galaxy property measurements , the interpretation of such results in terms of evolution or non - evolution is challenging due to the sample selection effects . in this paper we present a practical solution to this dilemma . we build on the framework presented in sw11 and give a quantitative assessment of , and a correction procedure for sample selection effects on observations of the @xmath0-bulge relations . we outline a maximum likelihood fitting approach that directly incorporates information on the selection function into the fitting . we also restrict prior knowledge of the underlying distribution functions to a minimum . the method is able to reconstruct the true unbiased @xmath0-bulge relation for an observed sample and thus largely overcomes the limitations imposed by the sample selection . the paper is organized as follows : in section [ sec : method ] we present the general maximum likelihood fitting method and its application to the @xmath0-bulge relation for agn . in section [ sec : predict ] we verify the robustness and uncertainties of the method via monte carlo simulations . section [ sec : application ] illustrates the method by its application to representative previous studies and discusses its ramifications for the evolution of the @xmath0-bulge relation . we conclude in section [ sec : conclusions ] . while the main goal of the method presented here is to study the @xmath1bulge relation , it is also applicable to more general cases . identifying and characterizing correlations between two observables is a fundamental aspect of astrophysical research . given a joint distribution of two parameters with non - zero intrinsic scatter , this correlation will often be affected by the specific sample selection . we therefore first outline the general concept of the approach , before applying it to the @xmath1bulge relation . let us assume that there exists a correlation between two observational parameters @xmath6 and @xmath7 for a class of objects . the space density of these objects is given by a bivariate distribution function @xmath8 . marginalizing @xmath9 over @xmath6 gives the distribution function of the variable @xmath7 , and vice versa . in the case of the @xmath1bulge relation @xmath6 is for example the stellar velocity dispersion @xmath10 and @xmath7 is the black hole mass @xmath11 . thus marginalizing their bivariate distribution function over @xmath10 gives the bhmf , while marginalizing over @xmath11 will provide the galaxy stellar velocity dispersion function ( svdf ) . we here assume the following parameterization for the bivariate distribution @xmath12 where @xmath13 is the distribution function in @xmath6 , e.g. the galaxy svdf , and @xmath14 gives the probability of finding the value @xmath7 given @xmath6 . this is the intrinsic correlation between @xmath6 and @xmath7 . we here specifically assume a linear relation @xmath15 with log - normal intrinsic scatter @xmath16 , i.e. @xmath17 the bivariate distribution function @xmath8 describes the intrinsic joint distribution function of @xmath6 and @xmath7 in the universe . in practice , this will not be equal to the observed bivariate distribution for a sample , since this sample is almost always affected by implicit or explicit selection effects . we incorporate these selection criteria into a selection function @xmath18 , where @xmath19 refers to the set of additional parameters present as selection criteria , such as redshift or luminosity . the multivariate distribution @xmath20 for an observed sample is then given by @xmath21 where @xmath22 is a set of normalized distribution functions of the parameters @xmath19 . the respective bivariate distribution @xmath23 is given by integration over @xmath19 . in general @xmath24 . this means the correlation between @xmath6 and @xmath7 , e.g. the @xmath25 relation , fitted directly to the observations is not equal to the intrinsic correlation . ignoring this inequality will lead to a bias ( see sw11 ) . in practice , we are mainly interested in the intrinsic correlation between @xmath6 and @xmath7 , i.e. @xmath14 and less in the full joint distribution function . the information of this intrinsic correlation is contained in the conditional probability function , but in general is not equal to it . there are two options to define the observed correlation between @xmath6 and @xmath7 . first , the relation of @xmath7 at a given @xmath6 , and second the relation of @xmath6 at a given @xmath7 . these are given by the conditional probabilities @xmath26 @xmath27 where @xmath28 gives the probability of finding the property @xmath7 ( e.g. black hole mass ) in a galaxy with properties @xmath6 ( e.g. @xmath29 ) and @xmath19 , while @xmath30 is the probability of finding a galaxy with @xmath6 for given @xmath7 and @xmath19 . we will refer to these as forward and inverse probabilities , respectively . let us assume the special case there are no selection effects on @xmath7 . when we use the bivariate distribution given by equation ( [ eq : psi ] ) the forward conditional probability is given by @xmath31 , i.e. the intrinsic underlying relation is directly recovered in the conditional probability . for the inverse conditional probability it follows in the absence of any selection effects @xmath32 this is already a deviation from the intrinsic relation @xmath14 , even without applying any selection criteria to the sample . in the context of the @xmath1bulge relation the stellar mass function exponentially decreases at the high mass end , which together with the intrinsic scatter leads to a deviation of @xmath33 at the high mass end ( * ? ? ? this can already qualitatively explain the steeper slope found in the @xmath25 relation for galaxies with dynamical black hole mass measurements when fitting the inverse @xmath25 relation @xmath34 instead of the relation @xmath35 @xcite . in practice selection effects are present and can produce a bias , i.e. @xmath36 . however , if we know the selection function of our sample , we can incorporate this knowledge directly into the fitting of the conditional probability to recover the intrinsic relation @xmath14 . the maximum likelihood technique provides a conceptually simple way to achieve this . this method aims at minimizing the likelihood function @xmath37 , with @xmath38 being the product of the likelihoods for the individual measurements . both conditional probabilities , @xmath39 and @xmath40 , can serve as likelihoods , i.e. we either minimize @xmath41 @xmath42 to obtain the best - fitting parameters of @xmath14 . we here refer to the minimization of @xmath43 as forward ml regression and of @xmath44 as inverse ml regression . we emphasize that these maximum likelihood fits go beyond a simple linear regression , since we are not only fitting a model to the data , but incorporate additional information about the data set and the underlying distributions . thereby , the used conditional probability function can have an arbitrarily complex form . we will refer to the more standard approach of fitting the conditional probability @xmath45 without including the effects of sample selection as simple regression. which of the two maximum likelihood fit options ( forward / inverse ) is the preferred choice depends on the specific selection effects , and on the underlying distributions . for example , when there are no selection effects then forward regression is preferable , since the fitted conditional probability is equal to the intrinsic relation @xmath14 . in this case the maximum likelihood fit is equal to a simple regression approach , used in other studies to fit the @xmath25 relation for quiescent galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . the inverse ml regression ( using equation [ eq : cond_smu_int ] ) requires additional information about the distribution function @xmath13 and is therefore more complicated to use however , in the presence of selection effects on @xmath7 the inverse ml regression might become preferable , since it can be unaffected by the selection effects , in contrast to the forward ml regression . we will illustrate this below for the @xmath1bulge relation for active galaxies . while other applications of this method are possible , in the following we focus on the specific case of studying the @xmath1bulge relation at high redshift via broad line agn samples . in sw11 we discussed the qualitative influence of sample selection effects on observations of the @xmath1bulge relation for broad line agn samples in detail within our framework of the bivariate distribution function . the @xmath1bulge relation is given by @xmath46 , with @xmath47 being the logarithmic black hole mass and @xmath48 is the logarithm of the host galaxy property , e.g. @xmath10 or @xmath49 , and @xmath50 and @xmath51 are the normalization and the slope of the relation . thus in the above terminology @xmath52 and @xmath53 . with the parametrization of equation ( [ eq : psi ] ) , the bivariate distribution function is @xmath54 where @xmath55 is the spheroid distribution function and @xmath56 is the intrinsic @xmath1bulge relation . integrating this bivariate distribution function over @xmath48 gives the total bhmf . for an agn sample the main selection criteria will be the agn flux . therefore , the selection function will depend on agn luminosity and redshift . the agn luminosity is determined by the black hole mass @xmath0 and the accretion rate , the latter is represented by the eddington ratio @xmath57 , i.e. @xmath58 . this implies a selection function @xmath59 , depending on black hole mass , eddington ratio and redshift . the corresponding distribution function is the bivariate distribution function of black hole mass and eddington ratio for active black holes @xmath60 , with its projections the active bhmf and the erdf . we here make the assumption that the bivariate distribution function of black hole mass and eddington ratio is separable into an independent bhmf and erdf , i.e. @xmath61 where @xmath62 is the erdf , normalized to one and @xmath63 is the active bhmf . this assumption is motivated by results on the local bivariate distribution function of @xmath64 and @xmath65@xcite and by the observation of an eddington ratio distribution of agn that is independent of stellar mass @xcite , suggesting also an independence of @xmath64 [ but see @xcite for evidence of a @xmath64 dependence of the erdf ] . we comment on the consequences of relaxing this assumption below . the active bhmf is related to the total bhmf via the active fraction , which gives the probability for a black hole to be in an active ( type 1 agn ) state as a function of black hole mass and redshift . it is defined by the ratio of active bhmf to total bhmf , @xmath66 . the latter is directly linked to the bivariate distribution of @xmath64 and @xmath48 ( equation ( [ eq : psi_agn ] ) ) . following equation ( [ eq : psio ] ) , the multivariate probability distribution @xmath67 of galaxy property , black hole mass , eddington ratio and redshift is then given by @xmath68 where @xmath59 is the agn selection function of the respective sample . the conditional probability for the forward ml regression then follows as @xmath69 the erdf @xmath70 and the galaxy distribution function @xmath71 cancel out and are not required for the fitting . we require only knowledge of the selection function and of the mass dependence of the active fraction to obtain the intrinsic @xmath1bulge relation from the fit to the conditional probability . the free parameters @xmath50 , @xmath51 and @xmath16 in the intrinsic relation @xmath56 are obtained by minimizing equation [ eq : sx ] . if the erdf is mass dependent it will not cancel out and knowledge of the erdf is additionally required . second , the conditional probability for the inverse ml regression is @xmath72 this is identical to the case without agn selection effects ( equation [ eq : cond_smu_int ] ) , i.e. the conditional probability @xmath73 is not affected by the agn selection . we already discussed this fact in sw11 , but we will use it here to recover the intrinsic relation by the maximum likelihood fit . the only quantity that has to be known to recover the intrinsic relation is the galaxy distribution function . both fitting options have their pros and cons . while the latter is largely unaffected by agn selection effects , it is susceptible to the shape of the galaxy distribution function and also more susceptible to the intrinsic scatter in the relation itself . the former does not require any knowledge about @xmath71 , but demands a proper understanding of the sample selection and also of the underlying distribution function of agn or at least the active fraction for the agn sample . both regression methods constitute an improvement for the determination of the intrinsic @xmath1bulge relation , compared to a standard regression that does not include knowledge of the selection function or galaxy distribution function , as we will illustrate below . in practice , the situation becomes more complicated , since black hole mass and galaxy property measurements have measurement uncertainties . virial black hole masses are associated with a non - negligible uncertainty of @xmath74 dex @xcite , while also agn host galaxy properties at high @xmath75 have significant measurement errors . these lead to second - order dependencies of the conditional probabilities . in sw11 we already studied the consequences of measurement uncertainties on the bivariate distribution and the @xmath1bulge relations . here , we discuss the ramifications of measurement uncertainties on the presented fitting approach . we assume that the black hole mass estimated by the virial method @xmath76 is drawn from a log - normal probability distribution around the true black hole mass @xmath64 , with dispersion @xmath77 , representing the uncertainty in the virial mass estimate , @xmath78 for the galaxy property @xmath48 we also assume a log - normal error distribution @xmath79 with dispersion @xmath80 . the bivariate distribution function for bulge property and virial black hole mass is then @xmath81 the bivariate distribution function is smoothed out by the measurement uncertainties . the conditional probability for the forward ml regression is @xmath82 assuming we know the measurement uncertainties @xmath77 and @xmath80 , these can be directly incorporated into the fit . however , we now introduce a dependence on the galaxy distribution @xmath71 into the forward ml regression which was canceled out before . to avoid this dependence , we can make the simplifying assumption @xmath83 . this approximation is justified for a small measurement error in @xmath48 and in particular in the flat part of the galaxy distribution function . it will break down for high stellar velocity dispersion or high stellar mass , i.e. at the exponential decline of the galaxy distribution function . it can be applied to samples dominated by lower mass galaxies . equation [ eq : condprob_agn_mus ] then simplifies to @xmath84 we will refer to both options as forward ml regression with measurement errors . the conditional probability for the inverse ml regression is given by @xmath85 for this case , the dependence of the conditional probability on the active fraction and on the selection function is not canceled out as it was the case without measurement uncertainty in @xmath64 . this effectively prevents ignoring the agn selection function for the fit of the @xmath0-bulge relations for observational samples . we can again derive an approximation to the probability distribution by making the simplifying assumptions @xmath86 and @xmath87 , for which equation ( [ eq : condprob_agn_smu ] ) reduces to @xmath88 for this approximation no knowledge of the agn selection function and the active fraction is required anymore . the assumption on the selection function depends on the details of the sample selection , but in general will be true in most cases , unless the object is close to the selection threshold ( e.g. to the flux limit of the survey ) . the assumption on the active fraction requires an active fraction with a weak or no black hole mass dependence . this is , for example , the case at @xmath2 , as discussed in section [ sec : app_z15 ] , but not in the local universe @xcite . therefore , in practice some idea about the active fraction is required in any case to judge the validity of this approximation . we will refer to these maximum likelihood fits as inverse ml regression with measurement errors . we here presented several options to fit for the underlying relation using a given sample with measurement uncertainties . which of these is the best choice for the specific sample will depend on the details of the sample selection and our knowledge of this selection function and the galaxy distribution function . formally , the most precise results are obtained using the conditional probabilities for either the forward or inverse ml regression given by equations [ eq : condprob_agn_mus ] and [ eq : condprob_agn_smu ] . this requires knowledge of the selection function , the active fraction and of the galaxy distribution function . to restrict the assumptions for the respective fit , we can also use the approximations given by equations [ eq : condprob_agn_mus_simp ] and [ eq : condprob_agn_smu_simp ] . while not precise , this approach can be justified if the selection function and/or the underlying distributions are only poorly known . to illustrate the method and test its reliability we carried out monte carlo simulations of the @xmath4 relation . for our simulations we chose as representative redshift @xmath89 . for the galaxy spheroid mass function we assume the estimate given in sw11 , which is based on the stellar mass function from @xcite , applying a rough correction for the bulge - to - total ratio to translate it into a spheroid mass function . we set up the simulations by first drawing a large number of objects from the spheroid mass function at @xmath89 . black hole masses are attributed according to the @xmath90 relation from @xcite with log - normal intrinsic scatter @xmath91 . we here assumed a constant active fraction , consistent with recent results on the active bhmf at @xmath2 ( * ? ? ? * ; * ? ? ? * schulze et al . in prep . ) and empirical models @xcite . for the active black holes an eddington ratio was drawn from the @xmath2 erdf from @xcite . black hole mass and eddington ratio define the bolometric agn luminosity . we convert @xmath92 into @xmath93-band magnitude using the agn spectral energy distribution ( sed ) from @xcite and the @xmath94-correction from @xcite . we applied three flux limits to the simulated samples , @xmath95=20 , 22 and 24 mag , respectively , and generated samples of various sizes . for each flux limit and sample size we generated 1000 monte carlo realizations . for simplicity , we first discuss the results without measurement errors . we will discuss the effect of adding measurement errors further below . each simulation realization was fitted by a forward and an inverse ml regression . furthermore , for each regression we fitted the realizations with a different set of free parameters . we first kept all three parameters ( @xmath50 , @xmath51 , @xmath16 ) free . secondly we fixed @xmath16 and finally we also fixed @xmath51 to the input value , leaving only the normalization as free parameter . additionally , we fitted each realization with a simple linear regression , i.e. not including the selection effects , both with @xmath64 and with @xmath48 as dependent variable . in fig . [ fig : mcfit ] , we illustrate the strength of our proposed fitting approach ( shown as black histograms , left for the forward ml regression and right for the inverse ml regression ) , compared to a simple linear regression ( shown as green histograms ) . we use 1000 realizations of a sample of @xmath96 objects at @xmath89 with @xmath97 mag . while the dispersion on the fitting parameters is increased , the bias present in the simple regression approach is removed . in this example we leave all three parameters free in the fit . in fig . [ fig : statsx ] we show the median and 68% confidence limits of the 1000 realizations for a forward ml regression for different flux limits and sample sizes . the magnitude limits of @xmath98 mag correspond to bolometric luminosity limits of @xmath99 , respectively . the left panel shows that we systematically underestimate @xmath16 for a small sample size when keeping it as a free parameter . this is due to the fact that the fitting procedure determines the _ sample _ standard deviation , which is known to be an underestimate of the _ population _ standard deviation . this underestimate for small sample sizes is already present in the conventional fit without including the selection function . since the intrinsic scatter is an important parameter , when the selection effects are included this underestimate is enhanced . also the best - fitting solutions for the normalization @xmath50 and the slope @xmath51 will be biased . therefore , we suggest to fix the intrinsic scatter for small sample sizes , @xmath100 , to avoid this bias . in the following we fix the intrinsic scatter @xmath16 in the maximum likelihood fit . in this case , the slope @xmath51 and normalization @xmath50 are recovered unbiased irrespective of sample size , as shown in the other three panels of fig . [ fig : statsx ] . increasing the sample size mainly reduces the statistical errors , shown by the dashed lines . fixing also the slope and keeping the normalization as only free parameter further reduces the statistical error , while the input is recovered in the mean . we further see that the error is basically independent of the flux limit . however , this assumes perfect knowledge of the selection function and underlying distribution . deeper samples are less susceptible to uncertainties in these and therefore come closer to the statistical errors shown here . in fig . [ fig : statsy ] we show the same for the inverse ml regression . keeping the intrinsic scatter as a free parameter also leads to an underestimate for small sample sizes and we again suggest to fix it for small samples , @xmath100 , to avoid a bias in the slope and normalization . but even when fixing @xmath16 we find in our simulations a small bias in @xmath50 and @xmath51 for very small samples , @xmath101 . however , for such small samples the errors on the best fit are also large . a joint determination of slope and normalization of the relation requires a reasonably large sample . when fixing the slope @xmath51 and only determining the normalization @xmath50 , an unbiased result with small error can be achieved even for relatively small sample sizes , as demonstrated in the right panel of fig . [ fig : statsy ] . we now investigate the consequences of adding measurement errors in the virial black hole masses and in the host galaxy properties to the simulations . we here test our prescription for including these uncertainties , as presented in section [ sec : mu ] and compare them to a regression without measurement errors . we added a log - normal random error of dispersion @xmath102 dex to the black hole mass and a log - normal error of dispersion @xmath103 dex to the spheroid mass of our monte carlo realizations at @xmath89 . for each realization we fixed @xmath51 and @xmath16 and only fit for the normalization @xmath50 . we fitted these with the forward ml regression with measurement errors , i.e. using equation [ eq : condprob_agn_mus ] . the difference to the inverse ml regression with measurement errors is marginal , in particular when the errors in @xmath64 and @xmath48 are of the same magnitude . to test the importance of the full correction for measurement errors , we also fitted the same realisations with the forward and inverse ml regression without measurement errors . the results are shown in fig . [ fig : stats_error ] . we show the median ( filled symbols ) , the 68% confidence limits ( solid lines ) and also the 68% confidence limits for the realizations with zero error in @xmath64 and @xmath48 ( dashed lines ) . in the left panel the result for the forward ml regression with measurement errors is presented . the input value is recovered in the median , while the 1@xmath16 confidence region ( solid lines ) is increased by an approximately constant factor of @xmath104 for our simulation , compared to the case without measurement errors ( dashed lines ) . the fitting method provides unbiased results if selection function , underlying distributions and the value of the measurement error are all known . in the middle and right panels of fig . [ fig : stats_error ] we show the results of applying the forward ml regression and inverse ml regression without measurement errors to the simulations that have errors added . while in the forward ml regression the zero - point @xmath50 is slightly biased towards a lower value , the inverse ml regression results are fully consistent with the input value , only showing an increase in the 1@xmath16 confidence range . thus , not fully including the measurement errors has only a small , second order effect on the best - fitting solution . the monte carlo simulations verify the reliability of the presented maximum likelihood fitting approach to recover the intrinsic @xmath4 relation in the median from a sample affected by sample selection effects . a simple linear regression would provide a biased result in these cases . the method is still reliable if there are measurement uncertainties in the black hole mass and galaxy property . while the most precise results are obtained when the uncertainty is well known and taken into account , over- or underestimating or even ignoring the measurement uncertainty will only have a small effect on the recovery of the intrinsic relation . we now apply our method to a few observational studies from the literature that investigated the evolution in the @xmath1bulge relations . this is meant for illustration of the methodology and constraining power of the approach . an exhaustive investigation of the literature on evolution in the @xmath1bulge relations in consideration of selection effects is beyond the scope of this paper . we note that such an investigation would also be hampered by the fact that many previous literature studies did not use well - defined samples , or at least samples for which a selection function ( however complicated ) could be defined . first , we reinvestigated the qso sample from @xcite . they studied the @xmath4 relation for 89 broad line agn from zcosmos in the redshift range @xmath105 with estimated stellar masses for the host galaxies ( including upper limits ) . the selection function is to first order defined by an @xmath106-band flux limit of 22.5 mag . to fit their @xmath4 relation we need additional information about the underlying distribution functions . we use the same assumptions as above for the monte carlo realizations , i.e. the stellar mass function from @xcite , and a constant active fraction . to investigate the degree of evolution , we assume a redshift evolution in the zero - point @xmath107 . for the local relation we use first the relation by @xcite , with an intrinsic scatter of @xmath108 , consistent with the work by @xcite , and second the recent results by @xcite , with @xmath91 . upper limits are incorporated as described in appendix [ sec : upperlimits ] . for the forward ml regression ( using equation [ eq : condprob_agn_mus ] ) , assuming measurement errors of @xmath109 dex in the stellar mass and @xmath110 dex in the black hole mass estimates , we get @xmath111 , using the @xcite local relation . using the relation from @xcite as zero - point , we find @xmath112 ( i.e. formally a negative evolution ) . for the inverse ml regression and when using approximations for the probability distribution ( equations [ eq : condprob_agn_mus_simp ] and [ eq : condprob_agn_smu_simp ] ) we get consistent results . the null hypothesis of no evolution in the @xmath113relation for both cases lies within @xmath114 of our results , thus it is not rejected with statistical significance . the best fit relation , using the @xcite local relation , is shown in fig . [ fig : fitz_merloni10 ] . note that the given errors are statistical errors from the @xmath115 contour , while uncertainties in the assumptions for the zero - point , the flux limit and the distribution functions are not included . this implies that the uncertainty estimates above are rather optimistic , since they only include statistical uncertainties and not additional uncertainties in the underlying assumptions . the likelihood for the null hypothesis of no evolution is then even higher than the @xmath114 mentioned above when those uncertainties are also included in the confidence determination . to get an estimate for the effect of uncertainties in the underlying distribution function , we varied the assumed stellar mass function from @xcite within their quoted uncertainties . furthermore , we varied the bulge - to - total ratio between our default correction and no correction at all , to investigate the importance of the transformation of the galaxy mass function to a spheroid mass function . the uncertainty in the stellar mass function adds an uncertainty of @xmath116 to the best - fitting @xmath117 , while not applying an bulge - to - total correction increases the best - fitting solution by @xmath118 . even more important is a precise calibration of the local zero - point , as it is illustrated by the opposite trends of evolution found for the local relations by @xcite and @xcite , which is mainly caused by the increase in the zero - point of the @xmath119 relation of @xcite . while the local zero - point of the @xmath119 relation increased in the most recent revised determinations of the relation @xcite , the @xmath120 factor and thus the calibration to the agn samples is not strongly affected by this revision @xcite . using this revised @xmath119 relation significantly reduces the apparent trend of positive evolution , a fact recently noted by @xcite . combined with the correction for the bias caused by sample selection this even suggests or is at least consistent with a negative evolution of the @xmath119 relation where galaxy bulges grow first and the black holes have to catch up . in summary , we see no statistically significant evidence for evolution in the @xmath113relation based on this data set , consistent with our more qualitative analysis in sw11 . studies of the most luminous qsos , observed at high redshift found the largest apparent offsets from the @xmath1bulge relations , by a factor of 10 or more @xcite . at first glance this seems to provide the most convincing evidence for strong evolution in the @xmath1bulge relations . however , these luminous qso studies are only sampling from the bright end of the luminosity function . for this kind of objects the bias caused by selection effects is most severe , as discussed in sw11 . furthermore , current sample sizes at @xmath121 are small , inhibiting firm statistical conclusions . we now endeavor to test quantitatively if the strong apparent offset found in these studies provides any statistically significant evidence for an intrinsic offset , once we account for sample selection . we focus on the highest redshift regime currently studied @xmath3 and the constraints that can be drawn on evolution in the @xmath1bulge relations . we used the sample from @xcite , and restricted it to the six objects selected from the sloan digital sky survey ( sdss ) @xmath122 quasar survey @xcite , which approximately might be considered a well - defined sample . for this sample we assume a selection function given by the survey magnitude limit of @xmath123 mag . black hole masses are for some of the objects estimated from the broad lines and for the rest from the bolometric luminosity , assuming eddington limited accretion . bulge masses are derived from dynamical mass measurements ( we assume an average inclination angle of @xmath124 ) , and stellar velocity dispersions were estimated from the width of molecular co line emission . there are still significant systematic uncertainties associated with these measurements and how they relate to galaxy masses and velocity dispersion measured for local galaxies . however , we here only discuss the statistical aspects of the sample . the dynamical range in black hole mass and agn luminosity covered by this sample is small and there is basically no apparent correlation between @xmath0 and either @xmath29 or @xmath125 . we therefore fixed the slope and intrinsic scatter and only fit for an evolution in the normalization , again parametrized by a parameter @xmath117 . for the stellar mass function we assume the mass function for star forming galaxies at @xmath126 from @xcite . we again assume a constant active fraction . using the forward regression with measurement errors we find @xmath127 at @xmath128 confidence , with a best fit of @xmath129 for the @xmath113relation , using the local relation from @xcite . we also varied the mass function according to their uncertainties , as given in @xcite . the best - fitting @xmath130 ranges from @xmath131 to @xmath132 with an @xmath128 confidence interval @xmath133 $ ] . for the local relation from @xcite as zero - point we find a best fit of @xmath134 with @xmath135 at @xmath128 confidence . at any rate the inferred value of @xmath117 is always consistent with zero . assuming a svdf is more problematic . the svdf is well determined for @xmath136 @xcite and there are estimates out to @xmath137 @xcite , but it is observationally unconstrained at higher redshifts . on the other hand , it is not clear at all to what degree the co line width traces the stellar velocity distribution at these redshift . we here simply try to bracket our ignorance about the svdf at @xmath3 and investigate their consequences on the results for the @xmath25 relation . note that for the correction only the shape of the distribution function is important , in particular the location of the schechter function cutoff , not its normalization . we started from the evolution of the stellar mass function between @xmath136 and @xmath126 , comparing the mass functions from @xcite and @xcite . they show a decrease of the break of the mass function of @xmath138 dex . assuming virialized systems , we can convert this into a shift for the svdf . however , this also requires a parameterization of the size evolution of galaxies . at @xmath5 massive galaxies are more compact by a factor of @xmath139 , compared to local galaxies of similar mass ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? assuming similar compact galaxies at @xmath140 this approximately balances out the evolution in the mass function , i.e. no evolution in the break of the svdf would be required , consistent with observations out to @xmath89 @xcite . we use the local svdf from @xcite and converted it to a rough guess for the shape of the svdf at @xmath141 , testing two extreme evolution scenarios : ( 1 ) no effective evolution in the svdf , due to an approximate balance between mass evolution and size evolution of galaxies , ( 2 ) no size evolution , leading to a decrease in the break of the svdf , consistent with the mass function evolution . for the first case we find @xmath142 at @xmath128 confidence , with a best fit of @xmath143 , compared to the local @xmath25 relation of @xcite , while the second case gives @xmath144 at @xmath128 confidence , with a best fit of @xmath145 . in fig . [ fig : fitz_wang10 ] , we show the data and the best - fitting results for the @xmath146 and the @xmath25 relation . while the uncertainties in the parameters are large , due to both the small number statistics and the uncertainty in the underlying distributions , the results are fully consistent with no evolution in the @xmath1bulge relations . thus , even while these @xmath3 qsos show large apparent offsets from the local relation , this can not be taken as evidence for intrinsic evolution in the @xmath1bulge relations up to high @xmath75 . the apparent offset is consistent with being fully caused by agn sample selection effects . even negative evolution in the @xmath1bulge relations at @xmath3 is fully consistent with these data . besides current methodical challenges in determining black hole masses and galaxy properties at @xmath3 , the significance on the @xmath1bulge relation at high @xmath75 is currently limited by small sample sizes and their bright flux limits . ongoing and future surveys , like ukidss , pan - starrs , vista , sumire / hyper - suprime cam , lsst will significantly increase the number of known @xmath3 qsos and future sub - mm observations of their host galaxies , in particular using the atacama large millimeter / submillimetre array ( alma ) , will open a new window to the study of their host galaxies and are able to obtain dynamical masses from molecular line emission for a reasonably large sample of high-@xmath75 qsos @xcite . while we here do not aim at a full investigation of the literature on the evolution of the @xmath1bulge relation , we want to summarize the current constraints we have on evolution or non - evolution . we base this on samples that are representative for the best data at their respective redshift , in terms of depth and well defined sample selection . we include the samples from @xcite and @xcite discussed above . additionally , we augmented this by the samples from @xcite at @xmath147 and @xcite at @xmath148 . @xcite studied a sample of 32 agn at @xmath149 , selected from _ xmm_-cosmos and report basically no offset from the local relation from @xcite for their sample . using the _ xmm_-cosmos soft x - ray flux limit , the stellar mass function from @xcite , and assuming a constant active fraction , we find an insignificant bias for this sample . thus , our inferred intrinsic relation is also fully consistent with the local relation from @xcite . however , compared to the local relation from @xcite the sample is offset by @xmath150 dex . this might indicate a systematic offset in the normalization of the virial black hole mass estimates , for example due to selection effects ( sw11 ) . in the following we use the @xmath1bulge relation from @xcite as local zero - point as a conservative constraint on possible positive evolution . @xcite could resolve the host galaxies for two luminous qsos at @xmath151 , selected from the sdss , with an additional selection criteria @xmath152 mag . to account for the selection effects , we applied this absolute magnitude limit and assume the mass function for star - forming galaxies at @xmath148 from @xcite and a constant active fraction . the best - fitting result for the forward regression is @xmath153 . in fig . [ fig : bhevo_intrinsic ] we summarize our study of the _ intrinsic _ evolution in the @xmath1bulge relation for these four samples . the _ apparent _ offset from the local relation is shown by the open symbols , while our inferred _ intrinsic _ offset is given by the filled symbols . while there is an apparent evolution trend , accounting for selection effects we find an _ intrinsic _ offset that is fully consistent with a non - evolving @xmath1bulge relation out to @xmath3 . however , we note that at @xmath154 the uncertainties on the @xmath1bulge relation are still substantial and future work in particular in this redshift regime is necessary to obtain robust results . for reference we also indicate the case of @xmath155 , predicted by some numerical simulations @xcite , by the dashed dotted line in fig . [ fig : bhevo_intrinsic ] . current data are not able to robustly distinguish between a @xmath155 scenario and a non - evolving @xmath1bulge relation . while the redshift evolution of the @xmath1bulge relation provides important constraints on black hole - galaxy co - evolution , its observational study is hampered by severe selection effects . we presented a novel fitting method to the @xmath1bulge relations and their evolution . this method corrects for the bias on the @xmath1bulge relation caused by sample selection and is able to recover the intrinsic relation in the fitting . it does so by incorporating the selection function of the observational sample and the underlying distribution function into the conditional probability distribution of black hole mass at given bulge property or vice versa . this probability distribution is derived from the bivariate distribution function of @xmath0 and bulge property . the intrinsic relation is reconstructed via a maximum likelihood fit of this conditional probability distribution to the data . since we are not modeling or fitting the full bivariate distribution function but only the conditional probability distribution , we can reduce the set of prior knowledge or assumptions to correct for the selection bias , compared to our discussion in sw11 . we presented two routes to fit the conditional probability : ( 1 ) using the probability of finding @xmath0 at a given bulge property ( forward regression ) , and ( 2 ) using the probability of finding a specific value for the bulge property at a given @xmath0 ( inverse regression ) . for the first case it is essential to properly understand the selection of the sample and have a fair estimate of the black hole mass dependence of the active fraction . for the second case the bulge distribution function and the intrinsic scatter of the @xmath1bulge relation need to be known . when measurement errors on @xmath0 and the bulge property can not be ignored , all of these prior assumptions have to be known for a full correction . we extensively tested the fitting method on monte carlo simulations and verified the reliability of the approach to recover evolution in the normalization unbiased . we also illustrated the method by applying it to literature data from @xmath147 to @xmath3 . the sample of @xcite at @xmath2 already puts significant constraints on the redshift evolution of the @xmath4 relation . assuming the local relation from @xcite , we can constrain the evolution in the @xmath4 relation to @xmath156 out to redshift @xmath5 at @xmath157 confidence . this is in general agreement with our conclusions in sw11 and with recent results at lower redshifts ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? taking these empirical clues together , it becomes apparent that there is currently no evidence for significant positive evolution in the @xmath1bulge relation out to @xmath5 . either black holes and galaxies grow almost co - eval since @xmath5 , or galaxies grow faster and the black holes would have to catch up to end up at the local relation at @xmath136 . the latter is also suggested by observations of @xmath5 smgs @xcite . within @xmath158 most of the black hole growth is governed by secular processes and not by major mergers @xcite . these processes probably grow the black hole at the same rate as the spheroid component . this changes at @xmath154 , beyond the peak of agn activity , where major mergers likely become the dominating black hole growth mechanism @xcite . therefore also a change in the evolution of the @xmath4 relation could be expected at @xmath154 . currently , this redshift regime is still poorly covered by observations , and usually restricted to the most luminous , and hence most biased , qsos . we illustrated the poor constraints that can be drawn from current observations , employing a @xmath3 qso sample based on @xcite . while the sample shows a large _ apparent _ positive offset from the local relation , we find here that there is no _ intrinsic _ offset , i.e. it is fully consistent with no or even negative evolution when sample selection is taken into account . we therefore conclude that even at @xmath154 and in particular at @xmath3 we currently have no statistically significant evidence for evolution in the @xmath4 relation . we confirmed these results by additionally studying the low redshift sample from @xcite and the small @xmath148 sample from @xcite . future studies will reduce the observational uncertainties and increase the sample size and will allow a better assessment of the @xmath4 relation at these early cosmic epochs . we will discuss the requirements on the design of these future studies of the @xmath4 relation in this redshift regime in a companion paper . while selection effects severely complicate the observational study of the @xmath1bulge relation at high redshift , these can be included in the analysis as we have demonstrated here . we anticipate that the method presented here will contribute to an improved understanding of black hole - galaxy co - evolution . however , essential prerequisites are ( 1 ) the use of a well defined sample , i.e. the selection function for the sample has to be under control , and ( 2 ) knowledge of the underlying demographics of galaxies and active black holes over the black hole mass and galaxy property range covered by the observations . given the expected future increase in the size of well defined samples and a better understanding of the underlying black hole and galaxy demographics , we anticipate that we will be able to trace back the cosmic history of the @xmath1bulge relations in the future . we thank knud jahnke for fruitful discussions on this work . we also thank the anonymous referee for an efficient and constructive report that strengthened our conclusions . we acknowledge support by the deutsche forschungsgemeinschaft under its priority programme spp1177 , grant wi 1369/23 . as acknowledges support by the china postdoctoral scientific foundation , grant 2012m510256 . aird , j. , coil , a. l. , moustakas , j. , et al . 2012 , apj , 746 , 90 alexander , d. m. , brandt , w. n. , smail , i. , et al . 2008 , aj , 135 , 1968 angls - alczar , d. , zel , f. , & dav , r. 2013 , apj , 770 , 5 , e. f. , mcintosh , d. h. , katz , n. , & weinberg , m. d. 2003 , apjs , 149 , 289 , v. n. , treu , t. , woo , j. , et al . 2010 , apj , 708 , 1507 bezanson , r. , van dokkum , p. g. , franx , m. , et al . 2011 , apj , 737 , l31 bongiorno , a. , merloni , a. , brusa , m. , et al . 2012 , mnras , 427 , 3103 , c. m. & schaye , j. 2011 , mnras , 413 , 1158 borys , c. , smail , i. , chapman , s. c. , et al . 2005 , apj , 635 , 853 canalizo , g. , wold , m. , hiner , k. d. , et al . 2012 , apj , 760 , 38 carrera , f. j. , page , m. j. , stevens , j. a. , et al . 2011 , mnras , 413 , 2791 chae , k .- h . 2010 , mnras , 402 , 2031 cisternas , m. , jahnke , k. , bongiorno , a. , et al . 2011 , apj , 741 , l11 cisternas , m. , jahnke , k. , inskip , k. j. , et al . 2011 , apj , 726 , 57 , d. j. 2006 , mnras , 369 , 1808 daddi , e. , renzini , a. , pirzkal , n. , et al . 2005 , apj , 626 , 680 , r. , falomo , r. , treves , a. , et al . 2010 , mnras , 402 , 2453 , t. , colberg , j. , springel , v. , hernquist , l. , & sijacki , d . 2008 , apj , 676 , 33 , t. , springel , v. , & hernquist , l. 2005 , nature , 433 , 604 draper , a. r. , & ballantyne , d. r. 2012 , apj , 751 , 72 dubois , y. , devriendt , j. , slyz , a. , & teyssier , r. 2012 , mnras , 420 , 2662 , x. , narayanan , v. k. , lupton , r. h. , et al . 2001 , aj , 122 , 2833 , x. , strauss , m. a. , richards , g. t. , et al . 2006 , aj , 131 , 1203 , x. , strauss , m. a. , schneider , d. p. , et al . 2003 , aj , 125 , 1649 , x. , white , r. l. , davis , m. , et al . 2000 , aj , 120 , 1167 , l. & merritt , d. 2000 , apj , 539 , l9 , a. , salimbeni , s. , grazian , a. , et al . 2006 , a&a , 459 , 745 , k. , bender , r. , bower , g. , et al . 2000 , apj , 539 , l13 gebhardt , k. , kormendy , j. , ho , l. c. , et al . 2000 , apj , 543 , l5 graham , a. w. , onken , c. a. , athanassoula , e. , & combes , f. 2011 , mnras , 412 , 2211 , k. , richstone , d. o. , gebhardt , k. , et al . 2009 , apj , 698 , 198 , n. & rix , h .- w . 2004 , apj , 604 , l89 hiner , k. d. , canalizo , g. , wold , m. , brotherton , m. s. , & cales , s. l. 2012 , apj , 756 , 162 , l. c. 2007 , apj , 669 , 821 , p. f. , robertson , b. , krause , e. , hernquist , l. , & cox , t. j. 2006 , apj , 652 , 107 , k. , bongiorno , a. , brusa , m. , et al . 2009 , apj , 706 , l215 , k. , elbaz , d. , pantin , e. , et al . 2009 , apj , 700 , 1820 , k. & macci , a. v. 2011 , apj , 734 , 92 , g. & haehnelt , m. 2000 , mnras , 311 , 576 kelly , b. c. , & shen , y. 2013 , apj , 764 , 45 kocevski , d. d. , faber , s. m. , mozena , m. , et al . 2012 , apj , 744 , 148 kormendy , j. , & ho , l. c. 2013 , arxiv:1304.7762 , a. , menci , n. , maiolino , r. , fiore , f. , & merloni , a. 2010 , mnras , 405 , 29 , t. r. , tremaine , s. , richstone , d. , & faber , s. m. 2007 , apj , 670 , 249 lee , k .- s . , ferguson , h. c. , wiklind , t. , et al . 2012 , apj , 752 , 66 , p. , letawe , g. , courbin , f. , et al . 2005 , nature , 437 , 381 , j. , tremaine , s. , richstone , d. , et al . 1998 , aj , 115 , 2285 maiolino , r. , neri , r. , beelen , a. , et al . 2007 , a&a , 472 , l33 , a. & hunt , l. k. 2003 , apj , 589 , l21 , a. , risaliti , g. , gilli , r. , et al . 2004 , mnras , 351 , 169 mcconnell , n. j. , & ma , c .- p . 2013 , apj , 764 , 184 , r. j. & jarvis , m. j. 2002 , mnras , 337 , 109 , r. j. , jarvis , m. j. , targett , t. a. , dunlop , j. s. , & best , p. n. 2006 , mnras , 368 , 1395 , a. , rudnick , g. , & di matteo , t. 2004 , mnras , 354 , l37 , a. , bongiorno , a. , bolzonella , m. , et al . 2010 , apj , 708 , 137 , a. & heinz , s. 2008 , mnras , 388 , 1011 , n. p. h. , de breuck , c. , lehnert , m. d. , et al . 2011 , a&a , 525 , a43 + nobuta , k. , akiyama , m. , ueda , y. , et al . 2012 , apj , 761 , 143 park , d. , kelly , b. c. , woo , j .- h . , & treu , t. 2012 , apjs , 203 , 6 , c. y. 2007 , apj , 671 , 1098 , c. y. , impey , c. d. , rix , h. , et al . 2006 , apj , 649 , 616 portinari , l. , kotilainen , j. , falomo , r. , & decarli , r. 2012 , mnras , 420 , 732 richards , g. t. , lacy , m. , storrie - lombardi , l. j. , et al . 2006 , apjs , 166 , 470 richards , g. t. , strauss , m. a. , fan , x. , et al . 2006 , aj , 131 , 2766 , d. a. , walter , f. , brewer , b. j. , et al . 2008 , apj , 686 , 851 , b. , hernquist , l. , cox , t. j. , et al . 2006 , apj , 641 , 90 , s. , shields , g. a. , gebhardt , k. , & bonning , e. w. 2007 , apj , 662 , 131 salviander , s. , & shields , g. a. 2013 , apj , 764 , 80 schawinski , k. , treister , e. , urry , c. m. , et al . 2011 , apj , 727 , l31 , m. , wisotzki , l. , & jahnke , k. 2008 , a&a , 478 , 311 schramm , m. , & silverman , j. d. 2013 , apj , 767 , 13 schulze , a. , & gebhardt , k. 2011 , apj , 729 , 21 , a. & wisotzki , l. 2010 , a&a , 516 , a87 + , a. & wisotzki , l. 2011 , a&a , 535 , a87 , f. , bernardi , m. , & haiman , z. 2009 , apj , 694 , 867 , y. & kelly , b. c. 2010 , apj , 713 , 41 , r. k. , bernardi , m. , schechter , p. l. , et al . 2003 , apj , 594 , 225 , g. a. , gebhardt , k. , salviander , s. , et al . 2003 , apj , 583 , 124 , d. , springel , v. , di matteo , t. , & hernquist , l. 2007 , mnras , 380 , 877 , j. & rees , m. j. 1998 , a&a , 331 , l1 , r. s. , hopkins , p. f. , cox , t. j. , robertson , b. e. , & hernquist , l. 2008 , mnras , 391 , 481 targett , t. a. , dunlop , j. s. , & mclure , r. j. 2012 , mnras , 420 , 3621 , s. , gebhardt , k. , bender , r. , et al . 2002 , apj , 574 , 740 trujillo , i. , conselice , c. j. , bundy , k. , et al . 2007 , mnras , 382 , 109 van dokkum , p. g. , franx , m. , kriek , m. , et al . 2008 , apj , 677 , l5 , m. & peterson , b. m. 2006 , apj , 641 , 689 volonteri , m. , & stark , d. p. 2011 , mnras , 417 , 2085 , f. , carilli , c. , bertoldi , f. , et al . 2004 , apj , 615 , l17 , r. , carilli , c. l. , neri , r. , et al . 2010 , apj , 714 , 699 wang , r. , wagg , j. , carilli , c. l. , et al . 2013 , apj , 773 , 44 willott , c. j. , omont , a. , & bergeron , j. 2013 , apj , 770 , 13 , j .- h . , treu , t. , malkan , m. a. , & blandford , r. d. 2008 , apj , 681 , 925 woo , j .- h . , schulze , a. , park , d. , et al . 2013 , apj , 772 , 49 zhang , x. , lu , y. , & yu , q. 2012 , apj , 761 , 5 for agn samples , per definition the black hole is always detected and its mass is estimated via the virial method . however , the host galaxy is not always detected , as it can be outshined by the agn . but , if at least upper limits on the host galaxy property can be set , these objects can be included in the fitting . otherwise their exclusion needs to be reflected in the selection function . we include upper limits on @xmath6 in the following manner , similar to @xcite . first , we compute the probability @xmath159 that a black hole with mass @xmath64 is located in a galaxy with their respective property greater than @xmath160 . for the forward regression without measurement errors we simply assume a uniform distribution in @xmath48 , @xmath161 for the inverse regression without measurement errors we get @xmath162 when we include measurement errors in @xmath48 and @xmath64 , the probability distribution for the forward regression is @xmath163 the likelihood for an upper limit at @xmath128 confidence level is @xmath164 where @xmath159 is one of the above conditional probabilities and @xmath165 is the probability to find the galaxy having @xmath48 greater than @xmath160 . for an upper limit at a different confidence level this probability changes accordingly .

synchronization is a dynamical process where two or more interacting oscillatory systems end up with identical movement . in 1665 huygens experimented with maritime pendulum clocks and discovered the anti - phase synchronization of two pendulums clocks mounted on the common frame @xcite . since then , synchronization has become a basic concept in nonlinear and complex systems @xcite . such systems include , but are not limited to , musical instruments , electric power systems , and lasers . there are numerous applications in mechanical @xcite and electrical @xcite engineering . new applications are found in mathematical biology such as synchronous variation of cell nuclei , firing of neurons , forms of cooperative behavior of animals and humans @xcite . recently , huygen s experiment has been widely discussed and several experimental devices were built @xcite . it was shown that two real mechanical clocks when mounted to a horizontally moving beam can synchronize both in - phase and anti - phase @xcite . in all these experiments synchronization was achieved due to energy transfer via the oscillating beam , supporting huygen s intuition @xcite . one of the rapidly developing areas in between physics and mathematics , is the topic of @xmath1-symmetry , which has started as a way to characterize non - hermitian hamiltonians in quantum mechanics @xcite . the key idea is that a linear schrdinger operator with a complex - valued potential , which is symmetric with respect to combined parity ( @xmath2 ) and time - reversal(@xmath3 ) transformations , may have a real spectrum up to a certain critical value of the complex potential amplitude . in nonlinear systems , this distinctive feature may lead to existence of breathers ( time - periodic solutions localized in space ) as continuous families of their energy parameter . the most basic configuration having @xmath1 symmetry is a dimer , which represents a system of two coupled oscillators , one of which has damping losses and the other one gains some energy from external sources . this configuration was studied in numerous laboratory experiments involving electric circuits @xcite , superconductivity @xcite , optics @xcite and microwave cavities @xcite . in the context of synchronization of coupled oscillators in a @xmath1-symmetric system , one of the recent experiments was performed by bender et al . these authors considered a @xmath1-symmetric hamiltonian system describing the motion of two coupled pendula whose bases were connected by a horizontal rope which moves periodically in resonance with the pendula . the phase transition phenomenon , which is typical for @xmath1-symmetric systems , happens when some of the real eigenvalues of the complex - valued hamiltonian become complex . the latter regime is said to have _ broken _ @xmath1 symmetry . on the analytical side , dimer equations were found to be completely integrable @xcite . integrability of dimers is obtained by using stokes variables and it is lost when more coupled nonlinear oscillators are added into a @xmath0-symmetric system . nevertheless , it was understood recently @xcite that there is a remarkable class of @xmath0-symmetric dimers with cross - gradient hamiltonian structure , where the real - valued hamiltonians exist both in finite and infinite chains of coupled nonlinear oscillators . analysis of synchronization in the infinite chains of coupled oscillators in such class of models is a subject of this work . in the rest of this section , we describe how this paper is organized . we also describe the main findings obtained in this work . section [ model ] introduces the main model of coupled pendula driven by a resonant periodic movement of their common strings . see figure [ pendulapic ] for a schematic picture . by using an asymptotic multi - scale method , the oscillatory dynamics of coupled pendula is reduced to a @xmath0-symmetric discrete nonlinear schrdinger ( dnls ) equation with gains and losses . this equation generalizes the dimer equation derived in @xcite . section [ symmetry ] describes symmetries and conserved quantities for the @xmath0-symmetric dnls equation . in particular , we show that the cross - gradient hamiltonian structure obtained in @xcite naturally appears in the asymptotic reduction of the original hamiltonian structure of newton s equations of motion for the coupled pendula . section [ breathers ] is devoted to characterization of breathers , which are time - periodic solutions localized in the chain . we show that depending on parameters of the model ( such as detuning frequency , coupling constant , driving force amplitude ) , there are three possible types of breather solutions . for the first type , breathers of small and large amplitudes are connected to each other and do not extend to symmetric synchronized oscillations of coupled pendula . in the second and third types , large - amplitude and small - amplitude breathers are connected to the symmetric synchronized oscillations but are not connected to each other . see figure [ branches ] with branches ( a ) , ( b ) , and ( c ) , where the symmetric synchronized oscillations correspond to the value @xmath4 and the breather amplitude is given by parameter @xmath5 . section [ zero - equilibrium ] contains a routine analysis of linear stability of the zero equilibrium , where the phase transition threshold to the broken @xmath0-symmetry phase is explicitly found . breathers are only studied for the parameters where the zero equilibrium is linearly stable . section [ variational ] explores the hamiltonian structure of the @xmath0-symmetric dnls equation and characterizes breathers obtained in section [ breathers ] from their energetic point of view . we show that the breathers for large value of parameter @xmath6 appear to be saddle points of the hamiltonian function between continuous spectra of positive and negative energy , similar to the standing waves in the dirac models . therefore , it is not clear from the energetic point of view if such breathers are linearly or nonlinearly stable . on the other hand , we show that the breathers for smaller values of parameter @xmath6 appear to be saddle points of the hamiltonian function with a negative continuous spectrum and finitely many ( either three or one ) positive eigenvalues . section [ stability ] is devoted to analysis of spectral and orbital stability of breathers . for spectral stability , we use the limit of small coupling constant between the oscillators ( the same limit is also used in sections [ breathers ] and [ variational ] ) and characterize eigenvalues of the linearized operator . the main analytical results are also confirmed numerically . see figure [ spectra ] for the three types of breathers . depending on the location of the continuous spectral bands relative to the location of the isolated eigenvalues , we are able to prove nonlinear orbital stability of breathers for branches ( b ) and ( c ) . we are also able to characterize instabilities of these types of breathers that emerge depending on parameters of the model . regarding branch ( a ) , nonlinear stability analysis is not available by using the energy method . our follow - up work @xcite develops a new method of analysis to prove the long - time stability of breathers for branch ( a ) . the summary of our findings is given in the concluding section [ conclusion ] , where the main results are shown in the form of table 1 . a simple yet universal model widely used to study coupled nonlinear oscillators is the frenkel - kontorova ( fk ) model @xcite . it describes a chain of classical particles coupled to their neighbors and subjected to a periodic on - site potential . in the continuum approximation , the fk model reduces to the sine - gordon equation , which is exactly integrable . the fk model is known to describe a rich variety of important nonlinear phenomena , which find applications in solid - state physics and nonlinear science @xcite . we consider here a two - array system of coupled pendula , where each pendulum is connected to the nearest neighbors by linear couplings . figure [ pendulapic ] shows schematically that each array of pendula is connected in the longitudinal direction by the torsional springs , whereas each pair of pendula is connected in the transverse direction by a common string . newton s equations of motion are given by @xmath7 where @xmath8 correspond to the angles of two arrays of pendula , dots denote derivatives of angles with respect to time @xmath9 , and the positive parameters @xmath10 and @xmath11 describe couplings between the two arrays in the longitudinal and transverse directions , respectively . the type of coupling between the two pendula with the angles @xmath12 and @xmath13 is referred to as the _ direct coupling _ between nonlinear oscillators ( see section 8.2 in @xcite ) . we consider oscillatory dynamics of coupled pendula under the following assumptions . * the coupling parameters @xmath10 and @xmath11 are small . therefore , we can introduce a small parameter @xmath14 such that both @xmath10 and @xmath11 are proportional to @xmath15 . * a resonant periodic force is applied to the common strings for each pair of coupled pendula . therefore , @xmath11 is considered to be proportional to @xmath16 , where @xmath17 is selected near the unit frequency of linear pendula indicating the @xmath18 parametric resonance between the force and the pendula . mathematically , we impose the following representation for parameters @xmath10 and @xmath19 : @xmath20 where @xmath21 are @xmath14-independent parameters , whereas @xmath14 is the formal small parameter to characterize the two assumptions ( a1 ) and ( a2 ) . in the formal limit @xmath22 , the pendula are uncoupled , and their small - amplitude oscillations can be studied with the asymptotic multi - scale expansion @xmath23 + \mu^3 x_n(t;\mu ) , \\[3pt ] y_n(t ) = \mu \left [ b_n(\mu^2 t ) e^{i \omega t } + \bar{b}_n(\mu^2 t ) e^{-i \omega t } \right ] + \mu^3 y_n(t;\mu ) , \end{array } \right . \label{expansion}\ ] ] where @xmath24 are amplitudes for nearly harmonic oscillations and @xmath25 are remainder terms . in a similar context of single - array coupled nonlinear oscillators , it is shown in @xcite how the asymptotic expansions like ( [ expansion ] ) can be justified . from the conditions that the remainder terms @xmath25 remain bounded as the system evolves , the amplitudes @xmath24 are shown to satisfy the discrete nonlinear schrdinger ( dnls ) equations , which bring together all the phenomena affecting the nearly harmonic oscillations ( such as cubic nonlinear terms , the detuning frequency , the coupling between the oscillators , and the amplitude of the parametric driving force ) . a similar derivation for a single pair of coupled pendula is reported in @xcite . using the algorithm in @xcite and restricting the scopes of this derivation to the formal level , we write the truncated system of equations for the remainder terms : @xmath26 where @xmath27 and @xmath28 are uniquely defined . bounded solutions to the linear inhomogeneous equations ( [ linear - equation ] ) exist if and only if @xmath29 for every @xmath30 . straightforward computations show that the conditions @xmath29 are equivalent to the following evolution equations for slowly varying amplitudes @xmath24 : @xmath31 the system ( [ dnls ] ) takes the form of coupled parametrically forced dnls equations ( see @xcite for references on parametrically forced nls equations ) . there exists an invariant reduction of system ( [ dnls ] ) given by @xmath32 to the scalar parametrically forced dnls equation . the reduction ( [ synchr-1 ] ) corresponds to the symmetric synchronized oscillations of coupled pendula of the model ( [ oscillators ] ) with @xmath33 in what follows , we consider a more general class of synchronized oscillations of coupled pendula . it turns out that the model ( [ dnls ] ) can be cast to the form of the parity time reversal ( @xmath0 ) dnls equations @xcite . using the variables @xmath34 the system of coupled dnls equations ( [ dnls ] ) is rewritten in the equivalent form @xmath35 , \\[3pt ] 2 i \dot{v}_n = \epsilon \left ( u_{n+1 } - 2 u_n + u_{n-1 } \right ) + \omega u_n - i \gamma v_n + 2\left [ \left ( |u_n|^2 + 2 |v_n|^2 \right ) u_n + \bar{u}_n v_n^2 \right ] , \end{array } \right . \label{pt - dnls}\end{aligned}\ ] ] which is the starting point for our analytical and numerical work . the invariant reduction ( [ synchr-1 ] ) for system ( [ dnls ] ) becomes @xmath36 without loss of generality , one can scale parameters @xmath37 , @xmath38 , and @xmath39 by a factor of two in order to eliminate the numerical factors in the system ( [ pt - dnls ] ) . also in the context of hard nonlinear oscillators ( e.g. in the framework of the @xmath40 theory ) , the cubic nonlinearity may have the opposite sign compared to the one in the system ( [ pt - dnls ] ) . however , given the applied context of the system of coupled pendula , we will stick to the specific form ( [ pt - dnls ] ) in further analysis . the system of coupled dnls equations ( [ pt - dnls ] ) is referred to as the @xmath0-symmetric dnls equation because the solutions remain invariant with respect to the action of the parity @xmath2 and time - reversal @xmath3 operators given by @xmath41 = \left [ \begin{array}{c } v \\ u \end{array } \right ] , \qquad \mathcal{t } \left [ \begin{array}{c } u(t ) \\ v(t ) \end{array } \right ] = \left [ \begin{array}{c } \bar{u}(-t ) \\ \bar{v}(-t ) \end{array } \right].\ ] ] the parameter @xmath39 introduces the gain loss coefficient in each pair of coupled oscillators due to the resonant periodic force . in the absence of all other effects , the @xmath39-term of the first equation of system ( [ pt - dnls ] ) induces the exponential growth of amplitude @xmath42 , whereas the @xmath39-term of the second equation induces the exponential decay of amplitude @xmath43 , if @xmath44 . the system ( [ pt - dnls ] ) truncated at a single site ( say @xmath45 ) is called the @xmath0-symmetric dimer . in the work of barashenkov _ et al . _ @xcite , it was shown that all @xmath0-symmetric dimers with physically relevant cubic nonlinearities represent hamiltonian systems in appropriately introduced canonical variables . however , the @xmath0-symmetric dnls equation on a lattice does not typically have a hamiltonian form if @xmath46 . nevertheless , the particular nonlinear functions arising in the system ( [ pt - dnls ] ) correspond to the @xmath0-symmetric dimers with a cross gradient hamiltonian structure @xcite , where variables @xmath47 are canonically conjugate . as a result , the system ( [ pt - dnls ] ) on the chain @xmath48 has additional conserved quantities . this fact looked like a mystery in the recent works @xcite . here we clarify the mystery in the context of the derivation of the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) from the original system ( [ oscillators ] ) . indeed , the system ( [ oscillators ] ) of classical newton particles has a standard hamiltonian structure with the energy function @xmath49 since the periodic movement of common strings for each pair of pendula result in the time - periodic coefficient @xmath19 , the energy @xmath50 is a periodic function of time @xmath9 . in addition , no other conserved quantities such as momenta exist typically in lattice differential systems such as the system ( [ oscillators ] ) due to broken continuous translational symmetry . after the system ( [ oscillators ] ) is reduced to the coupled dnls equations ( [ dnls ] ) with the asymptotic expansion ( [ expansion ] ) , we can write the evolution problem ( [ dnls ] ) in the hamiltonian form with the standard straight - gradient symplectic structure @xmath51 where the time variable @xmath9 stands now for the slow time @xmath52 and the energy function is @xmath53 the energy function @xmath54 is conserved in the time evolution of the hamiltonian system ( [ hamiltonian - dnls ] ) . in addition , there exists another conserved quantity @xmath55 which is related to the gauge symmetry @xmath56 with @xmath57 for solutions to the system ( [ dnls ] ) . when the transformation of variables ( [ variables ] ) is used , the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) is cast to the hamiltonian form with the cross - gradient symplectic structure @xmath58 where the energy function is @xmath59 the gauge - related function is written in the form @xmath60 the functions @xmath61 and @xmath62 are conserved in the time evolution of the system ( [ pt - dnls ] ) . these functions follow from ( [ energy - dnls ] ) and ( [ charge - dnls ] ) after the transformation ( [ variables ] ) is used . thus , the cross - gradient hamiltonian structure of the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) is inherited from the hamiltonian structure of the coupled oscillator model ( [ oscillators ] ) . we characterize the existence of breathers supported by the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) . in particular , breather solutions are continued for small values of coupling constant @xmath38 from solutions of the dimer equation arising at a single site , say the central site at @xmath45 . we shall work in a sequence space @xmath63 of square integrable complex - valued sequences . time - periodic solutions to the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) are given in the form @xcite : @xmath64 where the parameter @xmath6 is considered to be real , the factor @xmath65 is introduced for convenience , and the sequence @xmath66 is time - independent . the breather ( [ stationary ] ) is a localized mode if @xmath67 , which implies that @xmath68 as @xmath69 . the breather ( [ stationary ] ) is considered to be @xmath0-symmetric with respect to the operators in ( [ pt - symmetry ] ) if @xmath70 . the reduction ( [ reduction ] ) for symmetric synchronized oscillations is satisfied if @xmath71 the time - periodic breathers ( [ stationary ] ) with @xmath72 generalize the class of symmetric synchronized oscillations ( [ reduction - e ] ) . the time - independent sequence @xmath67 can be found from the stationary @xmath0-symmetric dnls equation : @xmath73 , \\[3pt ] e v_n = \epsilon \left ( u_{n+1 } - 2 u_n + u_{n-1 } \right ) + \omega u_n - i \gamma v_n + 2\left [ \left ( |u_n|^2 + 2 |v_n|^2 \right ) u_n + \bar{u}_n v_n^2 \right ] . \end{array } \right . \label{eq : state}\end{aligned}\ ] ] the @xmath0-symmetric breathers with @xmath70 satisfy the following scalar difference equation @xmath74 note that the reduction ( [ reduction - e ] ) is compatible with equation ( [ eq : statpt ] ) in the sense that if @xmath4 and @xmath75 , then @xmath76 satisfies a real - valued difference equation . let us set @xmath77 for now and consider solutions to the dimer equation at the central site @xmath45 : @xmath78 the parameters @xmath39 and @xmath37 are considered to be fixed , and the breather parameter @xmath6 is thought to parameterize continuous branches of solutions to the nonlinear algebraic equation ( [ dimer - states ] ) . the solution branches depicted on figure [ branches ] are given in the following lemma . ) . ] [ lemma - dimer ] assume @xmath46 . the algebraic equation ( [ dimer - states ] ) admits the following solutions depending on @xmath39 and @xmath37 : * @xmath79 - two symmetric unbounded branches exist for @xmath80 , * @xmath81 - an unbounded branch exists for every @xmath82 , * @xmath83 - a bounded branch exists for @xmath84 , where @xmath85 . substituting the decomposition @xmath86 with @xmath87 and @xmath88 into the algebraic equation ( [ dimer - states ] ) , we obtain @xmath89 excluding @xmath90 by using the fundamental trigonometric identity , we obtain the explicit parametrization of the solutions to the algebraic equation ( [ dimer - states ] ) by the amplitude parameter @xmath5 : @xmath91.\ ] ] the zero - amplitude limit @xmath92 is reached if @xmath93 , in which case @xmath94 , where @xmath85 . if @xmath95 , the solution branches ( if they exist ) are bounded away from the zero solution . now we analyze the three cases of parameters @xmath39 and @xmath37 formulated in the lemma . * if @xmath79 , then the parametrization ( [ branch - e - a ] ) yields a monotonically increasing map @xmath96 because @xmath97 > 0.\ ] ] in the two asymptotic limits , we obtain from ( [ branch - e - a ] ) : @xmath98 see figure [ branches](a ) . * if @xmath81 , the parametrization ( [ branch - e - a ] ) yields a monotonically increasing map @xmath99 , where @xmath100 indeed , we note that @xmath101 and @xmath102 so that the derivative in ( [ positivity - a ] ) is positive for every @xmath103 . we have @xmath104 see figure [ branches](b ) . * if @xmath83 , then the parametrization ( [ branch - e - a ] ) yields a monotonically decreasing map @xmath105 , where @xmath106 in ( [ a - minus ] ) , the first choice is made if @xmath107 and the second choice is made if @xmath108 . both choices are the same if @xmath109 . we note that @xmath110 , therefore , the derivative ( [ positivity - a ] ) needs to be rewritten in the form @xmath111 < 0,\ ] ] where @xmath112 for both @xmath107 and @xmath113 . in the two asymptotic limits , we obtain from ( [ branch - e - a ] ) : @xmath114 see figure [ branches](c ) . note that branches ( b ) and ( c ) coexist for @xmath83 . the reduction ( [ reduction - e ] ) corresponds to the choice : @xmath115 if @xmath44 , this choice corresponds to @xmath116 for @xmath117 , that is , the point @xmath4 on branch ( c ) . if @xmath118 , it corresponds to @xmath119 for any @xmath81 , that is , the point @xmath4 on branch ( b ) . every solution of lemma [ lemma - dimer ] can be extended to a breather on the chain @xmath48 which satisfies the spatial symmetry condition in addition to the @xmath0 symmetry : @xmath120 in order to prove the existence of the symmetric breather solution to the difference equation ( [ eq : statpt ] ) , we use the following implicit function theorem . + * implicit function theorem * ( theorem 4.e in @xcite ) . _ let @xmath121 and @xmath122 be banach spaces and let @xmath123 be a @xmath124 map on an open neighborhood of the point @xmath125 . assume that @xmath126 and that @xmath127 there are @xmath128 and @xmath129 such that for each @xmath130 with @xmath131 there exists a unique solution @xmath132 of the nonlinear equation @xmath133 with @xmath134 . moreover , the map @xmath135 is @xmath124 near @xmath136 . _ with two applications of the implicit function theorem , we prove the following main result of this section . [ theorem - existence - soliton ] fix @xmath46 , @xmath137 , and @xmath138 , where @xmath139 if @xmath93 . there exists @xmath140 sufficiently small and @xmath141 such that for every @xmath142 , there exists a unique solution @xmath143 to the difference equation ( [ eq : statpt ] ) satisfying the symmetry ( [ spatial - symmetry ] ) and the bound @xmath144 where @xmath5 and @xmath90 are defined in lemma [ lemma - dimer ] . moreover , the solution @xmath145 is smooth in @xmath38 . in the first application of the implicit function theorem , we consider the following system of algebraic equations @xmath146 where @xmath147 is given , in addition to parameters @xmath39 , @xmath37 , and @xmath6 . let @xmath148 , @xmath149 , @xmath150 , @xmath151 , and @xmath152 . then , we have @xmath153 and the jacobian operator @xmath154 is given by identical copies of the matrix @xmath155,\ ] ] with the eigenvalues @xmath156 . by the assumption of the lemma , @xmath157 , so that the jacobian operator @xmath154 is one - to - one and onto . by the implicit function theorem , for every @xmath147 and every @xmath158 sufficiently small , there exists a unique small solution @xmath159 of the system ( [ fixed - point-1 ] ) such that @xmath160 where the positive constant @xmath161 is independent from @xmath38 and @xmath162 . thanks to the symmetry of the difference equation ( [ eq : statpt ] ) , we find that @xmath163 , @xmath164 satisfy the same system ( [ fixed - point-1 ] ) with @xmath165 , with the same unique solution . in the second application of the implicit function theorem , we consider the following algebraic equation @xmath166 where @xmath167 depends on @xmath162 , @xmath39 , @xmath37 , and @xmath6 , satisfies the bound ( [ eq : u ] ) , and is uniquely defined by the previous result . let @xmath168 , @xmath169 , @xmath150 , @xmath151 , and @xmath170 . then , we have @xmath171 , where @xmath5 and @xmath90 are defined in lemma [ lemma - dimer ] . the jacobian operator @xmath172 is given by the matrix @xmath173 \biggr|_{u_0 = a e^{i \theta } } \\ \label{matrix - jacobian } & = & \left [ \begin{array}{cc } e - i \gamma - \frac{12 e a^2}{\omega + 8 a^2 } & -\omega - 12 a^2 \\ -\omega - 12 a^2 & e + i \gamma - \frac{12 e a^2}{\omega + 8 a^2 } \end{array } \right].\end{aligned}\ ] ] we show in lemma [ proposition - unbounded ] below that the matrix given by ( [ matrix - jacobian ] ) is invertible under the conditions @xmath46 and @xmath174 . by the implicit function theorem , for every @xmath158 sufficiently small , there exists a unique solution @xmath147 to the algebraic equation ( [ fixed - point-2 ] ) near @xmath175 such that @xmath176 where the positive constant @xmath177 is independent from @xmath38 . the bound ( [ bound - soliton ] ) holds thanks to the bounds ( [ eq : u ] ) and ( [ eq : u0 ] ) . since both equations ( [ fixed - point-1 ] ) and ( [ fixed - point-2 ] ) are smooth in @xmath38 , the solution @xmath145 is smooth in @xmath38 . in the following result , we show that the matrix given by ( [ matrix - jacobian ] ) is invertible for every branch of lemma [ lemma - dimer ] with an exception of a single point @xmath4 on branch ( c ) for @xmath178 . [ proposition - unbounded ] with the exception of the point @xmath4 on branch ( c ) of lemma [ lemma - dimer ] for @xmath178 , the matrix given by ( [ matrix - jacobian ] ) is invertible for every @xmath46 . the matrix given by ( [ matrix - jacobian ] ) has zero eigenvalue if and only if its determinant is zero , which happens at @xmath179 eliminating @xmath180 by using parametrization ( [ branch - e - a ] ) and simplifying the algebraic equation for nonzero @xmath181 , we reduce it to the form @xmath182 we now check if this constraint can be satisfied for the three branches of lemma [ lemma - dimer ] . * if @xmath79 , the constraint ( [ constraint - parameters ] ) is not satisfied because the left - hand side @xmath183 exceeds the right - hand side @xmath184 . * if @xmath81 and @xmath103 , where @xmath185 is given by ( [ a - plus ] ) , the constraint ( [ constraint - parameters ] ) is not satisfied because the left - hand side @xmath186 exceeds the left - hand side @xmath184 both for @xmath187 and for @xmath188 . * if @xmath83 and @xmath189 , where @xmath190 is given by ( [ a - minus ] ) , the constraint ( [ constraint - parameters ] ) is not satisfied because the left - hand side is estimated by @xmath191 in the first case , we have @xmath192 , so that the left - hand side is strictly smaller than @xmath193 . in the second case , we have @xmath194 , so that the left - hand side is also strictly smaller than @xmath193 . only if @xmath109 , the constraint ( [ constraint - parameters ] ) is satisfied at @xmath4 , when @xmath195 and @xmath196 hence , the matrix ( [ matrix - jacobian ] ) is invertible for all parameter values with one exceptional case . in the asymptotic limit @xmath197 as @xmath198 , see lemma [ lemma - dimer ] , the matrix ( [ matrix - jacobian ] ) is expanded asymptotically as @xmath199 + \mathcal{o}(1 ) \quad \mbox{\rm as } \quad |e| \to \infty , \label{matrix - jacobian - asymptotics}\end{aligned}\ ] ] with the two eigenvalues @xmath200 and @xmath201 . thus , the matrix given by ( [ matrix - jacobian - asymptotics ] ) is invertible for every branch extending to sufficiently large values of @xmath6 . here we discuss the linear stability of the zero equilibrium in the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) . the following proposition yields a simple result . the zero equilibrium of the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) is linearly stable if @xmath202 , where @xmath203 . [ prop - zero - equilibrium ] truncating the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) at the linear terms and using the fourier transform @xmath204 we obtain the linear homogeneous system @xmath205 = \left [ \begin{array}{c } 0 \\ 0 \end{array } \right ] , \quad \mbox{\rm where } \quad \hat{d}(k ) : = \left [ \begin{matrix } -2 \omega(k ) - i \gamma & -\omega + 4 \epsilon \sin^2(k/2 ) \\ -\omega + 4 \epsilon \sin^2(k/2 ) & -2 \omega(k ) + i \gamma \end{matrix } \right].\ ] ] the determinant of @xmath206 is zero if and only if @xmath207 is found from the quadratic equation @xmath208 for any @xmath202 , where @xmath209 is given by ( [ gamma-0 ] ) , the two branches @xmath210 found from the quadratic equation ( [ dispersion - relation ] ) are real - valued and non - degenerate for every @xmath211 $ ] . therefore , the zero equilibrium is linearly stable . on the other hand , for any @xmath212 , the values of @xmath207 are purely imaginary either near @xmath213 if @xmath214 or near @xmath215 if @xmath188 . therefore , the zero equilibrium is linearly unstable . the value @xmath209 given by ( [ gamma-0 ] ) represents the phase transition threshold and the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) is said to have broken @xmath0-symmetry for @xmath212 . if @xmath77 , the zero equilibrium is only linearly stable for @xmath216 . since the localized breathers can not be stable when the zero background is unstable , we shall study stability of breathers only for the case when @xmath216 , that is , in the regime of unbroken @xmath0-symmetry . it follows from theorem [ theorem - existence - soliton ] that each interior point on the solution branches shown on figure [ branches ] generates a fundamental breather of the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) . we shall now characterize these breathers as relative equilibria of the energy function . thanks to the cross - gradient symplectic structure ( [ hamiltonian - pt - dnls ] ) , the stationary @xmath0-symmetric dnls equation can be written in the gradient form @xmath217 keeping in mind the additional conserved quantity @xmath62 given by ( [ charge - pt - dnls ] ) , we conclude that the stationary solution @xmath66 is a critical point of the combined energy function given by @xmath218 if we want to apply the lyapunov method in order to study nonlinear stability of stationary solutions in hamiltonian systems , we shall investigate convexity of the second variation of the combined energy functional @xmath219 at @xmath66 . using the expansion @xmath220 , @xmath221 and introducing extended variables @xmath222 and @xmath223 with the blocks @xmath224 we can expand the smooth function @xmath219 up to the quadratic terms in @xmath223 : @xmath225 where @xmath226 is the self - adjoint ( hessian ) operator defined on @xmath63 and the scalar product was used in the following form : @xmath227 using ( [ energy - pt - dnls ] ) and ( [ charge - pt - dnls ] ) , the hessian operator can be computed explicitly as follows @xmath228 where blocks of @xmath229 at each lattice node @xmath30 are given by @xmath230\end{aligned}\]]and @xmath231 is the discrete laplacian operator applied to blocks of @xmath223 at each lattice node @xmath30 : @xmath232 in the expression for @xmath233 , we have used the @xmath0-symmetry condition @xmath70 for the given stationary solution @xmath66 . we study convexity of the combined energy functional @xmath219 at @xmath66 . since the zero equilibrium is linearly stable only for @xmath216 ( if @xmath77 ) , we only consider breathers of theorem [ theorem - existence - soliton ] for @xmath216 . in order to study eigenvalues of @xmath226 for small values of @xmath38 , we use the following perturbation theory . + * perturbation theory for linear operators * ( theorem vii.1.7 in @xcite ) . _ let @xmath234 be a family of bounded operators from banach space @xmath235 to itself , which depends analytically on the small parameter @xmath38 . if the spectrum of @xmath236 is separated into two parts , the subspaces of @xmath235 corresponding to the separated parts also depend analytically on @xmath38 . in particular , the spectrum of @xmath234 is separated into two parts for any @xmath158 sufficiently small . _ with an application of the perturbation theory for linear operators , we prove the following main result of this section . [ theorem - critical - point ] fix @xmath46 , @xmath37 , and @xmath6 along branches of the @xmath0-symmetric breathers @xmath66 given by theorem [ theorem - existence - soliton ] such that @xmath93 and @xmath138 , where @xmath139 . for every @xmath237 sufficiently small , the operator @xmath226 admits a one - dimensional kernel in @xmath63 spanned by the eigenvector @xmath238 due to the gauge invariance , where the blocks of the eigenvector are given by @xmath239 in addition , * if @xmath240 , the spectrum of @xmath226 in @xmath63 includes infinite - dimensional positive and negative parts . * if @xmath241 and @xmath83 , the spectrum of @xmath242 in @xmath63 includes an infinite - dimensional negative part and either three or one simple positive eigenvalues for branches ( b ) and ( c ) of lemma [ lemma - dimer ] respectively . if @xmath77 , the breather solution of theorem [ theorem - existence - soliton ] is given by @xmath243 for every @xmath244 and @xmath86 , where @xmath5 and @xmath90 are defined by lemma [ lemma - dimer ] . in this case , the linear operator @xmath245 decouples into @xmath246-by-@xmath246 blocks for each lattice node @xmath30 . for @xmath45 , the @xmath246-by-@xmath246 block of the linear operator @xmath229 is given by @xmath247.\]]using relations ( [ parameterization - a ] ) and ( [ branch - e - a ] ) , as well as symbolic computations with maple , we found that the @xmath246-by-@xmath246 matrix block @xmath248 admits a simple zero eigenvalue and three nonzero eigenvalues @xmath249 , @xmath250 , and @xmath251 given by @xmath252 for each branch of lemma [ lemma - dimer ] with @xmath46 and @xmath138 , we have @xmath253 , so that @xmath254 . furthermore , either @xmath255 or @xmath256 if and only if @xmath257 expanding this equation for nonzero @xmath5 yields constraint ( [ constraint - parameters ] ) . with the exception of a single point @xmath4 at @xmath178 , we showed in lemma [ proposition - unbounded ] that the constraint ( [ constraint - parameters ] ) does not hold for any of the branches of lemma [ lemma - dimer ] . therefore , @xmath258 and @xmath259 along each branch of lemma [ lemma - dimer ] and the signs of @xmath249 , @xmath250 , and @xmath251 for each branch of lemma [ lemma - dimer ] can be obtained in the limit @xmath198 for branches ( a ) and ( b ) or @xmath260 for branch ( c ) . by means of these asymptotic computations as @xmath198 or @xmath260 , we obtain the following results for the three branches shown on figure [ branches ] : * @xmath261 . * @xmath261 . * @xmath262 , @xmath263 , and @xmath264 . for @xmath265 , the @xmath246-by-@xmath246 block of the linear operator @xmath229 is given by @xmath266.\ ] ] each block has two double eigenvalues @xmath267 and @xmath268 given by @xmath269 since there are infinitely many nodes with @xmath244 , the points @xmath267 and @xmath268 have infinite multiplicity in the spectrum of the linear operator @xmath229 . furthermore , we can sort up the signs of @xmath267 and @xmath268 for each point on the three branches shown on figure [ branches ] : * if @xmath270 , then @xmath271 and @xmath272 . * if @xmath241 and @xmath83 , then @xmath273 . by using the perturbation theory for linear operators , we argue as follows : * since @xmath226 is hermitian on @xmath63 , its spectrum is a subset of the real line for every @xmath158 . * the zero eigenvalue persists with respect to @xmath158 at zero because the eigenvector ( [ kernel ] ) belongs to the kernel of @xmath226 due to the gauge invariance for every @xmath158 . * the other eigenvalues of @xmath229 are isolated away from zero . the spectrum of @xmath226 is continuous with respect to @xmath38 and includes infinite - dimensional parts near points @xmath267 and @xmath268 for small @xmath237 ( which may include continuous spectrum and isolated eigenvalues ) as well as simple eigenvalues near @xmath274 ( if @xmath274 are different from @xmath275 ) . the statement of the theorem follows from the perturbation theory and the count of signs of @xmath274 and @xmath275 above . in the asymptotic limit @xmath197 as @xmath198 , we can sort out eigenvalues of @xmath226 asymptotically as : @xmath276 where the remainder terms are @xmath277 as @xmath278 . the values @xmath249 , @xmath251 , and @xmath267 are close to each other as @xmath278 . it follows from theorem [ theorem - critical - point ] that for @xmath240 , the breather @xmath66 is a saddle point of the energy functional @xmath219 with infinite - dimensional positive and negative invariant subspaces of the hessian operator @xmath226 . this is very similar to the hamiltonian systems of the dirac type , where stationary states are located in the gap between the positive and negative continuous spectrum . this property holds for points 1 and 3 on branches ( a ) and ( b ) shown on figure [ branches ] . no branches other than @xmath240 exist for @xmath79 . on the other hand , points 2 and 4 on branches ( b ) and ( c ) shown on figure [ branches ] satisfy @xmath241 and @xmath83 . the breather @xmath66 is a saddle point of @xmath219 for these points and it only has three ( one ) directions of positive energy in space @xmath63 for point 2 ( point 4 ) . spectral stability of breathers can be studied for small values of coupling constant @xmath38 by using the perturbation theory @xcite . first , we linearize the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) at the breather ( [ stationary ] ) by using the expansion @xmath279 , \quad v(t ) = e^{-\frac{1}{2 } i et } \left [ v + { \bf v}(t ) \right],\ ] ] where @xmath280 is a small perturbation satisfying the linearized equations @xmath281 , \\ 2 i \dot{{\bf v}}_n + e { \bf v}_n = \epsilon \left ( { \bf u}_{n+1 } - 2 { \bf u}_n + { \bf u}_{n-1 } \right ) + \omega { \bf u}_n - i \gamma { \bf v}_n \\ \phantom{text } + 2 \left [ 2 \left ( |u_n|^2 + |v_n|^2 \right ) { \bf u}_n + ( u_n^2 + v_n^2 ) \bar{{\bf u}}_n + 2 ( \bar{u}_n v_n + u_n \bar{v}_n ) { \bf v}_n + 2 u_n v_n \bar{{\bf v}}_n \right ] . \end{array } \right . \label{pt - dnls - lin}\end{aligned}\ ] ] the spectral stability problem arises from the linearized equations ( [ pt - dnls - lin ] ) after the separation of variables : @xmath282 where @xmath283 is the eigenvector corresponding to the spectral parameter @xmath284 . note that @xmath285 and @xmath286 are no longer complex conjugate to each other if @xmath284 has a nonzero imaginary part . the spectral problem can be written in the explicit form @xmath287 , \\ ( e - i \lambda + i \gamma ) \psi_n - \omega \nu_n = \epsilon \left ( \nu_{n+1 } - 2 \nu_n + \nu_{n-1 } \right ) \\ \phantom{texttext } + 2 \left [ 2 |u_n|^2 ( \varphi_n + 2\nu_n ) + ( u_n^2 + \bar{u}_n^2 ) ( 2 \psi_n + \chi_n ) \right ] , \\ ( e + i \lambda + i \gamma ) \chi_n - \omega \varphi_n = \epsilon \left ( \varphi_{n+1 } - 2 \varphi_n + \varphi_{n-1 } \right ) \\ \phantom{texttext } + 2 \left [ 2 |u_n|^2 ( 2 \varphi_n + \nu_n ) + ( u_n^2 + \bar{u}_n^2 ) ( \psi_n + 2 \chi_n ) \right ] , \\ ( e - i \lambda - i \gamma ) \nu_n - \omega \psi_n = \epsilon \left ( \psi_{n+1 } - 2 \psi_n + \psi_{n-1 } \right ) \\ \phantom{texttext } + 2 \left [ 2 |u_n|^2 ( 2 \psi_n + \chi_n ) + ( u_n^2 + \bar{u}_n^2 ) ( \varphi_n + 2 \nu_n ) \right ] , \end{array } \right.\ ] ] where we have used the condition @xmath70 for the @xmath0-symmetric breathers . recalling definition of the hessian operator @xmath226 in ( [ pt - dnls - var ] ) , we can rewrite the spectral problem ( [ spectral - problem ] ) in the hamiltonian form : @xmath288 where @xmath289 is a symmetric matrix with the blocks at each lattice node @xmath30 given by @xmath290.\ ] ] we note the hamiltonian symmetry of the eigenvalues of the spectral problem ( [ eq : eigenval ] ) . [ prop - ham - sym ] eigenvalues of the spectral problem ( [ eq : eigenval ] ) occur either as real or imaginary pairs or as quadruplets in the complex plane . assume that @xmath291 is an eigenvalue of the spectral problem ( [ eq : eigenval ] ) with the eigenvector @xmath292 . then , @xmath293 is an eigenvalue of the same problem with the eigenvector @xmath294 , whereas @xmath295 is also an eigenvalue with the eigenvector @xmath296 . if @xmath83 and @xmath297 ( points 2 and 4 shown on figure [ branches ] ) , theorem [ theorem - critical - point ] implies that the self - adjoint operator @xmath226 in @xmath63 is negative - definite with the exception of either three ( point 2 ) or one ( point 4 ) simple positive eigenvalues . in this case , we can apply the following hamilton krein index theorem in order to characterize the spectrum of @xmath298 . + * hamilton krein index theorem * ( theorem 3.3 in @xcite ) . _ let @xmath299 be a self - adjoint operator in @xmath300 with finitely many negative eigenvalues @xmath301 , a simple zero eigenvalue with eigenfunction @xmath302 , and the rest of its spectrum is bounded from below by a positive number . let @xmath303 be a bounded invertible skew - symmetric operator in @xmath300 . let @xmath304 be a number of positive real eigenvalues of @xmath305 , @xmath306 be a number of quadruplets @xmath307 that are neither in @xmath308 nor in @xmath309 , and @xmath310 be a number of purely imaginary pairs of eigenvalues of @xmath305 whose invariant subspaces lie in the negative subspace of @xmath299 . let @xmath311 be finite and nonzero . then , @xmath312 _ [ theorem - index ] fix @xmath46 , @xmath83 , and @xmath313 , where @xmath139 . for every @xmath237 sufficiently small , @xmath314 for branch ( b ) of lemma [ lemma - dimer ] and @xmath315 for branch ( c ) of lemma [ lemma - dimer ] with @xmath316 . for branch ( c ) with @xmath317 , there exists a value @xmath318 such that @xmath319 for @xmath320 and @xmath315 for @xmath321 . if @xmath46 , @xmath83 , @xmath241 , and @xmath237 is sufficiently small , theorem [ theorem - critical - point ] implies that the spectrum of @xmath226 in @xmath63 has finitely many positive eigenvalues and a simple zero eigenvalue with eigenvector @xmath238 . therefore , the hamilton krein index theorem is applied in @xmath63 for @xmath322 , @xmath323 , and @xmath324 . we shall verify that @xmath325 where @xmath238 is given by ( [ kernel ] ) and @xmath326 denotes derivative of @xmath222 with respect to parameter @xmath6 . the first equation @xmath327 follows by theorem [ theorem - critical - point ] . by differentiating equations ( [ eq : state ] ) in @xmath6 , we obtain @xmath328 for every @xmath6 , for which the solution @xmath222 is differentiable in @xmath6 . for @xmath77 , the limiting solution of lemma [ lemma - dimer ] is differentiable in @xmath6 for every @xmath72 and @xmath138 . due to smoothness of the continuation in @xmath38 by theorem [ theorem - critical - point ] , this property holds for every @xmath237 sufficiently small . by using ( [ generalized - kernel - lemma ] ) with @xmath329 , we obtain @xmath330 where we have used the definition of @xmath62 in ( [ charge - pt - dnls ] ) . we compute the slope condition at @xmath77 : @xmath331,\ ] ] where relations ( [ parameterization - a ] ) and ( [ branch - e - a ] ) have been used . for branch ( b ) of lemma [ lemma - dimer ] with @xmath83 , we have @xmath332 and @xmath333 , so that @xmath334 . by continuity , @xmath335 remains strictly positive for small @xmath237 . thus , @xmath336 and @xmath314 by the hamilton krein index theorem . for branch ( c ) of lemma [ lemma - dimer ] with @xmath83 , we have @xmath337 and @xmath338 . therefore , we only need to inspect the sign of the expression @xmath339 . if @xmath316 , then for every @xmath340 , we have @xmath341 therefore , @xmath336 and @xmath315 by the hamilton krein index theorem . on the other hand , if @xmath342 , we have @xmath343 at @xmath92 ( @xmath344 ) and @xmath345 at @xmath116 ( @xmath4 ) . since the dependence of @xmath5 versus @xmath6 is monotonic , there exists a value @xmath318 such that @xmath319 for @xmath320 and @xmath315 for @xmath321 . if @xmath315 and @xmath346 , orbital stability of a critical point of @xmath219 in space @xmath63 can be proved from the hamilton krein theorem ( see @xcite and references therein ) . orbital stability of breathers is understood in the following sense . we say that the breather solution ( [ stationary ] ) is orbitally stable in @xmath63 if for every @xmath347 sufficiently small , there exists @xmath348 such that if @xmath349 satisfies @xmath350 , then the unique global solution @xmath351 , @xmath352 to the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) satisfies the bound @xmath353 [ def - stability ] the definition of instability of breathers is given by negating definition [ def - stability ] . the following result gives orbital stability or instability for branch ( c ) shown on figure [ branches ] . [ theorem - bounded ] fix @xmath46 , @xmath83 , and @xmath313 . for every @xmath237 sufficiently small , the breather @xmath66 for branch ( c ) of lemma [ lemma - dimer ] is orbitally stable in @xmath63 if @xmath316 . for every @xmath317 , there exists a value @xmath318 such that the breather @xmath66 is orbitally stable in @xmath63 if @xmath321 and unstable if @xmath320 . the theorem is a corollary of lemma [ theorem - index ] for branch ( c ) of lemma [ lemma - dimer ] and the orbital stability theory from @xcite . orbital stability of breathers for branches ( a ) and ( b ) of lemma [ lemma - dimer ] does not follow from the standard theory because @xmath354 for @xmath240 and @xmath355 for branch ( b ) with @xmath241 . nevertheless , by using smallness of parameter @xmath38 and the construction of the breather @xmath66 in theorem [ theorem - existence - soliton ] , spectral stability of breathers can be considered directly . spectral stability and instability of breathers is understood in the following sense . [ def - spectral ] we say that the breather solution ( [ stationary ] ) is spectrally stable if @xmath356 for every bounded solution of the spectral problem ( [ eq : eigenval ] ) . on the other hand , if the spectral problem ( [ eq : eigenval ] ) admits an eigenvalue @xmath357 with an eigenvector in @xmath358 , we say that the breather solution ( [ stationary ] ) is spectrally unstable . the following theorem gives spectral stability of breathers for branches ( a ) and ( b ) shown on figure [ branches ] . [ theorem - stability ] fix @xmath46 , @xmath93 , and @xmath6 along branches ( a ) and ( b ) of lemma [ lemma - dimer ] with @xmath72 and @xmath138 . for every @xmath237 sufficiently small , the spectral problem ( [ eq : eigenval ] ) admits a double zero eigenvalue with the generalized eigenvectors @xmath359 where the eigenvector @xmath238 is given by ( [ kernel ] ) and the generalized eigenvector @xmath326 denotes derivative of @xmath222 with respect to parameter @xmath6 . for every @xmath6 such that the following non - degeneracy condition is satisfied , @xmath360 the breather @xmath66 is spectrally stable . if @xmath77 , the breather solution of theorem [ theorem - existence - soliton ] is given by @xmath243 for every @xmath244 and @xmath86 , where @xmath5 and @xmath90 are defined by lemma [ lemma - dimer ] . in this case , the spectral problem ( [ spectral - problem ] ) decouples into @xmath246-by-@xmath246 blocks for each lattice node @xmath30 . recall that @xmath361 at @xmath77 . for @xmath45 , eigenvalues @xmath284 are determined by the @xmath246-by-@xmath246 matrix block @xmath362 . using relations ( [ parameterization - a ] ) and ( [ branch - e - a ] ) , as well as symbolic computations with maple , we found that the @xmath246-by-@xmath246 matrix block @xmath362 has a double zero eigenvalue and a pair of simple eigenvalues at @xmath363 , where @xmath364 for @xmath265 , eigenvalues @xmath284 are determined by the @xmath246-by-@xmath246 matrix block @xmath365 , where @xmath233 is given by ( [ block - l - n ] ) . if @xmath216 , @xmath72 , and @xmath138 , where @xmath85 , each block has four simple eigenvalues @xmath366 and @xmath367 , where @xmath368 so that @xmath369 . since there are infinitely many nodes with @xmath244 , the four eigenvalues are semi - simple and have infinite multiplicity . if @xmath237 is sufficiently small , we use perturbation theory for linear operators from section [ variational ] . * the double zero eigenvalue persists with respect to @xmath158 at zero because of the gauge invariance of the breather @xmath66 ( with respect to rotation of the complex phase ) . indeed , @xmath327 follows from the result of theorem [ theorem - critical - point ] . the generalized eigenvector is defined by equation @xmath370 , which is equivalent to equation @xmath371 . differentiating equations ( [ eq : state ] ) in @xmath6 , we obtain @xmath372 . since @xmath373 = 1 $ ] and @xmath374 the second generalized eigenvector @xmath375 exists as a solution of equation @xmath376 if and only if @xmath377 . it follows from the explicit computation ( [ slope - explicit ] ) that if @xmath77 , then @xmath378 for every @xmath6 along branches ( a ) and ( b ) of lemma [ lemma - dimer ] . by continuity , @xmath378 for small @xmath237 . therefore , the zero eigenvalue of the operator @xmath379 is exactly double for small @xmath237 . * using the same computation ( [ slope - explicit ] ) , it is clear that @xmath380 for every @xmath6 along branches ( a ) and ( b ) of lemma [ lemma - dimer ] . assume that @xmath381 and @xmath382 , which is expressed by the non - degeneracy condition ( [ assumption - non - degeneracy ] ) . then , the pair @xmath383 is isolated from the rest of the spectrum of the operator @xmath379 at @xmath77 . since the eigenvalues @xmath363 are simple and purely imaginary , they persist on the imaginary axis for @xmath158 because they can not leave the imaginary axis by the hamiltonian symmetry of proposition [ prop - ham - sym ] . * if @xmath216 , @xmath72 , and @xmath138 , the semi - simple eigenvalues @xmath366 and @xmath367 of infinite multiplicity are nonzero and located at the imaginary axis at different points for @xmath77 . they persist on the imaginary axis for @xmath158 according to the following perturbation argument . first , for the central site @xmath45 , the spectral problem ( [ spectral - problem ] ) can be written in the following abstract form @xmath384 where @xmath385 denotes a continuation of @xmath248 in @xmath38 . thanks to the non - degeneracy condition ( [ assumption - non - degeneracy ] ) as well as the condition @xmath157 , the matrix @xmath386 is invertible . by continuity , the matrix @xmath387 is invertible for every @xmath38 and @xmath284 near @xmath77 and @xmath388 . therefore , there is a unique @xmath389 given by @xmath390 which satisfies @xmath391 near @xmath77 and @xmath388 , where @xmath10 is a positive @xmath38- and @xmath284-independent constant . next , for either @xmath164 or @xmath165 , the spectral problem ( [ spectral - problem ] ) can be represented in the form @xmath392 where @xmath393 denotes a continuation of @xmath233 given by ( [ block - l - n ] ) in @xmath38 , whereas the operator @xmath231 is applied with zero end - point condition at @xmath45 . we have @xmath394 and @xmath395 near @xmath77 and @xmath388 . therefore , up to the first order of the perturbation theory , the spectral parameter @xmath284 near @xmath396 is defined from the truncated eigenvalue problem @xmath397 which is solved with the discrete fourier transform ( [ discrete - fourier ] ) . in order to satisfy the dirichlet end - point condition at @xmath45 , the sine fourier transform must be used , which does not affect the characteristic equation for the purely continuous spectrum of the spectral problem ( [ truncated - problem ] ) . by means of routine computations , we obtain the characteristic equation in the form , see also equation ( [ dispersion - relation ] ) , @xmath398 where @xmath211 $ ] is the parameter of the discrete fourier transform ( [ discrete - fourier ] ) . solving the characteristic equation ( [ continuous - spectrum ] ) , we obtain four branches of the continuous spectrum @xmath399 where the two sign choices are independent from each other . if @xmath93 is fixed and @xmath237 is small , the four branches of the continuous spectrum are located on the imaginary axis near the points @xmath366 and @xmath367 given by ( [ lambda - plus - minus - eig ] ) . + in addition to the continuous spectrum given by ( [ continuous - spectrum ] ) , there may exist isolated eigenvalues near @xmath366 and @xmath367 , which are found from the second - order perturbation theory @xcite . under the condition @xmath72 and @xmath138 , these eigenvalues are purely imaginary . therefore , the infinite - dimensional part of the spectrum of the operator @xmath400 persists on the imaginary axis for @xmath158 near the points @xmath366 and @xmath367 of infinite algebraic multiplicity . the statement of the lemma follows from the perturbation theory and the fact that all isolated eigenvalues and the continuous spectrum of @xmath401 are purely imaginary . in the asymptotic limit @xmath197 as @xmath198 , the eigenvalues @xmath402 and @xmath396 defined by ( [ lambda-0-eig ] ) and ( [ lambda - plus - minus - eig ] ) are given asymptotically by @xmath403 where the remainder terms are @xmath277 as @xmath278 . the values @xmath402 , @xmath404 , and @xmath405 are close to each other as @xmath406 . computations in the proof of theorem [ theorem - stability ] can be extended to the branch ( c ) of lemma [ lemma - dimer ] . indeed , @xmath380 for branch ( c ) with either @xmath316 or @xmath317 , and @xmath6 near @xmath407 . on the other hand , @xmath408 if @xmath317 and @xmath6 near @xmath409 . as a result , branch ( c ) is spectrally stable in the former case and is spectrally unstable in the latter case , in agreement with theorem [ theorem - bounded ] . observe in the proof of theorem [ theorem - stability ] that @xmath410 if @xmath95 . in this case , branch ( b ) of lemma [ lemma - dimer ] is spectrally unstable . this instability corresponds to the instability of the zero equilibrium for @xmath95 , in agreement with the result of proposition [ prop - zero - equilibrium ] . before presenting numerical approximations of eigenvalues of the spectral problem ( [ eq : eigenval ] ) , we compute the krein signature of wave continuum . this helps to interpret instabilities and resonances that arise when isolated eigenvalues @xmath383 cross the continuous bands near points @xmath366 and @xmath367 . the krein signature of simple isolated eigenvalues is defined as follows . [ def - krein ] let @xmath411 be an eigenvector of the spectral problem ( [ eq : eigenval ] ) for an isolated simple eigenvalue @xmath380 . then , the energy quadratic form @xmath412 is nonzero and its sign is called the krein signature of the eigenvalue @xmath402 . definition [ def - krein ] is used to simplify the presentation . similarly , one can define the krein signature of isolated multiple eigenvalues and the krein signature of the continuous spectral bands in the spectral problem ( [ eq : eigenval ] ) @xcite . the following lemma characterizes krein signatures of the spectral points arising in the proof of theorem [ theorem - stability ] . [ lemma - krein ] fix @xmath46 , @xmath93 , and @xmath413 with @xmath138 . assume the non - degeneracy condition ( [ assumption - non - degeneracy ] ) . for every @xmath237 sufficiently small , we have the following for the corresponding branches of lemma [ lemma - dimer ] : * the subspaces of @xmath379 in @xmath63 near @xmath366 , @xmath367 , and @xmath383 have positive , negative , and positive krein signature , respectively ; * the subspaces of @xmath379 in @xmath63 near @xmath366 , @xmath367 , and @xmath383 have negative , positive ( if @xmath414 ) or negative ( if @xmath415 ) , and positive krein signature , respectively ; * all subspaces of @xmath379 in @xmath63 near @xmath366 , @xmath367 , and @xmath383 ( if @xmath380 ) have negative krein signature . we proceed by the perturbation arguments from the limit @xmath77 , where @xmath416 is a block - diagonal operator consisting of @xmath417 blocks . in particular , we consider the blocks for @xmath265 , where @xmath233 is given by ( [ block - l - n ] ) . solving ( [ spectral - problem ] ) at @xmath77 and @xmath388 , we obtain the eigenvector @xmath418 as a result , we obtain for the eigenvector @xmath419 , @xmath420 for branch ( a ) , @xmath79 and @xmath414 . therefore , @xmath421 for @xmath422 and @xmath423 for @xmath424 . for branch ( b ) , @xmath83 and either @xmath414 or @xmath425 . in either case , @xmath423 for @xmath422 . on the other hand , for @xmath424 , @xmath421 if @xmath414 and @xmath423 if @xmath425 . for branch ( c ) , @xmath83 and @xmath425 . in this case , @xmath423 for either @xmath422 or @xmath424 . finally , the krein signature for the eigenvalue @xmath426 denoted by @xmath427 follows from the computations of eigenvalues @xmath274 in the proof of theorem [ theorem - critical - point ] . we have @xmath428 for branches ( a ) and ( b ) because @xmath429 and we have @xmath430 for branch ( c ) because @xmath431 , whereas the eigenvalue @xmath263 is controlled by the result of lemma [ theorem - index ] . the signs of all eigenvalues are nonzero and continuous with respect to parameter @xmath38 . therefore , the count above extends to the case of small nonzero @xmath38 . the spectrum of @xmath401 is shown at figure [ spectra ] . panels ( a ) , ( b ) and ( c ) correspond to branches shown at figure [ branches ] . for different branches of breathers . ] * we can see on panel ( a ) of figure [ spectra ] that @xmath432 do not intersect for every @xmath414 and are located within fixed distance @xmath277 , as @xmath433 . note that the upper - most @xmath402 and @xmath404 have positive krein signature , whereas the lowest @xmath405 has negative krein signature , as is given by lemma [ lemma - krein ] . * we observe on panel ( b ) of figure [ spectra ] that @xmath434 intersects @xmath402 , creating a small bubble of instability in the spectrum . the insert shows that the bubble shrinks as @xmath435 , in agreement with theorem [ theorem - stability ] . there is also an intersection between @xmath436 and @xmath402 , which does not create instability . these results are explained by the krein signature computations in lemma [ lemma - krein ] . instability is induced by opposite krein signatures between @xmath404 and @xmath402 , whereas crossing of @xmath405 and @xmath402 with the same krein signatures is safe of instabilities . note that for small @xmath6 , the isolated eigenvalue @xmath402 is located above both the spectral bands near @xmath404 and @xmath405 . the gap in the numerical data near @xmath344 indicates failure to continue the breather solution numerically in @xmath38 , in agreement with the proof of theorem [ theorem - existence - soliton ] . * we observe from panel ( c ) of figure [ spectra ] that @xmath402 and @xmath437 intersect but do not create instabilities , since all parts of the spectrum have the same signature , as is given by lemma [ lemma - krein ] . in fact , the branch is both spectrally and orbitally stable as long as @xmath380 , in agreement with theorem [ theorem - bounded ] . on the other hand , there is @xmath318 , if @xmath317 , such that @xmath408 for @xmath438 , which indicates instability of branch ( c ) , again , in agreement with theorem [ theorem - bounded ] . as we see on panel ( b ) of figure [ spectra ] , @xmath402 intersects @xmath404 for some @xmath439 . in the remainder of this section , we study whether this crossing point is always located on the right of @xmath440 . in fact , the answer to this question is negative . we shall prove for branch ( b ) that the intersection of @xmath402 with either @xmath404 or @xmath441 occur either for @xmath442 or for @xmath443 , depending on parameters @xmath39 and @xmath37 . fix @xmath46 , @xmath83 , and @xmath413 along branch ( b ) of lemma [ lemma - dimer ] . there exists a resonance @xmath444 at @xmath445 with @xmath446 if @xmath447 and @xmath448 if @xmath449 , where @xmath450 moreover , if @xmath451 , there exists a resonance @xmath452 at @xmath445 with @xmath448 . let us first assume that there exists a resonance @xmath444 at @xmath453 and find the condition on @xmath39 and @xmath37 , when this is possible . from the definitions ( [ lambda-0-eig ] ) and ( [ lambda - plus - minus - eig ] ) , we obtain the constraint on @xmath181 : @xmath454 after canceling @xmath455 since @xmath456 with @xmath185 given by ( [ a - plus ] ) , we obtain @xmath457 which has two roots @xmath458 since @xmath103 , the lower sign is impossible because this leads to a contradiction @xmath459 the upper sign is possible if @xmath460 . using the parametrization , we substitute the root for @xmath181 to the equation @xmath461 and simplify it : @xmath462 \\ & = & \frac{2 \sqrt{\omega^2 - 4 \gamma^2 } \left(2 |\omega| + \sqrt{\omega^2 - 4 \gamma^2}\right)^2}{\left(|\omega| + \sqrt{\omega^2 - 4 \gamma^2}\right)}.\end{aligned}\ ] ] this equation further simplifies to the form : @xmath463 squaring it up , we obtain @xmath464 which has only one positive root for @xmath465 given by @xmath466 this root yields a formula for @xmath467 in ( [ omega - star ] ) . since there is a unique value for @xmath468 , for which the case @xmath469 is possible , we shall now consider whether @xmath446 or @xmath470 for @xmath471 or @xmath447 . to inspect the range @xmath470 , we consider a particular case , for which the intersection @xmath472 happens at @xmath4 . in this case , @xmath473 given by ( [ a - plus ] ) , so that the condition @xmath474 can be rewritten as @xmath475 there is only one negative root for @xmath37 and it is given by @xmath476 . by continuity , we conclude that @xmath444 for @xmath477 and @xmath452 for @xmath478 , both cases correspond to @xmath448 . finally , we verify that the case @xmath444 occurs for @xmath446 if @xmath447 . indeed , @xmath479 and @xmath480 as @xmath198 , so that @xmath481 as @xmath482 . on the other hand , the previous estimates suggest that @xmath483 for every @xmath425 if @xmath447 . therefore , there exists at least one intersection @xmath444 for @xmath446 if @xmath447 . we have reduced newton s equation of motion for coupled pendula shown on figure [ pendulapic ] under a resonant periodic force to the @xmath0-symmetric dnls equation ( [ pt - dnls ] ) . we have shown that this system is hamiltonian with conserved energy ( [ energy - pt - dnls ] ) and an additional constant of motion ( [ charge - pt - dnls ] ) . we have studied breather solutions of this model , which generalize symmetric synchronized oscillations of coupled pendula that arise if @xmath4 . we showed existence of three branches of breathers shown on figure [ branches ] . we also investigated their spectral stability analytically and numerically . the spectral information on each branch of solutions is shown on figure [ spectra ] . for branch ( c ) , we were also able to prove orbital stability and instability from the energy method . the technical results of this paper are summarized in table 1 and described as follows . for branch ( a ) , we found that it is disconnected from the symmetric synchronized oscillations at @xmath4 . along this branch , breathers of small amplitudes @xmath5 are connected to breathers of large amplitudes @xmath5 . every point on the branch corresponds to the saddle point of the energy function between two wave continua of positive and negative energies . every breather along the branch is spectrally stable and is free of resonance between isolated eigenvalues and continuous spectrum . in the follow - up work @xcite , we will prove long - time orbital stability of breathers along this branch . for branch ( b ) , we found that the large - amplitude breathers as @xmath482 are connected to the symmetric synchronized oscillations at @xmath4 , which have the smallest ( but nonzero ) amplitude @xmath119 . breathers along the branch are spectrally stable except for a narrow instability bubble , where the isolated eigenvalue @xmath402 is in resonance with the continuous spectrum . the instability bubble can occur either for @xmath414 , where the breather is a saddle point of the energy function between two wave continua of opposite energies or for @xmath415 , where the breather is a saddle point between the two negative - definite wave continua and directions of positive energy . when the isolated eigenvalue of positive energy @xmath402 is above the continuous spectrum near @xmath404 and @xmath367 , orbital stability of breathers can be proved by using the technique in @xcite , which was developed for the dnls equation . finally , for branch ( c ) , we found that the small - amplitude breathers at @xmath484 are connected to the symmetric synchronized oscillations at @xmath4 , which have the largest amplitude @xmath116 . breathers are either spectrally stable near @xmath344 or unstable near @xmath4 , depending on the detuning frequency @xmath37 and the amplitude of the periodic resonant force @xmath39 . when breathers are spectrally stable , they are also orbitally stable for infinitely long times . * acknowledgements . * the authors thank i.v . barashenkov and p.g . kevrekidis for useful discussions regarding this project . the work of a.c . is supported by the graduate scholarship at mcmaster university . the work of d.p . is supported by the ministry of education and science of russian federation ( the base part of the state task no . 2014/133 , project no . 2839 ) . c. huygens , _ the pendulum clock or geometrical demonstrations concerning the motion of pendula as applied to clocks _ , iowa state university press series in the history of technology and science ( iowa state university press , ames , ia , 1986 ) . mohamed , m.a . rahman and r. seethapathy , _ robust line - voltage sensorless control and synchronization of lcl - filtered distributed generation inverters for high power quality grid connection _ , power electronics , ieee transactions * 27 * , 87 ( 2012 ) .

two of the most enigmatic types of high redshift galaxy are the extremely red objects ( eros , eg . * ? ? ? * ; * ? ? ? * ) , and the ly@xmath1 blobs ( eg . * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? eros are defined as having extremely red observed - frame optical / near - ir colors , and may be dusty starburst galaxies or agn ( eg . * ? ? ? * ; * ? ? ? * ) , or spheroidal galaxies ( eg . * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the ly@xmath1 blobs are characterized by 100 kpc scale low surface brightness ly@xmath1 nebulae , with total ly@xmath1 luminosities @xmath6 . all ly@xmath1 blobs identified to date appear to lie in proto - cluster environments . it has been proposed that these blobs are a form of agn , that they are associated with cooling flows @xcite or that they are generated by galaxy superwinds @xcite . can a galaxy be both an ero and a ly@xmath1 blob ? many high redshift radio galaxies contain ero nuclei embedded within dense gaseous regions , which are often associated with extended diffuse ly@xmath1 emission ( eg . * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in this paper we present a detailed study of a _ radio quiet _ ly@xmath1 blob that contains two compact eros . @xcite identified 2142@xmath04420 b1 , which is a luminous ly@xmath1 and emitting source at redshift @xmath7 . the enormous ly@xmath1 luminosity of this source comes from an extended @xmath8 kpc low surface brightness nebula . ground - based near - ir observations showed an unresolved , luminous , extremely red source lying within this blob . the whole system appears to lie within a cluster of lyman - limit qso absorption - line systems @xcite . the optical properties of this source closely resemble those of many high redshift radio galaxies , but @xcite obtained an upper limit on its radio flux of 0.27 mjy at 1344 and 2378 mhz . we describe our observations in section [ obs ] , and show the results in section [ results ] . we then discuss the nature of the ero ( section [ continuum ] ) and of the ly@xmath1 blob surrounding it ( section [ lines ] ) . in section [ conclusions ] we conclude that this object is probably a pair of young merging elliptical galaxies , though a compact dusty starburst model is hard to completely exclude . we further conclude that the ly@xmath1 nebula surrounding them is either photoionized by an agn or shock excited by the energy of gravitational infall or of a superwind . we assume a flat universe with @xmath9 , @xmath10 and @xmath11 . at @xmath7 , given this cosmology and redshift , one arcsecond corresponds to @xmath12 proper kpc , and the luminosity distance is @xmath13 . our previous observations of this object can be found in @xcite and @xcite . the observation log is in table [ obstab ] . hst imaging was used to determine the morphology and colors of b1 . ground - based imaging measured its integrated colors and diffuse ly@xmath1 emission . slitless spectroscopy and fabry - perot imaging was used to constrain the velocity structure . in addition to our own observations , we searched the rosat all sky survey @xcite for data on this object . there is only a 308 sec exposure toward the field , giving a 90% upper limit x - ray count rate of 0.0075 counts sec@xmath14 . assuming a galactic column density of @xmath15 @xmath16 along this line of sight ( which is the weighted mean of the four nearest 21 cm measurements ) and a power law spectrum with an energy index of 1.0 , we obtain a flux @xmath17 e@xmath18 photons @xmath16 sec@xmath14 kev@xmath14 where e is the energy in kev . lccclc hst / wfpc2 & f410 m & @xmath19 & 67200 & 1999 aug 27sep 8 & + hst / wfpc2 & f450w & @xmath20 & 19400 & 1999 aug 31sep 2 & + hst / wfpc2 & f702w & @xmath21 & 14400 & 1999 aug 23 & + hst / nicmos & f110w & @xmath22 & 5120 & 1998 jun 27 & + hst / nicmos & f160w & @xmath23 & 10112 & 1998 oct 30 & + hst / nicmos & f164n & @xmath24 & 10236 & 1997 oct 30 & + hst / stis & clear & & 3120 & 1998 oct 2527 & + hst / stis & g430l & & 17000 & 1998 oct 2527 & + ntt / sofi & @xmath25 & @xmath26 & 3540 & 1999 aug 16 & + ntt / sofi & @xmath27 & @xmath28 & 4740 & 1999 aug 16 & + ntt / sofi & @xmath29 & @xmath30 & 3480 & 1999 aug 16 & + ctio / mosaic & 4107 & @xmath31 & 18000 & 1999 aug 7 & + ctio / mosaic & @xmath32 & @xmath33 & 9000 & 1999 aug 8 & + ctio / mosaic & @xmath34 & @xmath35 & 7200 & 1999 aug 78 & + ctio / mosaic & @xmath36 & @xmath37 & 7800 & 1999 aug 78 & + ctio / mosaic & @xmath38 & @xmath39 & 2400 & 1999 aug 8 & + ctio / mosaic & @xmath40 & @xmath41 & 4800 & 1999 aug 8 & + aat / ttf & 4110 & @xmath42 & 5400 & 1999 sep 11 & + aat / ttf & 4114 ? & @xmath43 & 5400 & 1999 sep 11 & + aat / ttf & 4118 & @xmath44 & 9000 & 1999 sep 12 & + the cluster region was observed for 38 orbits with the wide field and planetary camera ii ( wfpc2 , * ? ? ? * ) on the hubble space telescope . galaxy b1 was placed on the wf3 chip . exposures were dithered on a sub - pixel grid , to allow reconstruction of a better sampled image . broad - band images were obtained through the f450w and f702w filters . relatively narrow - band images were taken through the f410 m filter , which matches the wavelength of ly@xmath1 at the cluster wavelength . the f410 m images suffer from a very low surface brightness , time - varying mottling . we hypothesise that this is caused by scattered earth - light . the images were pipeline processed , then cosmic rays were removed and the frames co - added using the drizzle algorithm @xcite to minimise undersampling . the space telescope imaging spectrograph ( stis , * ? ? ? * ) observations ( pairs of slitless spectra and images ) were taken at two different orientation angles separated by 11.5 deg , in an attempt to separate slitless spectra from objects aligned along the same row for a given angle . for both the direct clear - aperture and the dispersed images , the ccd was binned @xmath45 , producing a plate scale of @xmath46 arcsec / pixel , and spectral resolution @xmath47 for point sources . there were 20 exposures each for the direct and dispersed images , with dithering of @xmath48 arcsec between each exposure in order to minimize the effects of cosmic rays and hot pixels . the direct and dispersed images were combined using the stis gto team software calstis @xcite . the spectra were extracted using both the stis_extract @xcite and slwidget @xcite software . results from both were consistent . the near infrared camera and multi object spectrometer ( nicmos , * ? ? ? * ) observations were made with camera 3 , through the f110w , f160w and f164n filters . all exposures were dithered for each filter over 4 ( or 8 in the case of f160w ) positions spanning a @xmath45 arcsec box , to minimize the effects of bad pixels . the f160w and f164n observations are affected by vignetting from the field offset mirror , and were taken before the focus became optimal . the nicmos data were reduced using the same methods as for sofi ( section [ irim ] ) . the data were also reduced using nicred @xcite ; there is little difference between final images reduced by the two methods . a comparison of broadband sofi and nicmos data show the photometry to be consistent between the two , with the exception of a vignetted band along one end of the chip in the f160w and f164w images . near - ir photometry was obtained with the sofi camera on the eso new technology telescope ( ntt , * ? ? ? conditions were photometric , but seeing was @xmath49 1.1 and the telescope was subject to considerable wind shake . the data were reduced using a modified version of iraf scripts written by peter mcgregor . nicmos standards @xcite were used . we obtained broad- and narrow - band ( ly@xmath1 ) imaging of this field with the mosaic camera @xcite , at the prime focus of the cerro tololo inter - american observatory ( ctio ) 4 m blanco telescope . the narrow - band images were taken through a specially purchased filter , which working in the f/2.8 beam has a bandpass of 54 , centered at 4107 . conditions were mostly photometric , with seeing @xmath50 . our anglo - australian telescope imaging was obtained with the taurus tunable filter ( ttf , * ? ? ? * ) , which is a fabry - perot etalon system . conditions were photometric on the first night , and the typical image quality was @xmath51 . a mit - lincoln labs ccd was used behind the ttf . the ttf was used with a full - width at half maximum ( fwhm ) spectral resolution of between 3.35 and 4.4 . images were obtained at central wavelengths of 4110 , 4114 , and 4118 . an electronics fault was subsequently found to induce significant wavelength drifts on timescales of a few hours . the 4110 and 4118 observations were taken within four hours of a calibration , but the 4114 observations were taken between four and eight hours after the calibration . thus the wavelength calibration of all three images are somewhat uncertain , particularly that of the 4114 image . we obtained a long - slit spectrum of b1 with the low dispersion survey spectrograph ( ldss , * ? ? ? * ) . these observations are described in @xcite our images of b1 are shown in figures [ wfpc_frame ] & [ zoom_frame ] . the appearance of b1 varies dramatically as a function of wavelength and resolution . in the observed - frame red and near - ir , it breaks up into two compact components : b1a and b1b ( fig [ zoom_frame ] ) , separated by 0.8 . in this respect it is similar to another strong , high redshift ly@xmath1 source , the coup - foure galaxy @xcite . a third continuum source , b1c , appears as chain of diffuse knots extending about 1 to the south from b1a and b1b . in ly@xmath1 , the picture is different again . when viewed with wfpc2 , we see a knot of ly@xmath1 emission from b1b , but no emission from b1a . diffuse patches of low surface brightness ly@xmath1 emission are found @xmath52 to the south and north - east of b1b . the ground - based ctio image has much greater sensitivity to low surface brightness ly@xmath1 emission , and confirms that it extends at least 10 north - east of b1a and b1b , as reported by @xcite and @xcite . the non - detection of this emission in the wfpc2 image indicates that it is truly diffuse , and not coming from a series of compact sources separated by less than the ground - based resolution limit ( as was the case for the ly@xmath1 nebula observed by @xcite ) . the ly@xmath1 contribution to the blue continuum images , and the blue continuum contribution to the narrow - band ly@xmath1 images , were removed . b1 has a total ly@xmath1 flux of about @xmath53 : a more exact number is hard to calculate as the diffuse flux fades gradually into the sky ( this number is for a rectangular aperture 10 long and 5 wide , aligned along the major axis of the ly@xmath1 flux ) . b1b has a ly@xmath1 flux of @xmath54 , which is only @xmath55% of the total . the remaining flux is diffuse : the average ly@xmath1 surface brightness of the region within 2 of b1b is @xmath56 . 4further east , the average surface brightness drops to @xmath57 . continuum fluxes ( table [ magtab ] ) were calculated for the three components of b1 . small circular apertures were used : 0.3 in radius for b1a and b1b , and 0.65 for b1c . these small apertures will miss much of the flux of these components , but were chosen to avoid contamination from the other components . no aperture corrections were attempted . a total flux for b1 was also measured , using a 2radius aperture . error estimates were made using apertures placed randomly on sky regions . a copy of the image was made with all the sources detected by the sextractor @xcite package masked out . we randomly placed 1000 apertures of identical radius on the image . any aperture that fell on a masked region was rejected , and the procedure continued until 1000 clean measurements had been obtained . the standard deviation of these measurements is our sky error estimate . a poisson noise model was combined with the sky error estimate to calculate object flux errors . the ly@xmath1 contribution to the f450w flux has been removed , but the probable emission - line contributions to f110w , f160w and @xmath29 have not been removed : these contributions are discussed in section [ linesub ] . finally , the fluxes were corrected for the local milky way extinction ( @xmath58 ) , as estimated by @xcite . the extinction curve of @xcite was used . integrated near - ir magnitudes for b1 were measured from the ntt image , using sextractor with 4.2 radius circular apertures . we measure @xmath59 , @xmath60 and @xmath61 . the measured coordinates of the components of b1 are shown in table [ coords ] . lc b1a & 21:42:27.45@xmath044:20:28.69 + b1b & 21:42:27.51@xmath044:20:28.99 + b1c & 21:42:27.46@xmath044:20:30.2 + lcccc f450w ( @xmath34 ) & @xmath62 & @xmath63 & @xmath64 & @xmath65 + f702w ( @xmath38 ) & @xmath66 & @xmath67 & @xmath68 & @xmath69 + f110w ( @xmath25 ) & @xmath70 & @xmath71 & @xmath72 & @xmath73 + f160w ( @xmath27 ) & @xmath74 & @xmath75 & @xmath76 & @xmath77 + the near - ir filters are subject to possible emission - line contamination at this redshift , z=2.38 . [ ] ( 3727 ) is redshifted into the @xmath25-band , h@xmath78 and the [ ] ( 4959 and 5007 ) doublet are shifted into the @xmath27-band , and h@xmath1 and [ ] ( 6583 ) are shifted into the @xmath79 band . these lines could in principle be very strong ( eg . * ? ? ? * ) . in this section , we estimate their strength , and conclude that it is probably small . as described in @xcite , we imaged the field with a narrow - band filter of wavelength coverage @xmath80 m , which includes both h@xmath1 and [ ] . using a 5 radius aperture , we determine that these lines contribute @xmath81% of the integrated @xmath29 flux of b1 . the f164n nicmos image covers the wavelength of redshifted h@xmath78 . no significant narrow - band excess was detected : we can place a @xmath82 upper limit on the fraction of the integrated @xmath27-band light coming from h@xmath78 in b1 of 3% . we have no direct measurements of the other line fluxes , so we estimate their importance by making assumptions about their ratios with measured balmer lines . typical line ratios were taken from @xcite , @xcite and @xcite , for a variety of assumed chemical compositions and ionization sources , and including both narrow- and broad - line agn ratios . we conclude that line emission probably contributes only @xmath83% at most of the total continuum flux of b1 in @xmath27 and @xmath29 . in @xmath25 , the contribution is probably @xmath84% . furthermore , only about 10% of the line flux will likely come from within the photometric aperture of b1b , and @xmath83% from within the aperture of the b1c . the remainder of the optical emission lines will come from the region of the diffuse blob . as b1a shows no ly@xmath1 emission , its rest - frame optical line flux is also probably small . the line corrections will thus probably be minimal . we have not , therefore , applied any emission - line corrections to the near - ir fluxes . the ly@xmath1 line is generally extremely optically thick and often strongly self absorbed , making it a very unreliable tracer of dynamics . it is , however , the only sufficiently strong line available to us , and can be used to place some limits on the dynamics of the ly@xmath1 nebula surrounding b1 . the velocity dispersion of the gas around b1 ( measured at the peak of the surface brightness , as seen in ground - based images ) is @xmath85 @xcite . ly@xmath1 is redshifted with respect to , indicating probable self absorption . the deep multislit spectroscopy of @xcite marginally spatially resolves this emission along the slitlet . fig [ longslit ] shows the spectra extracted from different ccd columns : notice the extra component at @xmath86 in the southernmost spectrum ( and another possible component at @xmath87 ) . the diffuse ly@xmath1 structure differs significantly between the three ttf images . this confirms the existence of multiple velocity components separated by @xmath88 . it is hard to say more , given the uncertainty in the wavelength calibration of the ttf data ( section [ ttf ] ) . the stis spectrum detected the ly@xmath1 flux from b1b . a flux of @xmath89 was measured , consistent with the value derived from the wfpc2 imaging . the line was not significantly spectrally resolved ; an upper limit on the velocity dispersion of @xmath90 can be placed . the spectral energy distributions ( seds ) of the three components of b1 are shown in fig [ seds ] . b1a and b1b are quite red , while the sed of b1c is very blue . b1c s colors are similar to those of lyman - break galaxies ( eg . * ? ? ? . a power - law fit to its colors gives a slope of the form @xmath91 , which is close to the median slope of lyman break galaxies ( 1.5 , * ? ? ? our ntt photometry demonstrates that while b1 is red at wavelengths shortward of @xmath27 , it has a relatively blue @xmath92 color . this confirms the result of @xcite . b1 s sed thus peaks in the @xmath27-band . as we discussed in section [ linesub ] , this is unlikely to be an artifact of emission - line contamination . we modelled the seds of b1 using the 1997 version of the spectral synthesis models of @xcite . models both with continuous uniform star formation and models in which all the stars were formed in an instantaneous burst ( a simple stellar population ) were used . all models had salpeter stellar initial mass functions , with no mass cut - off . dust was modelled using the empirical absorption curves of @xcite . our predicted colors are compared with the observations in figures [ brhmodel ] and [ jhkmodel ] . the red components , b1a and b1b , can be fit by two models : * an unreddened burst model of age @xmath93 myr ( 1@xmath94 limits ) and stellar mass @xmath95 . this naturally fits the @xmath27-band peak ( fig [ jhkmodel ] ) and the photometry from @xmath96 thru @xmath29 . it greatly underpredicts the observed f450w flux of b1a . we do not consider this to be a problem : a star formation rate of only @xmath97 will produce ob stars sufficient to produce the observed blue flux . * a dusty starburst model . the best fit is a model with a continuous star formation rate of @xmath98 , an extinction of @xmath99 mag , and an age of 500 myr . a slightly worse fit ( but still acceptable at the @xmath100 level ) is younger ( 100 myr ) , dustier ( @xmath101 mag ) and has a star formation rate of @xmath102 . the dusty starburst model is hard to reconcile with the strong ly@xmath1 emission of b1b , as ly@xmath1 is resonantly scattered and thus very strongly absorbed by even small amounts of dust . b1c is much bluer , and can be fit either by an unreddened instantaneous burst model of age @xmath8 myr and mass @xmath103 , or by a slightly dusty continuous star formation model , with a star formation rate of @xmath104 and @xmath105 . in this section , we show that b1a and b1b are spatially extended . they are , however , very compact objects : too compact to be easily modelled as spiral galaxy disks . they appear to show colour gradients : bluer in their central regions . in figures [ rpsf ] and [ jpsf ] we compare the radial profiles of b1a and b1b with the relevant point spread functions ( psfs ) measured from a bright star in the same image . b1a and b1b are marginally resolved in the f702w ( rest - frame 2100 ) and f110w ( rest - frame 3250 ) images . the f160w ( rest - frame 4730 ) image was taken with nicmos out of focus , and it does not significantly resolve b1a and b1b . the observed psf does not vary significantly across the images , and is consistent with analytic predictions . although b1a and b1b are clearly resolved , both are very compact , with the observed surface brightness dropping to half its peak value within about 0.1 ( @xmath52 kpc ) . this places a strong constraint on the size of the galaxies . consider a disk galaxy , with a typical radial surface brightness profile of the form @xmath106 @xcite . the surface brightness of such a disk drops to half its peak value at @xmath107 ( fig [ newfig ] ) . thus our observations require that @xmath108 kpc . only @xmath109% of modern galaxy disks are this compact @xcite . now consider a spheroidal galaxy , with a radial surface brightness profile of the form @xmath110\}\ ] ] @xcite . this profile is much more sharply peaked than the exponential disk profile , and its surface brightness drops to half its peak value at @xmath111 ( fig [ newfig ] ) . thus @xmath112 kpc . over 50% of modern elliptical galaxies have effective radii this small @xcite . more detailed modelling confirms these results . a wide variety of model galaxy profiles were convolved with psfs and fit by @xmath113 minimisation to the observed radial profiles . the modelling confirms that b1a and b1b are significantly extended . both exponential disk models and spheroidal ( de vaucouleurs ) models give acceptable fits to the data , as do models with a point source embedded in a fainter halo . pure exponential disk models only give acceptable fits if @xmath114 ( @xmath115 pc ) , which would place b1a and b1b in the most compact 1% of present - day disk galaxies . spheroidal models give best fits for effective ( half - light ) radii @xmath116 of around @xmath117 ( @xmath118 kpc ) , which are typical of modern elliptical galaxies . models which combine a nuclear spheroid or point source with a disk can give acceptable fits , but only if the disk is unusually small and/or contributes only a small fraction ( @xmath119 10% ) of the light . the compact radial profiles of b1a and b1b are typical of those of high redshift lyman - break galaxies @xcite . the inferred surface brightnesses in the rest - frame uv are comparable to those of faint lyman - break galaxies , but the rest - frame optical surface brightnesses are an order of magnitude higher . as @xcite note , these compact radial profiles are consistent with present - day elliptical galaxy profiles or bulges , but are much more compact than present - day disks . curiously , both b1a and b1b appear to be significantly larger at longer wavelengths . this can be seen in figures [ rpsf ] and [ jpsf ] : despite the wider psf for the f110w filter , both components are clearly more extended than as seen through the f702w filter . our modelling confirms this result : even after convolution with the relevant psfs , no single model gives a good fit to either b1a or b1b at both f702w and f110w . the central @xmath52 kpc of both components is @xmath120 magnitudes bluer in f702w@xmath0f110w than the outer regions . the change in size as a function of wavelength could be explained by a colour gradient within a single spheroidal component . it could also be explained by a two component model : a compact blue component that dominates at rest - frame 2100 and a more diffuse ( but still compact ) red component that dominates at 3250 and beyond . this red component , if disk - like , must have @xmath121 , which would still place it in the most compact 1% of low redshift disks . what are the continuum components of b1 ? b1c , with its blue colors , appears relatively straightforward : it is a region of extended moderate star formation , obscured by little dust . this star formation may be triggered by the interaction of b1a and b1b . the nature of b1a and b1b is less straightforward . we consider four models in turn . could b1a and b1b be young elliptical galaxies ? this hypothesis fits the data well . * the radial profiles of b1a and b1b are well fit with de vaucouleurs profiles , and the inferred radii are typical of present - day elliptical galaxies . * the red colors are most easily explained by a @xmath122 stellar population which completed its star formation about 750 myr before we observe it . it would thus be similar to modern e@xmath123a galaxies . * b1 lies in an overdense region of the early universe @xcite , which may be the ancestor of a galaxy cluster . * the velocity field around b1 , if virial , implies masses of @xmath124 . if b1a and b1b are elliptical galaxies , how do their surface brightnesses and sizes fit on the kormendy relation @xcite ? the f160w filter corresponds quite closely to the rest - frame @xmath34-band . the radial profile fitting suggests that both components have effective radii in the observed - frame near - ir of @xmath118kpc . integrating a de vaucouleurs profile over our photometric aperture , we can convert our observed @xmath125 magnitudes into rest - frame @xmath34-band surface brightnesses at the effective radius . were these components at low redshift , their surface brightnesses at the effective radius would be @xmath126 . this is roughly an order of magnitude higher than the kormendy relation would predict for an elliptical galaxy of this radius . this is consistent with passive evolution from our unreddened burst model ( section [ sed ] ) . if b1a and b1b are elliptical galaxies , why do they apparently show color gradients ? the uv emission from the central @xmath52 kpc could be caused by a starburst , a hole in the dust or ( for b1b ) an agn . alternatively , some chemodynamical models predict color gradients in young elliptical galaxies ( eg . * ? ? ? * ; * ? ? ? a dusty agn can certainly have colors as red as b1a and b1b ( eg . * ? ? ? the red components of b1a and b1b are however spatially extended , and hence can not be produced by an agn . could the observed color gradient , however , be caused by the superposition of a compact blue agn on a more extended , lower surface brightness red galaxy ? even in the relatively blue central regions , the continuum slope is redder than @xmath127 , while radio - quiet qsos have typical continuum slopes of @xmath128 @xcite . any central agn would thus have to be reddened by dust with @xmath129 . this would make it hard for the ly@xmath1 emission from b1b to escape . in addition , the agn in b1a and b1b would need remarkably equal luminosities and dust extinctions to explain the very similar observed colors and radial profiles of the two components . this fine tuning seems implausible . furthermore , the high measured surface brightness was measured in the f160w band , to which the central blue component does not significantly contribute . we therefore conclude that while agn may be present , they are unlikely to be responsible for the observed colors . sensitive hard x - ray or far - ir observations will be required to determine whether a dusty agn is present . the more extended red component of b1a and b1b can be fit by a very compact exponential disk . could these two objects then be compact spiral galaxies ? the more compact bluer central component could be caused by agns or nuclear starbursts . with sufficient dust , a spiral galaxy can certainly appear as red as these two components ( section [ sedmodel ] ) . two lines of reasoning however oppose this hypothesis . 1 . even the extended red components are extremely compact by the standards of modern disk galaxies . this could , however , be explained if disk galaxies form from the inside outwards . this model requires that the outer ( red ) components of b1a and b1b be greatly reddened . this would make it difficult for the observed uv light and ly@xmath1 emission from the bluer central components to escape . could b1a and b1b be the bulges of two disk galaxies ? the disks themselves may not have formed , or might be too low in surface brightness to be detected by the hubble space telescope . two lines of reasoning oppose this hypothesis . firstly , the red colors and @xmath27-band peak of b1 are seen in our ground - based photometry with large ( @xmath130 ) apertures . thus any extended component must have a similar stellar population to that inferred here . secondly , the inferred masses and sizes for b1a and b1b are far larger than those typical of the bulges of disk galaxies . b1 has a total ly@xmath1 luminosity of @xmath131 , spread over a region at least @xmath132 kpc in size . the emission consists of a number of diffuse , discrete components with relative velocities of @xmath88 . what can produce such a luminous , diffuse , fast moving ly@xmath1 nebula ? the bulk motions of the ly@xmath1 nebula suggest that fast shocks must be present . we will show that such shocks are quite capable of producing the observed ly@xmath1 luminosity . the puzzle then becomes one of explaining the origins of the bulk motions . it has also been suggested that photoionzation by an agn , or a cooling flow may power such nebulae . we discuss these possibilities , which are hard to exclude . if we assume that the ly@xmath1 velocities are representative of the gas , and not an artifact of the high optical depth in ly@xmath1 , then the dynamical timescale ( crossing time ) of the nebula is @xmath133 years . this is interestingly similar to the inferred age of the stellar population ( section [ sed ] ) . what is the density of the nebula ? the sight - line to background qso 2139@xmath04434 passes 20 from b1 ( though it does not pass through the ly@xmath1 nebula ) . the qso spectrum shows a strong absorption - line system at the redshift of b1 , with a column density @xmath134 @xcite . if we assume that the column density through the ly@xmath1 nebula is at least as great as this , and that the nebula is @xmath83 kpc thick ( ie . as thick as it is wide ) along the line of sight ( both big assumptions ) , then its density must be at least @xmath135 . if , however , the density were this low , the gas would be fully ionized by the metagalactic uv background at this redshift . if the nebula is @xmath83 kpc thick , it must have a density of @xmath136 for its recombination rate to balance the photoionization from the uv background @xcite . the diffuse star formation of b1c , taking place within part of the ly@xmath1 nebula , also suggests that densities may be higher . in reality , of course , the cloud may be highly inhomogeneous , with dense neutral ` bullets ' embedded in low density ionized gas . if we take this lower limit on the density of the nebula ( @xmath137 ) , and assume that the nebula is 100 kpc long , 30 kpc wide and 10 kpc thick , then its hydrogen mass is @xmath138 . whenever there are blobs of gas moving at supersonic relative speeds , as are observed in the nebula , shocks are inevitable . we thus know that shocks are present . could the emission from these shocks produce the observed ly@xmath1 emission ? we used the fast shock models computed with the mappings iii code to calculate the line emission from the shocks . mappings iii is an updated version of the mappings ii code @xcite and uses improved radiation continuum and hydrogenic calculations to calculate the line and continuum radiation from shocks . the new version of the code enabled the calculation of a grid of shocks for a range of metallicities , up to 1000 km / s , although a small number of low metallicity models above @xmath139 failed to converge after several global iterations . a fiducial model with a velocity of @xmath140 was used here , having achieved a converged solution over a range of metallicities . if the shocks are radiative , and if we assume , for example , a density of @xmath141 , then the ly@xmath1 emissivity will be @xmath142 , almost independent of the metallicity of the gas . a quarter of this comes from the shock itself , the rest coming from photoionization of the assumed neutral gas in front of the shock ( the precursor region ) . if the shock is face - on to our sight - line , this would correspond to an observed surface brightness of @xmath143 : nearly two orders of magnitude greater than the observed ly@xmath1 surface brightness ( section [ results ] ) . much of this may be self absorbed , but if even 1% escapes , it can explain our observations . the predicted ly@xmath1/ ratio is also consistent with the observations @xcite . how does this vary with the assumed density of the gas ? if the shocks are radiative , the emissivity will scale roughly with the density . thus even with our lower limit on the density @xmath144 ( section [ lines ] ) , shocks can produce all the ly@xmath1 we observe . would these shocks be radiative ? for our lower limit on the density , @xmath145 shocks become radiative on a timescale of @xmath133 years , which is comparable to the dynamical timescale . thus it seems likely that at least some fraction of the shocks will be radiative . thus shocks are probably present , and can easily produce the ly@xmath1 emission we observe . a number of factors could , however , suppress the ly@xmath1 emission from shocks . 1 . the high optical depth in ly@xmath1 may have caused us to overestimate the velocity dispersion of the ly@xmath1 nebula . the emission from shocks is a strong function of their velocity : @xmath146 shocks emit two orders of magnitude less ly@xmath1 per unit area than our fiducial model . 2 . the density of the ly@xmath1 nebula may be close to our lower limit of @xmath144 , in which case the ly@xmath1 emission , after the probable self absorption losses , could be an order of magnitude below the observed ly@xmath1 surface brightness . 3 . the ly@xmath1 nebula could consist of small dense ` bullets ' of neutral gas , moving at high speeds through a hot , highly ionized , low density medium . the bowshocks in the ionized medium would not be radiative , while the reverse shocks within the ` bullets ' would be slow , and hence have low surface brightnesses ( as well as having a small area ) . if the shock model is correct , the bulk gas motions explain the ly@xmath1 luminosity . what then could be the energy source for the bulk gas motions ? if we divide the total kinetic energy of the nebula ( assuming a mass of @xmath147 ) by the ly@xmath1 luminosity , we can derive an upper limit on the damping time , which is @xmath148 years . this is less than the dynamical timescale and stellar age of b1 , suggesting that continued energy input is required . note that the ly@xmath1 we observe may only be a small fraction of the ly@xmath1 emitted , which in turn will only be a small fraction ( @xmath52% ) of the total energy dissipation in the shocks . on the other hand , the gas mass may be greater than we are assuming . if continued energy input is required , what can the source be ? perhaps the most natural energy source is the gravitational potential energy of b1 . if we assume that b1 has a total mass of @xmath149 , then the gravitational potential energy released in its formation would be @xmath150 ergs , which is more than sufficient to power the ly@xmath1 luminosity for the dynamical timescale . such a mass for b1 would imply virial velocities in the ly@xmath1 nebula of @xmath151 , which are comparable to the bulk gas motions observed . this energy could be released by the merger of b1a and b1b , or by a continuing mass infall rate of @xmath152 . @xcite point out that chemically primordial gas at a temperature of @xmath153 k has few cooling mechanisms other than ly@xmath1 emission . thus if most of the gas around b1 is primordial , a ly@xmath1 luminosity comparable to the gravitational binding energy ( section [ energy ] ) of the system must be radiated . alternatively , @xcite suggest that galactic superwinds driven by a starburst could power extended ly@xmath1 nebulae . this would naturally explain the similarity between the dynamical timescale and the inferred age of the stellar population in b1a and b1b . the maximum plausible amount of energy which star formation can inject into the intergalactic medium is @xmath154 ( eg . as b1a and b1b have a combined stellar mass of @xmath155 , the total energy liberated would be @xmath156 ergs , which could in principle drive the observed ly@xmath1 luminosity for @xmath157 years . the ly@xmath1 emission in high redshift radio galaxies has sometimes been ascribed to mechanical energy deposition by a radio jet ( eg . * ? ? ? could such a jet be present in b1 ? we do not detect any radio emission from b1 @xcite . our @xmath82 upper limit of 0.23 mjy at 2.4 ghz corresponds to a luminosity limit of @xmath158 at rest - frame 1.4 ghz ( assuming a typical radio galaxy radio spectrum of the form @xmath159 ) . typical ratios of the total jet mechanical power to the monochromatic radio flux at 1.4 ghz are @xmath160@xmath161 @xcite . thus we can place an upper limit of @xmath162 on the energy injection from a radio jet . this is sufficient to power the observed ly@xmath1 luminosity only if @xmath163% of the mechanical energy is converted into observable ly@xmath1 emission , which seems implausible . deeper radio observations should test this hypothesis . if the shock emission is suppressed , is there an alternative mechanism capable of producing the ly@xmath1 emission ? stellar ionisation ( by hot young stars ) is incapable of producing the observed emission . photoionisation by the uv continuum radiation from a qso can , however , produce both the ly@xmath1 and , as discussed by @xcite . we do not see strong uv continuum emission from any component of b1 , so such an agn would have to be concealed from our sight - line . @xcite detected strong sub - mm emission from a ly@xmath1 blob at @xmath164 , suggesting that a dusty agn or starburst was present , but we have no evidence for such a source in b1 . extended emission - line nebulosities are associated with many low redshift cooling flow clusters , such as ngc 1275 . as all high - redshift ly@xmath1 blobs seem to lie in cluster environments , could the ly@xmath1 luminosity be associated with a cooling flow ( eg . * ? ? ? @xcite suggest that ly@xmath1 nebulae are caused by gas cooling from higher temperatures , in such a cooling flow . if each cooling hydrogen atom produces one ly@xmath1 photon , however , we require that @xmath165 of hydrogen be cooling onto b1 . this would build the observed baryonic mass of b1 in @xmath166 years . this is far smaller than either the dynamical time or the age of the stellar population in b1a and b1b , so the simple cooling model seems implausible . note that the nebular emission - line luminosities of low redshift cooling - flow clusters are also orders of magnitude too great to be explained by this simple recombination mechanism ( eg . * ? ? ? the cause of this discrepancy is still controversial even at low redshifts . on purely energetic grounds , could the ly@xmath1 luminosity be powered by the thermal energy in the nebula ? if we divide the thermal energy of @xmath167 of hydrogen at @xmath168 k by the observed ly@xmath1 luminosity , we get a timescale of @xmath148 years : shorter than the dynamical timescale or the stellar age . @xmath167 may however be an underestimate of the true gaseous mass of the ly@xmath1 nebula , particularly if the neutral gas we observe is immersed in a much more massive cloud of hot , highly ionised x - ray emitting gas . so what is b1 ? on balance , we conclude that b1a and b1b are probably young elliptical galaxies , perhaps analogues of e@xmath123a galaxies today . both their red colors and compact sizes are naturally explained by this model . an extremely dusty starburst in the central regions of a disk galaxy can however , with some fine tuning , reproduce the colors and sizes . the bluer colors of the centres of both components , and the ly@xmath1 emission from the core of b1b are , however , arguments against the presence of large amounts of dust in these systems . near - ir h - band spectroscopy on an 8m - class telescope should resolve this ambiguity , by detecting the spectral signatures of a balmer/4000break . b1c appears to be a region of relatively unobscured star formation , perhaps induced by the interaction of b1a and b1b . if b1a and b1b are elliptical galaxies , their surface brightnesses are an order of magnitude greater than those that the kormendy relation would predict for galaxies of this size , and they show strong radial color gradients , with blue cores . the very existence of massive elliptical galaxies at this redshift , if confirmed , is interesting ( eg . * and references therein ) . these galaxies would give us a chance to study young elliptical galaxies for the first time . our observations would pose several puzzles : * why is the surface brightness of these galaxies so great ? is this simply due to passive evolution ? * why do both galaxies show strong , but very similar color gradients ? * is the remarkable similarity and proximity of the two galaxies coincidental , or is it telling us something about how cluster elliptical galaxies form ? * if these galaxies are so massive and red at @xmath169 , what would they have looked like at @xmath170 when they were forming ? the nature of the ly@xmath1 nebula is less well constrained by our observations . the nebula could be excited by photoionization from an agn , by whatever powers low redshift cooling flow nebulae , or by fast shocks powered by either superwinds or gravitational potential energy . spatially resolved spectroscopy of other emission lines will help discriminate between these models . we wish to thank joss bland - hawthorn and catherine drake for their assistance at with the ttf observations , and all the staff at ctio for heroic efforts to get the mosaic camera working for us . we are grateful to mike dopita , megan donahue and martijn de kool and ian george for helpful discussions . hsiao - wen chen was instrumental in getting the slitless stis spectrum extraction software working . david hogg s cosmology crib sheet ( astro - ph/9905116 ) was very helpful to us .

over the past few years , the prospect of realizing majorana bound states ( mbss ) in superconducting hybrid devices has attracted a lot of attention , see refs . @xcite for reviews . mbss are predicted to emerge , for instance , as end states in proximitized one - dimensional rashba nanowires in a magnetic zeeman field , where signatures for their existence have already been reported in transport experiments @xcite . for these topologically superconducting ( ts ) nanowires , mbss are located at the respective ends of the wire , and the overlap between their wavefunctions becomes exponentially small for large nanowire length @xmath0 . mbss can then intuitively be thought of as spatially separated `` half fermions '' . they are under discussion as basic ingredients for `` topological '' qubits in quantum information applications , where information is nonlocally stored and , thus , should be quite robust against local decoherence processes . systems with mbss are also predicted to allow for non - abelian ising anyon braiding statistics . since ising anyons do not allow for universal quantum computation by themselves , it is essential to thoroughly understand the physics of majorana states coupled to conventional qubits , where the latter can be realized in terms of nanoscale quantum dots @xcite . majorana bound states are described by self - adjoint operators , @xmath1 , which anticommute with each other , @xmath2 , and with all other fermion operators . a pair of mbss effectively forms a single nonlocal fermion with annihilation operator @xmath3 where the eigenvalue @xmath4 of the number operator @xmath5 describes the state of the mbs pair @xcite . for large @xmath0 , the hybridization energy between the mbss forming this pair becomes exponentially small , @xmath6 the length @xmath7 refers to the spatial size of a mbs and is related to the superconducting coherence length in the proximitized nanowire @xcite . for @xmath8 , the mbs pair is then equivalent to a single zero - energy fermion . in this work , we consider the simplest case of a single mbs pair and study the concurrence of entanglement , @xmath9 , between two single - level quantum dots @xcite tunnel - coupled to the mbss , see fig . [ fig1 ] for a schematic illustration of the setup . entanglement is a key concept in quantum mechanics and represents an essential resource for many quantum computation schemes @xcite . two spatially separated systems are said to be entangled when their quantum states can not be described independently . one may naively expect that because each mbs amounts to half of the same fermion state , the mbss themselves are strongly entangled . however , this is a meaningless statement since only the mbs pair has a well - defined state representation , see eq . . on the other hand , the nonlocality of the @xmath10 fermion level suggests that the two dots coupled to the mbss could be fully entangled . for a grounded superconducting island , where the charging energy @xmath11 is negligible , this question has been studied in refs . @xcite , and one finds that precisely the opposite statement holds true in the long - distance limit @xmath12 , i.e. , both dots are perfectly disentangled . this result is consistent with the absence of correlations among currents flowing through normal - conducting leads in contact to different mbss @xcite . schematic illustration of the setup . large circles correspond to two quantum dots ( qd@xmath13 and qd@xmath14 ) , where gate voltages @xmath15 allow one to vary the energy levels @xmath16 . a rashba nanowire is deposited on a superconducting island with charging energy @xmath11 , where the proximity effect in combination with a zeeman field induces a ts phase . the dimensionless gate parameter @xmath17 ( see main text ) is proportional to a gate voltage @xmath18 regulating the average charge on the ts island . small ( red ) circles represent mbs operators @xmath19 and @xmath20 , where the corresponding wavefunctions are located near the ends of the nanowire and the hybridization @xmath21 corresponds to the distance @xmath0 between both mbss , see eq . . tunnel couplings @xmath22 and @xmath23 connect the dots to the respective mbss . in addition , the josephson energy @xmath24 describes the coupling to another grounded bulk superconductor ( sc ) . , width=283 ] in what follows , we are especially interested in the possibility of long - range entanglement generation , such that the concurrence @xmath25 does not decay up to very long dot distance @xmath0 @xcite . we shall determine the detailed conditions for the realization of such a `` majorana entanglement bridge '' in the setup shown in fig . in particular , we find that the charging energy @xmath11 plays an important role in allowing for long - range entanglement generation , i.e. , one should work with a mesoscopic floating ( not grounded ) superconducting island . we note in passing that standard notions of long - range entanglement refer to an exponential decay , @xmath26 , and merely require that the lengthscale @xmath27 is long @xcite . in particular , for the topologically trivial limit of our setup , i.e. , when the superconducting island does not host mbss , such an exponential decay of @xmath25 is expected , see also ref . @xcite , where @xmath27 is equivalent to the superconducting coherence length . the length - independence of @xmath25 predicted for the device in fig . [ fig1 ] holds as long as the charging energy , which scales approximately as @xmath28 in the ts nanowire geometry , remains the dominant energy scale . in particular , @xmath11 should exceed both the thermal energy @xmath29 and the mbs hybridization scale @xmath21 in eq . . indeed , charging effects can change the physics in a profound way by coupling mbss to the dynamics of the cooper pair condensate on the island . in ref . @xcite , fu studied the transmission through a setup as in fig . [ fig1 ] but with normal - conducting leads replacing the dots . for large @xmath11 and when the gate voltage is tuned to a coulomb blockade conductance peak , i.e. , the dimensionless parameter @xmath17 regulating the average island charge is half - integer , only two degenerate island charge states are accessible and the model effectively maps to a spinless resonant tunneling problem . in a slight abuse of terminology , the resulting nonlocal phase - coherent transfer of electrons from one lead to the other has been dubbed `` teleportation '' ( tp ) by fu @xcite , see also refs . however , it has been argued that entanglement may be achieved more efficiently under coulomb blockade valley conditions , i.e. , when @xmath17 stays close to an integer @xcite . as we discuss in detail below , entanglement is then a consequence of elastic cotunneling @xcite processes . in our setup , elastic cotunneling has a strongly nonlocal character due to the underlying mbs realization of the @xmath10 fermion , and therefore this process has also been referred to as `` teleportation '' @xcite , see sec . [ sec3a ] below . when more than two mbss are present on the superconducting island , a nonperturbative version of the tp mechanism triggers the so - called topological kondo effect @xcite . however , in the present work we shall focus on just one mbs pair , where topological kondo physics is absent . besides tp processes , however , it is well - known that entanglement can also be generated through crossed andreev reflection ( car ) processes . so far , car - induced entanglement in a system as shown in fig . [ fig1 ] has only been discussed for the grounded device , where @xmath30 . as mentioned above , when also @xmath31 , the dots will be perfectly disentangled , but for finite @xmath21 ( and hence not too large @xmath0 ) , car processes generate entanglement both for the case of normal - conducting leads @xcite and for the dot case at hand @xcite . this entanglement can be probed through the violation of bell inequalities when several mbs pairs are present @xcite . in general terms , car refers to the splitting of a cooper pair on the island which produces entangled electrons on different dots ( or leads ) @xcite . similarly , in the reverse process , a cooper pair is created by the combined in - tunneling of electrons from different sides . for normal - conducting leads , local andreev reflection competes with car , since both electrons can be created ( or annihilated ) also in the same lead , and hence perfect entanglement is not possible @xcite . for the single - level dots studied here , however , local andreev reflection is absent and perfect entanglement due to car processes may be realized . below we address the fate of car - generated entanglement as well as the interplay of car and tp processes when interactions ( @xmath32 ) are important . in addition , we have also studied the time - dependent concurrence , @xmath33 , found after one tunnel coupling is suddently switched on . such a `` quench '' can be experimentally implemented by changing the voltage on a finger gate in close proximity to the respective tunnel contact , cf . @xcite . by monitoring the time - dependent concurrence , one can achieve a better understanding of the timescales on which entanglement is built up in such a setting @xcite . for the setup in fig . [ fig1 ] with @xmath30 , this problem has recently been studied in ref . we here confirm and analytically explain the observations of ref . @xcite for the noninteracting case . more importantly , we also address the entanglement dynamics for finite charging energy , and reveal the underlying timescales governing this interacting case . before going into a detailed discussion , let us briefly summarize the main results of our work : ( i ) we consider the full crossover behavior of the concurrence all the way from vanishing to large charging energy , ( ii ) allow for arbitrary @xmath17 , ( iii ) carefully address issues related to fermion parity , ( iv ) include the effects of a josephson coupling @xmath24 to an additional bulk @xmath34-wave superconductor , ( v ) address the @xmath0-dependence of the concurrence by including the direct overlap @xmath21 between the mbss , and ( vi ) study the timescales of entanglement generation after the sudden change of a tunnel coupling . throughout the paper , we provide closed analytical expressions for the concurrence valid in different parameter regimes , both for the static and the time - dependent case , which are shown to give highly accurate approximations for our numerically exact results . importantly , our results can be quantitatively interpreted in terms of an interplay of tp and car processes , thereby providing a comprehensive physical framework to describe entanglement in mbs devices . in contrast to ref . @xcite , we provide a rather complete analysis of this problem , including both tp and car processes , and providing detailed analytical results for the concurrence . the structure of this paper is as follows . in sec . [ sec2 ] , we briefly summarize the model for the setup in fig . [ fig1 ] . for the detailed derivation of the model , we refer the reader to refs . we then turn to a discussion of the entanglement - generating tunneling mechanisms in sec . the concurrence , @xmath9 , provides a quantitative measure for entanglement of the two dots . we compute @xmath9 analytically in several parameter regimes of interest , and validate the results by comparing to exact numerical diagonalization in sec . [ sec4 ] . the time - dependent concurrence obtained after a quench of tunnel couplings will then be studied in sec . finally , we offer some concluding remarks in sec . [ sec6 ] . throughout the paper , we focus on the most interesting zero - temperature limit and use units with @xmath35 . we start by briefly summarizing the low - energy hamiltonian , @xmath36 , describing the setup shown in fig . [ fig1 ] , see also refs . @xmath37 contains ( i ) the part @xmath38 for the central ts island , where a mesoscopic superconductor with charging energy @xmath11 causes proximity - induced pairing in the nanowire , ( ii ) the josephson term @xmath39 coupling the ts island to a second bulk superconductor , ( iii ) the dot hamiltonian @xmath40 , and ( iv ) a tunneling hamiltonian @xmath41 connecting the dots and the island . we consider energy scales well below the proximity - induced superconducting gap , such that it is justified to neglect quasiparticle excitations of the ts ( see also the discussion in sec . [ sec6 ] ) . the state of the island is then fully described by specifying the integer cooper pair number @xmath42 on the central island , where the number operator @xmath43 is canonically conjugate to the superconductor s phase , @xmath44_-=-i$ ] , and the occupation number @xmath45 for the @xmath10 fermion built from the mbs pair , see eq . . including the mbs hybridization ( [ epsfl ] ) and a coulomb charging term , with the single - electron charging energy @xmath11 and a dimensionless gate - voltage parameter @xmath17 , the island hamiltonian is given by @xcite @xmath46 where @xmath47 . the nonlocal @xmath10 fermion representing the mbs pair thus couples to the condensate dynamics through coulomb charging effects . furthermore , the josephson coupling @xmath24 to another bulk superconductor ( which is held at constant phase @xmath48 and different from the superconductor responsible for the proximity - induced pairing in the nanowire ) is described by @xmath49 where the operator @xmath50 ( @xmath51 ) raises ( lowers ) the cooper pair number by one unit , @xmath52 . next , each quantum dot in fig . [ fig1 ] is assumed to be in the coulomb blockade regime such that the two dots can be modeled as single fermion levels at energies @xmath53 and @xmath54 , see ref . these levels can be tuned by electrostatic gates ( @xmath55 in fig . [ fig1 ] ) and by the zeeman field which induces the ts phase of the wire and also breaks spin degeneracy on the dots . as a consequence , we can use the effectively spinless fermion annihilation operator , @xmath56 , which is connected by a tunnel coupling @xmath57 to @xmath58 , see eq . this tunnel coupling also captures a possible spin dependence of microscopic transition amplitudes and can be taken as real - valued positive , see ref . the hamiltonian describing both dots then reads @xmath59 where the number operator @xmath60 has eigenvalues @xmath61 . equation represents a pair of charge qubits that can be entangled through the majorana island . while charge qubits are more susceptible to detrimental noise than spin qubits @xcite , it is more transparent to analyze the concurrence within the present formulation . the extension of the above model to the case of spin qubits , where local qubit operations are also easier to implement , is rather straightforward by following the route sketched in ref . @xcite . however , this case is considerably more involved on a technical level , since one then effectively needs two ts nanowires . we here discuss the simpler spinless case , but due to the spatial separation of the dots , the results reported below represent `` useful entanglement '' @xcite that could be exploited in quantum computation schemes . finally , we come to the tunneling hamiltonian @xmath41 . note that within our low - energy model , no ts quasiparticles are available for single - electron tunneling processes , and tunneling therefore has to involve mbss . we consider sufficiently large @xmath0 such that @xmath62 is tunnel - coupled only to the mbs described by @xmath63 , see eq . and fig . similarly , the mbs coupled to @xmath64 corresponds to @xmath65 . it is now crucial to take into account charge conservation , since a floating mesoscopic superconductor can not simply absorb or emit charge @xmath66 without energy cost this is possible only in the grounded case where @xmath30 . this consideration implies that tunneling terms @xmath67 must include a factor @xmath51 , which annihilates a cooper pair and thus restores charge balance . with the gauge choice in ref . @xcite , we obtain @xmath68 + { \rm h.c . } , \ ] ] where `` normal '' tunneling terms like @xmath69 do not affect the condensate but `` anomalous '' tunneling terms such as @xmath70 change the cooper pair number by one unit . in what follows , it will be convenient to choose occupation number basis states , @xmath71 to represent the hamiltonian . for this basis choice , the quantum numbers take the values @xmath72 and @xmath73 , where @xmath74 restricts the number of cooper pairs . here @xmath74 is taken relative to @xmath75 , and the @xmath76 limit of interest is rapidly approached when @xmath11 is finite . the basis size is then @xmath77 , and @xmath37 becomes a @xmath78 matrix . diagonalization of this matrix yields the ground state @xmath79 , which in turn determines the amount of entanglement between the two dots as explained below . in order to quantify entanglement , one has to identify a suitable measure . for our bipartite case with two single - level quantum dots , a convenient and reliable measure is given by the concurrence , even though its construction is rather formal and abstract @xcite . the concurrence is expressed in terms of the reduced density matrix @xmath80 for the dots , which follows from the full density matrix after tracing over the island degrees of freedom . in the zero temperature limit , this yields the hermitian @xmath81 matrix @xmath82 with trace equal to unity . the concurrence is then defined by @xcite @xmath83 with the eigenvalues @xmath84 of the matrix @xmath85 where @xmath86 denotes the complex conjugate of @xmath87 and the @xmath88 pauli matrices act in the respective @xmath89 ( with @xmath61 ) space corresponding to each of the two dots . equation implies @xmath90 , with @xmath91 for separable ( non - entangled ) states and @xmath92 for maximally entangled states . we note that the concurrence defined in eq . is a gauge - invariant quantity , i.e. , we can use the above gauge choice with real - valued @xmath57 in eq . . we have also computed the negativity @xcite , which is an alternative measure for entanglement . while detailed results for the negativity differ from those for the concurrence , the physical conclusions are the same . from now on , we therefore discuss the concurrence only . numerically , @xmath9 follows from eq . once the full ground state of the total system , @xmath79 , has been determined by diagonalization of the @xmath78 hamiltonian matrix . analytical progress is possible when @xmath79 has a simple form . an important example is given by @xmath93 with arbitrary cooper pair numbers @xmath94 and complex coefficients @xmath95 subject to normalization , @xmath96 . calculating @xmath9 in eq . for this state , we find that at least one pair of number states in eq . must have identical @xmath97 in order to produce entanglement . there are two such possibilities , ( i ) @xmath98 with @xmath99 , or ( ii ) @xmath100 with @xmath101 . in both cases , the concurrence is given by @xmath102 see also ref . @xcite . suppressing the @xmath97 indices , maximally entangled states with @xmath92 thus correspond to the bell states @xmath103 . let us now address symmetries and conserved quantities for our model . we first note that the total fermion parity defined as @xmath104 is a conserved quantity since @xmath37 contains no terms mixing states with even and odd total electron number . the even - parity ( odd - parity ) sector has @xmath105 ( @xmath106 ) , where the parity - resolved ground states , @xmath107 , follow by diagonalization of a hamiltonian matrix of size @xmath108 only . conservation of @xmath109 will generally be weakly broken in concrete physical realizations due to quasiparticles neglected in our model . in the presence of parity relaxation , the true ground state then corresponds to the @xmath107 state with lower energy . however , we here assume that @xmath109 is conserved on all timescales of interest , such that both states @xmath107 are experimentally relevant and the concurrence depends on total fermion parity , @xmath110 . although this assumption represents an experimental challenge due to the inevitable presence of residual quasiparticle poisoning , very recent experiments indicate that the corresponding timescales may reach minutes in similar devices as considered here @xcite . we now turn to the @xmath17-dependence of the parity - constrained concurrence , @xmath111 . since a shift @xmath112 can be absorbed by shifting @xmath113 in the hamiltonian , see eq . , all observables have to be periodic in @xmath17 with period @xmath114 . ( in the presence of parity relaxation , the period will in general be reduced to @xmath115 . ) we thus obtain a first symmetry relation for the concurrence , @xmath116 this relation allows us to restrict @xmath17 to the window @xmath117 throughout . a second relation follows from an electron - hole - like symmetry property of @xmath37 which relates the odd- and even - parity concurrences , @xmath118 in order to derive eq . , we first note that @xmath37 is invariant under the replacement @xmath119 with @xmath120 , accompanied by a `` particle - hole transformation '' @xmath121 that exchanges creation and annihilation operators , @xmath122 etc . , and @xmath123 yielding @xmath124 . this symmetry implies that if we have the ground state @xmath125 , we obtain another ground state from the relation @xmath126 now the point is that these two ground states have different total parities . indeed , the original state has @xmath127 , while the transformed one has @xmath128 . we thus arrive at eq . . in passing we note that for an island hosting more than one pair of mbss , if an odd number of dots is tunnel - coupled to mbss , we instead would have @xmath129 , and eq . does not connect different parity sectors anymore . moreover , the above arguments also show that the relative sign between @xmath42 and @xmath45 in eq . is irrelevant . for our island with a single mbs pair , eq . states that the concurrence in the odd - parity sector directly follows from the even - parity result by simply inverting the energy scales @xmath130 and letting @xmath131 . from now on , unless noted otherwise , we therefore describe results for the even - parity sector @xmath105 only , where it is sufficient to consider gate - voltage parameters in the window @xmath117 . before turning to detailed results for the concurrence of the two dots in fig . [ fig1 ] , let us first discuss the entanglement - generating processes at work in such a system , where the nonlocality of the @xmath10 fermion built from the mbs pair turns out to be crucial . when searching for maximally entangled states of the two dots , eq . suggests two candidate sets for suitable superpositions of dot states @xmath132 , @xmath133 in this section , we shall focus on the large-@xmath11 case , assuming that @xmath17 is not half - integer such that the equilibrium island state is uniquely defined . hence the cooper pair number corresponds to @xmath134 , and for @xmath105 , we have @xmath135 ( @xmath136 ) for set a ( b ) . to realize the maximally entangled bell states , ( virtual ) tunneling processes through the island now have to provide the necessary coupling between states within a given set . the resulting superpositions should ideally have equal weight , which is possible when the dot energy levels are adjusted to fulfill the condition @xmath137 ( @xmath138 ) for states in set a ( b ) . perturbation theory in the tunneling amplitudes then shows the existence of two different entanglement - generating mechanisms , namely elastic cotunneling ( i.e. , tp ) and car . in fact , our quantitative results will find a natural interpretation in terms of these two mechanisms . the tp mechanism refers to the nonlocal transfer of an electron across the ts island . for example , let us consider the state @xmath139 , where an electron resides in the left dot and the right dot is empty . we now study how a coupling to the partner state @xmath140 in set a is established by @xmath41 . after the in - tunneling process from the left dot , the island charge will change by one unit . however , this change typically comes with an energy cost of order @xmath11 and , therefore , is possible only as a virtual process . the corresponding out - tunneling event restores the equilibrium charge state again and thereby may transfer an electron to the right dot . in effect , this process is similar to elastic cotunneling @xcite but with two major differences . first , tp has a highly nonlocal character inherited from the nonlocality of the @xmath10 fermion . second , because of the existence of `` normal '' and `` anomalous '' tunneling terms in @xmath41 , there are two different contributions . using the notation in eq . ( [ statelabel ] ) , the first one ( denoted by tp@xmath141 ) proceeds solely by `` normal '' tunneling ( @xmath142 ) , while the second contribution ( tp@xmath143 ) only employs `` anomalous '' tunneling processes , @xmath144 by virtue of the processes in eq . , the @xmath139 and @xmath140 states are coupled together . for @xmath137 , both states enter the superposition with equal weight . they may then form the maximally entangled bell state @xmath145 in eq . , where tp is responsible for long - range entanglement . finally , we note that the tp phenomenon discussed by fu @xcite is a resonant version of the above process , which takes place near half - integer values of @xmath17 . we will address this case in detail in sec . [ sec4a ] , where we show that perfect entanglement is not possible under coulomb blockade peak conditions . car provides a distinct second entanglement - generating mechanism , where for @xmath138 , the two dot states in set b enter the superposition with equal weight and may yield the bell state @xmath146 in eq . . since entangled states have to involve a pair of number states with matching @xmath97 entries , see eq . , the two states of interest now correspond to @xmath147 and @xmath148 . their total particle numbers are different , and hence not only the tunnel couplings @xmath149 , but also a finite josephson coupling @xmath24 to another bulk superconductor is needed in order to yield non - zero concurrence . we now show that this condition arises even though car processes can connect states with different island particle number @xmath150 already for @xmath151 . indeed , starting from @xmath147 , there are six different car transitions corresponding to all permutations of the three elementary transfer processes @xmath152 . as an example , we show two of these car sequences , @xmath153 where the first ( second ) step for the car@xmath13 ( car@xmath14 ) process describes anomalous tunneling , the respective third step corresponds to normal tunneling , and the second ( first ) step involves the josephson coupling in eq . . in the latter step , a cooper pair is transferred to the ts island . it is only because of this step that the two states @xmath147 and @xmath148 can be connected . finally , we note that in our quantitative analysis below , not only the two processes in eq . but all six car sequences will be taken into account . in this section , we quantitatively discuss the concurrence @xmath9 , describing the amount of entanglement between the two dots in fig . [ fig1 ] , by comparing analytical expressions in different parameter regimes to the corresponding numerically exact results obtained from the ground state of the full system , see eq . . since the entire behavior of the concurrence in the odd - parity sector follows from the even - parity results , see eq . , we shall only discuss the case @xmath105 . furthermore , because of eq . , the gate parameter @xmath17 can be taken in the window @xmath117 . for convenience , we will employ a compact notation for the number states by writing @xmath154 in eq . . without the josephson coupling and the tunnel amplitudes , i.e. , for @xmath155 , the eigenenergies of the full system are then given by @xmath156 with the integer island particle number @xmath157 . the dependence of @xmath158 on @xmath159 includes the well - known coulomb charging energy parabola @xcite and is sketched in fig . we shall return to fig . [ fig2 ] below when discussing our analytical results for the concurrence . moreover , the symbol @xmath160 appearing in some equations below denotes a normalization constant for the respective state . schematic illustration of the energy @xmath158 , see eq . , vs the island particle number @xmath157 for @xmath161 . panel ( a ) shows the case @xmath162 for @xmath163 . panel ( b ) is also for @xmath162 but with @xmath164 , while panel ( c ) is for @xmath165 . the respective even - parity states @xmath166 are also indicated , where tunneling @xmath167 and/or josephson @xmath168 processes cause the shown transitions . , width=226 ] schematic illustration of the energy @xmath158 , see eq . , vs the island particle number @xmath157 for @xmath161 . panel ( a ) shows the case @xmath162 for @xmath163 . panel ( b ) is also for @xmath162 but with @xmath164 , while panel ( c ) is for @xmath165 . the respective even - parity states @xmath166 are also indicated , where tunneling @xmath167 and/or josephson @xmath168 processes cause the shown transitions . , width=226 ] schematic illustration of the energy @xmath158 , see eq . , vs the island particle number @xmath157 for @xmath161 . panel ( a ) shows the case @xmath162 for @xmath163 . panel ( b ) is also for @xmath162 but with @xmath164 , while panel ( c ) is for @xmath165 . the respective even - parity states @xmath166 are also indicated , where tunneling @xmath167 and/or josephson @xmath168 processes cause the shown transitions . , width=226 ] before describing our results for the concurrence in detail , we pause to offer some guidance for focused readers . in sec . [ sec4a ] , we analyze the limit of strong coulomb blockade , where the charging energy is the dominant energy scale and only a few island charge states are accessible . we derive analytical expressions for the concurrence @xmath9 for gate - voltage parameters near the coulomb blockade valleys centered around @xmath162 , see eq . , and @xmath165 mod 2 , see eq . in addition , we also provide closed results valid near a coulomb blockade peak , see eq . , and compare our analytical results to full numerical diagonalization . we then briefly discuss the noninteracting ( @xmath30 ) limit in sec . [ sec4b ] , where the concurrence follows as given in eq . . finally , we return to the strong-@xmath11 limit in sec . [ sec4c ] but now also allow for a strong mbs hybridization @xmath21 , i.e. , arbitrary length @xmath0 in eq . . however , for simplicity , we only study the case of integer @xmath17 in sec . [ sec4c ] . we start our quantitative analysis of the concurrence with the limit of large @xmath11 , @xmath169 where the charging energy dominates over all other energy scales except for the ts gap , which implicitly sets the ultraviolet cutoff for our model . the parameter regime in eq . has also been addressed in refs . it is convenient to distinguish four different , partially overlapping , gate - voltage parameter windows defined by @xmath170 where @xmath171 parametrizes the deviation from the respective center value @xmath172 . for both half - integer values of @xmath172 , the dominant charging energy contribution , given by the first term in eq . , is precisely degenerate and thus corresponds to a coulomb blockade conductance peak @xcite . we present our results for this case later on , but first turn to the coulomb blockade valley at @xmath173 . for @xmath173 in eq . , we observe from fig . [ fig2](a ) that the ground state has island occupation number @xmath174 in order to minimize the large coulomb energy , such that @xmath135 and @xmath134 . for @xmath105 , one additional electron then has to occupy the dots . as a consequence , the low - energy sector is spanned by @xmath175 with the two states denoted by @xmath176 . the energetically closest set of states , @xmath177 , has @xmath178 or @xmath179 , see fig . [ fig2](a ) , where the @xmath10-electron level representing the mbs pair is empty , @xmath136 , with @xmath134 or @xmath180 . the two dots are then either both occupied or both empty , resulting in the subspace @xmath181 these states are separated from the @xmath182 sector by an energy of order @xmath11 . this separation of energy scales allows us to derive a low - energy hamiltonian , @xmath183 , describing the parameter regime in eq . for gate - voltage parameters close to @xmath173 . this reduced hamiltonian is obtained by projecting the full hamiltonian to the subspace @xmath182 only . using the two basis states @xmath176 in eq . , @xmath183 corresponds to a @xmath184 matrix , such that it is straightforward to determine the ground state and the concurrence analytically . the above projection is implemented by disregarding all high - energy states beyond @xmath177 in eq . and by treating the coupling between the subspaces @xmath182 and @xmath185 through a schrieffer - wolff transformation @xcite , which here is equivalent to second - order perturbation theory . for @xmath163 , this coupling is due to tunneling ( @xmath41 ) only , see fig . [ fig2](a ) , and with the energies @xmath158 in eq . , we obtain @xmath186 where the second term describes virtual excursions to the energetically higher @xmath187 states . using eq . , @xmath130-terms in the respective denominators are small against the large charging energy contribution , and up to an overall energy shift , we arrive at @xmath188 where @xmath189 parametrizes @xmath17 around @xmath173 , see eq . . the ground state of eq . is @xmath190,\ ] ] which matches the general form in eq . ( [ redmats ] ) . the concurrence then directly follows from eq . , @xmath191 stating that @xmath9 is a universal function of a single parameter @xmath192 which depends on basically all the microscopic parameters according to eq . . for @xmath193 , eq . yields the ideal value , @xmath92 , characterizing perfect entanglement . indeed , eq . then reduces to the bell state @xmath145 in eq . , and we conclude that entanglement is here established by the tp mechanism . this condition for perfect entanglement is fulfilled for all symmetric systems , i.e. , for identical dot level energies and tunneling strengths , @xmath194 the approximations leading to the reduced hamiltonian ( [ reducedham ] ) , and hence to the concurrence in eq . , eventually break down near a coulomb peak with @xmath195 . nonetheless , eq . shows that @xmath92 can persist over an extended gate - voltage parameter window around @xmath173 . in a similar manner , we next study what happens around @xmath196 . in fig . [ fig2](c ) , we show a subset of the relevant states , implementing two of the possible six car processes . the energy @xmath158 is then minimized for @xmath197 , and the @xmath105 low - energy sector is spanned by @xmath198 where both dot levels are either occupied or empty . the excited states relevant for car processes , cf . [ sec3b ] , then correspond to @xmath199 where the first and the last state have island charge @xmath200 and thus are highest in energy , @xmath201 , while the remaining states ( with @xmath202 ) are separated from the @xmath182 sector by an energy scale @xmath203 . the two states in eq . have different total particle number . as discussed in sec . [ sec3b ] , see also fig . [ fig2](c ) , the car mechanism connecting those states has to involve a finite josephson coupling @xmath24 and therefore constitutes a third - order process . in particular , now a transition within the high - energy sector in eq . is necessary , and we thus have to include one normal tunneling , one anomalous tunneling , and one josephson process , i.e. , we have to apply @xmath41 twice and @xmath39 once , see eqs . and . the corresponding third - order perturbation theory can again be implemented by a schrieffer - wolff transformation @xcite , which effectively yields @xmath204 for the reduced hamiltonian in eq . . with @xmath205 , we find @xcite @xmath206 are the virtually occupied higher - energy states in eq . and terms @xmath207 are small against the charging energy contributions in the denominator . using the two states in eq . to represent @xmath183 , we obtain the same @xmath184 matrix as in eq . but with the replacements @xmath208 where @xmath189 now parametrizes @xmath17 around @xmath196 . we observe that second - order processes only give corrections to the state energies ( @xmath209 ) , while third - order car processes produce the crucial coupling of both states ( @xmath210 ) needed for entanglement . we note in passing that second - order corrections @xmath211 only give an irrelevant constant energy shift that has been dropped in eq . . the ground state finally follows as in eq . , but with the basis states in eq . and the replacements in eq . . hence the concurrence is again given by eq . but with @xmath212 . although these results look very similar , there are crucial differences in the entanglement properties when compared to the @xmath173 case . in particular , although we may have @xmath213 for a specific gate - voltage parameter ( @xmath189 ) , such that perfect entanglement is realized , it is not possible to have @xmath214 over a wide parameter range anymore . in that sense , within the large-@xmath11 regime defined in eq . , entanglement produced by the car mechanism is less stable than the one caused by tp . to demonstrate this point , consider dot energies @xmath138 where car processes are most effective , cf . [ sec3b ] , keeping symmetric tunnel couplings as before , @xmath215 we then obtain @xmath216 describing a narrow concurrence peak centered around @xmath165 , with height @xmath217 and width @xmath218 . this concurrence peak is a clear signature of car processes and can easily be distinguished from the broad and robust @xmath92 plateau generated by tp around @xmath162 . in particular , the car entanglement peak quickly disappears with increasing @xmath219 , while it remains stable against variations with @xmath220 . we thus obtain precisely the opposite behavior as for the tp - generated concurrence plateau . next we study the concurrence behavior near half - integer values of @xmath17 . for simplicity , we discuss the most interesting symmetric case defined in eq . . with @xmath17 near @xmath221 , there are four low - energy states corresponding to the subspace latexmath:[\[\label{lowen4 } { \cal h}_0^{(n_{g,0}=1/2)}=\ { these states are already directly coupled by tunneling , the reduced hamiltonian @xmath183 can simply be obtained by projecting @xmath37 to the subspace in eq . , thereby neglecting all virtual excursions to higher energy states . when represented in the basis , up to an irrelevant overall constant , @xmath183 then takes the form @xmath223 with the @xmath17-dependent energy scale @xmath224 the ground state of @xmath183 is either given by @xmath225,\end{aligned}\ ] ] or it corresponds to the separable @xmath91 state @xmath226 . the respective energies are @xmath227 since @xmath228 is of the form in eq . , we can again use eq . to directly read off the concurrence , but @xmath229 is necessary to have a finite concurrence . with the heaviside step function @xmath230 , we thus obtain the concurrence in analytical form , @xmath231 ^ 2 } , \ ] ] which now depends on two parameters . in order to obtain a non - zero value in eq . , the gate parameter must be above a critical value , @xmath232 , with @xmath233 this relation is only relevant when @xmath234 , since the condition @xmath229 is always met for @xmath235 . for @xmath234 , the concurrence thus exhibits an abrupt jump from @xmath91 to a finite value as @xmath17 increases through @xmath236 . on the other hand , it can happen that at some gate parameter @xmath237 ( above @xmath236 for @xmath234 ) , the turnover value @xmath238 separating the car - dominated regime ( near the @xmath165 valley ) from the tp - dominated regime ( near @xmath162 ) is reached . using eq . , this point is at @xmath239 , corresponding to @xmath240 note that this expression holds in particular for @xmath241 . we therefore see that the gate parameter @xmath242 determining the transition through @xmath238 is given by @xmath243 where the last expression becomes exact for @xmath244 . for @xmath245 , the concurrence transition is an abrupt jump . otherwise , it is a smooth transition , and the concurrence jump only happens at a gate parameter @xmath246 where @xmath9 has already dropped to a small value . the actual value for @xmath242 in eq . is close to @xmath247 for large @xmath11 , but it exhibits a systematic shift towards @xmath248 when @xmath249 increases . we also note that as @xmath17 moves from @xmath242 towards @xmath162 , the concurrence given by eq . approaches the perfect entanglement regime , @xmath250 , as predicted by our expansion around @xmath251 . consequently , our results for @xmath252 obtained within different @xmath17 windows smoothly match onto each other . finally , a similar analysis can be carried out for @xmath17 near @xmath253 . the result is given by eq . again but with the replacements @xmath254 and @xmath255 . for @xmath256 , the transition corresponding to eq . now takes place at the gate parameter @xmath257 which again shifts towards @xmath248 with increasing @xmath249 . concurrence @xmath9 vs @xmath17 for the even - parity sector with @xmath258 . we consider a symmetric system , see eq . , with @xmath259 and @xmath260 , for two different values of @xmath11 . solid curves are obtained by numerical diagonalization of the full model . dashed curves represent the analytical result derived in the large-@xmath11 limit , see eqs . and . with @xmath261 , see eq . , we only show one period.,width=226 ] with analytical results for the concurrence at our disposal , we now compare them to numerically exact results . the @xmath17-dependence of the concurrence for @xmath258 is shown in fig . [ fig3 ] , where we consider a symmetric system with @xmath259 . for @xmath262 , the analytical curve almost perfectly matches the numerically exact result , where we have @xmath92 for practically the full valley range @xmath263 , with a sharp crossover around @xmath247 to @xmath91 near even @xmath17 . for @xmath264 , the qualitative behavior is still captured by our analytical expressions but there are quantitative differences , as expected when moving away from the strong-@xmath11 limit . we note in passing that for odd parity , the concurrence behavior in both valleys will effectively be interchanged . same as fig . [ fig3 ] but for finite @xmath24 and @xmath265 . the dashed curves again give analytical results , using eq . near @xmath165 and @xmath266 . , width=226 ] let us next address the effects of adding a small but finite josephson coupling @xmath24 , where we again study a symmetric system with @xmath267 . the concurrence is shown in fig . [ fig4 ] for @xmath262 . we first notice that near @xmath162 , the effects of @xmath24 are negligible and the tp - induced perfect value @xmath92 is robust against adding a small josephson coupling . however , we now encounter the predicted narrow concurrence peak with width @xmath268 centered at @xmath165 mod 2 , see eq . , which provides a very characteristic signature of car processes . same as fig . [ fig4 ] but for finite @xmath269 , using @xmath270.,width=226 ] in figs . [ fig3 ] and [ fig4 ] , we have considered a symmetric system with @xmath267 , where the concurrence exhibits the additional symmetry @xmath271 , and the turnover between both concurrence regimes is perfectly smooth and happens precisely at half - integer @xmath17 . however , both of these features will not be present in general , and a counter - example with finite @xmath269 is shown in fig . [ fig5 ] . in accordance with eqs . and , we find that the turnover points between the car- and tp - dominated regimes move closer to @xmath248 with increasing @xmath269 . as expected from our above discussion , we also confirm that for @xmath234 , the crossover in @xmath252 ( at @xmath17 slightly above @xmath247 ) deforms into an abrupt jump , while the corresponding crossover near @xmath272 stays smooth . hence fig . [ fig5 ] clearly shows that @xmath273 . we mention in passing that for finite but very small @xmath269 , the narrow entanglement peak seen in fig . [ fig4 ] may survive , but with the center position shifted to finite @xmath17 . however , this peak is found to quickly disappear with increasing @xmath249 , cf . [ sec3b ] , and is altogether absent for the parameters in fig . [ fig5 ] . let us now briefly address the noninteracting case , @xmath30 , where the cooper pair sector decouples from the fermionic sector spanned by the @xmath274 states . the @xmath30 case with normal metallic leads instead of our dots was studied in ref . @xcite , while the dot case was also considered in ref . an alternative way to reach the noninteracting limit is to let @xmath275 , where all @xmath276 states are strongly mixed and charge quantization effects are washed out @xcite . in both limits , @xmath277 or @xmath275 , one arrives at the same effective hamiltonian @xmath183 , where the parameters @xmath11 , @xmath17 and @xmath24 do not appear anymore . in the even - parity sector , @xmath183 has a @xmath81 matrix representation , which can easily be diagonalized to yield the ground state @xmath79 . for the symmetric case with @xmath259 , see eq . , and omitting the redundant @xmath42 indices , we find @xmath278 \\ \nonumber & - & \sqrt{8}\lambda \left [ |101\rangle-|011\rangle \right]\bigr\},\end{aligned}\ ] ] which is of the form in eq . . we can therefore infer the concurrence from eq . , @xmath279 for @xmath280 , using eq . , we thus obtain the conventional exponential decay of the concurrence with increasing distance between the mbss , @xmath281 . for @xmath31 , the car contribution is precisely cancelled by the tp contribution corresponding to the second pair of states in eq . , see also eq . , and the concurrence vanishes identically . on the other hand , for @xmath282 , we find @xmath92 due to car processes , since entanglement is now dominated by the first pair of states in eq . . the noninteracting system thus allows for car - mediated ideal entanglement as well , but @xmath92 is reached only for rather small @xmath0 where @xmath283 , see eq . . for the realization of long - range entanglement , it is therefore essential to work with a floating island where @xmath32 . we conclude that the noninteracting case can also be understood quantitatively in terms of a competition between tp and car processes . for even parity , we have seen that car processes dominate when a sizeable mbs hybridization @xmath21 is present , which effectively corresponds to a rather short distance @xmath0 between the mbss . we note that for odd parity , the role of tp and car processes is effectively interchanged , again yielding @xmath92 for @xmath284 . since we consider single - level dots and not metallic leads , otherwise competing processes such as normal reflection or local andreev reflection @xcite are absent . as a consequence , the entanglement found in eq . is generally more robust . for small @xmath0 , the mbs hybridization @xmath21 may exceed the charging energy @xmath11 significantly . we then consider the parameter regime but with arbitrary ratio @xmath285 , i.e. , both @xmath21 and @xmath11 may be large compared to all other energy scales in the problem . let us start with the case @xmath162 , assuming a symmetric setup , see eq . . as shown before , for @xmath286 , we then have @xmath92 due to the tp mechanism . now for finite ( but large ) @xmath287 , the energy gap between the ground and excited state sectors in eqs . and is given by @xmath288 , see fig . [ fig2](a ) . however , as long as the condition in eq . continues to hold with the replacement @xmath289 , the situation remains conceptually as before , and for @xmath290 ( i.e. , @xmath162 ) and @xmath289 in eqs . , we again find @xmath214 . on the other hand , for @xmath291 , our @xmath30 discussion suggests that the car mechanism generates full entanglement of the dots with @xmath92 as well . we therefore now consider a finite ( large ) charging energy with @xmath292 , where the previous ground and excited state sectors are effectively interchanged , @xmath293 in eqs . and . this situation is illustrated in fig . [ fig2](b ) , where we can read off the energy gap separating the two sectors , @xmath294 . with the replacement @xmath295 , we then assume that the condition in eq . holds again . in particular , the josephson coupling @xmath24 now directly connects the respective states in the low - energy sector such that the car processes needed for generating entanglement , cf . [ sec3b ] , are already captured by second - order perturbation theory , see eq . . considering the case @xmath296 and @xmath297 , and diagonalizing the remaining @xmath298 hamiltonian in the basis , we obtain the ground state @xmath299 \\ \nonumber & + & \left(\gamma+\sqrt{\gamma^2+(e_j/2)^2}\right ) \left[|110,0\rangle+|000,1\rangle \right]\bigr\},\end{aligned}\ ] ] where the rate @xmath300 connects @xmath148 and @xmath301 through the virtual occupation of excited states . although the concurrence formula is not applicable for @xmath79 in eq . , @xmath9 can be computed analytically , with the result @xmath302 in accordance with sec . [ sec3b ] , the concurrence vanishes for @xmath303 , in particular for @xmath304 . color - scale plot of the concurrence for @xmath162 in the @xmath21-@xmath11 plane , taking @xmath305 and a symmetric system with @xmath259 , cf . .,width=226 ] these analytical results are nicely recovered from a numerically exact calculation of the concurrence for @xmath162 . we show a color - scale plot of @xmath9 as a function of @xmath21 and @xmath11 in fig . [ fig6 ] , confirming that lines of constant concurrence are present when @xmath306 is held constant . we took @xmath307 in fig . [ fig6 ] , such that both tp and car processes can generate stable entanglement , see sec . [ sec3 ] . in the crossover regime @xmath308 , we find a @xmath91 line right at @xmath309 , see fig . this limit is correctly captured from the car point of view ( even though perturbation theory breaks down ) , since then the rate @xmath310 in eq . becomes large and hence @xmath311 in eq . . approaching this limit from the tp - dominated side is more subtle . the energy diagrams in fig . [ fig2](a , b ) reveal that there are six low - energy states at play for @xmath308 . these states are directly coupled via @xmath149 and @xmath24 , which can be viewed as a mixing of tp and car processes . in that case , neither tp nor car are effective in generating entanglement anymore since the previously excited states may now be occupied as well . same as fig . [ fig6 ] but for @xmath165.,width=226 ] finally , we turn to the case @xmath165 , where we consider optimal conditions for the car mechanism by choosing dot level energies with @xmath138 and symmetric tunnel couplings , @xmath312 . the corresponding energy diagram in fig . [ fig2](c ) shows two out of the six possible car processes , see also sec . [ sec3b ] . with the mbs hybridization @xmath21 , the first and the last state in eq . are still separated by the energy gap @xmath313 from the ground state sector , while all other excited states in eq . have the gap @xmath314 . depending on the ratio @xmath285 , different sequences out of the six possible car processes then become important . nonetheless , as long as eq . holds for @xmath315 , we may effectively apply eq . again . we thereby find @xmath183 as in eq . but with @xmath316 where we set @xmath317 ( i.e. , @xmath318 ) . the first term comes from the two car processes with excursions to @xmath202 states , see fig . [ fig2](c ) , while the second term describes the remaining four car processes which also include a virtual occupation of @xmath319 states . regardless of which term dominates , throughout the regime @xmath320 , we observe that car processes dominate the concurrence both for @xmath321 and for @xmath291 . this fact can be rationalized by noting that the inversion of ground and excited state sectors ( discussed for @xmath162 above ) does not take place anymore , since for arbitrary ratio @xmath285 , the states @xmath147 and @xmath148 are lowest in energy . we therefore obtain @xmath92 for almost the entire @xmath21-@xmath11 plane . the color - scale plot of the numerically exact solution in fig . [ fig7 ] confirms this result , with the region @xmath322 representing the only exception . in the latter region , one approaches the perfectly disentangled limit @xmath323 , where car and tp processes interfere destructively . finally , we remark that lines of constant concurrence are visible in fig . [ fig7 ] when @xmath324 in eq . is held fixed . although the above results follow with the replacements @xmath325 in eq . , we stress again that @xmath11 itself has to remain large , @xmath326 . otherwise the description via charging energy parabolas as in fig . [ fig2 ] is no longer sufficient and we expect deviations from the analytical results . such deviations can be seen for @xmath327 in figs . [ fig6 ] and [ fig7 ] , where one effectively approaches the noninteracting limit @xmath277 . up to this point , we have studied the stationary case , with time - independent parameters in @xmath37 . in this section , we address the entanglement dynamics , @xmath33 , observed after a quench of a tunnel coupling where , say , @xmath23 is suddenly switched on from zero to a finite value at time @xmath328 . we assume that at times @xmath329 , we then have a symmetric system with @xmath259 and @xmath260 in eq . , and we again consider the even - parity sector , @xmath105 . studies of the entanglement dynamics can be very useful in revealing important timescales of the problem @xcite . we will see below that similar to the static case in sec . [ sec4 ] , perturbation theory is helpful in elucidating the underlying physical mechanisms and the relevant timescales governing @xmath33 . in particular , our results for @xmath33 can again be interpreted in terms of tp and car processes . time - dependence of the concurrence , @xmath33 , obtained after a quench of the tunnel coupling , @xmath330 at @xmath328 , for three different values of @xmath331 . we consider the case of finite @xmath11 with @xmath162 , taking @xmath332 and @xmath333 . exact results ( black solid curves ) follow from eqs . and . the envelope functions ( red dashed curves ) in eq . accurately capture the slow part of the dynamics.,width=188 ] time - dependence of the concurrence , @xmath33 , obtained after a quench of the tunnel coupling , @xmath330 at @xmath328 , for three different values of @xmath331 . we consider the case of finite @xmath11 with @xmath162 , taking @xmath332 and @xmath333 . exact results ( black solid curves ) follow from eqs . and . the envelope functions ( red dashed curves ) in eq . accurately capture the slow part of the dynamics.,width=188 ] time - dependence of the concurrence , @xmath33 , obtained after a quench of the tunnel coupling , @xmath330 at @xmath328 , for three different values of @xmath331 . we consider the case of finite @xmath11 with @xmath162 , taking @xmath332 and @xmath333 . exact results ( black solid curves ) follow from eqs . and . the envelope functions ( red dashed curves ) in eq . accurately capture the slow part of the dynamics.,width=188 ] let us begin with the case of finite @xmath11 and arbitrary @xmath21 , keeping @xmath334 , see sec . [ sec4c ] . for simplicity , we put @xmath335 and @xmath162 , where we expect that entanglement is mediated by tp processes . we note in passing that in the presence of weak parity - breaking processes , e.g. , due to quasiparticle poisoning , the even total parity ( @xmath105 ) sector is the true ground state at @xmath162 . suddenly switching on the tunnel coupling @xmath336 at time @xmath328 results in a time - dependence of the concurrence , where our numerical results for @xmath33 are shown in fig . here we start out from the initial state @xmath337 , where the decoupled dot ( qd@xmath14 in fig . [ fig1 ] ) is empty before the quench . the post - quench state @xmath338 is of the form @xmath339 where the time - dependent coefficients @xmath340 are subject to normalization , @xmath341 , and obey the initial condition @xmath342 . the solution is given by @xmath343 , \\ c_3(t ) & = & e^{-i\epsilon_f t/2}-c_2(t ) , \nonumber\end{aligned}\ ] ] where @xmath344 . the concurrence @xmath33 then follows by inserting the coefficients in eq . into eq . . since for @xmath335 , total particle number is conserved , only the given four states in eq . with same total particle number as in the initial state are accessible . for that reason , eqs . and represent the exact solution for arbitrary @xmath345 as long as @xmath335 . indeed , the @xmath33 results in fig . [ fig8 ] are identical to those obtained from numerically solving the full schrdinger equation . to gain physical intuition , it is instructive to compare these exact ( but not very illuminating ) results for @xmath33 to the corresponding predictions obtained from the reduced @xmath184 hamiltonian @xmath183 in eq . , with the replacement @xmath346 as in sec [ sec4c ] . using the parameters @xmath347 and @xmath348 in @xmath183 , only a single oscillation frequency , @xmath349 , appears within that approach , where for our symmetric system , we find @xmath350 . the low - energy hamiltonian @xmath183 captures only the slow part of the dynamics , i.e. , the envelope function @xmath351 . moreover , its perturbative derivation requires the condition @xmath352 . on the other hand , for @xmath353 , we can directly separate the slow and fast variations in the full analytical expression for @xmath33 described above , which gives @xmath351 also for @xmath354 . combining both cases , we obtain a sinusoidal envelope function , @xmath355 with the frequency @xmath356 notice that for @xmath357 , the slow part of the dynamics becomes universal ( independent of the tunnel coupling @xmath358 ) . for @xmath359 , @xmath33 is governed by two almost resonant frequencies and the envelope function in eq . becomes less meaningful . figure [ fig8 ] illustrates that away from this ( quite narrow ) transition regime , eqs . and accurately describe the slow envelopes of all @xmath33 curves . for @xmath360 , the frequency @xmath361 in eq . coincides with the perturbative scale @xmath362 extracted from @xmath183 . in this limit , eq . becomes exact , @xmath363 , and the fast oscillations in @xmath33 disappear altogether , cf . [ fig8 ] . by comparing the characteristic frequencies appearing in the exact solution for @xmath33 to perturbative estimates valid for @xmath364 , we thus have complemented our picture of the system in fig . [ fig2 ] . such an analysis of the entanglement dynamics becomes more difficult for other parameters within the large-@xmath11 regime , we here briefly consider the case @xmath318 with @xmath365 , where car processes can establish entanglement . the josephson coupling now makes the inclusion of higher - energy states with different total particle numbers necessary , and exact expressions for the envelope curves as in eqs . and are more challenging to obtain . nonetheless , analytical estimates are still possible in the perturbative regime , where we obtain @xmath351 as in eq . but with @xmath366 , using @xmath324 in eq . . the car - induced entanglement dynamics at @xmath165 is thus characterized by a sinusoidal envelope function again but has the rather long timescale @xmath367 . another case of interest that allows for an exact solution is given by @xmath30 . since the cooper pair sector decouples , see sec . [ sec4b ] , we then have a set of four states describing the full dynamics again . the concurrence dynamics after a tunnel coupling quench was recently addressed for this limit of our setup in ref . @xcite , where oscillations in @xmath33 were reported as well . with @xmath368 , taking the pre - quench ground state @xmath369 , and expanding @xmath370 in terms of the four basis states available for even parity and @xmath30 , the post - quench ( @xmath329 ) state is as in eq . but with the initial condition @xmath371 . by solving the time - dependent schrdinger equation as detailed in ref . @xcite , @xmath33 follows again from eq . . in fact , even though car processes are now responsible for generating entanglement , finding @xmath33 for @xmath30 is fully equivalent to the tp - mediated case in sec . [ sec5a ] after replacing @xmath372 . we thus obtain the same concurrence dynamics as in eqs . and after this replacement @xcite , including the results shown in fig . [ fig8 ] , where now @xmath373 separates the perturbative regime ( @xmath284 ) from the regime where calculation of @xmath33 needs to account for all four states ( @xmath374 ) , see eqs . and . in this paper , we have provided a detailed study of entanglement in a floating topological superconducting island hosting a pair of mbss , each of which is tunnel - coupled to a single - level dot . the concurrence then provides a convenient measure to quantify entanglement of the two dots . we have shown that this setup offers a robust route towards the implementation of a `` majorana entanglement bridge '' where the concurrence does not exhibit any decay with increasing separation of the mbss . the only restriction for the separation of the quantum dots then is given by requiring a finite charging energy on the island , yielding much longer ranges of entanglement than in usual notions of exponential decay . this unique behavior is due to the intrinsic nonlocality of the fermion state representing the pair of mbss . the underlying mechanisms for entanglement generation have been identified in an intuitive yet quantitative manner . in particular , we have shown that entanglement is created by the interplay of teleportation and crossed andreev reflection processes . with this understanding , we were then able to discuss the more complicated entanglement dynamics after the quench of a tunnel coupling in terms of simple envelope functions . we have studied entanglement for the majorana device in fig . [ fig1 ] by employing the concurrence as entanglement measure . although the concurrence has a rather formal definition , the predicted entanglement features should be observable in mesoscopic transport experiments , where the two dots are weakly coupled to additional leads as described , e.g. , in ref . @xcite . since current - current correlation functions can detect the violation of bell inequalities @xcite , signatures of entanglement are expected to appear in shot noise measurements , see also refs . @xcite for related proposals . in principle also suitably designed conductance measurements could probe entanglement @xcite . recent experimental work @xcite indicates that devices corresponding to the model studied here are in close reach . for instance , charging energies of order @xmath375 mev and ts gaps @xmath376 mev have been reported in ref . @xcite , where quantum dots can be formed spontaneously near the ts wire ends when contacting them by au leads @xcite . in order to reach the regime @xmath377 assumed in our model , the charging energy could be lowered by simply increasing the wire length . on the other hand , it may be possible to achieve significant entanglement between the dots even when @xmath378 , where fermionic ts quasiparticles should be taken into account . when such states are localized in the bulk of the ts such that they do not have significant overlap with the majorana bound states , they have no effect on entanglement . when low - lying quasiparticle states are localized near the ts nanowire ends , the physics discussed here may change , but entanglement may nonetheless be possible according to our preliminary analysis , see also ref . @xcite . using nbtin @xcite or inas / al @xcite to proximity - induce superconductivity in a device similar to the one in fig . [ fig1 ] , extremely small poisoning rates have been reported as well . in any case , a detailed study of quasiparticle effects on entanglement is beyond the scope of the present work . to conclude , we are confident that future theoretical and experimental studies of similar setups utilizing such a majorana entanglement bridge will benefit from the detailed physical understanding supplied here . we thank s. albrecht , s. bose , d. bruss , e. eriksson , k. flensberg , h. kampermann , t. martin , g. semenoff , and a. levy yeyati for useful discussions . r.e . acknowledges support by the network spp 1666 of the deutsche forschungsgemeinschaft ( germany ) and by the program `` science without borders '' ( swb ) of cnpq ( brazil ) . thanks the ministry of science , technology , and innovation of brazil and cnpq for granting a `` bolsa de produtividade em pesquisa '' , and acknowledges support by the cnpq swb program and from mcti and ufrn / mec ( brazil ) . 99 j. alicea , _ new directions in the pursuit of majorana fermions in solid state systems _ , rep . . phys . * 75 * , 076501 ( 2012 ) . m. leijnse and k. flensberg , _ introduction to topological superconductivity and majorana fermions _ , semicond . * 27 * , 124003 ( 2012 ) . beenakker , _ search for majorana fermions in superconductors _ , annu . rev . condens . matter phys . * 4 * , 113 ( 2013 ) . v. mourik , k. zuo , s.m . frolov , s.r . plissard , e.p.a.m . bakkers , and l.p . kouwenhoven , _ signatures of majorana fermions in hybrid superconductor - semiconductor nanowire devices _ , science * 336 * , 1003 ( 2012 ) . l. rokhinson , x. liu , and j. furdyna , _ the fractional a.c . josephson effect in a semiconductor - superconductor nanowire as a signature of majorana particles _ , nature physics * 8 * , 795 ( 2012 ) . a. das , y. ronen , y. most , y. oreg , m. heiblum , and h. shtrikman , _ zero - bias peaks and splitting in an al - inas nanowire topological superconductor as a signature of majorana fermions _ , nat . * 8 * , 887 ( 2012 ) deng , c.l . huang , m. larsson , p. caroff , and h.q . xu , _ anomalous zero - bias conductance peak in a nb - insb nanowire - nb hybrid device _ , nano lett . * 12 * , 6414 ( 2012 ) . churchill , v. fatemi , k. grove - rasmussen , m.t . deng , p. caroff , h.q . xu , and c.m . marcus , _ superconductor - nanowire devices from tunneling to the multichannel regime : zero - bias oscillations and magnetoconductance crossover _ , phys . b * 87 * , 241401(r ) ( 2013 ) . lee , x.c . jiang , m. houzet , r. aguado , c.m . lieber , and s. de franceschi , _ spin - resolved andreev levels and parity crossings in hybrid superconductor - semiconductor nanostructures _ , nature nanotech . * 267 * , 79 ( 2014 ) . s. nadj - perge , i.k . drozdov , j. li , h. chen , s. jeon , j. seo , a.h . macdonald , b. andrei bernevig , and a. yazdani , _ observation of majorana fermions in ferromagnetic atomic chains on a superconductor _ , science * 346 * , 602 ( 2014 ) . f. hassler , a.r . akhmerov , c.y . hou , and c.w.j . beenakker , _ anyonic interferometry without anyons : how a flux qubit can read out a topological qubit _ , new j. phys . * 12 * , 125002 ( 2010 ) . sau , s. tewari , and s. das sarma , _ universal quantum computation in a semiconductor quantum wire network _ , a * 82 * , 052322 ( 2010 ) . k. flensberg , _ non - abelian operations on majorana fermions via single - charge control _ lett . * 106 * , 090503 ( 2011 ) . l. jiang , c.l . kane , and j. preskill , _ interface between topological and superconducting qubits _ lett . * 106 * , 130504 ( 2011 ) . p. bonderson and r.m . lutchyn , _ topological quantum buses : coherent quantum information transfer between topological and conventional qubits _ , phys . * 106 * , 130505 ( 2011 ) . m. leijnse and k. flensberg , _ quantum information transfer between topological and spin qubit systems _ , phys . lett . * 107 * , 210502 ( 2011 ) . m. leijnse and k. flensberg , _ hybrid topological - spin qubit systems for two - qubit - spin gates _ , phys . b * 86 * , 104511 ( 2012 ) . t. hyart , b. van heck , i.c . fulga , m. burrello , a.r . akhmerov , and c.w.j . beenakker , _ flux - controlled quantum computation with majorana fermions _ , phys . b * 88 * , 035121 ( 2013 ) . yu.v . nazarov and ya.m . blanter , _ quantum transport : introduction to nanoscience _ ( cambridge university press , cambridge , uk , 2009 ) . r. horodecki , p. horodecki , m. horodecki , and k. horodecki , _ quantum entanglement _ , phys . * 81 * , 865 ( 2009 ) . g. semenoff and p. sodano , _ stretching the electron as far as it will go _ , electron . . phys . * 10 * , 157 ( 2006 ) . bolech and e. demler , _ observing majorana bound states in @xmath379-wave superconductors using noise measurements in tunneling experiments _ , phys . 98 * , 237002 ( 2007 ) . j. nilsson , a.r . akhmerov , and c.w.j . beenakker , _ splitting of a cooper pair by a pair of majorana bound states _ , * 101 * , 120403 ( 2008 ) . s. tewari , c. zhang , s. das sarma , c. nayak , and d.h . testable signatures of quantum nonlocality in a two - dimensional chiral @xmath379-wave superconductor _ , phys . lett . * 100 * , 027001 ( 2008 ) . j. li , t. yu , h.q . lin , and j.q . you , _ probing the non - locality of majorana fermions via quantum correlations _ , sci . * 4 * , 4930 ( 2014 ) . a. zazunov , p. sodano , and r. egger , _ even - odd parity effects in majorana junctions _ , new j. phys . * 15 * , 035033 ( 2013 ) . s. bose , _ quantum communication through an unmodulated spin chain _ , phys . lett . * 91 * , 207901 ( 2003 ) . a. bayat , s. bose , and p. sodano , _ entanglement routers using macroscopic singlets _ , lett . * 105 * , 187204 ( 2010 ) . m. leijnse and k. flensberg , _ coupling spin qubits via superconductors _ , * 111 * , 060501 ( 2013 ) . l. fu , _ electron teleportation via majorana bound states in a mesoscopic superconductor _ , phys . lett . * 104 * , 056402 ( 2010 ) . a. zazunov , a.l . yeyati , and r. egger , _ coulomb blockade of majorana - fermion - induced transport _ , phys . b * 84 * , 165440 ( 2011 ) . r. htzen , a. zazunov , b. braunecker , a.l . yeyati , and r. egger , _ majorana single - charge transistor _ , phys . lett . * 109 * , 166403 ( 2012 ) . z. wang , x.y . liang , and x. hu , _ detecting majorana fermions by nonlocal entanglement between quantum dots _ , b * 87 * , 214513 ( 2013 ) . s. walter and j.c . budich , _ teleportation - induced entanglement of two nanomechanical oscillators coupled to a topological superconductor _ , phys . b * 89 * , 155431 ( 2014 ) . in particular , the authors of ref . @xcite studied the large-@xmath11 limit with @xmath31 in a coulomb valley ( integer @xmath17 ) by fixing the number of cooper pairs , and reported long - range entanglement in this limit . by means of a master equation approach , ref . @xcite arrived at similar conclusions when the dots are replaced by nanomechanical oscillators . b. bri and n.r . cooper , _ topological kondo effect with majorana fermions _ , lett . * 109 * , 156803 ( 2012 ) . a. altland and r. egger , _ multiterminal coulomb - majorana junction _ , lett . * 110 * , 196401 ( 2013 ) . b. bri , _ majorana - klein hybridization in topological superconductor junctions _ , * 110 * , 216803 ( 2013 ) . a. altland , b. bri , r. egger , and a.m. tsvelik , _ multichannel kondo impurity dynamics in a majorana device _ , lett . * 113 * , 076401 ( 2014 ) . a. zazunov , a. altland , and r. egger , _ transport properties of the coulomb - majorana junction _ , new j. phys . * 16 * , 015010 ( 2014 ) drummond , a.a . kovalev , c.y . hou , k. shtengel , and l.p . pryadko , _ demonstrating entanglement by testing bell s theorem in majorana wires _ , phys . b * 90 * , 115404 ( 2014 ) . lesovik , t. martin , and g. blatter , _ electronic entanglement in the vicinity of a superconductor _ , eur . j. b * 24 * , 287 ( 2001 ) . p. recher , e.v . sukhorukov , and d. loss , _ andreev tunneling , coulomb blockade , and resonant transport of nonlocal spin - entangled electrons , _ phys . rev . b * 63 * , 165314 ( 2001 ) . p. sodano , a. bayat , and s. bose , _ kondo cloud mediated long - range entanglement after local quench in a spin chain _ b * 81 * , 100412(r ) ( 2010 ) . van woerkom , a. geresdi , and l.p . kouwenhoven , _ 1 minute parity lifetime of a nbtin cooper - pair transistor , _ preprint arxiv:1501.03855 . a.p . higginbotham , s.m . albrecht , g. kirsanskas , w. chang , f. kuemmeth , p. krogstrup , t.s . jespersen , j. nygrd , k. flensberg , and c.m . marcus , _ parity lifetime of bound states in a proximitized semiconductor nanowire , _ preprint arxiv:1501.05155 . s. bravyi , d.p . divincenzo , and d. loss , _ schrieffer - wolff transformation for quantum many - body systems , _ ann . phys . * 326 * , 2793 ( 2011 ) . the general expression for @xmath380 @xcite is of symmetrized form with respect to the initial and final state energies @xmath381 and @xmath382 , cf . also eq . . since those two energies here coincide , we obtain the simpler result in eq . . p.a . oliveira and l. sanz , _ bell states and entanglement dynamics on two coupled quantum molecules _ , ann . ( in press ) , see arxiv:1503.00928 . at first sight , it might be confusing that even though entanglement was generated by tp in sec . [ sec5a ] , car processes here yield practically the same results ( with the above replacements ) . however , note that for @xmath383 , the cooper pair sector decouples completely . therefore , tp and car effectively become equivalent virtual transition processes that only act on different sets of states , see secs . [ sec4b ] and [ sec4c ] . s. gustavsson , r. leturcq , m. studer , i. shorubalko , t. ihn , k. ensslin , d.s . driscoll , and a.c . gossard , _ electron counting in quantum dots _ , surf . rep . * 64 * , 191 ( 2009 ) . chtchelkatchev , g. blatter , g.b . lesovik , and t. martin , _ bell inequalities and entanglement in solid - state devices _ , phys . b * 66 * , 161320(r ) ( 2002 ) . d. loss and e.v . sukhorukov , _ probing entanglement and nonlocality of electrons in a double - dot via transport and noise _ , phys . rev . lett . * 84 * , 1035 ( 2000 ) . p. samuelsson , e.v . sukhorukov , and m. bttiker , _ orbital entanglement and violation of bell inequalities in mesoscopic conductors _ , phys . * 91 * , 157002 ( 2003 ) . m. blaauboer and d.p . divincenzo , _ detecting entanglement using a double - quantum - dot turnstile _ , phys . lett . * 95 * , 160402 ( 2005 ) . b. braunecker , p. burset , and a. levy yeyati , _ entanglement detection from conductance measurements in carbon nanotube cooper pair splitters _ , lett . * 111 * , 136806 ( 2013 ) . albrecht and c.m . marcus , private communication .

_ globally coupled maps _ are collections of individual discrete - time dynamical systems ( their _ units _ ) which act independently on their respective phase spaces , except for the influence ( the _ coupling _ ) of a common parameter that is updated , at each time step , as a function of the _ mean field _ of the whole system . systems of this type have received some attention through the work of kaneko @xcite in the early 1990s , who studied systems of @xmath0 quadratic maps acting on coordinates @xmath1 $ ] , and coupled by a parameter depending in a simple way on @xmath2 . his key observation , for huge _ system size _ @xmath0 , was the following : if @xmath3 denotes the time series of mean field values of the system started in a random configuration @xmath4 , then , for many parameters of the quadratic map , and even for very small coupling strength , pairs @xmath5 of consecutive values of the field showed complicated functional dependencies plus some noise of order @xmath6 , whereas for uncoupled systems of the same size the @xmath7 , after a while , are constant up to some noise of order @xmath6 . while the latter observation is not surprising for independent units , the complicated dependencies for weakly coupled systems , a phenomenon kaneko termed _ violation of the law of large numbers _ , called for closer investigation . the rich bifurcation structure of the family of individual quadratic maps may offer some explanations , but since a mathematically rigorous investigation of even a small number of coupled quadratic maps in the chaotic regime still is a formidable task , there seem to be no serious attempts to tackle this problem . a model which is mathematically much easier to treat is given by coupled tent maps . indeed , for tent maps with slope larger than @xmath8 and moderate coupling strength , a system of @xmath0 mean field coupled units has an ergodic invariant probability density with exponentially decreasing correlations @xcite . this is true for all @xmath0 and for coupling strengths that can be chosen to be the same for all @xmath0 . nevertheless , ershov and potapov @xcite showed numerically that ( albeit on a much smaller length scale than in the case of coupled quadratic maps ) also mean field coupled tent maps exhibit a violation of the law of large numbers in the aforementioned sense . they also provided a mathematical analysis which demonstrated that the discontinuities of the invariant density of a tent map are at the heart of the problem . their analysis was not completely rigorous , however , as chawanya and morita @xcite could show that there are indeed ( exceptional ) parameters of the system for which there is no violation of the law of large numbers - contrary to the predictions in @xcite . on the other hand , references @xcite contain further simulation results on systems violating the law of large numbers . ( but at present , a mathematically rigorous treatment of globally coupled tent maps that is capable of classifying and explaining the diverse dynamical effects that have been observed does not seem to be in sight either . ) these studies were complemented by papers by jrvenp @xcite and keller @xcite , showing ( among other things ) that globally coupled systems of smooth expanding circle maps do not display violation of the law of large numbers at small coupling strength , because their invariant densities are smooth . given this state of knowledge , the present paper investigates specific systems of globally coupled piecewise fractional linear maps on the interval @xmath9 $ ] , where each individual map has a smooth invariant density . for small coupling strength , theorem 4 in @xcite extends easily to this setup and proves the absence of a violation of the law of large numbers . for larger coupling strength , however , we are going to show that this phenomenon does occur in the following sense : bifurcation : : the nonlinear _ self - consistent perron - frobenius operator ( pfo ) _ @xmath10 on @xmath11 , which describes the dynamics of the system in its thermodynamic limit , undergoes a supercritical pitchfork bifurcation as the coupling strength increases . ( here and in the sequel @xmath12 denotes lebesgue measure . ) mixing : : at the same time , all corresponding finite - size systems have unique absolutely continuous invariant probability measures @xmath13@xmath14 on their @xmath0-dimensional state space , and exhibit exponential decay of correlations under this measure . stable behaviour : : in the _ stable regime _ , i.e. for fixed small coupling strength below the bifurcation point of the infinite - size system , the measures @xmath13@xmath14 converge weakly , as the system size @xmath15 , to an infinite product measure @xmath16 , where @xmath17 is the unique fixed point of @xmath10 . bistable behaviour : : in the _ bistable regime _ , i.e. for fixed coupling strength above the bifurcation point of the infinite - size system , all possible weak limits of the measures @xmath13@xmath14 are convex combinations of the three infinite product measures @xmath18 , @xmath19 , where now @xmath17 is the unique unstable fixed point of @xmath10 and @xmath20 are its two stable fixed points . ( we conjecture that the measure @xmath16 is not charged in the limit . ) this scenario clearly bears some resemblance to the curie - weiss model from statistical mechanics and its dynamical variants . we also stress that a simple modification of our system leads to a variant where , instead of two stable fixed points , one stable two - cycle for @xmath10 is created at the bifurcation point . this may be viewed as the simplest possible scenario for a violation of the law of large numbers in kaneko s original sense . + in the next section we describe our model in detail , and formulate the main results . section [ sec : finite - sys - proofs ] contains the proofs for finite - size systems . in section [ sec : ifs ] we start the investigation of the infinite - size system via the self - consistent pfo @xmath10 . we observe that this operator preserves a class of probability densities which can be characterised as derivatives of herglotz - pick - nevanlinna functions . integral representations of these functions reveal a hidden order structure , which is respected by the operator @xmath10 , and allows us to describe the pitchfork bifurcation . in section [ sec : infinite - sys - proofs ] this dynamical picture for @xmath21 is extended to arbitrary densities . finally , in section [ sec : noisy ] , we discuss the situation when some noise is added to the dynamics . throughout , all measures are understood to be borel , and we let @xmath22probability measures on @xmath23 . lebesgue measure will be denoted by @xmath12 . we introduce a @xmath24-parameter family of piecewise fractional - linear transformations @xmath25 on @xmath26 $ ] , which will play the role of the local maps . to facilitate manipulation of such maps , we use their standard matrix representation , letting @xmath27{cc}a & b\\ c & d \end{array } \right ) \text{,}\ ] ] so that @xmath28 and @xmath29 . specifically , we consider the function @xmath30 , depending on a parameter @xmath31 , given by the coefficient matrix @xmath32{cc}r+4 & r+1\\ 2r & 2 \end{array } \right ) \text{.}\ ] ] one readily checks that @xmath33 , @xmath34 , @xmath35 for @xmath36 , and that ( the infimum being attained on @xmath37 ) @xmath38 the latter shows that @xmath30 is uniformly expanding if and only if @xmath39 , and we define our _ single - site maps _ @xmath40 with parameter @xmath41 by letting @xmath42{ll}f_{m_{r}}(x ) & \text{on } [ -\frac{1}{2},\alpha_{r})\text{,}\\ f_{m_{r}}(x)-1=f_{n_{r}}(x ) & \text{on } ( \alpha_{r},\frac{1}{2}]\text{,}\end{array } \right.\ ] ] where @xmath43 we thus obtain a family @xmath44 of uniformly expanding , piecewise invertible maps @xmath40 , each having two increasing covering branches . note also that this family is symmetric in that @xmath45 according to well - known folklore results , each map @xmath25 , @xmath41 , has a unique invariant probability density @xmath46 , and @xmath25 is exact ( hence ergodic ) w.r.t . the corresponding invariant measure . due to ( [ eq_symmetry ] ) , we have @xmath47 mod @xmath12 . we denote the _ perron - frobenius operator _ ( _ pfo_),w.r.t . lebesgue measure @xmath12 , of a map @xmath48 by @xmath49 , abbreviating @xmath50 . in our construction below we will exploit the fact that 2-to-1 fractional linear maps like @xmath25 in fact enable a fairly explicit analysis of their pfos on a suitable class of densities . in particular , the @xmath51 are known explicitly : let @xmath52 , @xmath53 . then @xmath54 is an integrable invariant density for @xmath55 , see @xcite . its normalised version @xmath56 is the unique @xmath55-invariant probability density . the key point in the choice of this family of maps is that for @xmath57 , @xmath55 is steeper in the positive part of @xmath58 than in its negative part , hence typical orbits spend more time on the negative part , which is confirmed by the invariant density ( see figure [ fig : trandhr ] ) . if @xmath59 , then @xmath55 favours the positive part . ( left ) , and @xmath51 ( right ) , for @xmath60 . ] ( left ) , and @xmath51 ( right ) , for @xmath60 . ] the heuristics of our construction is that for sufficiently strong coupling this effect of `` polarisation '' is reinforced and gives rise to bistable behaviour . for any probability measure @xmath61 , we denote its mean by @xmath62 and call this the _ field _ of @xmath63 . with a slight abuse of notation we also write , for @xmath64 , @xmath65 and , for @xmath66 @xmath67 to define the system of globally coupled maps ( both in the finite- and the infinite - size case ) we will , at each step of the iteration , determine the actual parameter as a function of the present field . this is done by means of a _ feedback function _ @xmath68 $ ] which we always assume to be real - analytic . everything else remains true if @xmath69 is merely of class @xmath70 . ] and _ s - shaped _ in that it satisfies @xmath71 and @xmath72 for all @xmath73 , while @xmath74 if @xmath75 . the most important single parameter in our model is going to be @xmath76 which quantifies the _ coupling strength_. [ remark : tanh ] the following will be our standard example of a suitable feedback function @xmath69 : @xmath77 where @xmath78 and @xmath79 . ( this requires some numerical effort . for @xmath80 and @xmath81 , elementary estimates suffice . ) for the results to follow we shall impose a few additional constraints on the feedback function @xmath69 , made precise in assumptions i and ii below . we consider a system @xmath82 of @xmath0 coupled copies of the parametrised map , defined by @xmath83 with @xmath84 . for the following theorem , which we prove in section [ sec : finite - sys - proofs ] , we need the following assumption ( satisfied by the example above ) : @xmath85{0.88\linewidth}\text{\textbf{assumption i:}}\quad\text { $ g'(x){\leqslant}25 - 50|g(x)|$ for all $ x\in x$. } \end{minipage}}}\ ] ] [ theo : finite - systems ] suppose the s - shaped function @xmath69 satisfies . then , for any @xmath86 , the map @xmath82 has a unique absolutely continuous invariant probability measure @xmath13@xmath14 . its density is strictly positive and real analytic . the systems @xmath87 @xmath13@xmath88 are exponentially mixing in various strong senses , in particular do hlder observables have exponentially decreasing correlations . the key to the proof is an estimate ensuring uniform expansion . after establishing the latter in section [ sec : finite - sys - proofs ] , the theorem follows from `` folklore '' results whose origins are not so easy to locate in the literature . in a @xmath89-setting , existence , uniqueness and exactness of an invariant density were proved essentially by krzyzewski and szlenk @xcite . exponential mixing follows from the compactness of the transfer operator first observed by ruelle @xcite . for a result which applies in our situation and entails theorem 1 , we refer to the main theorem of @xcite . since the coupling we defined is of mean - field type , we can adapt from the probabilistic literature ( see for example @xcite ) the classical method of taking the _ thermodynamic limit _ of our family of finite - size systems @xmath90 , as @xmath15 . to do so , consider the set @xmath91 of borel probability measures on @xmath58 , equipped with the topology of weak convergence and the resulting borel @xmath92-algebra on @xmath91 . define @xmath93 by @xmath94 we can then represent the evolution of any finite - size system using @xmath95 . indeed , if @xmath96 is the empirical measure of @xmath97 , then @xmath98 satisfies @xmath99 . furthermore , when restricted to the set of probability measures absolutely continuous with respect to @xmath12 , @xmath100 is represented by the _ self - consistent perron frobenius operator _ , which is the nonlinear positive operator @xmath101 defined as @xmath102 clearly , this map satisfies @xmath103 and preserves the set @xmath104 of probability densities . note , however , that it does not contract , i.e. there are @xmath105 such that @xmath106 . one may finally join these two aspects , the action of @xmath95 on means of dirac masses , or on absolutely continuous measures , via the following observation : [ pr : propchaos ] let @xmath107 , with @xmath108 . if @xmath109 is chosen according to @xmath110 , then , for any @xmath111 , the empirical measures @xmath112 converge weakly to @xmath113 as @xmath15 . this result confirms the point of view that the self - consistent pfo @xmath114 represents the infinite - size thermodynamic limit @xmath15 of the finite - size systems @xmath90 . its proof is reasonably simple ( easier than for stochastic evolutions ) . the only difficulty is that @xmath100 is not a continuous map on the whole of @xmath91 . this can be overcome with the following lemma , which proposition [ pr : propchaos ] is a direct consequence of , and whose proof is given in section [ se : conttil ] . [ le : contttil ] assume that a sequence @xmath115 in @xmath91 converges weakly to some non - atomic @xmath63 . then @xmath116 converges weakly to @xmath117 . here is an immediate consequence of this lemma that will be used below . [ coro : contttil ] assume that a sequence @xmath118 of @xmath95-invariant borel probability measures _ on _ @xmath91 converges weakly to some probability @xmath119 _ on _ @xmath91 . if there is a borel set @xmath120 with @xmath121 which only contains non - atomic measures , then @xmath119 is also @xmath95-invariant . our goal is to analyse the asymptotics of @xmath101 on @xmath104 . some basic features of @xmath101 can be understood considering the dynamics of @xmath122,}\ \quad { h}(r):=g(\phi(u_{r})),\ ] ] which governs the action of @xmath101 on the densities @xmath123 introduced in [ subsec : family - def ] , as @xmath124 in studying @xmath101 , we will always presuppose the following : @xmath125 this assumption can be checked numerically for specific feedback functions @xmath69 , like that of remark [ remark : tanh ] , cf . [ ssec : infinite - sys - proofs-1 ] below . by ( [ eq_symmetry ] ) , @xmath126 . note , however , that @xmath127 itself is not s - shaped ( see figure [ fig : plots-1 ] ) so that the s - shapedness of @xmath69 alone is not sufficient for that of @xmath128 . + assumption ii will enter our arguments only via the following dichotomy which it entails : either @xmath129 has a unique fixed point at @xmath130 + ( the * stable regime * with @xmath131 and @xmath130 stable ) , + or + @xmath129 has exactly three fixed points @xmath132 + ( the * bistable regime * with @xmath133 and @xmath134 stable ) . we will see that @xmath135 and @xmath136 , so that the stable regime corresponds to the condition @xmath137 . observe now that @xmath138 ( since @xmath139 for @xmath140 , and each @xmath25 is ergodic ) . we are going to show that the fixed points @xmath141 , and @xmath142 dominate the long - term behaviour of @xmath143 on @xmath104 completely , and that they inherit the stability properties of the corresponding parameters @xmath132 . therefore , the stable / bistable terminology for @xmath144 introduced above also provides an appropriate description of the asymptotic behaviour of @xmath101 . [ * long - term behaviour of * @xmath101 * * on * * @xmath104 ] [ theo : infinite - system ] consider @xmath145 , @xmath146 equipped with the metric inherited from @xmath11 . assuming ( i ) and ( ii ) , we have the following : 1 . [ theo2:item1 ] in the stable regime , @xmath17 is the unique fixed point of @xmath101 , and attracts all densities , that is , @xmath147 2 . in the bistable regime , @xmath148 are the only fixed points of @xmath101 . now @xmath17 is unstable , while @xmath149 and @xmath150 are stable . more precisely : a. @xmath142 are stable fixed points for @xmath101 in the sense that their respective basins of attraction are @xmath151-open . b. if @xmath152 is not attracted by @xmath149 or @xmath150 , then it is attracted by @xmath153 . c. @xmath153 is not stable . indeed , @xmath17 can be @xmath151-approximated by convex analytic densities from either basin . it is a hyperbolic fixed point of @xmath10 in the sense made precise in proposition [ prop : hyperbolic - fixed ] of section [ sec : infinite - sys - proofs ] . in case @xmath154 with @xmath78 and @xmath79 , both theorems apply . the infinite - size system is stable iff @xmath155 , and bistable otherwise , while all finite - size systems have a unique a.c.i.m . in this parameter region . the theorem summarises the contents of propositions [ prop : summary2 ] , [ prop : basin - boundary ] and [ prop : hyperbolic - fixed ] of section [ sec : infinite - sys - proofs ] ( which , in fact , provide more detailed information ) . the proofs rest on the fact that pfos of maps with full fractional - linear branches leave the class of herglotz - pick - nevanlinna functions invariant . this observation can be used to study the action of @xmath101 in terms of an iterated function system on the interval @xmath156 $ ] with two fractional - linear branches and place dependent probabilities . in the bistable regime the system is of course not contractive , but it has strong monotonicity properties and special geometric features which allow to prove the theorem . + our third theorem , which is essentially a corollary to the previous ones , describes the passage from finite - size systems to the infinite - size system . below , weak convergence of the @xmath13@xmath157 to some @xmath158@xmath159 means that @xmath160@xmath13@xmath161@xmath13 for all continuous @xmath162 which only depend on finitely many coordinates . ( so that @xmath163@xmath13@xmath14 is defined , in the obvious fashion , for @xmath0 large enough . ) [ theo : passage - to - infinity ] the @xmath90-invariant probability measures @xmath13@xmath14 of theorem [ theo : finite - systems ] correspond to the @xmath95-invariant probability measures @xmath13@xmath164 on @xmath91 . all weak accumulation points @xmath119 of the latter sequence are @xmath165-invariant probability measures concentrated on the set of measures absolutely continuous w.r.t . @xmath12 . furthermore : 1 . in the stable regime , the sequence @xmath166@xmath13@xmath167 converges weakly to the point mass @xmath168 . in other words , the sequence @xmath166@xmath13@xmath169 converges weakly to the pure product measure @xmath170 on @xmath171 . 2 . in the bistable regime , each weak accumulation point @xmath119 of the sequence @xmath166@xmath13@xmath167 is of the form @xmath172 for some @xmath173 $ ] . in other words , each weak accumulation point of the sequence @xmath166@xmath13@xmath169 is of the form @xmath174 . we can not prove , so far , that @xmath175 , which is to be expected because @xmath17 is an unstable fixed point of @xmath10 . in section [ sec : noisy ] we show that @xmath175 indeed , if some small noise is added to the system . as @xmath99 , the @xmath90-invariant probability measures @xmath13@xmath14 of theorem [ theo : finite - systems ] correspond to @xmath95-invariant probability measures @xmath13@xmath164 on @xmath176 . their possible weak accumulation points are all concentrated on sets of measures from @xmath176 with density w.r.t . @xmath12 , see theorem 3 in @xcite . ( the proof of that part of the theorem we refer to does not rely on the continuity of the local maps that is assumed in that paper . ) therefore corollary [ coro : contttil ] shows that all these accumulation points are @xmath165-invariant probability measures concentrated on measures with density w.r.t . @xmath12 . in other words , they can be interpreted as @xmath177-invariant probability measures on @xmath146 . now theorem [ theo : infinite - system ] implies that the sequence @xmath166@xmath13@xmath167 converges weakly to the point mass @xmath178 in the stable regime , whereas , in the bistable regime , each such limit measure is of the form @xmath172 for some @xmath173 $ ] ( observe the symmetry of the system ) . now the corresponding assertions on the measures @xmath13@xmath14 follow along known lines , for a reference see e.g. ( * ? ? ? * proposition 1 ) . we assume throughout this section that @xmath179 in order to apply the main theorem of mayer @xcite we must check his assumptions ( a1 ) ( a4 ) for the map @xmath180 . to that end define @xmath181^n$ ] by @xmath182 . obviously @xmath183 , and ( a1 ) ( a4 ) follow readily from the following facts that we are going to prove : [ lemma : homeo ] @xmath181^n$ ] is a homeomorphism which extends to a diffeomorphism between open neighbourhoods of @xmath184 and @xmath185^n$ ] . [ lemma : analytic - diffeo ] the inverse @xmath186 of @xmath187 is real analytic and can be continued to a holomorphic mapping on a complex @xmath188-neighbourhood @xmath189 of @xmath185^n$ ] such that @xmath190 is contained in a @xmath191-neighbourhood of @xmath184 for some @xmath192 . to verify these two lemmas we need the following uniform expansion estimate which we will prove at the end of this section . ( here @xmath193 denotes the euclidean norm . ) [ lemma : inv - contraction ] there is a constant @xmath194 such that @xmath195 for all @xmath196 and @xmath197 . obviously @xmath198^n$ ] . hence it is sufficient to prove the assertions of the lemma for the map @xmath199 . as each @xmath200 is differentiable on @xmath201 ( recall that @xmath202 ) , @xmath203 extends to an analytic mapping from @xmath204 . by lemma [ lemma : inv - contraction ] , it is locally invertible on @xmath205 for each sufficiently small @xmath206 . ( note that @xmath207 . ) all we need to show is that this implies global invertibility of @xmath208 , because then the possibility to extend @xmath203 diffeomorphically to a small open neighbourhood of @xmath184 in @xmath209 follows again from the local invertibility on @xmath210 for some @xmath211 . so we prove the global invertibility of @xmath208 . as each @xmath212 is a homeomorphism that leaves fixed the endpoints of the interval @xmath58 , we have @xmath213 and @xmath214 . observing the simple fact that @xmath215 is a paracompact connected smooth manifold without boundary and with trivial fundamental group , we only need to show that @xmath208 is proper in order to deduce from ( * ? ? ? * corollary 1 ) that @xmath216 is a diffeomorphism of @xmath215 . so let @xmath217 be a compact subset of @xmath215 . as @xmath216 extends to the continuous map @xmath218 and as @xmath213 , the set @xmath219 is closed and hence compact . therefore @xmath208 is indeed proper . as @xmath187 is real analytic on a real neighbourhood of @xmath184 , the real analyticity of @xmath186 on a real neighbourhood of @xmath185^n$ ] follows from the real analytic inverse function theorem ( * ? ? ? * theorem 18.1 ) . it extends to a holomorphic function on a complex @xmath188-neighbourhood @xmath189 of @xmath185^n$ ] see e.g. the discussion of complexifications of real analytic maps in@xcite . if @xmath220 is sufficiently small , lemma [ lemma : inv - contraction ] implies that @xmath186 is a uniform contraction on @xmath189 . hence the @xmath191 in the statement of the lemma can be chosen strictly smaller than @xmath188 . recall that @xmath221 where @xmath222 , @xmath223 and we write @xmath224 instead of @xmath200 . denote @xmath225 , @xmath226 and observe that @xmath227 simplifies to @xmath228 so that @xmath229 then the derivative of the coupled map @xmath230 is @xmath231 with @xmath232 denoting the identity matrix . letting @xmath233 and observing that @xmath234 so that @xmath235 , the inverse of @xmath236 is @xmath237 in order to check conditions under which @xmath187 is uniformly expanding in all directions , it is sufficient to find conditions under which @xmath238 _ uniformly _ in @xmath239 . observe first that @xmath240 for @xmath241 . from now on we fix a point @xmath239 and suppress it as an argument to all functions . then , if @xmath242 is any vector in @xmath243 , some scalar multiple of it can be decomposed in a unique way as @xmath244 where @xmath245 is perpendicular to @xmath246 . denote @xmath247 , @xmath248 , and observe that @xmath249 as @xmath246 has only nonnegative entries . we estimate the euclidian norm of @xmath250 : @xmath251 where @xmath252 . as @xmath253 and @xmath254 ( observe that all entries of @xmath246 are bounded by @xmath24 ) , we can continue the above estimate with @xmath255 to estimate this expression we abbreviate temporarily @xmath256 . then @xmath257 , and straightforward maximisation yields : @xmath258 so we can continue the above estimate by @xmath259 where @xmath260 hence @xmath261 and therefore @xmath262 observing @xmath263 one finds numerically that the norm is bounded by @xmath264 uniformly for all @xmath265 , if @xmath266 and @xmath267 . hence the map @xmath187 is uniformly expanding in all directions provided ( [ eq : finite - ass ] ) holds . the pfos @xmath268 allow a detailed analysis since their action on certain densities has a convenient explicit description : consider the family @xmath269 of probability densities on @xmath58 given by @xmath270 as pointed out in @xcite ( using different parametrisations ) , perron - frobenius operators @xmath271 of fractional - linear maps @xmath272 act on these densities via their _ duals _ @xmath273 , where @xmath274{cc}0 & 1\\ 1 & 0 \end{array } \right ) \cdot m\cdot\left ( \begin{array } [ c]{cc}0 & 1\\ 1 & 0 \end{array } \right ) = \left ( \begin{array } [ c]{cc}d & c\\ b & a \end{array } \right ) \text { \quad for } m=\left ( \begin{array } [ c]{cc}a & b\\ c & d \end{array } \right ) \text{,}\ ] ] in that @xmath275 for matrices @xmath276 and intervals @xmath277 for which @xmath278 . ( this can also be verified by direct calculation ) . since @xmath279 , one can compute the duals @xmath280 of the individual branches of @xmath25 , then express @xmath281 as the convex combination @xmath282}\cdot w_{y})\nonumber\\ & = p_{r}(y)\cdot w_{\sigma_{r}(y)}+(1-p_{r}(y))\cdot w_{\tau_{r}(y)}\text{,}\label{eq_pr_gy_in_terms_of_sigma_tau}\ ] ] with weights @xmath283 it is straightforward to check that for every @xmath31 the functions @xmath284 , @xmath285 are continuous and strictly increasing on @xmath156 $ ] with images @xmath286)=[-2,2/3]$ ] and @xmath287)=[-2/3,2]$ ] . from this remark and ( [ eq_pr_gy_in_terms_of_sigma_tau ] ) it is clear that the @xmath268 preserve the class of those @xmath288 which are convex combinations @xmath289 of the special densities @xmath290 for some _ representing measure _ @xmath13 from @xmath291 . we find that @xmath268 acts on representing measures according to @xmath292 where @xmath293 denotes the measure with density @xmath294 w.r.t . @xmath13 . to continue , we need to collect several facts about the dual maps @xmath284 and @xmath285 . we have @xmath295 showing that @xmath284 and @xmath285 are strictly concave , respectively convex , on @xmath156 $ ] . one next gets readily from that @xmath296 wherever defined , and @xmath297\diagdown\{-r\}\ ] ] while ( and this observation will be crucial later on ) @xmath298 and @xmath299 have a common zero @xmath300 and @xmath301 in the following , we restrict our parameter @xmath302 to the set @xmath303 $ ] . direct calculation proves that , letting @xmath304 $ ] , we have @xmath305 so that @xmath306 is an invariant set for all such @xmath284 and @xmath285 , and that @xmath307 which provides us with a common contraction rate on @xmath306 for the @xmath284 and @xmath285 from this parameter region . all these features of @xmath298 and @xmath299 are illustrated by figure [ fig : sigma - tau ] . and @xmath299 for @xmath308 . the small invariant box has endpoints @xmath309 and @xmath299 . ] we denote by @xmath310 the field of the density @xmath311 , and find by explicit integration that @xmath312 for @xmath313 . in particular , @xmath314 and @xmath315 , so the field depends monotonically on @xmath316 . note also that we have , for all @xmath317 , @xmath318 we focus on densities @xmath319 with representing measure @xmath13 supported on @xmath306 , i.e. on the class @xmath320 : @xmath321 . writing @xmath322 and recalling ( [ eq_pr_acting_on_representing_measures ] ) , we find that our nonlinear operator @xmath101 acts on the representing measures from @xmath323 via @xmath324 as @xmath325 , the support of @xmath326 , is contained in @xmath327 , it is immediate from ( [ eq_invariance_of_y ] ) that @xmath328 for @xmath329 $ ] we find that @xmath284 and @xmath285 each have a unique stable fixed point in @xmath306 , given by @xmath330 respectively . note that the interval @xmath331 $ ] of width @xmath332 between these stable fixed points is invariant under both @xmath284 and @xmath285 , see the small boxed region in figure [ fig : sigma - tau ] . furthermore , each of @xmath333 and @xmath334 is mapped to @xmath302 under the branch not fixing it , i.e. @xmath335 meaning that , restricted to @xmath336 , @xmath284 and @xmath285 are the inverse branches of some 2-to-1 piecewise fractional linear map @xmath337 . the explicit @xmath25-invariant densities @xmath51 from ( [ eq : inv - dens-2 ] ) can be represented as @xmath338 with @xmath339 given by @xmath340 our goal in this section is to study the asymptotic behaviour of @xmath101 on @xmath341 , using its representation by means of the ifs @xmath342 . we will prove [ * long - term behaviour of * @xmath101 * * on * * @xmath341 ] [ prop : summary ] take any @xmath343 , @xmath344 for some @xmath317 . the following is an exhaustive list of possibilities for the asymptotic behaviour of the sequence @xmath345 : 1 . @xmath346 for some @xmath111 . then @xmath345 converges to @xmath347 and hence @xmath348 converges to @xmath349 in @xmath350 . [ item : summary-2 ] the interval @xmath351 contains @xmath352 for all @xmath111 . then @xmath345 converges to @xmath353 and hence @xmath354 converges to @xmath17 in @xmath350 . in this case also the length of @xmath351 tends to @xmath352 3 . @xmath355 for some @xmath111 . then @xmath356 converges to @xmath357 and hence @xmath348 converges to @xmath358 in @xmath350 . in the stable regime , only scenario ( 2 ) is possible , so that we always have convergence of @xmath345 to @xmath353 . our arguments will rely on continuity and monotonicity properties of the ifs , that we detail below , before proving proposition [ prop : summary ] in section [ se : propro ] . _ convergence in _ @xmath323 , @xmath359 , will always mean weak convergence of measures , @xmath360 for bounded continuous @xmath361 . since @xmath306 is a bounded interval , this is equivalent to convergence in the _ wasserstein - metric _ @xmath362 on @xmath323 . if @xmath363 and @xmath364 are the distribution functions of @xmath13 and @xmath365 , then @xmath366 the kantorovich - rubinstein theorem ( e.g. ( * ? ? * ch.11 ) ) provides an additional characterisation : @xmath367{\leqslant}1}\int_{y}\psi\,d(\mu-\nu)\ ] ] for any @xmath368 . here , @xmath369 : = \sup_{y , y^{\prime}\in y;y\neq y^{\prime}}\left\vert \psi(y)-\psi(y^{\prime})\right\vert /\left\vert y - y^{\prime}\right\vert $ ] for any @xmath370 ( and analogously for functions on other domains ) . we now see that there is a constant @xmath371 ( the common lipschitz bound for the functions @xmath372 on @xmath306 , where @xmath73 ) such that @xmath373 this means that , for densities from @xmath341 , convergence of the representing measures implies @xmath151-convergence of the densities . we will also use the following estimate . [ lemma : stability ] there are constants @xmath374 such that @xmath375 for all @xmath368 and all @xmath376 . for lipschitz functions @xmath377 on @xmath306 we have @xmath378 \cdot{\mathsf{d}_w}(\mu,\nu)\nonumber\\ + \sup_{y\in y}\,\mathrm{lip}_{r}\left [ \psi(\sigma_{.}(y))\,p_{. } ( y)+\psi(\tau_{.}(y))(1-p_{.}(y))\right ] \cdot|r - s| \end{gathered}\ ] ] suppose that @xmath377 is @xmath379 , with @xmath380 . then the first lipschitz constant is bounded by @xmath381 } < \infty,\ ] ] and the second one by @xmath382 } < \infty,\ ] ] and the lemma follows from the kantorovich - rubinstein theorem , since these @xmath383 functions @xmath377 uniformly approximate the lipschitz functions appearing there . [ co : contifs ] the operators @xmath384 , @xmath385 , and @xmath342 are uniformly lipschitz - continuous on @xmath323 for the wasserstein metric . the lipschitz - continuity of @xmath384 is immediate from lemma [ lemma : stability ] . for @xmath342 , we recall from and that @xmath386 with @xmath387 so that @xmath388 which allows to conclude with lemma [ lemma : stability ] . rigorous numerical bounds give @xmath389 and @xmath390 . these estimates can be used to show that not only the individual @xmath384 are uniformly contracting on @xmath323 , but also ( under suitable restrictions on the function @xmath69 ) @xmath342 is a uniform contraction on @xmath391 where @xmath392 is a suitable neighbourhood of the support of @xmath347 , and @xmath347 is the representing measure of @xmath349 with @xmath393 the unique positive fixed point of the equation @xmath394 . our treatment of @xmath342 , however , does not rely on these estimates , because it is based on monotonicity properties explained below . on the space @xmath323 of probability measures @xmath395 representing densities from @xmath396 we introduce an order relation by defining @xmath397 the symbols @xmath398 , @xmath399 and @xmath400 designate the usual variants of @xmath401 . we collect a few elementary facts on this order relation : @xmath402{0.88\linewidth}$\mu\preceq\nu$ if and only if $ \textstyle{\int_y } u\,d\mu{\leqslant}\int_y u\,d\nu$ for each bounded and non - decreasing $ u : y\to{{\mathbb r}}$.\end{minipage}}}\ ] ] in particular , if @xmath403 , then @xmath404 , and hence @xmath405 as well . @xmath406{0.88\linewidth}if $ \mu\preceq\nu$ and if $ \rho_1,\rho_2:y\to y$ are non - decreasing and such that $ \rho_1(y){\leqslant}\rho_2(y)$ for all $ y$ , then $ \mu\circ\rho_1^{-1}\preceq\nu\circ\rho_2^{-1}$.\end{minipage}}}\ ] ] @xmath407{0.88\linewidth}if $ \mu\preceq\nu$ and if $ { \textstyle{\int_y } } u\,d\mu = { \textstyle{\int_y } } u\,d\nu$ for some strictly increasing\\ $ u : y\to{{\mathbb r}}$ , then $ \mu=\nu$.\end{minipage}}}\ ] ] @xmath408{0.88\linewidth}let $ z\in y$. then $ \delta_z\preceq\mu$ if and only if $ { \operatorname{supp}}(\mu)\subseteq[z,\infty)$.\end{minipage}}}\ ] ] we also observe that the representing measures @xmath409 of the @xmath55-invariant densities @xmath51 form a linearly ordered subset of @xmath323 . routine calculations based on ( [ eq : mu - r - density ] ) show that @xmath410 our analysis of the asymptotic behaviour of the ifs will crucially depend on the fact that the operators @xmath411 and @xmath412 respect this order relation on @xmath323 , as made precise in the next lemma : [ lemma : pre - monotonicity ] let @xmath368 and @xmath413 . a. [ item : pre - monotonicity - a ] if @xmath414 , then @xmath415 . b. [ item : pre - monotonicity - b ] if @xmath416 , then @xmath417 . c. [ item : pre - monotonicity - c ] if @xmath414 , then @xmath418 . let @xmath419 be bounded and non - decreasing and recall that @xmath420}\,d\mu.\end{gathered}\ ] ] [ item : pre - monotonicity - a ] ) let @xmath403 . in view of ( [ eq : mono - fact-1 ] ) we can prove @xmath421 by showing that the integrand on the right - hand side of is non - decreasing . for this we use the facts that @xmath298 and @xmath299 are strictly increasing with @xmath422 , and that @xmath423 is non - increasing , since @xmath424 . one gets then , for @xmath425 , @xmath426.}_{{\geqslant}0 } \end{split}\ ] ] hence @xmath403 implies @xmath427 . now , if @xmath319 is strictly increasing , then ( [ eq : monotonicity - estimate ] ) always is a strict inequality , i.e. the integrand on the right - hand side of is strictly increasing . therefore , @xmath414 implies @xmath428 by ( [ eq : mono - fact-2a ] ) . + [ item : pre - monotonicity - b ] ) we must show that is non - decreasing as a function of @xmath302 . to this end note first that @xmath429 hence , if @xmath430 , then @xmath431 and for strictly increasing @xmath319 we have indeed a strict inequality . + [ item : pre - monotonicity - c ] ) this follows from [ item : pre - monotonicity - a ] ) and [ item : pre - monotonicity - b ] ) : @xmath432 as @xmath433 by . in [ subsec : stablemfd ] below , we will also make use of a more precise quantitative version of statement a ) . it is natural to state and prove it at this point . [ lemma : gap - estimate ] suppose that @xmath434 are such that @xmath435 and @xmath436 . then , for @xmath403 , @xmath437 observing that @xmath438 and @xmath439 are @xmath440 , we find for the last expression in that @xmath441 { \geqslant}\alpha \beta y - \alpha \beta x.\ ] ] this turns into a chain of inequalities which shows that the function given by @xmath442 is non - decreasing . hence , by , @xmath403 entails @xmath443 , which is . we are now going to clarify the asymptotic behaviour of @xmath342 on @xmath323 . in view of ( [ eq : norm_w_estimate ] ) , this also determines the asymptotics of @xmath101 on @xmath341 , and hence proves proposition [ prop : summary ] . our argument depends on monotonicity properties which we can exploit since the topology of weak convergence on @xmath323 , conveniently given by the wasserstein metric , is consistent with the order relation introduced above . indeed , one easily checks : @xmath444{0.88\linewidth}if $ ( \nu_n)$ and $ ( \bar{\nu}_n)$ are weakly convergent sequences in $ \mathsf{p}(y)$ with $ \nu_n \preceq \bar{\nu}_n$ for all $ n$ , then $ \lim \nu_n \preceq \lim \bar{\nu}_n$.\end{minipage } } } } \ ] ] recall from [ subsec : infinite - sys ] that , in the bistable regime , @xmath393 is the unique positive fixed point of the equation @xmath394 . for convenience , we now let @xmath445 then , in either case , @xmath51 with representing measure @xmath409 is fixed by @xmath101 iff @xmath446 . by ( [ eq : mono - mu - r - family ] ) we have @xmath447 with strict inequalities in the bistable regime . [ lemma : monotonicity ] if @xmath448 , then @xmath449 , and the sequence @xmath345 converges weakly to a measure @xmath450 with @xmath446 . the same holds for @xmath399 instead of @xmath401 . the monotonicity of the sequence @xmath345 follows immediately from lemma [ lemma : pre - monotonicity][item : pre - monotonicity - c ] ) . because of ( [ eq : mono - cty - fact-1 ] ) , it implies that the sequence can have at most one weak accumulation point . compactness of @xmath323 and continuity of @xmath342 therefore ensure that @xmath345 converges to a fixed point of @xmath342 , i.e. to one of the measures @xmath409 with @xmath451 , and ( [ eq : mono - cty - fact-1 ] ) entails @xmath452 . the proof for decreasing sequences is the same . the following lemma strengthens the previous one considerably . it provides uniform control , in terms of the wasserstein distance , on the asymptotics of large families of representing measures . [ lemma : controlled - convergence ] we have the following : a. [ item : controlled - convergence - a ] in the stable regime , there exists a sequence @xmath453 of positive real numbers converging to zero such that @xmath454 b. [ item : controlled - convergence - b ] in the bistable regime , for every @xmath455 there exists a sequence @xmath453 of positive real numbers converging to zero such that @xmath456 an analogous assertion holds for measures @xmath457 . as @xmath458 $ ] , we trivially have @xmath459 for all @xmath460 . in particular , @xmath461 , and lemma [ lemma : monotonicity ] ensures that @xmath462 converges . due to lemma [ lemma : pre - monotonicity][item : pre - monotonicity - c ] ) , we have @xmath463 for all @xmath111 , showing , via ( [ eq : mono - cty - fact-1 ] ) , that @xmath464 . in the same way one proves that @xmath465 converges to @xmath357 . for the stable regime this means that both sequences converge to @xmath466 . + [ item : controlled - convergence - a ] ) assume we are in the stable regime . by the above discussion , @xmath467 tends to zero . for any @xmath13 , ( [ eq : mono - cty - fact-1 ] ) guarantees @xmath468 for all @xmath111 . hence @xmath469 for all @xmath316 , proving @xmath470 . + [ item : controlled - convergence - b ] ) now consider the bistable regime . note first that if there is a suitable sequence @xmath453 for some @xmath455 , then it also works for all @xmath471 . therefore , there is no loss of generality if we assume that @xmath455 is so small that @xmath472 ( use , and to see that this can be achieved . ) since @xmath473 and @xmath474 , we then have @xmath475 , recall lemma [ lemma : pre - monotonicity ] . lemma [ lemma : monotonicity ] then implies that @xmath476 converges to @xmath347 . in view of the initial discussion , @xmath462 converges to @xmath347 as well , so that @xmath477 defines a sequence of reals converging to zero . now take any @xmath317 with @xmath478 , then @xmath479 for all @xmath111 , and @xmath480 follows as in the proof of [ item : controlled - convergence - a ] ) above . this observation enables us to determine the asymptotics of @xmath481 for any @xmath460 which is completely supported on the positive half @xmath482 $ ] of @xmath306 ( meaning that @xmath483 , cf . ( [ eq : mono - fact-3 ] ) ) , or on its negative half @xmath484 . [ coro : monotone - convergence ] let @xmath317 . a. [ item : monotone - convergence - a ] in the stable regime , the sequence @xmath345 converges to @xmath353 . b. [ item : monotone - convergence - b ] in the bistable regime , if @xmath483 , then the sequence @xmath345 converges to @xmath347 . if @xmath485 , it converges to @xmath357 . [ item : monotone - convergence - a ] ) follows immediately from lemma [ lemma : controlled - convergence][item : controlled - convergence - a ] ) . we turn to [ item : monotone - convergence - b ] ) : let @xmath486 . then @xmath59 because @xmath487 on @xmath482 $ ] and @xmath483 . therefore @xmath488 . fix some @xmath489 as in lemma [ lemma : controlled - convergence][item : controlled - convergence - b ] ) , w.l.o.g . @xmath490 . then @xmath491 since @xmath284 and @xmath285 map @xmath492 into @xmath493 $ ] , so that indeed @xmath494 as @xmath495 by the lemma . it remains to investigate the convergence of sequences @xmath345 when none of these measures can be compared ( in the sense of @xmath398 ) to @xmath353 . to this end let @xmath496:=y$ ] . given a sequence of parameters @xmath497 define @xmath498 for @xmath499 , and , for any @xmath500 $ ] , consider the measures @xmath501 then @xmath502 @xmath503 ) \cup \tau_{r_{n } } ( [ a_{n-1},b_{n-1 } ] ) \subseteq$ ] @xmath504 $ ] by induction . write @xmath505_{\varepsilon}:=[a-\varepsilon , b+\varepsilon]$ ] , where @xmath506 . the next lemma exploits the crucial observation that the two branches @xmath298 and @xmath299 have tangential contact at their common zero @xmath507 , see figure [ fig : sigma - tau ] . [ lemma : crucial ] there exists some @xmath508 such that the following holds : suppose that @xmath509 is any given sequence in @xmath510 . if for some @xmath206 and @xmath511 we have @xmath512_{\varepsilon}\quad \text{for}\ n{\geqslant}\bar{n}(\varepsilon)\text{,}\tag{${\clubsuit}_\varepsilon$}\ ] ] then @xmath513 @xmath514 and , in case @xmath515 , @xmath516\quad \text{for}\ n{\geqslant}\bar{n}(0)+1 \text{.}\ ] ] let @xmath206 and assume ( @xmath517 ) . note that , for @xmath518 , @xmath519 the first implication holds because @xmath520 as @xmath521 increases strictly , and since by ( @xmath517 ) we have @xmath522 $ ] , whence @xmath523 due to ( [ eq : less3/4 ] ) . analogously for the second implication . the third is immediate from monotonicity . now , as @xmath284 and @xmath285 share a common zero @xmath507 , ( [ eq : less3/4 ] ) ensures @xmath524 in case @xmath525 $ ] . otherwise , note that @xmath507 is @xmath526-close to one of the endpoints , w.l.o.g . to @xmath527 . since @xmath284 and @xmath285 are tangent at @xmath528 , there is some @xmath529 s.t . @xmath530 in this case , while ( [ eq : less3/4 ] ) controls the rest of @xmath531 . in view of @xmath532 , we thus obtain , for @xmath533 , @xmath534 statement ( [ eq : club-1 ] ) follows immediately . for the asymptotic estimate ( [ eq : club-2 ] ) on @xmath535 , use the above inequality plus the observation that , by the first two implications stated in this proof , @xmath536 and @xmath537 never exceed @xmath538 . finally , if @xmath515 , ( [ eq : club-3 ] ) is straightforward from ( @xmath517 ) and the third implication above . while the full strength of this lemma will only be required in the next subsection , the @xmath515 case enables us to now conclude the the conclusions of ( 1 ) and ( 3 ) follow from corollary [ coro : monotone - convergence ] . if neither of these two cases applies , then the assumption of ( 2 ) must be satisfied , and so condition ( @xmath539 ) of lemma [ lemma : crucial ] is satisfied with @xmath540 . hence @xmath541 by ( [ eq : club-2 ] ) . as the @xmath481 are supported in @xmath542 $ ] , these measures must converge to @xmath353 . we are now going to clarify the asymptotics of the self - consistent pfo on the set @xmath104 of all densities , proving [ * long - term behaviour of * @xmath101 * * on * * @xmath104 ] [ prop : summary2 ] for every @xmath64 , the sequence @xmath543 converges in @xmath11 , and @xmath544 the basins @xmath545 of the stable fixed points @xmath20 are @xmath151-open . ( the set of densities attracted to @xmath17 in the bistable regime will be discussed in [ subsec : stablemfd ] below . ) + we begin with some notational preparations . throughout , we fix some @xmath288 . the iterates @xmath354 define parameters @xmath546 ( @xmath499 ) . with this notation , @xmath547 . we let @xmath548 , @xmath549 , denote the partition of @xmath58 into monotonicity intervals of @xmath550 . note that each branch of this map is a fractional linear bijection from a member of @xmath548 onto @xmath58 . since the @xmath55 , @xmath385 , have a common uniform expansion rate , we see that @xmath551 , and hence , by the standard martingale convergence theorem , @xmath552\rightarrow u$ ] in @xmath350 , that is , @xmath553-u\right\| _ { l_{1}({x},\lambda)}\longrightarrow0\quad \text { as } n\rightarrow\infty\text{.}\ ] ] write @xmath554\right)\quad \text{for}\ { k}{\geqslant}0\ \text{and}\ n{\geqslant}1 , \ ] ] and observe that @xmath555 because it is a weighted sum of images of the constant function @xmath24 under various fractional linear branches ( recall ( [ eq : transfer - by - fract - lin ] ) and ( [ eq_invariance_of_y ] ) ) . for @xmath556 we let @xmath557 and write , in analogy to the notation introduced for @xmath549 , @xmath558 and @xmath559 . obviously , @xmath560 for all @xmath561 . hence there are measures @xmath562 such that @xmath563 . observe also that @xmath564 as @xmath565 for all @xmath302 , so that in particular @xmath566 in addition , we need to understand the distances @xmath567 which , in fact , admit some control which is uniform in @xmath568 : [ lemma : shadow - estimate ] there is a non - decreasing sequence @xmath569 in @xmath570 , not depending on @xmath571 , such that @xmath572 let @xmath573 , and observe that entails @xmath574 applying lemma [ lemma : stability ] repeatedly , we therefore see that @xmath575 where the last inequality uses . ( recall that @xmath101 does not contract on @xmath576 , whence the need for the @xmath577-term . ) letting @xmath578 and @xmath579 , we thus obtain @xmath580 which proves our assertion . we can now complete the we begin with the easiest situation : + [ [ the - stable - regime ] ] * the stable regime * + + + + + + + + + + + + + + + + + + + we have to show that @xmath581 . take any @xmath582 . let @xmath453 be the sequence provided by lemma [ lemma : controlled - convergence][item : controlled - convergence - a ] ) , and @xmath217 the constant from ( [ eq : norm_w_estimate ] ) . there is some @xmath583 ( henceforth fixed ) for which @xmath584 . in view of ( [ eq : eta - n - cge-0 ] ) , there is some @xmath585 such that @xmath586 whenever @xmath587 . we then find , using ( [ eq : shadow - estimate ] ) , lemma [ lemma : shadow - estimate ] , and ( [ eq : norm_w_estimate ] ) together with lemma [ lemma : controlled - convergence][item : controlled - convergence - a ] ) that @xmath588 which completes the proof in this case . + [ [ the - bistable - regime ] ] * the bistable regime * + + + + + + + + + + + + + + + + + + + + + given the sequence @xmath589 as before , we let @xmath542\subseteq y$ ] be the sequence of parameter intervals from lemma [ lemma : crucial ] . observe that the measures representing the @xmath590 satisfy @xmath591 $ ] for all @xmath583 and @xmath0 . we now distinguish two cases : + _ first case : _ for all @xmath211 we have ( @xmath592 ) from lemma [ lemma : crucial ] . then , for any @xmath211 , the lemma ensures that there is some @xmath583 ( henceforth fixed ) with @xmath593 , so that also @xmath594 , whatever @xmath0 . due to ( [ eq : eta - n - cge-0 ] ) , @xmath595 for @xmath587 , and we find , using ( [ eq : shadow - estimate ] ) and ( [ eq : norm_w_estimate ] ) , @xmath596 showing that indeed @xmath597 . + _ second case : _ there is some @xmath598 s.t . ( @xmath599 ) is violated in that , say , @xmath600 for infinitely many @xmath583 . we show that this implies @xmath601 . ( if ( @xmath599 ) is violated in the other direction , @xmath602 then follows by symmetry . ) + in view of ( [ eq : no - good - name ] ) , and since ( due to @xmath603 , ( [ eq : mono - fact-1 ] ) , and ( [ eq_mean_for_convex_combi ] ) ) @xmath604 , we have @xmath605 and hence , observing that @xmath606 and writing @xmath607 for @xmath313 , @xmath608 for all @xmath0 and @xmath568 . note that @xmath609 , see and . therefore , if we fix some @xmath610 , there exists some @xmath611 such that @xmath612 for all @xmath613}$ ] . without loss of generality , @xmath614 . + now fix @xmath0 such that @xmath615 , and let @xmath616 satisfy ( [ eq : exceptional : n ] ) . due to , we have @xmath617 now , if @xmath618 , then , by ( [ eq : sigma - tilde ] ) , @xmath619 otherwise , @xmath620 , and again @xmath621 it follows inductively that @xmath622 . more precisely : if @xmath623 and @xmath624 are integers such that @xmath625 and @xmath626 , then @xmath627 in particular , if the initial density @xmath319 is such that @xmath628 , then we can take @xmath629 . + next , fix @xmath630 , and choose a sequence @xmath453 according to lemma [ lemma : controlled - convergence][item : controlled - convergence - b ] ) . then @xmath631 and hence @xmath632 for @xmath633 so that the lemma implies @xmath634 . hence , by ( [ eq : norm_w_estimate ] ) , @xmath635 we then find , using ( [ eq : shadow - estimate ] ) and lemma [ lemma : shadow - estimate ] , @xmath636 for @xmath633 and all @xmath637 . now @xmath638 follows as in the stable case . + it remains to prove that the basin of attraction of @xmath349 is @xmath151-open . ( then , by symmetry , the same is true for @xmath358 . ) as @xmath10 is @xmath151-continuous , it suffices to show that this basin contains an open @xmath151-ball centered at @xmath349 . to check the latter condition , first notice that @xmath639 so that there is some @xmath640 such that @xmath641 . as we can assume w.l.o.g . that @xmath642 , we have @xmath643 , and as @xmath10 is @xmath151-continuous , there is some @xmath644 such that @xmath645 whenever @xmath646 . therefore we can continue to argue as in the previous paragraph ( using the present @xmath624 and @xmath629 ) to conclude that @xmath647 . we just proved a bit more than what is claimed in proposition [ prop : summary2 ] : another look at equation reveals that , in the bistable regime , the stable fixed point @xmath349 of @xmath10 is even lyapunov - stable ( and the same is true for @xmath358 ) . indeed , fix @xmath648 , @xmath649 and @xmath650 as in the preceding paragraph . that choice was completely independent of the particular initial densities investigated there , and the same is true of the choice of the constants @xmath217 , @xmath651 and @xmath652 occuring in estimate . now let @xmath220 . choose @xmath653 such that @xmath654 and then @xmath655 . then equation , applied with @xmath556 , shows that for each @xmath656 with @xmath657 and for each @xmath111 holds @xmath658 let @xmath659 denote the _ stable manifold _ of @xmath17 in the space of _ all _ probability densities on @xmath58 . clearly , all symmetric densities @xmath319 ( i.e. those satisfying @xmath660 ) belong to @xmath661 , because symmetric densities have field @xmath662 so that also the parameter @xmath663 , and symmetry is preserved under the operator @xmath664 . + however , @xmath661 is not a big set . in the present section we prove [ prop : basin - boundary ] each density in @xmath665 belongs to the boundaries of the basins of @xmath349 and of @xmath358 . we start by providing more information on the fields @xmath666 of orbits in @xmath665 . recall that for @xmath667 we have @xmath668 @xmath669 . given such a density , we denote by @xmath670 the `` radius '' of the support of @xmath481 , i.e. @xmath671\}$ ] , and let @xmath672 . [ lemma : roland ] in the bistable regime , for each @xmath673 there exists a constant @xmath674 such that @xmath675 in view of the explicit formula , we have @xmath676 and @xmath677 , and therefore see that there is some @xmath678 such that for every @xmath679 and all @xmath680 $ ] , @xmath681 where @xmath682 and @xmath683 , too , is a positive constant which only depends on the function @xmath69 . in addition , elementary calculations based on and show that letting @xmath684 , @xmath685 can be chosen such that , for every @xmath679 and @xmath686 , also @xmath687\text { , and}\\ 0>\tau_{r}(y ) & { \geqslant}\sigma_{r}(y){\geqslant}\left ( \frac{1}{2}+\varepsilon\right ) ( y+r)>- b \varepsilon\text { \quad for } y\in[-\varepsilon ,- r)\text{. } \end{split } \label{eq_estimate_tau_sigma_near_zero}\ ] ] ( recall that @xmath284 and @xmath285 share a zero at @xmath688 . ) finally , note that we can w.l.o.g . take @xmath685 so small that @xmath689 $ ] . ( due to assumption i we have @xmath690 . ) + consider some @xmath691 with @xmath692 . we claim that for @xmath693 @xmath694.\ ] ] denote @xmath695 which by s - shapedness of @xmath69 satisfies @xmath696 . in view of our system s symmetry , we may assume w.l.o.g . that @xmath697 . according to ( [ eq_mean_for_convex_combi ] ) and ( [ eq_p_tilda_acting_on_representing_measures ] ) we have @xmath698 so that , due to ( [ eq_estimate_f_near_zero ] ) and ( [ eq_estimate_tau_sigma_near_zero ] ) , @xmath699}\circ\sigma_{r}(y)}{6}\right ) \cdot \sigma_r(y)\,p_{r}(y)\,d\nu(y)\\ & { \geqslant}\int_{{y}}\left ( \frac{1_{[-\frac23,-r)}(y)}{6 - 6\varepsilon^2}\left ( \frac{1}{2 } + \varepsilon\right)+\frac{1_{[-r,\frac23]}(y)}{6}\left ( \frac{1}{2 } -\kappa\varepsilon\right)\right)\cdot ( y+r ) \,p_{r}(y)\,d\nu(y)\text{. } \end{aligned}\ ] ] combining this with the parallel estimate for @xmath700 , we get @xmath701}\frac{\left ( \frac{1}{2}-\kappa\varepsilon\right ) ( y+r)}{6}\,d\nu(y).\ ] ] continuing , we find that @xmath702}\frac{\frac{1}{2}-\kappa\varepsilon}{6}\cdot y\,d\nu(y)+\frac{\frac{1}{2}-\kappa\varepsilon}{6}\,\cdot r\\ & { \geqslant}\int_{[-\frac{2}{3},0)}\frac{\frac{1}{2}+\varepsilon}{6 - 6\varepsilon^2}\cdot y\,d\nu(y)+\int_{[0,\frac{2}{3}]}\frac{\frac{1}{2}-\kappa\varepsilon}{6}\cdot y\,d\nu(y)+\frac{\frac{1}{2}-\kappa\varepsilon}{6}\,\cdot r\\ & { \geqslant}\int_{[-\frac{2}{3},0)}k\cdot\overline{w}(y)\,d\nu(y ) + \int_{[0,\frac{2}{3}]}k^*\cdot\overline{w}(y)\,d\nu(y ) + \frac{\frac{1}{2 } -\kappa\varepsilon}{6}\,\cdot r\ , \end{aligned}\ ] ] where @xmath703 . as , because of ( [ eq_estimate_f_near_zero ] ) , @xmath704 , so that @xmath705 , we conclude @xmath706 since @xmath707 and @xmath708 whenever @xmath709 . this proves . + now take any @xmath673 . then @xmath710 , and the second alternative of proposition [ prop : summary ] applies , so that @xmath711 and @xmath712 for all @xmath583 larger than some @xmath713 . in particular , @xmath714 for these @xmath583 in view of . applying , for @xmath715 , the estimate to @xmath716 and @xmath717 , we obtain @xmath718 suppose for a contradiction that @xmath719 for some @xmath720 . then @xmath721 , and therefore @xmath722 . we can thus continue inductively to see that @xmath723 which contradicts @xmath710 . therefore @xmath724 for all @xmath720 , and the assertion of our lemma follows . [ lemma : stable - alternative ] in the bistable regime , if @xmath725 and @xmath691 are densities in @xmath726 with @xmath414 , then at most one of @xmath319 and @xmath727 can belong to @xmath661 . suppose that @xmath728 . we are going to show that @xmath729 , i.e. @xmath730 as @xmath495 . assume for a contradiction that also @xmath731 . we denote the parameters obtained from @xmath319 by @xmath732 , and define @xmath733 analogously . then our assumption implies that @xmath734 . in view of , @xmath735 , and one checks immediately that @xmath736 so that there is @xmath737 such that @xmath738 for all @xmath739 . because of the strict monotonicity of @xmath342 ( lemma [ lemma : pre - monotonicity ] ) we have @xmath740 , so that ( replacing @xmath13 and @xmath365 by these iterates ) we can assume w.l.o.g . that @xmath741 . denote @xmath742 so that @xmath743 and ( @xmath744 being non - decreasing ) @xmath745 for @xmath746 . therefore @xmath747 in view of the lower bounds for @xmath748 and @xmath749 , repeated application of the estimate from lemma [ lemma : gap - estimate ] yields @xmath750 observe that the last integral is strictly positive because @xmath414 , cf . . on the other hand , due to proposition [ prop : summary ] there are @xmath751 such that @xmath752 , \ ] ] and as @xmath753 and @xmath754 ( whence also @xmath755 ) , there exists a constant @xmath529 such that @xmath756 for @xmath757 . hence @xmath758 for @xmath757 by lemma [ lemma : roland ] , and as @xmath759 , this contradicts the previous estimate . we can now conclude this section with the suppose that @xmath760 . for @xmath761 let @xmath762 . then @xmath763 , hence @xmath764 by the previous proposition . therefore , due to proposition [ prop : summary ] and monotonicity of @xmath765 , for any @xmath766 , @xmath767 converges to @xmath768 as @xmath495 . on the other hand , @xmath769 , so @xmath319 is in the boundary of the basin of @xmath349 . replacing @xmath770 by @xmath771 yields the corresponding result for the basin of @xmath358 . as @xmath10 is based on a parametrised family of pfos where the _ branches _ of the underlying map ( and not only their weights ) depend on the parameter , it is nowhere differentiable , neither as an operator on @xmath11 nor as an operator on the space @xmath772 of ( much more regular ) functions of bounded variation on @xmath58 . on the other hand , as the branches of the map and their parametric dependence are analytic , one can show that @xmath10 is differentiable as an operator on the space of functions that can be extended holomorphically to some complex neighbourhood of @xmath773 . here we will focus on a more general but slightly weaker differentiability statement . [ lemma : differentiability ] let @xmath774 be a probability density w.r.t . @xmath12 and let @xmath775 have @xmath776 . then @xmath777 where @xmath778 , @xmath779 , and @xmath780 . if we consider @xmath10 as an operator from @xmath772 to @xmath11 , then @xmath10 is even differentiable at each probability density @xmath781 and @xmath782 in order to simplify the notation define a kind of transfer operator @xmath783 by @xmath784 and note that @xmath785 . observing that @xmath786 , we have @xmath787 . define @xmath788 for a function @xmath774 denote by @xmath789 the antiderivative of @xmath319 . then @xmath790 where @xmath791 as @xmath792 , we have @xmath793 together with this yields @xmath794 and @xmath795 with a constant that involves only the first two derivatives of @xmath319 . now let @xmath774 be a probability density , and let @xmath775 be such that @xmath796 . let @xmath797 and @xmath798 . then @xmath799 this implies at once formula for the directional derivative , and as @xmath800 ( @xmath801 ) uniformly for @xmath802 in the unit ball of @xmath772 , also follows at once . [ prop : hyperbolic - fixed ] in the bistable regime , @xmath803 is a hyperbolic fixed point of @xmath804 in the following sense : the derivative of @xmath805 at @xmath803 has a one - dimensional unstable subspace and a codimension @xmath24 stable subspace . let @xmath806 . as @xmath807 and @xmath808=[x]$ ] , it follows from that @xmath809\otimes\phi$ ] . ( here @xmath810 $ ] denotes the function @xmath811 , etc . ) observe now that @xmath812)=\frac1{12}$ ] . then @xmath813=p_0[x]+\frac b{12}\,[x]=(\frac12+\frac b{12})[x]$ ] so that , for @xmath814 , @xmath63 has the unstable eigendirection @xmath815 $ ] with eigenvalue @xmath816 . on the other hand , as @xmath817 , we have @xmath818 , so the constant density @xmath24 is a neutral eigendirection , and finally , for @xmath819 , we have @xmath820 , so @xmath821 . in theorem [ theo : passage - to - infinity ] we proved that , in the bistable regime , each weak accumulation point of the sequence @xmath166@xmath13@xmath167 is of the form @xmath172 for some @xmath173 $ ] , i.e. that the stationary states of the finite - size systems approach a mixture of the stationary states of the infinite - size system . it is natural to expect that actually @xmath175 , meaning that any limit state thus obtained is a mixture of _ stable _ stationary states of @xmath10 . while we could not prove this for the model discussed so far , we now argue that this conjecture can be verified if we add some noise to the systems . at each step of the dynamics we perturb the parameter of the single - site maps by a small amount . to make this idea more precise , let @xmath822 let @xmath823 be i.i.d . symmetric real valued random variables with common distribution @xmath824 and @xmath825 . for @xmath826 and @xmath197 let us define the @xmath184-valued markov process @xmath827 by @xmath828 and @xmath829 assume now that the distribution of @xmath830 has density @xmath831 w.r.t . lebesgue measure on @xmath184 . then routine calculations show that the distribution of @xmath832 has density @xmath833 where @xmath834 is the pfo of the map @xmath835 , @xmath836 . it is straightforward to check that , for sufficiently small @xmath526 , lemmas [ lemma : homeo ] [ lemma : inv - contraction ] from section [ sec : finite - sys - proofs ] carry over to all @xmath837 ( @xmath838 ) with uniform bounds , and that @xmath839 so that the perturbation theorem of @xcite guarantees that the process @xmath827 has a unique stationary probability @xmath13@xmath840 whose density w.r.t . @xmath841 tends , in @xmath842 , to the unique invariant density of @xmath90 as @xmath843 . this convergence is not uniform in @xmath0 , however . nevertheless , folklore arguments show that there is some @xmath844 such that , for all @xmath845 and all @xmath196 the absolutely continuous stationary measure @xmath13@xmath840 is unique so that the symmetry properties of the maps @xmath55 and the random variables @xmath846 guarantee that @xmath13@xmath840 is symmetric in the sense that its density @xmath847 satisfies @xmath848 . on the other hand , for each fixed @xmath211 , all weak limit points of the measures @xmath13@xmath849 as @xmath15 are stationary probabilities for the @xmath850-valued markov process @xmath851 defined by @xmath852 compare the definition of @xmath93 in . the proof is completely analogous to the corresponding one for the unperturbed case ( see lemma [ le : contttil ] and corollary [ coro : contttil ] ) . for @xmath845 the symmetry of the @xmath13@xmath840 carries over to these limit measures @xmath63 in the sense that @xmath853 for each borel measurable set @xmath854 where @xmath855 for all borel subsets @xmath856 . the following proposition then shows that , in the bistable regime and for small @xmath211 and large @xmath0 , the measures @xmath13@xmath840 are weakly close to the mixture @xmath857 of the stable states for @xmath10 ; compare also theorem [ theo : passage - to - infinity ] . suppose @xmath858 so that we are in the bistable regime and recall that the @xmath846 are symmetric random variables . then , for every @xmath220 there is @xmath859 such that for each @xmath860 the stationary distribution @xmath861 of @xmath862 on @xmath863 is supported on the set of measures @xmath864 which have density @xmath865 with representing measures @xmath866 satisfying @xmath867 . let @xmath63 be a stationary distribution of @xmath868 that occurs as a weak limit of the measures @xmath13@xmath840 . so @xmath63 is symmetric . just as in the proof of theorem [ theo : passage - to - infinity ] , where the `` zero noise limit '' , namely the transformation @xmath95 is treated , one argues that @xmath63 is supported by the set of measures @xmath869 , @xmath870 . arguing as in the derivation of one shows that densities in the support of @xmath63 can be approximated in @xmath11 by densities from @xmath726 , and the stationarity of @xmath63 implies that @xmath63 is indeed supported by measures with densities from @xmath726 . therefore the process @xmath871 can be described by the transfer operator @xmath872 of an iterated function system on @xmath306 just as the self - consistent pfo @xmath177 is described by the operator @xmath873 in equation . the only difference is that in this case one first chooses the parameter @xmath302 randomly , @xmath874 and then the branch @xmath298 or @xmath299 with respective probabilities @xmath875 and @xmath876 . let @xmath877 be such that @xmath878 . suppose now that for some realisation of the process @xmath871 the numbers @xmath879 satisfy condition ( @xmath880 ) of lemma [ lemma : crucial ] for all @xmath211 . then it follows , as in the proof of proposition [ prop : summary ] , that @xmath881 and the measures @xmath868 converge weakly to @xmath12 so that also @xmath882 . as @xmath883 for infinitely many @xmath583 almost surely , both limit can not be zero at the same time , and we conclude that almost surely there is some @xmath211 such that ( @xmath884 ) is not satisfied . in particular , there are @xmath885 and @xmath886 such that ( @xmath887 ) is violated for @xmath888 with some positive probability @xmath889 . let @xmath890 with @xmath891 . ( so @xmath831 and @xmath892 are random objects . ) as in we conclude that @xmath893 or @xmath894 in this case . without loss of generality we assume that the latter happens with probability at least @xmath895 . next , as in we may choose @xmath896 so small that @xmath897 . hence , for reasons of continuity , there is @xmath898 such that also @xmath899 if @xmath900 . therefore , in view of the monotonicity of the operator @xmath901 , we can conclude that @xmath902 for all @xmath903 with probability at least @xmath895 . now fix @xmath220 . by lemma [ lemma : controlled - convergence][item : controlled - convergence - b ] there is some ( non - random ) @xmath649 such that @xmath904 for all @xmath905 with probability at least @xmath895 . but then , by continuity reasons again , there is @xmath906 such that @xmath907 for all @xmath903 with probability at least @xmath895 . the claim of the proposition follows now , because @xmath908 is a markov process and because the stationary distribution @xmath63 is symmetric . it suffices to prove the convergence for evaluations of any lipschitz continuous function @xmath909 defined on @xmath58 . let us denote @xmath910 ( resp . @xmath911 ) , and @xmath912 ( resp . @xmath913 ) the discontinuity point of @xmath914 ( resp . @xmath55 ) . recall that @xmath915 ( resp . @xmath916 ) . let us fix @xmath211 . @xmath63 being non - atomic , there exists @xmath220 such that the interval @xmath917 $ ] is of @xmath63-measure smaller that @xmath526 . the weak convergence of @xmath918 to @xmath63 implies that @xmath919 tends to @xmath683 , and that @xmath920 . let us choose @xmath921 such that for all @xmath739 , @xmath922 and @xmath923 . one then has @xmath924 since the application @xmath925 as a single discontinuity point , which is of zero @xmath63-measure , the first term converges to zero . the second one also goes to zero since it measures the dependence of @xmath55 on its parameter away from the discontinuity point ( one can make an explicit computation ) . we start with some observations on the function @xmath926 that are based on symbolic computations and on numerical evaluations . one finds @xmath927 from this numerical evidence ( see figure [ fig : plots-1 ] for a plot ) it is clear that , for @xmath928 $ ] , @xmath929 hence @xmath930 . as @xmath931 @xmath932 follows provided @xmath933 . therefore , assumption is fulfilled , if @xmath934 for @xmath935 , in which case @xmath936 and @xmath937 this can be checked numerically . ( observe that @xmath938 and distinguish the cases @xmath939 and @xmath814 . ) for an illustration see the rightmost plot of @xmath940 in figure [ fig : plots-1 ] . o. bandtlow , o. jenkinson , _ invariant measures for real analytic expanding maps _ , j. london math . * 75 * ( 2007 ) , 343 - 368 . t. chawanya , s. morita , _ on the bifurcation structure of the mean - field fluctuation in the globally coupled tent map systems _ , physica d * 116 * ( 1998 ) , 44 - 70 . g. chichilnisky , _ topology and invertible maps _ , adv . mathematics * 21 * ( 1998 ) , 113 - 123 . d.a . dawson , j. grtner , _ large deviations from the mckean - vlasov limit for weakly interacting diffusions _ , stochastics , * 20 * ( 1987 ) , 247 - 308 . r. de la llave , _ invariant manifolds associaed to invariant subspaces without invariant complements : a graph transform approach , _ mathematical physics electronic journal * 9 * ( 2003 ) , paper 3 . ( http://www.ma.utexas.edu/mpej/vol/9/3.ps ) r. dudley , real analysis and probability , cambridge univ . press ( 2002 ) . ershov , a.b . potapov , _ on mean field fluctuations in globally coupled maps _ , physica d * 86 * ( 1995 ) , 523 - 558 . e. jrvenp , _ an srb - measure for globally coupled analytic expanding circle maps , _ nonlinearity * 10 * ( 1997 ) , 1435 - 1469 . k. kaneko , _ globally coupled chaos violates the law of large numbers but not the central limit theorem _ , phys . letters * 65 * ( 1990 ) , 1391 - 1394 . k. kaneko , _ remarks on the mean field dynamics of networks of chaotic elements _ , physica d * 86 * ( 1995 ) , 158 - 170 . g. keller , _ stochastic stability in some chaotic dynamical systems _ , monatshefte math . * 94 * ( 1982 ) , 313 - 333 . g. keller , _ an ergodic theoretic approach to mean field coupled maps _ , progress in probability , vol . * 46 * ( 2000 ) , 183 - 208 . g. keller , _ mixing for finite systems of coupled tent maps _ . math . * 216 * ( 1997 ) , 315 - 321 . krantz , h.r . parks , a primer of real analytic functions , sec . ed . , birkhuser ( 1992 ) . k. krzyzewski , w. szlenk , _ on invariant measures for expanding differentiable mappings _ , stud . math . * 33 * ( 1969 ) , 83 - 92 . d.h . mayer , _ approach to equilibrium for locally expanding maps in @xmath945 _ , commun . * 95 * ( 1984 ) , 1 - 15 . n. nakagawa , t.s . komatsu , _ dominant collective motion in globally coupled tent maps _ , phys . e * 57 * ( 1998 ) 1570 . n. nakagawa , t.s . komatsu , _ confined chaotic behavior in collective motion for populations of globally coupled chaotic elements _ , phys . rev e * 59 * ( 1999 ) , 1675 - 1682 . c. ptzsche , s. siegmund , _ @xmath946-smoothness of invariant fiber bundles , _ topol . methods nonlinear anal . * 24 * ( 2004 ) , 107 - 145 . ( http://www.math.uni-frankfurt.de/ siegmund / preprints/03hpoesie.ps ) d. ruelle , _ zeta - functions for expanding maps and anosov flows _ , inventiones math . * 34 * ( 1976 ) , 231 - 242 . f. schweiger , _ invariant measures for piecewise fractional linear maps_. j. austral . a * 34 * ( 1983 ) , 55 - 59 . sznitman , alain - sol , _ topics in propagation of chaos . ecole dete de probabilites de saint - flour xix1989 _ , lecture notes in math . , * 1464 * , 165251,springer , berlin , 1991 .

the magnetic field of the milky way galaxy is the only galactic magnetic field that can be studied over @xmath5 5 orders of magnitude , on length scales less than a parcsec to several kiloparsec . the structure of this magnetic field is revealed by different observing techniques that are usually sensitive to either the line - of - sight component of the magnetic field ( faraday rotation , zeeman splitting ) , or the component of the magnetic field perpendicular to the line of sight ( stellar polarization , polarization of radio synchrotron emission ) . zeeman splitting provides a direct in - situ measurement of the strength of the magnetic field , while faraday rotation and stellar polarization are propagation effects that require additional information on the electron density or the extinction column of dust along the line of sight . galactic synchrotron emission at frequencies of a few ghz suffers from strong depolarization from faraday rotation in the galactic disk ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . even at high frequencies ( e.g. 20 ghz wmap ; * ) where faraday rotation is not significant , the summation of polarized emission with different polarization angles over a long line of sight can be a significant source of depolarization . the large - scale structure of the galactic magnetic field has been revealed mostly by faraday rotation of polarized emission from pulsars and distant active galactic nuclei ( agn ) ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * among others ) . the angle change @xmath6 from faraday rotation of polarized emission of a compact radio source with negligible internal faraday rotation is given by @xmath7 where @xmath8 if the observing wavelength , @xmath9 is the electron density , @xmath10 is the magnetic field vector , and @xmath11 is a line of sight segment pointing from the source to the observer so that @xmath12 has a positive sign if there is a net magnetic field component along the line of sight in the direction of the observer . the angle change @xmath6 can be determined from observations of the polarization angle of a radio source as a function of wavelength . the electron density along the line of sight must be determined independently to derive the strength of the magnetic field . for the large - scale magnetic field of the galaxy , a model of the distribution of the electron density @xmath9 along the line of sight has been derived from pulsar dispersion measures @xcite . however , for smaller length scales , an electron density distribution must be proposed as an assumption that is largely unconstrained by current observations , as significant angle changes can occur in an ionized region that can not be detected by its thermal emission @xcite . correlation between fluctuations in electron density and magnetic field strength can bias the rm upward or downward , depending whether there is a positive correlation or an anti - correlation respectively between thermal electron density and field strength @xcite . structure in the galactic magnetic field on smaller scales has been apparent in polarization data of stars within 1 kpc ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , and polarization of diffuse galactic synchrotron emission ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * landecker et al . , in prep . ) . diffuse polarized emission has the advantage that it provides continuous sampling of a foreground faraday screen , but the signal is usually a complicated integral of synchrotron emissivity and faraday rotation over the volume probed by the beam @xcite . equation [ fr - eq ] does not apply in this case , but imaging polarized intensity and polarization angle , and rm synthesis @xcite have revealed structures in diffuse polarized emission ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? rm structure functions ( sfs ) @xcite have been used to obtain a statistical description of rm variance as a function of angular scale . at high galactic latitudes the variance in rm was found to be independent of angular scale @xcite . at lower latitudes , rm variance is often observed to increase with angular scale @xcite , sometimes with a break around an angular scale of @xmath1 above which the variance remains constant or increases at a slower rate @xcite . the break in the rm sf is usually interpreted as an outer scale in the turbulence that gives rise to the rm variance . the smallest scale that can be probed depends on the density of rm measurements , and angular resolution of the radio data @xcite . @xcite and @xcite used depolarization of point sources as a measure of rm variations on the angular scale of extragalactic radio sources . a database of 37,543 rms that covers 80% of the sky @xcite was constructed recently from the nrao vla sky survey ( nvss ; * ? ? ? * ) , using the two frequency bands of the nvss separated by 70 mhz ( center to center ) . the nvss rms are reliable to @xmath13 . higher rms are subject to ambiguities because it was not possible to derive a unique rm using the available constraints and noise in the data . this occurs mainly at low galactic latitude ( @xmath14 ) . @xcite showed the signature of the large - scale magnetic field in the solar neighborhood in these data including a reversal of the sign of rm across the galactic plane consistent with a quadrupole magnetic field geometry . assuming a symmetric electron density , they found that the magnetic field strength south of the galactic plane is approximately two times the magnetic field strength in the north . towards the galactic poles , @xcite found a positive rm in the north and in the south , also stronger in the south . @xcite found a significant mean rm in the region around the south galactic pole ( gpp ) , but not around the north galactic pole ( ngp ) . here we discuss rm structures on angular scales @xmath15 and their relation the local interstellar medium . figures [ ha_rm - fig ] , [ drao_rm - fig ] , and [ rosat_rm - fig ] show the rm database as circles on a background of h@xmath0 intensity @xcite , diffuse polarized emission at 1.4 ghz @xcite , and diffuse x - ray emission @xcite , respectively . figure [ conditional_rm - fig ] shows a different representation of the rm sky towards the anticenter , excluding data with @xmath16 , and showing all circles at the same size . this representation emphasizes rm structure with amplitudes of a few tens @xmath17 , but also boundaries where the sign of rm changes . three large regions previously identified by @xcite as regions a , b , and c are responsible for most of the structure on angular scales @xmath18 . these structures are also represented in other all - sky rm images with resolution @xmath19 @xcite . the denser sampling of the present data reveals structure in more detail , and we adopt somewhat different boundaries for regions a and c of @xcite as discussed below . the boundaries adopted here are listed in table [ sfregions - tab ] , and we will refer to region a@xmath20 and region c@xmath20 instead . region b in @xcite refers to the gum nebula , and will be discussed as such . region a is a large area of negative rm for which @xcite listed the boundaries as @xmath21 , @xmath22 . with the superior sampling of the new data , we see substantial substructure in this region . the most negative rms occur at intermediate latitudes in the region @xmath23 , @xmath24 . distinctly smaller rm amplitudes are found in the area @xmath25 , @xmath26 , except for an area with higher rm amplitudes in the longitude range @xmath27 ( see also figure [ conditional_rm - fig ] ) . the mean rm in the area @xmath28 , @xmath24 is @xmath29 , with a standard deviation of @xmath30 . the mean h@xmath0 intensity in this area is 1.8 r. the mean rm in the area @xmath31 , @xmath26 is @xmath32 with a standard deviation of @xmath33 . the mean background h@xmath0 intensity in this area , avoiding bright hii regions is 5 r. the section of region a with higher rm amplitude is therefore separated from the galactic plane by a region with rm amplitudes a factor @xmath3 lower than those at more negative latitudes , even though the h@xmath0 intensity is a factor @xmath34 higher than at more negative latitudes . @xcite suggested that region a extends across the galactic plane , to @xmath35 , but this appears to be based on only two sources in their figure 1 , and there is no indication of an extension to positive latitudes in their figure 2 . the data of @xcite sampled the intermediate region with low rm amplitudes very sparsely . region a is bordered on the high - longitude side by a line where rm nearly vanishes , and then reverses sign to become positive ( e.g. * ? ? ? * ) . the boundary with near - zero rm runs approximately at an angle of @xmath36 inclined to the galactic equator from @xmath37 to @xmath38 ( figure [ conditional_rm - fig ] ) , consistent with the boundary of radio loop ii @xcite . a spur of bright polarized emission is visible in figure [ drao_rm - fig ] inside the region with negative rm , while strong depolarization is found at the boundary where rm changes sign . the h@xmath0 image shows @xmath19 long filaments with a similar orientation with respect to the galactic equator , approximately coincidental with the line where rm changes sign . region b of @xcite , centered near @xmath39 , @xmath40 , was identified with the gum nebula by these authors . structure in rm amplitude has since been noticed in this area in the form of a magnetized shell @xcite that is also visible in figure [ ha_rm - fig ] . we note also that there is no clear reversal of the sign of rm across the galactic equator in the area of the gum nebula . region c of @xcite is an area with positive rm bounded by @xmath41 , @xmath42 and was noticed for its wider range in rm amplitudes from @xmath43 to @xmath44 . region c is in the same area as the bottom of loop i ( a circle with diameter @xmath45 centered around @xmath46 , @xmath47 berkhuijsen et al . 1971 ) , but @xcite considered this a coincidence . figure [ ha_rm - fig ] shows a distinct area inside region c with consistently high positive rm in the area @xmath48 , @xmath49 . the median rm in this area is @xmath50 , while the median rm of the adjacent areas @xmath51 and @xmath52 , both with @xmath49 is @xmath53 . the excess rm amplitude over the immediate surroundings is therefore @xmath54 . the high - latitude boundary of this region of enhanced positive rm is not sharp . if it extends well north of @xmath55 at the level of @xmath56 , it may be responsible for the rm excess associated with the hercules supercluster by @xcite . figure [ ha_rm - fig ] also shows enhanced rm amplitude associated with some bright hii regions , e.g. at @xmath57 = @xmath58 , @xmath59 , @xmath60 ( @xmath61 oph ) , @xmath62 ( gum ) , and @xmath63 ( @xmath8 ori ) . smaller excesses of rm amplitude are associated with more diffuse , low - surface brightness h@xmath0 emission , e.g. in the orion - eridanus region ( @xmath64 , @xmath65 ) . we also see a small rm excess associated with the shell of the orion - eridanus superbubble @xcite . figure [ rosat_rm - fig ] shows rms on an image of 0.25 kev diffuse x - ray emission from @xcite . at low latitudes there is little emission because of strong absorption . at high latitudes , emission from the more distant halo is observed , while high - latitude clouds absorb the halo emission locally , and appear as dark silhouettes known as x - ray shadows @xcite . region c@xmath20 coincides with a large x - ray shadow that has an hi counterpart at @xmath66 , and that is part of the outer hi shell of radio loop i @xcite . structure in rm on angular scales of tens of degrees becomes visible because of the high sampling density ( @xmath67 rm per square degree ) . the rm density on the sky in our data is not sufficient to recognize rm structures on angular scales less than a few degrees , but structure on small angular scales does affect the variance of rm because it introduces differences from one line of sight to another . this addition to the variance of rm may depend on angular scale , and is conveniently expressed as the second - order sf @xmath68 defined as @xmath69_i^2 , \label{sf - eq}\ ] ] where the sum is over all pairs of sources with a mutual distance @xmath70 on the sky , and @xmath71 is the number of pairs included in the sum . the mean distance between an rm measurement and its closest neighbor in our data is @xmath72 . sfs can probe rm variance on smaller angular scales than this depending on the number of source pairs available to evaluate the variance with reasonable confidence . for our analysis , the smallest angular scale is taken to be the scale for which we have a minimum of 20 source pairs , or @xmath73 , whichever comes first . the largest angular scale we consider for the sfs is @xmath74 , so the rm gradient related to the large - scale magnetic field does not contribute significantly to d. all possible sources of variance contribute to the amplitude of the sf . we may write d as a sum of variance terms that we consider statistically independent @xmath75 with @xmath76 the variance of rm originating inside the source , @xmath77 the variance in rm originating in the intergalactic medium , for different lines of sight separated by angle @xmath78 , @xmath79 the variance of rm originating in the interstellar medium for lines of sight separated by angle @xmath78 , and @xmath80 the variance resulting from measurement errors and noise . the variance @xmath80 should be subtracted from @xmath68 to obtain the sum of the three terms of astrophysical interest ( e.g. * ? ? ? this term was calculated from the errors on the rms contributing to pairs separated by angular distance @xmath78 . each term in the summation of equation [ sf - eq ] contains differences @xmath81 ( @xmath82 indices of data in the rm catalog ) . for each pair of sources included in the summation , the variance added by noise is @xmath83 where @xmath84 and @xmath85 are the standard errors for @xmath86 and @xmath87 respectively . this assumes that the error bars in the rm data base provide a realistic estimate of the measurement errors . we defer this discussion to section [ discussion - sec ] . in principle @xmath80 can depend on @xmath78 , but for the sfs presented here , it is virtually constant . the median error in our complete rm catalog is @xmath88 , so the error term can be a significant source of variance and it affects the slope of the sf on small angular scales . unless stated otherwise , sfs in this paper or derived properties such as amplitude and slope have been corrected for variance associated with measurement errors . figure [ sf_sim - fig ] shows a simulation of the effect of noise on a sf , and the noise - subtracted sf . the simulated data consisted of 1978 points drawn at random from an image with a random field with input sf with slope 0.6 kindly provided by v. uritsky . for figure [ sf_sim - fig]a , measurement errors for each data point were simulated by first drawing a standard deviation from a uniform distribution between 0.4 and 2.0 ( data units ) , then drawing a number from a gaussian distribution with the chosen standard deviation , to be added to the actual data value . the standard deviation itself was recorded as the error bar for each rm . the figure shows the sf as observed ( upper solid curve with dots ) , the noise power @xmath80 evaluated from the error bars of rm ( dashed curve ) , and the sf corrected for noise power ( points with error bars ) , compared with the sf of the same 1978 noiseless data points ( lower solid curve ) . for figure [ sf_sim - fig]b , the same set of original data points was used , but the standard deviations of the rm errors were drawn independently from a uniform distribution between 0.8 to 4.0 data units . this on average doubles the measurement errors compared with the situation shown in figure [ sf_sim - fig]a , and quadruples the level of the noise power in equation [ variance - eq ] . we retrieve the input sf in the simulation shown in figure [ sf_sim - fig ] and other simulations that included sfs of various slopes and noise levels . these simulations confirmed our evaluation of the noise power even though the errors in individual rms varied by a factor 5 . on the smallest angular scales , we see occasional large variation in the data after noise subtraction . this is to be expected , because the contribution of the noise to the total variance is large , but our estimate of the noise power is uncertain because relatively few source pairs with small angular separation exist in the catalog . we can easily identify such points and remove them from consideration . the subtraction of the sf of the noise is a statistical correction that is more accurate for larger samples . the sf of the measurement errors is not necessarily flat , for instance in the case of periodic variations in the noise of a mosaicking survey with widely separated field centers , or in the case of a deep survey of a small area of the sky where the noise increases towards the edge of the survey area ( affecting all source pairs with large angular separation ) . in our analysis we do _ not _ assume that the sf of the noise is flat , or that all rm error are equal . we calculate the sf of the noise for each @xmath78 from the actual error bars of all included rms . however , we found no systematic variation of the sf of the noise with angular separation for the @xcite rm catalog . lrrrrrrrl ngp & @xmath89 & 180 & 75 & 90 & 2.40 ( 0.03 ) & 0.63 ( 0.39 ) & 0.05 ( 0.04 ) & [ sfpoles - fig]a , b + sgp & @xmath89 & 180 & @xmath90 & @xmath91 & 2.67 ( 0.03 ) & 0.99 ( 0.58 ) & 0.02 ( 0.04 ) & [ sfpoles - fig]c , d + orion & @xmath89 & @xmath92 & @xmath93 & @xmath94 & 2.92 ( 0.02 ) & 0.79 ( 0.08 ) & 0.50 ( 0.03 ) & [ sfregions - fig]a ( black ) + anti - orion & @xmath89 & @xmath92 & 10 & 50 & 2.67 ( 0.01 ) & 1.10 ( 0.39 ) & 0.31 ( 0.01 ) & [ sfregions - fig]a ( gray ) + region c@xmath20 & 33 & 68 & 10 & 35 & 3.06 ( 0.01 ) & 0.33 ( 0.20 ) & 0.39 ( 0.02 ) & [ sfregions - fig]b ( black ) + anti - region c@xmath20 & 33 & 68 & @xmath95 & @xmath94 & 3.15 ( 0.03 ) & 0.31 ( 0.48 ) & 0.58 ( 0.05 ) & [ sfregions - fig]b ( gray ) + region a@xmath20 & 80 & 150 & @xmath96 & @xmath97 & 2.81 ( 0.01 ) & 0.82 ( 0.30 ) & 0.51 ( 0.02 ) & [ sfregions - fig]c ( black ) + anti - region a@xmath20 & 80 & 150 & 20 & 40 & 2.68 ( 0.01 ) & 0.74 ( 0.26 ) & 0.38 ( 0.03 ) & [ sfregions - fig]c ( gray ) + fan & 115 & 160 & 5 & 25 & 2.78 ( 0.02 ) & @xmath980.02 ( 0.12 ) & 0.58 ( 0.03 ) & [ sfregions - fig]d ( black ) + anti - fan & 115 & 160 & @xmath99 & @xmath100 & 2.98 ( 0.02 ) & 0.77 ( 0.17 ) & 0.59 ( 0.04 ) & [ sfregions - fig]d ( gray ) + gum & 100 & 100 & 20 & 20 & 3.72 ( 0.04 ) & 1.08 ( 0.41 ) & 1.14 ( 0.09 ) & [ sfgum - fig ] + [ sfregions - tab ] before we derive an all - sky image of sf parameters , we examine sfs for a few areas of interest inspired by results discussed in the previous section . these regions visualize the significance of differences in sf parameters with direction . table [ sfregions - tab ] lists these regions , and sfs are shown in figures [ sfpoles - fig ] , [ sfregions - fig ] , and [ sfgum - fig ] . experience from the literature @xcite and from the present data @xcite shows that the slope of the sf is not always constant , but sometimes changes slope at an angular scale @xmath101 . the precise angular scale at which a break in the power law occurs is not necessarily the same everywhere , but visual inspection of sfs across the sky suggests that _ if _ a break occurs , the sf slope is considerably higher for @xmath102 than for @xmath103 . we therefore fit separate power laws to the sfs at angular scales @xmath104 and @xmath105 : @xmath106 with @xmath107 for @xmath108 and @xmath109 for @xmath110 . table [ sfregions - tab ] lists results of such fits for the sfs shown in figures [ sfpoles - fig ] , [ sfregions - fig ] , and [ sfgum - fig ] . the fits for small angular scales are much less constrained because of the limited number of close pairs , and a relatively larger contribution of @xmath80 to the total variance on small angular scales . we use the fit parameter @xmath111 as a measure of the amplitude of the sf . @xmath111 is the variance of rm for an angular scale of @xmath1 , corrected for variance due to measurement errors . figure [ sfpoles - fig ] shows rm sfs within 15 degrees of the north galactic pole ( ngp ; 1019 sources ) and the south galactic pole ( sgp ; 752 sources ) . parameters for these sfs are listed in table [ sfregions - tab ] . we have fewer rms in the sgp region because of the declination limit of the nvss ( @xmath112 ) , and because of less sky coverage with boths ifs required to derive a rm . the contribution of @xmath80 is indicated by the dashed curves , and subtracted in figure [ sfpoles - fig]b and d. subtracting the noise variance that is approximately independent of angular scale @xmath70 , reduces the amplitude of the sf significantly , and also steepens the slope of the sf on small angular scales . the amplitudes of the sfs at the ngp and the sgp are signifcantly different . the amplitude of the sf in figure [ sfpoles - fig]a on angular scales more than @xmath113 is consistent with the amplitude @xmath114 found by @xcite for this region . these authors did not subtract the @xmath80 term in equation [ variance - eq ] . the sf at the ngp also shows a slight increase in amplitude from angular scales @xmath115 to @xmath116 , but the error of @xmath117 in this region is large ( table [ sfregions - tab ] ) . @xcite reported a slope @xmath118 in the ngp region and @xmath119 in the sgp region . our sf at the sgp appears flatter than at the ngp . a difference in the slope and amplitude of the sfs of the two galactic pole regions suggests a significant galactic foreground in these regions . assuming @xmath77 is negligible , and @xmath76 is independent of @xmath78 for unrelated sources , a rise in the noise - corrected sf or differences in amplitude must be related to differences in @xmath79 . since the amplitude of the sf at the sgp is higher than at the ngp , and the sf at the ngp appears to have a positive slope on smaller angular scales , it follows that both galactic poles have significant foreground rm variations . an upper limit @xmath120 is suggested by the rm variance in the ngp region on angular scales less than @xmath1 . we do not expect @xmath121 to be much less than @xmath122 , because the variation of rm across individual sources at high latitude on angular scales less than @xmath123 @xcite is not much smaller than our upper limit determined at angular scales @xmath124 - @xmath125 from figure [ sfpoles - fig]b , while models of the foreground suggest lower values @xcite . we adopt the value @xmath126 and find @xmath127 at the ngp , and @xmath128 at the sgp . if this additional variance arises in a single faraday screen , the rms rm amplitude of this screen would be @xmath129 , but there is no evidence that the higher variance results from a separate additional faraday screen in the southern hemisphere . @xcite and @xcite reported that the mean rm near near the sgp is larger than near the ngp . the difference in rm variance found here is consistent with the difference in mean rm in the sense that both variance and mean are larger by a factor @xmath3 in the south . figure [ sfregions - fig ] and figure [ sfgum - fig ] show sfs of selected intermediate - latitude areas defined in table [ sfregions - tab ] . the pairs of sfs in each panel of figure [ sfregions - fig ] represent regions with the same longitude , mirrored with respect to the galactic plane . the gum region is shown separately in figure [ sfgum - fig ] because its mirror region is not covered by the nvss , and the amplitude of the sf is much larger than the sfs shown in figure [ sfregions - fig ] . the first impression of figure [ sfregions - fig ] is that the amplitude of the sfs is not symmetric with respect to the galactic plane . the highest sf in each panel of figure [ sfregions - fig ] is in the southern galactic hemisphere , showing the same asymmetry as figure [ sfpoles - fig ] . on average these sfs suggest that the variance of rm in different areas in the southern galactic hemisphere is @xmath130 larger than in the northern galactic hemisphere . @xcite found that the mean rm across the southern galactic hemisphere is consistently higher than in the northern galactic hemisphere . we will show that the asymmetry in rm variance is also global . figure [ sfregions - fig ] and figure [ sfgum - fig ] also show significant differences in the slope of the sfs . the sf of the gum region is among the steepest identified in the data . the orion region has a significantly steeper slope than its counterpart region across the galactic plane , and the same is true for region a@xmath20 . in other areas , such as the fan region , we see a difference in amplitude but no significant difference in slope . figure [ sf2_skydist_ha - fig ] the sky distribution of @xmath111 and @xmath117 derived from a dense grid of sfs , using sources inside a circular aperture with diameter @xmath131 . the slope and amplitude of the sfs for @xmath103 have a similar accuracy as those listed in table [ sfregions - tab ] , because of the similar number of sources involved . the uncertainty in the amplitude is therefore approximately 5% , and the uncertainty in the slope is approximately 0.03 . most of the structure visible in figure [ sf2_skydist_ha - fig ] represents significant differences in amplitude or slope of the rm sf . the asymmetry in sf amplitude between the ngp and sgp shown in figure [ sfpoles - fig ] is also visible in the top panel of figure [ sf2_skydist_ha - fig ] . there are only a few areas in the south where the amplitude drops below @xmath132 , while most of the ngp region has amplitudes less than @xmath132 . towards the galactic plane , rm fluctuations increase . the largest fluctuations ( @xmath133 ) are found in the galactic plane , in the direction @xmath134 . rm fluctuations may be approximately symmetric with respect to @xmath135 , but the declination limit of the nvss removes the fourth galactic quadrant and part of the third galactic quadrant . the amplitude of the sfs in figure [ sf2_skydist_ha - fig]a traces the brightest extended h@xmath0 emission , with significantly higher amplitudes towards the inner galaxy than towards the outer galaxy . this contrast shows the excentric location of the sun in the galaxy , and thus suggests that the low - latitude rm variance on an angular scale of @xmath1 originates from a line of sight distance of several kpc . structure at higher latitudes are probably more local in nature . around longitude @xmath136 , the contours of sf amplitude follow the outline of the gum nebula in the h@xmath0 image . the low - latitude peaks follow bright emission in the galactic plane , except perhaps the local maximum near @xmath137 . the brighter parts of the orion - eridanus region ( @xmath138 , @xmath139 ) are also associated with enhanced rm variance . the outlines of radio loops i - iv @xcite are also shown in figure [ sf2_skydist_ha - fig ] . only loop iii may have have a counterpart in sf amplitude . the slope of the sf ( figure [ sf2_skydist_ha - fig]b ) is highest in the galactic plane , but it does not show the same concentration toward the galactic plane as the amplitude . the largest slopes are @xmath140 in the direction of the @xmath61oph hii region ( @xmath141 , @xmath142 ) , and the gum nebula . at higher northern galactic latitude , we see a systematic trend that the slope of the sf is shallower towards the galactic center than towards the galactic anticenter at the same latitude . the contour of @xmath143 reaches @xmath144 towards @xmath145 , while it is found around @xmath146 towards @xmath147 . it is not clear whether the same behavior also occurs at southern galactic latitudes , because of the unsampled region @xmath148 . the @xmath149 contour is found mostly between @xmath150 and @xmath151 , except for @xmath152 where it is closer to the galactic plane similar to its northern counterpart . the over - all distribution of the sf slope at high latitudes is more symmetric with respect to the galactic plane than the amplitude . there is a suggestion of a steeper slope in the rim of radio loop iii . figure [ sf2_skydist_pi - fig ] shows the amplitude and slope of rm sfs as contours on polarized intensity of diffuse emission at 1.4 ghz from @xcite . the main features in this polarized intensity were described by @xcite . the gradual trend in slope if high - latitude sfs with longitude mimics the boundaries of the high - latitude polarized emission described by @xcite . in the first galactic quadrant ( @xmath153 ) , the drop in polarized intensity at @xmath154 and @xmath155 noted by @xcite corresponds with rm variance of @xmath156 north and south of the galactic plane , although we see extended polarized emission in some areas with rm variance up to @xmath157 . the polarized intensity drops abruptly by a factor @xmath158 at the boundary , while the average percentage polarization is a factor @xmath159 smaller at lower latitudes than for @xmath160 @xcite . some of this difference may result from depolarization at a larger distance . the 23 ghz wmap polarization image @xcite shows the north polar spur polarized at lower latitudes , implying some depolarization occurs for @xmath161 at 1.4 ghz . we take the abrupt decrease of a factor @xmath158 in polarized intensity as a measure for the amount of depolarization by a possible foreground screen . in the second galactic quadrant ( @xmath162 ) , the brightest polarized emission is observed at lower latitudes north of the galactic equator in the fan region . the brightest polarized emission of the fan region is intersected by the @xmath163 rm variance contour . in contrast to the first quadrant we do not see a clear anti - correlation between polarized intensity and rm variance , with the possible exception of radio loop iii . the third galactic quadrant ( @xmath164 ) is only partially sampled by the present data . the region of the gum nebula as traced by the @xmath165 contour appears depolarized . a narrow polarized feature between longitudes @xmath166 , and latitude @xmath167 is an extension of polarized emission in the second quadrant . it does not appear to be correlated with structure in rm variance . rms from polarized extragalactic sources probe the entire line of sight from the source to the observer . most extragalactic sources experience relatively little internal faraday rotation , and intergalactic faraday rotation is believed to be substantially smaller than the intrinsic rm variance of the extragalactic sources . with the exception of some sources with high internal faraday rotation , most of the rm originates from the section of the line of sight inside our own galaxy . rm amplitude also increases significantly towards the galactic plane . although the rm is an integral along the line of sight , a single object can contribute significantly . the gum nebula ( distance @xmath168 pc * ? ? ? * ) contributes significantly to the rm , even though the line of sight towards the gum nebula intersects the inner galaxy , where the electron density is high over a line of sight of several kpc @xcite . the orion star forming region , also at a distance of approximately 400 pc @xcite , is at higher latitude , so its background is a relatively short line of sight through the galaxy . the line of sight through the galaxy is much shorter for high - latitude sources , and structures in the local ism may affect the rm significantly . some high - latitude hii regions were identified as such , consistent with earlier results that showed that individual hii regions can have a significant impact on rm @xcite . we have found evidence that region c@xmath20 is associated with a nearby cloud revealed by its absorption of soft x - rays from the galactic halo that is at least partially neutral because it is visible in hi emission . this associates region c@xmath20 with the outer hi shell of loop i. the enhanced rm amplitude in region c@xmath20 spans only @xmath169 in galactic longitude , with its central longitude around @xmath170 ( see also * ? ? ? * ) . towards the galactic anticenter , where we expect almost no contribution from the large - scale galactic magnetic field , we find curved filamentary structures in rm with amplitudes of a few tens @xmath17 , positive and negative ( figure [ conditional_rm - fig ] ) . from equation [ fr - eq ] , we obtain an order - of - magnitude estimate for the rm imposed by a structure along the line of sight with electron density @xmath171 , internal regular magnetic field @xmath172 , and line - of - sight extent @xmath173 : @xmath174 the rm amplitudes and morphology of these structures resemble one or a superposition of several bubble walls with size of the order of @xmath5100 pc ( with a large uncertainty ) . one of these is the orion - eridanus superbubble ( @xmath175 , @xmath176 @xcite , visible in emission in figure [ ha_rm - fig ] , and in absorption in figure [ rosat_rm - fig ] . these filaments contribute to the amplitude of rm sfs on large angular scales , increasing the sf slope @xmath117 . the overarching picture that emerges is one of rm structure in the local ( @xmath177 few kpc ) ism on large scales is mostly associated with bubble walls , hii regions , and local clouds , region @xmath178 ( table [ sfregions - tab ] ) being a possible exception . if this is representative for an average few kpc line of sight through the galaxy , the rm variance added by structures on scales of a few hundred pc should contribute significantly to the total variance in rm at low galactic latitudes . if bubble walls contribute significantly , models of rm variance should include a correlation between field strength and density on scales of a few hundred pc . @xcite derived sfs at the ngp and sgp caps ( latitude @xmath179 ) based on 354 and 319 rms respectively . @xcite find almost identical sf amplitudes at both galactic poles , an apparent contradiction of the result in figure [ sfpoles - fig ] . these authors also presented a comparison of their rms with our rm catalog , and foud a small correlation coefficient between the two samples ( 0.39 towards the ngp , and 0.36 towards the sgp ) . @xcite suggested that the differences between the samples are related to multiple rm components in some sources that can not be resolved in the two - frequency data of @xcite , but they did not provide specific examples where this would be the case . while it is true that the two - frequency data in @xcite can not resolve non - linearity in the relation between polarization angle and @xmath180 , rotation measure synthesis with limited @xmath180 coverage , as in the sample of @xcite has its limitations it was recently pointed out by l. rudnick et al . ( unpublished ) that rm synthesis can provide erroneous solutions in a simple system with just two discrete rm components . in order to gain a better understanding of the difference between sfs derived by @xcite and those derived here , we must first compare the rm samples in more detail . the median error for sources common to both rm catalogs is @xmath181 in @xcite , @xmath122 in the ngp region of @xcite , and @xmath182 in the sgp region of @xcite . figure [ rm_rm_comp - fig ] shows scatter plots of rms matched between the two samples , subdivided into rms with error @xmath183 ( panels a and c ) , and all other rms ( panels b and d ) . when considering only rms with small errors , we see a correlation between the two datasets at both galactic poles ( figure [ rm_rm_comp - fig ] a and c ) . the larger scatter in the sgp area found by @xcite is related to rms with larger error bars , in either @xcite , or @xcite , or both . we do _ not _ calculate a correlation coefficient , because the correlation coefficient between independent rm surveys of low - rm sources _ must _ be small . unresolved multiple rm components may affect both surveys in a similar way , and thus introduce correlated systematic errors without increasing the scatter in figure [ rm_rm_comp - fig ] , but still raise the amplitude of sfs derived from both data sets . figure [ rm_rm_comp - fig ] a and c both show significant rm signal . the scatter in these relations provides an estimate of the variance introduced by measurement errors . we examine the subsets with small errors shown in figure [ rm_rm_comp - fig ] a and c , as well as the complete sample . for the subsets with small errors , the total variance is @xmath184 in the north and @xmath185 in the south . the error bars ( added in quadrature for each source ) account for a total variance of @xmath186 in figure [ rm_rm_comp - fig]a and @xmath187 in figure [ rm_rm_comp - fig]b . for the complete sample , the total variance is @xmath188 in the north and @xmath189 in the south , while the error bars account for @xmath190 and @xmath184 respectively . the stated errors underestimate the true variance in the correlation of rms common to both samples , but more so if the stated errors are small . the rm errors in @xcite were estimated by error propagation of the noise in the stokes @xmath191 and @xmath192 images , while the errors in @xcite were estimated as half the width of the rm spread function divided by the signal to noise ratio in polarization . the error estimates in both surveys converge to zero as the signal to noise ratio in polarization increases without bound . perhaps the true errors do not converge as rapidly as assumed . before we consider the effect of this underestimation of errors on the amplitude of sf at the galactic poles , we first consider another aspect of the error statistics of the data . figure [ rm_error - fig ] shows the stated errors as a function of rm amplitude for the ngp and sgp regions of the @xcite sample . black dots represent rms used in the construction of sfs , and gray crosses respresent rejected rms . while the median error in figure [ rm_error - fig]a is larger , figure [ rm_error - fig]b contains data with large rm errors that contribute to the increased scatter in figure [ rm_rm_comp - fig]d . rms with amplitude more than @xmath193 in figure [ rm_error - fig]b are much more likely to have a large error , introducing a correlation between rm error and rm amplitude : the sgp data of @xcite are heteroskedastic . the highest rm amplitudes ( a few percent of the total sample ) are rejected in both regions , but the distribution of rejected rms is more skewed toward high rm amplitude in the south ( figure [ rm_error - fig]b ) . some rms in @xcite were rejected in order to eliminate agns with strong internal faraday rotation , and to minimize the effect of known galactic and extragalactic structures . this approach is preferred if one wishes to determine the intrinsic rm variance of agn - powered radio sources , or investigate small - scale turbulence at high galactic latitude . this paper deals with the complete rm foregound . figure [ taylor_error - fig ] shows the rm errors in relation to rm amplitude for our data in the ngp and sgp regions . the larger rm errors in our data are compensated by the @xmath158 times larger sample size , especially when probing fluctuations on small angular scales . what does this all mean for the sf amplitudes at the galactic poles ? figure [ sf_poles_comp - fig ] shows rm sfs at the ngp and sgp . the gray sfs were made with the censored data of @xcite ( 354 rms at the ngp and 319 rms at the sgp ) . we find a smaller amplitude than @xcite , because these authors underestimated the noise power by approximately a factor 2 . the problem was identified and the authors confirmed they recover the sf shown in figure [ sf_poles_comp - fig ] ( a. mao , private communication ) . the black points in figure [ sf_poles_comp - fig ] show sfs made with the complete sample of @xcite . the origin of the large scatter for @xmath194 in figure [ sf_poles_comp - fig]a was discussed in section [ strucfunc - sec ] ( figure [ sf_sim - fig ] ) . using the complete sample of @xcite , we confirm that the sf in the sgp region is a factor @xmath3 higher than in the ngp region , but both sfs in figure [ sf_poles_comp - fig ] have a somewhat smaller amplitude than those in figure [ sfpoles - fig ] . both data sets suggest a small upward slope of the sf in the ngp region . in table [ sfregions - tab ] we list @xmath195 , while @xcite found @xmath196 . for the sgp region we find @xmath197 , while @xcite found @xmath198 . if we _ assume _ that the noise power in our sfs is really a factor 1.5 higher than that calculated from the rm errors , we would retrieve the black curves in figure [ sf_poles_comp - fig ] . this would be equivalent with the rm errors being underestimated by a factor 1.22 in @xcite , and account for the variance in the comparison of rms from the two surveys . while this appears a plausible explanation of the difference , the actual comparison may be more complicated . this analysis assumes that figure [ rm_rm_comp - fig ] reveals all rm errors , while it does not reveal possible systematic errors common to both surveys that may increase sf amplitude . another question is the nature of the sources in @xcite for which no reliable rm could be found ( 6% of their data in the ngp region and 28% in the sgp region . it is encouraging that we find consistency between sfs from independent data sets , after accounting for all noise power traced by direct comparison of rms common to both surveys . this consistency requires just a @xmath199 increase in the errors quoted by @xcite ( in a low - rm environment ) , corresponding with only a few @xmath17 . it suggests that either the effect of multiple rm components on the data of @xcite is quite small , or the data from @xcite are affected in the same way for most sources . however , such a systematic error should still increase the amplitude of the sfs , assuming it is uncorrelated between sources . from the total amplitude of the sfs at high latitude , we estimate that any such systematic error should be smaller than @xmath56 . our analysis shows that both rm surveys point indicate a larger rm variance in the direction of the sgp than in the direction of the ngp . our sfs at intermediate galactic latitude show the same trend , and these are less sensitive to the amplitude of the noise power because it contributes much less to the total variance . both @xcite and @xcite reported a higher mean rm in the south than in the north . @xcite found this to be the case in the entire southern galactic hemisphere . variance in rm originates from fluctuations in electron density and magnetic field on a range of scales in the ism , including hiiregions , bubbles , turbulence , and possibly also molecular clouds with a small ionized fraction but a strong magnetic field . the variance in rm should be approximately proportional to the length of the line of sight through the galaxy @xcite . figure [ sf2_skydist_ha - fig ] shows that the amplitude of sfs on an angular scale @xmath200 follows the distribution of h@xmath0 intensity in the galactic plane . the highest amplitudes at low galactic latitude are found in the direction of the local arm ( @xmath201 ) and other regions along the galactic plane . the peak in rm variance at @xmath202 ) occurs in a region of the galactic plane with strong extinction at visible wavelengths . the enhance variance in the gum region is clearly extented in figure [ sf2_skydist_ha - fig ] with morphology similar to h@xmath0 intensity . the low - latitude rm variance is smaller in the direction of the galactic anti - center , where the line of sight through the galaxy is shorter . these observations are consistent with the expectation that variance in rm increases with the length of the line of sight through the galaxy @xcite . higher electron density and stronger turbulence related to more intense star formation in the inner galaxy probably also contribute to the high rm variance for lines of sight through the inner galaxy . the outer scale in rm variance associated with interarm regions , but not with spiral arms reported by @xcite and @xcite indicates that besides line - of - sight distance , the local conditions are also important . the difference in rm variance between the galactic northern and southern hemispheres ( figure [ sf2_skydist_ha - fig ] and table [ sfregions - tab ] ) is particularly interesting . assuming the intrinsic rm variance of extragalactic sources to be @xmath203 ( equation [ variance - eq ] ) , the rm variance introduced by the galactic foreground is 202 @xmath204 at the ngp , and 419 @xmath204 at the sgp ( 120 @xmath204 and 307 @xmath204 respectively using sfs from figure [ sf_poles_comp - fig ] ) . the rm variance near the sgp originating from the galactic foreground is therefore a factor 2 larger than the galactic rm variance near the ngp . the difference in variance is equivalent with an additional source of rm variation with amplitude @xmath205 in the southern galactic hemisphere . the sun is located 20 pc from the galactic mid - plane in the direction of the ngp . this distance is too small compared with the scale height of the ionized gas to make a significant difference in rm variance . it is more likely that we see a local asymmetry of density or magnetic field , or both with respect to the galactic plane . @xcite also found a factor @xmath3 difference in the mean rm of the ngp and the sgp , while @xcite found a significant positive rm at the sgp but not at the ngp . the slope of the sf is higher near the galactic plane with most values in the range from 0.4 to 0.8 and excesses up to 1.3 in the direction of two nearby hii regions ( @xmath61 oph and the gum nebula ) . the slope of our sfs ( table [ sfregions - tab ] ) is generally smaller than the 5/3 slope expected for kolmogorov turbulence @xcite , to the smallest angular scales probed by our data ( @xmath206 ) . @xcite found rm sf slopes @xmath207 in the nearly face - on galaxy ngc 6946 on scales up to 6 kpc , which is in the range of values of @xmath117 in table [ sfregions - tab ] . if a change in slope occurs in the range of angular scales probed by the current data ( @xmath208 ) , the slope on larger angular scales is usually smaller . a change in sf slope is sometimes interpreted as a physical scale in the turbulence probed by rm variance , that could be an outer scale length if the sf becomes flat on large scales . an implicit assumption is that the rm variance traces stochastic variations of electron density and magnetic field related to turbulence in the magneto - ionized ism . if deterministic structure , e.g. an hii region that may or may not be detectable in current h@xmath0 surveys , or a meso - scale structure in the magnetic field , contributes significantly to the variation of rm with position over the area for which the sf is evaluated , the shape of the sf may reflect the angular scale of the ionized structure , not the turbulence along the line of sight . the present data suggest that this may be true in some directions . the steepest sfs at low latitude are associated with bright extended hii regions in figure [ sf2_skydist_ha - fig ] . figure [ conditional_rm - fig ] also shows rm structure on angular scales @xmath209 with amplitude of a few tens of @xmath17 , sufficient to affect the slope of the sf between @xmath210 and @xmath74 . sampling with a denser rm grid can reveal such structures , as in the case of the rm anomaly in cygnus @xcite . future wide - area rm surveys with the arecibo radio telescope ( galgacts ) , the australian ska pathfinder askap ( possum ) , and with the square kilometre array will provide denser sampling that may reveal underlying structure that affects the slope of our sfs , while not recognizable in the present data . model sfs by @xcite and @xcite that assume a turbulent kolmogorov power spectrum for density and magnetic field , predict a flat sf at low latitude on angular scales @xmath211 as the number of turbulent cells in the line of sight increases . these models explore angular scales much smaller than those probed here . we see a general increase in sf slope towards the galactic plane , contrary to the expectation of a smaller slope expected by @xcite . as noted in figure [ sf2_skydist_ha - fig ] , the lack of a strong longitude dependence of sf slope , and its wider distribution with galactic latitude suggest that the slope of sfs on larger angular scales is affected by the local interstellar medium . this suggests that local structure not represented in the models of @xcite may be responsible for the steep slope of the sfs at low latitude . rms s of extragalactic sources are affected by faraday rotation along the entire line of sight from the source to the observer , but mostly in the galaxy . depolarization of diffuse emission extends to @xmath212 . at this latitude , the line of sight through the galactic ism is approximately twice as long as the line of sight towards the galactic poles . adopting a scale height of the warm ionized medium @xmath213 kpc @xcite , most of the faraday rotation in a line of sight at @xmath214 should arise within a distance @xmath215 kpc . this is still more than an order of magnitude larger than the distance to the north polar spur , for which we adopt the distance to the scorpio - centaurus ob association ( 170pc ; * ? ? ? * ) , although the near side may be closer . assuming an exponential density profile , and @xmath216 , @xmath217 independent of density , approximately half of the faraday rotation and rm variance occurs in front of the north polar spur . large - scale inhomogeneity of the ism introduces significant uncertainty in this estimate . a foreground screen with rm variance @xmath218 on scales much smaller than the @xmath219 beam size of the drao 26-m telescope depolarizes emission observed at wavelength @xmath8 by a factor @xmath220 @xcite . the sudden drop of a factor @xmath158 in polarized intensity at @xmath221 observed at @xmath222 requires @xmath223 , where the error represents a factor 2 uncertainty in the depolarization factor . the observed total rm variance is @xmath224 . around the depolarization edge , the sf slope is between 0.3 and 0.4 ( figure [ sf2_skydist_pi - fig ] ) . assuming a typical slope 0.35 around @xmath221 , we find the observed rm variance on the scale of the beam of the drao 26-m telescope @xmath225 . subtracting the variance related to measurement errors @xmath226 and the intrinsic variance of background sources @xmath227 ( equation [ variance - eq ] ) , the rm variance attributed to the galactic ism along the line of sight is @xmath228 . approximately half of this variance is expected to arise behind the north polar spur , and would not depolarize emission from this region . the observed variance is a factor @xmath158 lower than what is required to depolarize the diffuse emission , but in view of the estimated errors , the difference is not very significant . within the errors , it is possibile that differential faraday rotation along the line of sight contributes to depolarization of the diffuse emission . this paper presents an analysis of structure in the magneto - ionized interstellar medium probed by 37,543 rms derived from the nvss @xcite . rm sfs are presented as a function of direction for 80% of the sky . we find that : \1 . large - scale structures visible in rm amplitude and sign appear related to structures in the local ism , such as radio loops i and ii , the gum nebula and some other hii regions . we associate region c@xmath20 ( table [ sfregions - tab ] ) with a soft x - ray shadow also visible in hi at @xmath229 . sfs near the galactic poles indicate the presence of a faraday screen at both galactic poles . the variance in rm from foregound ism is @xmath127 and the ngp , and @xmath128 at the sgp ( 120 @xmath204 and 307 @xmath204 respectively when using figure [ sf_poles_comp - fig ] ) . this difference is attributed to a local asymmetry of electron density and/or properties of the magnetic field with respect to the galactic plane . the distribution of rm variance over the sky shows a systematic trend with galactic longitude such that lines of sight through the inner galaxy have higher rm variance than lines of sight through the outer galaxy . this suggests that most or all of the line of sight through the galaxy contributes to rm variance . the slope of the sf does not show the same trends , suggesting its origin is related to local structures that dominate rm variance on larger angular scales . radio loop iii and the orion - eridanus bubble appear to enhance rm variance between latitudes @xmath230 and @xmath231 , and a higher sf slope is found in these areas . several other curved filamentary structures are found in direction of the galactic anti - center ( figure [ conditional_rm - fig ] ) . the steepest sfs ( @xmath232 ) are found in the direction of the gum nebula and the @xmath61 oph hiiregions , but the steep slope may be a result of a localized enhancement of the mean rm by these two hii regions . rm variance traced by compact extragalactic radio sources can account for depolarization of diffuse galactic emission given the uncertainties , but the over - all variance is on the low side allowed by the errors . differential faraday rotation along the line of sight may increase depolarization as well . jms thanks dr . vadim uritsky for providing a set of images with different sf parameters as part of a blind test of the sf analysis . this research was supported by a discovery grant of the natural sciences and engineering research council of canada to jeroen stil . de bruyn , a. g. , bernardi , g. & the lofar collaboration 2009 , in the low - frequency radio universe asp conference series , vol . 407 , proceedings of the conference held 8 - 12 december 2008 , at national centre for radio astrophysics ( ncra ) , tifr , pune , india . edited by d. j. saikia , d. a. green , y. gupta , and t. venturi . san francisco : astronomical society of the pacific , 3 haverkorn , m. 2007 , in sins - small ionized and neutral structures in the diffuse interstellar medium asp conference series , vol . 365 , proceedings of the conference held 21 - 24 may , 2006 at the national radio astronomy observatory in socorro , new mexico , usa . edited by m. haverkorn and w. m. goss . san francisco : astronomical society of the pacific , 242 johnston - hollitt , m. , hollitt , c. p. , & ekers , r. d. 2004 , in the magnetized interstellar medium , proceedings of the conference , held in antalya , turkey , september 8 - 12 , 2003 , eds : b. uyaniker , w. reich , and r. wielebinski , copernicus gmbh , katlenburg - lindau . , 13 sun , x. h. , & han , j. l. 2004 , in the magnetized interstellar medium , proceedings of the conference , held in antalya , turkey , september 8 - 12 , 2003 , eds : b. uyaniker , w. reich , and r. wielebinski , copernicus gmbh , katlenburg - lindau . , 25 taylor , a. r. , gibson , s. j. , peracaula , m. , martin , p. g. , landecker , t. l. , brunt , c. m. , dewdney , p. e. , dougherty , s. m. , gray , a. d. , higgs , l. a. , kerton , c. r. , knee , l. b. g. , kothes , r. , purton , c. r. , uyaniker , b. , wallace , b. j. , willis , a. g. , & durand , d. 2003 , , 125 , 3145

the question of the origin of stellar clusters is an old one . as early as 1785 , william herschel first considered this problem in the pages of these _ transactions _ when he speculated on the origin of the clusters messier 80 and messier 4 in ophiuchus . more than two centuries later , despite profound advances in astronomical science , we find that the question of the physical origin of stellar clusters remains largely a mystery . this is , in part , due to the fact that cluster formation is a complex physical process that is intimately linked to the process of star formation , for which there is as yet no complete theory . the development of a theoretical understanding of star and cluster formation therefore depends on first acquiring detailed empirical knowledge of these phenomena . this can be a very difficult task . consider , for example , the globular clusters , the most massive stellar clusters in our own galaxy , the milky way . these stellar systems are more than 12 billion years old and are no longer formed in the milky way . consequently , direct empirical study of their formation process is not possible . the situation is considerably better for galactic open clusters . these typically span ages from 1 myr to 1 gyr and thus must be continually forming in the milky way , making direct observational study of their formation process possible , at least in principle . however , such studies have been seriously hindered by the fact that open clusters are born in molecular clouds and , during their formation and early evolution , are completely embedded in molecular gas and dust . they are thus obscured from view at optical wavelengths , where the traditional astronomical techniques are most effective . fortunately , molecular clouds are considerably less opaque at infrared wavelengths and the development and deployment of infrared imaging cameras on optical- and infrared - optimized telescopes during the past two decades has provided astronomers with the ability to detect , survey and systematically study the extremely young embedded stellar clusters within nearby molecular clouds . such studies indicated that embedded clusters are quite numerous and that they account for a significant fraction , if not the majority , of all star formation presently taking place in the galaxy . similarly , as discussed in accompanying articles by de grijs ( 2010 ) and larsen ( 2010 ) , young clusters also appear to account for a significant fraction of star formation in other galaxies , such as in the very active and luminous starburst galaxies and in the vigorous star - formation episodes which accompany galaxy mergers and close interactions . in our galaxy , the embedded - cluster phase lasts only 24 myr and the vast majority of embedded clusters which form in molecular clouds dissolve within 10 myr or less of their birth . high cluster infant mortality has also been inferred for clusters in other galaxies ( see , e.g. , de grijs 2010 ; larsen 2010 ) . this early mortality of embedded clusters is likely a result of the low star - formation efficiency that characterizes the massive molecular - cloud cores within which the clusters form . the physical reason for the low formation efficiencies is not well understood and the origin of the massive cores themselves is one of the many mysteries of modern astrophysical research . the embedded clusters are the primary laboratory for research into the question of the physical origin of stellar clusters . at the present time , the answer to this question is shrouded by the dusty veils of the dark molecular clouds in which stars and clusters are born . the ultimate goal of a theoretical understanding of the physical process of cluster formation is far from being realized . the first step towards achieving this goal is to construct a solid empirical foundation upon which a physical theory can eventually be built . this then requires a detailed understanding of the physical nature of embedded clusters and the environments in which they form . in this article , i will review the empirical knowledge that is setting the stage for the eventual development of a theory of cluster formation . i will describe existing knowledge concerning many of the basic physical properties of the embedded - cluster population , including the cluster mass function , cluster birthrate , structural properties , etc . , as well as the nature of the gaseous cloud cores in which they form . i will ( at the end ) briefly speculate on the implications of these results for understanding the origin of stellar clusters . a more theoretical discussion of possible physical mechanisms for forming such clusters is contained in the accompanying article by clarke ( 2010 ) . it is useful to begin any discussion of embedded clusters with the definition of what constitutes such objects . following lada & lada ( 2003 ) , we first define a cluster to be a physically related group of stars that satisfy the following two requirements . first , the group must have a stellar - mass volume density sufficiently large to render it stable against tidal disruption by the galaxy ( i.e. , @xmath1 pc@xmath2 ; bok 1934 ) and by passing interstellar clouds ( i.e. , @xmath3 pc@xmath2 ; spitzer 1958 ) . a group of only 810 , 0.5 stars within a radius of 1 pc would satisfy this first requirement . second , the group should contain a sufficient number of members to ensure that , if in a state of virial equilibrium , its evaporation time ( i.e. , the time it takes for internal , gravitational stellar encounters to eject all of its members ) is greater than the typical lifetime of galactic open clusters of @xmath4 yr . for a cluster in virial equilibrium , this number is about 35 stars . an embedded cluster is then defined as a cluster that is fully or partially embedded in interstellar gas and dust . the above definition of a cluster includes both gravitationally bound and unbound stellar systems . bound stellar clusters are those whose total energy ( kinetic and potential ) is negative . when determining the total energy of a cluster , we take into account contributions from all forms of mass within the boundary of the cluster , including interstellar material . this is crucial in evaluating the dynamical states of embedded clusters , since generally the mass of interstellar material exceeds the stellar mass within the cluster boundaries . because most of their members are often optically obscured by interstellar dust , infrared imaging is generally required to identify and investigate embedded clusters . the existence of a cluster is established observationally by an excess density of stars over the background . consequently , the detectability of a cluster depends on its richness and compactness , the brightness of the cluster members and the density of background objects . figure 1 shows optical and infrared images of the famous orion nebula and its embedded cluster . it clearly illustrates the advantage of infrared imaging for revealing the stellar content of embedded clusters . a number of authors have compiled catalogues of embedded clusters based on infrared observations . ( 2003 ) compiled a list of 34 clusters , which is likely nearly complete for clusters within 1 kpc of the sun . lada & lada ( 2003 ) list 76 embedded clusters with well - determined properties within 2.5 kpc of the sun . they estimate that their list is representative but incomplete and that there are likely a total of about 200 such clusters within this volume of space , most of which are unobserved to date . ( 2003 ) compiled a highly incomplete list containing more than 300 infrared clusters , stellar groups and candidate clusters , most within about 5 kpc of the sun , some fraction of which are likely embedded . even cursory consideration of these studies suggests that embedded clusters are quite numerous in the galactic disc . although the existence of a cluster can be established by the increase of source density over the background , identification of the individual members is considerably more difficult , especially for intrinsically faint members whose numbers are only comparable or significantly lower than those of the background / foreground stellar field population . the size of the cluster s membership can be determined statistically through comparison with the background / foreground field - star surface density over the cluster area . however , determining whether or not a specific star is a cluster member requires independent information . since embedded clusters are very young stellar systems , independent indicators of stellar youth , such as the presence of circumstellar discs , variable emission lines , x - ray emission , etc . , can be employed to ascertain membership of individual stars . for typical embedded - cluster ages of 13 myr , at least half of the members will be surrounded by circumstellar discs , remnants of the star - formation process ( e.g. , haisch _ et al . _ 2001 ; hernandez _ et al . many of these discs will be accretion discs and exhibit strong , variable hydrogen recombination lines at optical and infrared wavelengths ( e.g. , green & lada 1996 ; white _ et al . infrared spectroscopic surveys , especially those done with multi - object spectrographs , can be used to identify accreting objects in a cluster field . however , such observations can be time consuming and even prohibitive for faint members , which can account for a significant portion of the cluster population . circumstellar discs can be detected directly with infrared broad - band photometry as a readily measurable excess above the expected emission from a normal stellar atmosphere . excesses due to circumstellar discs and envelopes are most robustly measured at longer infrared wavelengths ( i.e. , @xmath5 m ) . in recent years , the _ spitzer space telescope _ , operating at wavelengths between 3 and 70 @xmath6 m , has provided the capability to survey and measure infrared excesses and circumstellar - disc emission with unprecedented sensitivity across the complete mass range of stars in embedded - cluster regions ( e.g. , allen _ et al . these observations are providing the first statistically complete accountings of disc populations in individual young clusters ( e.g. , lada _ et al . _ 2006 ; sicilia aguilar _ et al . _ 2006 ) and important new insights into their structural properties ( gutermuth _ et al . _ 2009 ) . although infrared observations are sufficiently sensitive to detect the presence of infrared excess from stars across the entire spectrum of stellar mass , there are typically many cluster members which have no discs or detectable infrared excess , presumably because these stars have passed quickly through the disc phase of early stellar evolution . as many as 50% of all members of a young cluster may have lost their discs and any other accretion signatures at an age of 23 myr . observations made from space with x - ray observatories such as the _ chandra x - ray observatory _ have shown that young stars emit x - rays at levels @xmath7 times that of normal stars , particularly during the first 10 myr of their lives ( feigelson & prebisch 2005 ) . such emission appears to be detectable across almost the entire stellar - mass spectrum , whether or not the stars have discs ( e.g. , telleschi _ et al . _ 2007 ) . in principle , x - ray surveys can be used to identify the discless membership of a young stellar population and thus , in combination with infrared surveys , lead to identification of all members of an embedded cluster . for example , a deep _ chandra observatory _ observation of the trapezium cluster in the orion nebula resulted in the x - ray image shown in figure 2 , in which all cluster members ( with and without discs ) are detected as x - ray sources ( getman _ et al . _ 2005 ) . however , these _ chandra _ observations were very deep , and in general the degree to which x - ray surveys can sample the membership of a cluster depends on the sensitivity and depth of the observations . in typical situations , infrared observations are more sensitive to the faintest members of a cluster population ( e.g. , winston _ et al . _ 2007 ) . nonetheless , to accurately obtain a census of the membership of an embedded cluster requires a varied set of observations and such information is , at the present time , only available for a limited set of clusters . the frequency distribution of embedded - cluster masses is a fundamental property of these systems which any physical theory of cluster formation must explain . lada & lada ( 2003 ) used the data from their cluster catalogue to uniformly derive masses for the known embedded clusters within 2.5 kpc . they then constructed the embedded - cluster mass spectrum ( ecms ) and found it to be given by a power law of the form @xmath8 , where @xmath92.0 for masses in excess of about 50 . for a distribution with @xmath10 , the total mass of clusters in any logarithmic interval ( bin ) is constant . thus , even though there are many more 50 than 1000 clusters , the very rare 1000 clusters contribute the same fraction of the total mass in clusters as the more numerous 50 clusters . this , in turn , implies that if a star in the galaxy is born in a cluster , it is just as likely to form in a 1000 as in a 50 cluster . the ecms very likely represents the primordial mass function of galactic open clusters . in this context , it is interesting to note that a recent attempt to determine the mass function of classical open clusters also produced a power - law relation with an index of @xmath11 ( piskunov _ et al . _ 2008 ; see also discussion by larsen 2010 ) . observations also suggest that the ecms peaks near 50 and then falls off towards lower masses . indeed , in a detailed study of all known stellar groups and clusters within 1 kpc of the sun , porras _ ( 2003 ) found that although small stellar ( @xmath12 30 ) groups greatly outnumber clusters , clusters with 100 or more members account for 80% of the total number of stars contained in both groups and clusters . this is consistent with the notion that most stars that form in the galaxy form or have formed in embedded clusters with 100 or more members , and these are the clusters with masses of 50 or greater . is shown as the dotted histogram . massive embedded clusters from the list of ascenso ( 2008 ) are represented by the dashed trace . ( adapted from ascenso 2008 . ) ] the most massive cluster in the lada & lada ( 2003 ) sample was the orion nebula cluster , weighing in at a little above @xmath13 . more massive systems are known to exist , so it seems natural to assume that the lack of massive clusters in the lada & lada mass function is not caused by a truncation at high masses but is instead a result of the distance restriction of 2.5 kpc coupled with small - number statistics at higher cluster mass . recently , ascenso ( 2009 ) compiled a list of known clusters independent of distance in the galaxy with masses estimated as in excess of 10@xmath14 . the largest of these appears to be westerlund 1 , whose mass of about @xmath15 approaches that of globular clusters ( brandner _ et al . _ 2008 ) . of the nine clusters in her list , six have ages of 3 myr or less and can be considered as part of the embedded , young cluster population . to estimate the numbers of massive , embedded clusters that may inhabit the milky way , ascenso ( 2008 ) scaled the 2.5 kpc ecms to correspond to that expected for the entire area of the galactic disc ( @xmath16 kpc ) . figure 3 shows the scaled ecms , along with the six massive clusters from ascenso s list . the predicted distribution of clusters is also plotted , calculated for a mass spectrum index @xmath17 and normalized to the scaled observed distribution at cluster masses @xmath18 ( ascenso 2008 ) . a significant number of young clusters are predicted for cluster masses in excess of @xmath19 , but only a few such clusters have been identified . understanding the nature and origin of high - mass ( @xmath20 ) clusters is of considerable interest because observations indicate that such large clusters are a major component of the vigorous star - formation activity in merging and starburst galaxies ( e.g. , bastian _ _ 2005 ; see also accompanying reviews by de grijs 2010 ; larsen 2010 ) . are there examples of such clusters being produced in the milky way at the present epoch or is the ecms truncated at high mass ? the question of whether there exists a high - mass cutoff or truncation to the ecms is difficult to answer . there are no known embedded clusters with reliably determined masses in excess of @xmath21 . one would think that such clusters would be easily identified if they existed . however , the predicted numbers are low enough that small - number statistics could be responsible , in part , for the observed lack of such systems in the galaxy today . the present - day masses of classical open clusters offer additional constraints on the possibility of a truncation of the upper end of the primordial ecms . for example , consider the most massive open clusters in our galaxy with ages @xmath22 myr . those systems all have masses @xmath23 ( e.g. , bruch & sanders 1983 ; mermilliod 2000 ) . recently , astronomers identified three red - supergiant - rich clusters with ages between 10 and 20 myr and masses between 2 and @xmath24 ( figer _ et al . _ 2006 ; davies _ et al . _ 2007 ; clark _ et al . if the cluster - formation rate were constant over time , figure 4 suggests that we might expect at least 100 objects with masses @xmath25 and with ages of 300 myr or less . so far , only three such objects are known . therefore , either the ecms is truncated at masses @xmath26 or such clusters did not survive emergence from molecular clouds and dissolved very rapidly into the field . the theory of stellar structure and evolution predicts fairly precisely the mass - dependent evolution of stellar luminosities and temperatures with age . this enables the age of a stellar population to be determined by placing the stars on a hertzsprung russell ( hr ) diagram , which plots the distribution of stellar luminosities versus temperatures . the locations of stars are then compared to the predictions of stellar - evolution theory . this has proved to be a powerful tool for age dating mature clusters whose more massive stars have finished their main hydrogen - burning stage and experience significant luminosity and temperature evolution . however , age dating young , embedded clusters has proved to be considerably more difficult . the reason for this is twofold . first , the majority of stars in an embedded cluster are pre - main - sequence ( pms ) stars , that is , stars that have not yet evolved to the point where nuclear burning commences in their interiors . for these stars , the theoretical trajectories on the hr diagram can be highly uncertain , especially for stars with ages of 1 myr or less . second , it is difficult to measure accurate luminosities and temperatures for pms stars since they are obscured by interstellar dust , exhibit infrared excess emission and are variable . nonetheless , a number of embedded clusters have been age dated in this fashion and typical median ages of stellar members in clusters range between 1 and 3 myr , with the age spread in a given cluster of comparable magnitude to its age ( lada & lada 2003 ) . these results are consistent with the early work of leisawitz _ et al . _ ( 1989 ) , who found that clusters older than about 5 myr exhibit no evidence of interstellar molecular gas within or near their boundaries . thus , the embedded - cluster phase of evolution lasts somewhere between 3 and 5 myr . [ it is interesting to note here that recent analysis using only main - sequence cluster members on the hr diagram by naylor ( 2009 ) yields age estimates 1.5 to 2 times older than derived from the pms stars . ] assuming typical ages of embedded clusters of between 1 and 2 myr , lada & lada ( 2003 ) estimated their birthrate within 2 kpc from the sun at @xmath27 to 4 myr@xmath28 kpc@xmath29 . upon emergence from their dusty wombs , embedded clusters become optically visible open clusters . yet , the embedded - cluster birthrate is an order of magnitude higher than that ( 0.45 myr@xmath28 kpc@xmath29 ) , estimated for classical open clusters within 2 kpc of the sun ( battinelli & capuzzo dolcetta 1991 ) . the large difference in the estimated birthrates is key in understanding the formation and early evolution of galactic clusters . the distribution of ages for a combined sample of embedded and open clusters within 2 kpc is shown in figure 4 . embedded clusters fall into the first bin and classical open clusters have ages that span a range from roughly @xmath30 to @xmath31 yr . the distribution is surprisingly flat . also plotted is the predicted age distribution for a constant rate of cluster formation , assuming all clusters can live to ages of at least a few times @xmath31 yr . there is a large and increasing discrepancy between the expected and observed numbers . this figure suggests a high infant - mortality rate for embedded clusters . most of these clusters do not survive emergence from their parent molecular cloud . roughly 90% dissolve in less than 10 myr . less than about 4% of the initial cluster population reach an age of 100 myr . lada & lada ( 2003 ) suggest that only the most massive embedded clusters ( @xmath32 ) survive to ages of 100 myr or more . it has been understood for some time that passing massive interstellar clouds could tidally disrupt open clusters with mass densities @xmath33 pc@xmath2 within 200 myr ( spitzer 1958 ; gieles _ et al . _ 2006 ) . these tidal interactions with giant molecular clouds ( gmcs ) likely account for the flat age distribution of clusters with ages @xmath34 myr . however , the rapid disintegration of most clusters in the first 10 myr of life requires another mechanism . to understand this process , we need to first understand the conditions under which embedded clusters form . the intimate physical association with interstellar gas and dust is the defining characteristic of embedded clusters . such objects are immersed in dusty , gaseous material to varying degrees . the most deeply embedded clusters are believed to be the youngest and least evolved objects of this class . they are found within massive , dense structures or cores inside gmcs . gmcs are , along with globular clusters , the most massive objects in the milky way . they are complex structures , often defined by long filamentary features . such clouds are primarily composed of molecular hydrogen gas and characterized by a wide range of gas densities ( @xmath35 cm@xmath2 ) . star formation has long been known to take place in the dense ( @xmath37 ) component of molecular clouds ( lada 1992 ) and this component only accounts for between 1 and 10% of the total mass of gmcs observed in the vicinity of the sun . there are also indications that the star - formation rate in local clouds is closely related to their total content of dense gas ( lada _ et al . the latter is found in the form of filaments and also in more compact cores . individual stars can form along dense filaments , usually in compact , dense cores within the filamentary structure . clusters of stars typically form in the most massive and dense cores within a gmc . often , these massive cores are compact structures located at the ends or at an intersection of filaments , as illustrated in figure 5 ( myers 2009 ) . the dense cores that form clusters range in mass from @xmath38 to 2000 and have sizes on the order of 12 pc in diameter ( e.g. , higuchi et al . 2009 ) . the spatial extents of embedded clusters are comparable to the sizes of the dense cores within which they reside . this suggests that the stars may be in approximate virial equilibrium with the gas in the core . the cluster members may thus not be colder dynamically than the surrounding gas . on the other hand , there are some indications that the stars within clusters may have started out dynamically colder than the gas . starless dense molecular cores , the precursors to individual new stars , are characterized by subvirial velocities in a number of clouds ( e.g. , walsh _ et al . _ 2004 ; peretto _ et al . _ 2006 ; lada _ et al . _ 2008 ) and protostellar objects have also been found to be spatially more confined than more evolved young stellar objects ( e.g. , teixeira _ et al . _ 2006 ; muench _ et al . such observations suggest that , at birth , new stars in clusters are dynamically colder than the bulk of the gas surrounding them . such stars are then expected to violently relax and collapse to smaller configurations . in doing so , dynamical interactions can produce segregation of the most massive stars , similar to observations ( allison _ et al . comparison of the stellar content of embedded clusters with the mass of their surrounding cores yields star - formation efficiencies of typically between 10 and 30% . if the cluster stars were indeed initially dynamical colder than the bulk molecular gas , these measured ` final ' efficiencies will be higher than the initial birth efficiencies ( lada _ et al . _ 1984 ; goodwin & bastian 2006 ) . the observed low to modest final star - formation efficiencies are key to understanding the early dynamical evolution and infant mortality of such objects . m with the _ spitzer space telescope _ ( rathborne _ this cloud contains in excess of 1000 within a radius of a parsec and yet exhibits no indication of active star formation . it is possibly a future site of cluster formation . ] -1.0 in the most promising sites to investigate the formation of clusters may be found in a class of objects known as infrared dark clouds ( irdcs ) ( e.g. , rathborne _ et al . they are identified as high - extinction dark clouds when viewed against bright , diffuse mid - infrared ( 10@xmath39 m ) galactic background emission . to be regions of high extinction in the mid - infrared requires them to have enormous column densities ( @xmath40 cm@xmath29 ) and correspondingly high volume densities ( @xmath42 cm@xmath2 ) , just the type of conditions that are ripe for star and cluster formation . the more massive of these clouds tend to be concentrated in the so - called 5 kpc ring ( simon _ et al . _ 2006 ) , a large - scale galactic structure rich in massive molecular clouds and active star formation . millimetre - wave studies show many irdcs to contain cores of sufficient mass ( 1001000 ) to form modest to rich clusters . yet , many of these objects show no evidence of significant star or cluster formation . figure 6 shows one of these irdcs imaged with the _ spitzer space telescope_. this particular cloud contains more than @xmath13 , but exhibits no evidence for star formation . its compact size and large mass suggest that it is gravitationally bound and very likely a site of future cluster formation . ( 2009 ) estimate that a 300 cluster could ultimately emerge from this cloud . other massive irdcs have exhibited evidence for the formation of massive stars within their boundaries . since massive stars are rarely formed in isolation , these clouds may also be the sites of incipient cluster formation ( rathborne _ et al . _ 2006 ) . because these clouds are identified from mid - infrared observations , which can only be obtained from space , their existence has only been known for about a decade . since the massive irdcs tend to be relatively distant , they are challenging to observe and , as a consequence , little is known about their detailed physical properties . unfortunately , too little is known about the detailed conditions in the clusterless irdcs to provide any strong constraints on the problem of cluster formation . however , future observational capabilities provided by such facilities as the _ atacama large millimeter array _ and the _ james webb space telescope _ hold the promise for obtaining a more detailed description of the physical conditions in these objects and for deeper insights into the cluster - formation problem . knowledge of the internal structure of an embedded cluster can provide additional clues relating to its origin and early dynamical history . for the youngest clusters , the spatial distribution of member stars may be closely related to the primordial structure of the cold , dense molecular gas from which the clusters formed . in their review of embedded - cluster properties , lada & lada ( 2003 ) identified two structural types of clusters , those that exhibit extended , irregular surface - density distributions , often with multiple peaks , and those that appear to be relatively more compact and have well - defined centrally condensed surface - density distributions with one dominant peak . more recently , ferreira & lada ( 2009 , personal communication ) used ground - based infrared imaging observations to systematically investigate the structure of approximately 40 previously catalogued embedded clusters . they found that about 60% of these objects were of the centrally condensed variety . the remainder exhibit relatively flat surface - density profiles with more irregular boundaries and more elongated shapes than the centrally condensed clusters . observations with the _ spitzer space telescope _ at infrared wavelengths of 324 @xmath6 m provide an improved census of cluster membership and thus enable better measurements of cluster shapes and sizes . these observations have shown that most embedded clusters are elongated , with a typical aspect ratio of about 2 , very similar to the aspect ratios of the filamentary molecular clouds in which the clusters are born ( allen _ et al . _ 2007 ; gutermuth _ et al . these observations imply that the structures of most embedded clusters probably reflect those of the original molecular clouds from which they formed . they also indicate that understanding the origin of such clusters is intimately tied to understanding the origin and evolution of the dense molecular material that produces them . a potentially significant clue concerning the nature of the physical process of cluster formation is found in analysis of the distributions of spatial separations of the youngest members of some embedded clusters . figure 7 shows the distributions of the nearest - neighbour separations of protostars in an embedded cluster in the ngc 2264 star - forming complex ( teixeira _ et al . this distribution is strongly peaked at an angular separation of 20 arcsec and clearly different from that expected for a random distribution of separations . the characteristic separation corresponds to the jeans length for the density and temperature of the cloud core containing the cluster and suggests that jeans fragmentation may be the physical process that governs the formation of the cluster from the massive cloud core . _ spitzer space telescope _ studies of other clusters have found such evidence for jeans fragmentation to be a more common property of embedded clusters than previously thought ( gutermuth _ et al . because stars form in dense gas , it is the initial distribution of the dense material in a cloud that controls the degree to which a cloud forms an extended population of distributed stars and/or compact clusters . that the mass spectrum of embedded clusters resembles the shape of the mass spectrum of massive cores in gmcs appears to support this surmise ( lada & lada 2003 ) . once dense gas is present , star formation likely proceeds quickly . the timescale is somewhat uncertain but the fact that numerous examples of massive , clusterless cores in irdcs exist probably indicates a time for the onset of cluster formation on the order of a million years . on the basis of the ages of stars in embedded clusters , the time needed to construct the full cluster is likely 24 myr from the onset of significant star formation . the physical mechanism that causes a massive core to fragment and form a cluster of stars is not clear . that the jeans scale appears to describe the separations of the young stellar objects in embedded clusters suggests that a jeans - like fragmentation process is responsible for the formation of a stellar cluster from a massive , dense core . support for this idea also derives from measurements of the initial mass function ( imf ) of stars within individual clusters . the stellar imf has a characteristic mass of about 0.3 , which appears invariant among clusters and even stars in the field . this suggests a mass scale for star formation that is consistent with thermal jeans fragmentation ( larson 1992 ) . the jeans instability describes the situation where gravity overcomes the internal thermal pressure of a gas cloud and leads to both its fragmentation and the collapse of its individual fragments . however , it is not clear how this mechanism would operate in a very massive core where internal turbulence dominates thermal motions . moreover , since the critical conditions of density and temperature which determine the value of the jeans length and mass likely vary among regions , one would expect the characteristic stellar mass of clusters to likewise vary , but it does not . recent observations by the _ chandra _ and _ spitzer space telescopes _ have revived the idea that external triggering could represent another mode of cluster formation in the galaxy . in the cepheus b ( getman _ _ 2009 ) and w5 ( koenig _ et al . _ 2009 ) star - forming complexes , spatiotemporal patterns of star formation have been found in the distribution of recently formed stars . such patterns of sequential star formation have long been regarded as evidence for triggered star formation ( e.g. , elmegreen & lada 1977 ) . triggering produces enhanced compression of existing molecular clouds , raising their mean densities and , in doing so , increasing the fraction of cloud mass above the critical densities necessary for star formation . one way to compress a cloud is through increasing its external pressure . it is possible that conditions in merging and interacting galaxies , and in the nuclear regions of starburst galaxies , produce globally increased external pressures that significantly increase the fraction of dense material in a cloud and lead to the formation of massive dense cores and clusters . lack of such conditions ( i.e. , high external pressures ) may explain why very massive clusters are not presently forming in the milky way . although it is beyond the scope of this particular review to discuss the theory of cluster formation , it is useful to briefly mention the results of recent calculations since some insight into the physical process of cluster formation can be provided by numerical simulations . a more detailed discussion of the theory of cluster formation can be found in the accompanying review by clarke ( 2010 ) . simulations of cluster formation follow the collapse and fragmentation of massive , initially turbulent clouds from some specified initial gaseous state down to the formation of stellar objects . the results of these calculations yield impressive details concerning the clusters they form but are very sensitive to the input physics . in particular , calculations starting with roughly similar initial conditions ( i.e. , @xmath43 turbulent clouds ) can have drastically different outcomes , depending on whether feedback from newly formed stars is included . models without feedback ( e.g. , bate & bonnell 2005 ) produce highly filamentary cloud configurations reminiscent of observed clouds . these models produce extensive ( efficient ) fragmentation and form rich clusters of primarily low - mass stars whose characteristic mass is directly related to the jeans mass and thus varies with initial conditions ( i.e. , density ) . models that include feedback ( e.g. , krumholtz _ et al . _ 2007 ; bate 2009 ) produce more massive cloud cores , smaller clusters , more massive stars and an invariant characteristic stellar mass , all in better agreement with observations of embedded clusters . most importantly perhaps , the models which include feedback from the newly formed stars are also characterized by low star - formation efficiencies , again similar to observations . once embedded clusters are formed , the process of their subsequent dynamical evolution and emergence from molecular clouds is relatively well understood . analytic and numerical calculations have shown that the key factor in determining the evolution of an embedded cluster is the ( final ) star - formation efficiency characterizing its formation ( e.g. , lada _ et al . _ 1984 ; geyer & burkert 2001 ; kroupa & boily 2002 ) . observations indicate that the star - formation efficiencies for embedded clusters range between 10 and 30% ( lada & lada 2003 ; higuchi _ et al . _ 2009 ) with some indications that those with the highest efficiencies are the most evolved objects ( higuchi _ et al . numerical simulations of cluster formation suggest that these low efficiencies are caused by the radiative feedback of the recently formed stars on their surrounding dense gas . in any case , the gravitational glue that binds the stars and gas together in an embedded cluster system is largely provided by the gas . we expect that , as more and more stars form in the cluster , the feedback from the stars due to radiation and energetic jets and outflows , generated in the early stellar - accretion process , becomes destructive and begin to disperse the dust and gas in the core . massive stars are particularly destructive and a single o star can completely disrupt an entire cloud in a very short period of time . because of the low efficiency of conversion of gaseous to stellar mass in these systems , they can easily be completely disrupted if the gas is removed on timescales comparable to the dynamical time of the system . thus , low efficiency coupled with relatively rapid gas removal can explain why 90% of embedded clusters do not survive their emergence from molecular clouds as bound systems ( lada & lada 2003 ) . paradoxically , the clusters that survive emergence from molecular clouds as bound systems are likely the most massive systems at the upper end of the embedded - cluster mass function . this is despite the fact that these systems are the ones most likely to produce destructive o stars . simulations indicate that , as clusters are disrupted by gas - removal processes , this disruption is rarely 100% complete and bound remnants containing a small fraction of the original stellar mass are often left behind ( lada _ et al . _ 1984 ; kroupa & boily 2002 ) . the size of the bound remnant depends on the initial size of the cluster and on how rapidly gas removal occurs . lada & lada ( 2003 ) suggest that the field population of open clusters in the disc of the milky way was likely provided by embedded clusters with masses in excess of 500 . this may also explain the relative lack of truly massive ( i.e. , @xmath20 ) clusters in the milky way with ages in excess of 10 myr . the few that are formed in the milky way do not emerge completely unscathed . only smaller remnants of these giants survive as bound systems beyond 10 myr . more than two centuries after herschel ( 1785 ) first considered the question , we now understand that clusters are formed in cold , dark molecular clouds . stellar clusters begin their lives deeply buried in dense molecular gas and dust as embedded infrared protoclusters . over the past two decades , advances in detector and telescope technology have led to steady advances in our knowledge of embedded protoclusters . these extremely young stellar systems appear to account for a significant fraction of all star formation presently taking place in the milky way and , thus , they are tracers of the current epoch of star formation in the galaxy . although considerable progress has been made towards understanding the basic physical properties of embedded clusters , a physical theory of cluster formation eludes us . in the milky way , the primary mode of cluster formation at the present epoch appears to be one in which relatively compact , massive dense cores in gmcs undergo a process of jeans - like fragmentation that transforms cold , dense interstellar material into stellar form . triggering of clouds by shocks or other mechanisms that lead to an increase in external pressure may represent an additional mode of galactic cluster formation . although its relative contribution to present - day cluster formation in the milky way is unclear , this latter mechanism may be significant in interacting galaxies and nuclear starbursts , and perhaps even in the early history of the milky way itself . it is now clear that to develop a predictive theory of cluster formation will require both a better understanding of the process of star formation and a comprehensive understanding of the physical mechanisms that organize the dense material of a gmc into massive , compact cores which , for a typical gmc in the milky way , occupy only a small fraction ( @xmath44% ) of its volume and account for only a small fraction ( 110% ) of its total mass . i thank joana ascenso , ken janes , lisa townsley , eric feigelson , elizabeth lada and bruno ferreira for informative discussions , joana ascenso for preparing figure 3 and joo alves for providing figure 5 . i am grateful to simon goodwin and richard de grijs for constructive comments that improved this review . allen , l. , megeath , s. t. , gutermuth , r. , myers , p. c. , wolk , s. , adams , f. c. , muzerolle , j. , young , e. & pipher , j. l. 2007 the structure and evolution of young clusters , in _ protostars & planets v _ ( eds b. reipurth , d. jewitt & k. keil ) , pp . 361376 , tucson : university of arizona press . bastian , n. , trancho , g. , konstantopoulos , i. s. & miller , b. 2009 _ gemini _ spectroscopic survey of young star clusters in merging / interacting galaxies : iii . the antennae . _ astrophys . j. _ , * 701 * , 607619 . clark , j. s. , negueruela , i. , davies , b. , larionov , v. m. , ritchie , b. w. , figer , d. f. , messineo , m. , crowther , p. a. & arkharov , a. a. 2009 a third red supergiant rich cluster in the scutum crux arm . _ _ , * 498 * , 109114 . feigelson , e. , townsley , l. , gdel , m. & stassun , k. 2007 x - ray properties of young stars and stellar clusters . in _ protostars & planets v _ ( eds . b. reipurth , d. jewitt & k. keil ) , pp . 313328 , tucson : university of arizona press . figer , d. f. , mackenty , j. w. , massimo , r. , smith , k. , najarro , f. , kudritzki , r. p. & herrero , a. 2006 discovery of an extraordinarily massive cluster of supergiants . _ astrophys . j. _ , * 643 * , 11661179 . getman , k. v. , feigelson , e. d. , luhman , k. l. , sicilia aguilar , a. , wang , j. & garmire , g. p. 2009 protoplanetary disk evolution around the triggered star - forming region cepheus b. _ astrophys . j. _ , * 699 * , 14541472 . gieles , m. , portegies zwart , s. f. , baumgardt , h. , athanassoula , e. , lamers , h. j. g. l. m. , sipior , m. & leenaarts , j. 2006 star cluster disruption by giant molecular clouds . _ mon . not . r. astron . _ , * 371 * , 793804 . gdel , m. , padgett , d. l. & dougados , c. 2007 the taurus cloud : multiwavelength surveys with _ xmm newton _ , spitzer space telescope _ and the _ canada france hawaii telescope_. in _ protostars & planets v _ b. reipurth , d. jewitt & k. keil ) , pp . 329344 , tucson : university of arizona press . gutermuth , r. a. , megeath , s. t. , myers , p. c. , allen , l. e. , pipher , j. l. & fazio , g. g. 2009 a _ spitzer _ survey of young stellar clusters within one kiloparsec of the sun : cluster core extraction and basic structural analysis . _ astrophys . j. suppl _ , * 184 * , 1883 . higuchi , a. e. , kurono , y. , saito , m. & kawabe , r. 2009 a mapping survey of dense clumps associated with embedded clusters : evolutionary stages of cluster - forming clumps . _ astrophys . j. _ , in press ( arxiv:0909.1132 ) . hernndez , j. , hartmann , l. , calvet , n. , jeffries , r. d. , gutermuth , r. , muzerolle , j. & stauffer , j. 2008 a _ view of protoplanetary disks in the @xmath45 velorum cluster . _ astrophys . , * 686 * , 11951208 . koenig , x. p. , allen , l. e. , gutermuth , r. a. , hora , j. l. , brunt , c. m. & muzerolle , j. 2009 clustered and triggered star formation in w5 : observations with _ spitzer_. _ astrophys . j. _ , * 688 * , 11421158 . mermilliod , j .- c . 2000 massive clusters in the milky way and magellanic clouds . in _ massive stellar clusters _ a. lanon & c. boily ) , astron . * 211 * , 4348 , san francisco : astron . pac . piskunov , a. e. , kharchenko , n. v. , schilbach , e. , roser , s. , sholtz , r. d. & zinnecker , h. 2008 the initial luminosity and mass functions of the galactic open clusters . _ , * 487 * , 557566 . peretto , n. , andr , p. & belloche , a. 2006 probing the formation of intermediate - to - high mass stars in protoclusters . a detailed millimeter study of the ngc 2264 clumps . _ , * 445 * , 979998 . porras , a. , christopher , m. , allen , l. , di francesco , j. , megeath , s. t. & myers , p. c. 2003 a catalog of young stellar groups and clusters within 1 kiloparsec of the sun . j. _ , * 126 * , 19161924 . sicilia aguilar , a. , hartmann , l. , calvet , n. , megeath , s. t. , muzerolle , j. , allen , l. , dalessio , p. , mern , b. , stauffer , j. , young , e. & lada , c. 2006 disk evolution in cep ob2 : results from the _ spitzer space telescope_. _ astrophys . j. _ , * 638 * , 897919 . teixeira , p. s. , lada , c. j. , young , e. t. , marengo , m. , muench , a. , muzerolle , j. , siegler , n. , rieke , g. , hartmann , l. , megeath , s. t. & fazio , g. 2006 identifying primordial substructure in ngc 2264 . _ astrophys . j. _ , * 636 * , l45l48 . telleschi , a. , gdel , m. , briggs , k. r. , audard , m. & palla , f. 2007 x - ray emission from t tauri stars and the role of accretion : inferences from the _ xmm newton _ extended survey of the taurus molecular cloud . _ , * 468 * , 425442 . winston , e. , megeath , s. t. , wolk , s. j. , muzerolle , j. , gutermuth , r. , hora , j. l. , allen , l. e. , spitzbart , b. , myers , p. & fazio , g. g. 2007 a combined _ spitzer _ and _ chandra _ survey of young stellar objects in the serpens cloud core . _ astrophys . j. _ , * 669 * , 493518 . white , r. j. , greene , t. p. , doppmann , g. , covey , k. r. & hillenbrand , l. a. 2007 stellar properties of embedded protostars , in _ protostars & planets v _ b. reipurth , d. jewitt & k. keil ) , pp . 117130 , tucson : university of arizona press .

eclipsing binaries have historically contributed a wealth to stellar astrophysics . they have been used to determine distances , compute fundamental stellar parameters , and test stellar evolution models . the _ kepler _ mission @xcite and its unprecedented precise photometry of @xmath0 stars , has allowed us to create a catalog of 2605 eclipsing binaries ( hereafter ebs ) in the _ kepler _ field @xcite . this catalog is rich in interesting objects for individual study and also presents a large sample of ebs for statistical analysis . in studying this sample , we can attempt to determine the occurrence rate of ebs , circumbinary planets , and multiple star systems . some theories for short - period binary star formation call for the presence of a third - body . in these scenarios , the close binary was not created _ in situ _ , but rather at a larger separation as a part of a multiple star system @xcite . tidal friction and kozai cycles between the inner - binary and a companion can cause the inner - orbit to shrink over time @xcite , and result in a hierarchical multiple system @xcite . the spectroscopy and imaging studies by @xcite and @xcite have found 40% of binaries with periods less than 10 days , and 96% with periods less than 3 days , have a wide tertiary companion . the general interpretation of these findings is that the tightest binaries likely became hardened over time through interactions with the tertiary companion , and the system evolves toward an increasingly hierarchical configuration . indeed , the slowpokes study of ultra - wide binaries in the sloan digital sky survey @xcite found that the widest visual pairs with physical separations of 0.011 pc , in fact contain a tight binary @xmath180% of the time @xcite , again confirming the general picture that tight binaries are nearly always accompanied by wide tertiaries and that the tightest binaries are accompanied by the widest tertiaries . discovery and study of these multiple systems gives new insight into the physics of ebs . statistically , we can compare observed rates of multiple systems to theoretical models for short - period binary formation . we can also model each system individually to study the disruptive dynamical effects seen in some cases . @xcite determines ephemerides for the entire _ kepler _ eclipsing binary catalog . if there are no external effects , a linear ephemeris will correctly predict all eclipse times of an eb . by measuring the exact time of each eclipse for a particular binary and comparing it to the calculated time from the linear ephemeris , we can create an etv curve ( ` eclipse timing variations ' ; sometimes also referred to as an o - c diagram ) . any trend in these timing residuals may be the result of one or more physical effects occurring in the system . using transit timings and eclipse timings to find exoplanets is a well - known method @xcite . @xcite , @xcite and @xcite used transit timings to detect and study multiple planetary systems , while _ kepler _ 16 @xcite , 34 , and 35 @xcite were validated , in part , through their eclipse timing variations . the processes that can induce etvs , which are the focus of this paper , include the following : * light time travel effect ( ltte ) : a third - body perturbing the center of mass of the binary system creates a light - time delay along the line of sight which can cause eclipses to appear earlier or later than expected . * non - hierarchical third - body : the presence of a third - body actually changes the period of the binary over time . * mass transfer : mass transfer between the components in the binary changes the period . * gravitational quadrupole coupling ( applegate effect ) : spin - orbit transfer of angular momentum in a close binary due to one of the stars being active produces period changes up to @xmath2 times the binary period @xcite . * apsidal motion : the rotation of the line of apsides causes a change in the time between primary and secondary eclipses even though the period remains unchanged ( requires an eccentric orbit ) . * spurious signals : due to spots and other effects that distort the eb light curve . @xcite previously published a list of 39 candidate third - body _ kepler _ systems using eclipse times and @xcite published a preliminary study on timing variations in 41 _ kepler _ eclipsing binaries . @xcite will provide eclipse times for detached binaries , and this paper provides eclipse times for close binaries . together , these two papers will comprehensively cover all 2605 binaries in the catalog . s essentially uninterrupted observing over a long time baseline presents the opportunity to precisely time the eclipses and detect any underlying signals due to third bodies , apsidal motion , dynamical interaction , etc . due to the large number of ebs in the entire catalog , it is necessary to create an automated method for timing eclipses across the catalog . short period and overcontact systems present a particular challenge due to spot activity and data convolution , due to a relatively long integration time . in this paper we discuss our method for automating eclipse timings for close _ kepler _ ebs in section 2 . eclipse timings are reported for 1279 binaries in the catalog in section 3 . in section 4 , light time travel effect models for the 236 that are flagged as potential third - body candidates are also provided . we discuss our findings in section 5 in the context of binary formation and evolution theory , and summarize our conclusions as well as information for accessing the products of our comprehensive eclipsing timing measurements in section 6 . @xcite will update the _ kepler _ eclipsing binary catalog , raising the count of ebs from 2165 to 2605 . the database is kept up - to - date with future data and revisions at ` http://keplerebs.villanova.edu ` . as changes and updates are made to the catalog , etvs are being recomputed and updated automatically and made available in real - time through the online catalog . @xcite will provide eclipse times for binaries with flat out - of - eclipse regions , covering most of the detached binaries with periods greater than 1 day . there we locally detrend each eclipse and use a piecewise hermite spline template to determine the time of mid - eclipse . this technique performs well on the set of detached systems but is not optimal for overcontact systems , systems with strong reflection effects or tidal distortion , or short - period binaries with only a few points in each eclipse due to _ kepler _ s 30 minute cadence . for this reason , we divide the catalog based on the morphology parameter as described in @xcite . this parameter is a value between 0 and 1 which describes the `` detachedness '' of an eclipsing binary , with 0 being completely detached and 1 being overcontact or ellipsoidal . @xcite report timings for binaries with a morphology parameter less than 0.5 . our method addresses and determines eclipse times for the remainder of the _ kepler _ eclipsing binary catalog . the distribution of the catalog between these two methods is shown in fig . [ etvcat ] , with 1279 binaries in the sample for this paper . we detrend and phase `` sap '' ( simple aperture photometry ) _ kepler _ data through q16 as described in @xcite . the upper envelope of the raw data is fit with a chain of legendre polynomials using a sigma - clipping technique and manually setting the breaks between sections and orders of the polynomials . the data are then divided by this fit , resulting in a flat baseline . these detrended data are then phased on the linear ephemeris as reported in @xcite , and used as input into the etv code , described below . we fit a polynomial chain to the phased light curve data as described in @xcite . this analytic function is a chain of four polynomials that is continuous , but not necessarily differentiable , at knots which were optimized to find the best overall solution . this function does not represent a physical model , but rather analytically describes the mean phased shape of the binary light curve , an example of which can be seen in fig . [ ecl_bounds ] . we then take this analytical representation and , using a combination of heuristic and bisection approaches , determine the horizontal shift required to minimize the @xmath3 ( cost function ) for each individual eclipse as shown in fig . [ bisection ] . in order to minimize the effect due to spots or imperfect detrending , a vertical shift is first determined using linear least squares for each eclipse and is applied before computing cost functions for horizontal shifts . the cost function is initially sampled at 20 evenly - spaced phase shifts between -0.05 and 0.05 phase . the minimum of this sampling is then used as the center of the bisection algorithm to quickly find the local minimum of the cost function . the resulting @xmath4 values are unusually large because the errors on the _ kepler _ data are only formal and do not include any absolute calibration effect @xcite . therefore , for each eclipse , we normalize the entire cost function such that the minimum cost is set to @xmath5 , where @xmath6 is the number of data points used for that eclipse and @xmath7 is the degrees of freedom , which we take to be 1 . this reduced cost function is then used to compute 1-sigma errors on each timing to correspond to the @xmath8 contour . for the shortest binaries in the catalog , however , the long - cadence data result in significant phase - smearing and limits our method to a very minimal number of points per cycle to determine a fit . if there were to be a third - body , the signal would likely be buried in the noise induced by these factors . for this reason , we include as many data points as possible in each eclipse timing . each data point is considered to belong to an eclipse if its phase as determined by the initial linear ephemeris is within bounds . we initially set these bounds to be the mid - point between polyfit knots in the out - of - eclipse region as shown in fig . [ ecl_bounds ] . to improve results for particular objects being studied individually , changing these bounds to use the knots ( instead of the mid - points ) can sometimes lower the systematics in the signal . for any given eclipse , if the region between these bounds is not fully sampled or does not have at least 3 data points , then timings are not computed for that eclipse . eclipse timings are then compared to the values expected from the linear ephemeris as reported by @xcite to compute the residuals and test for the presence of an etv signal . due to a typically small number of points per eclipse , our timings are sensitive to various imperfections in the data processing , affecting the measured eclipse time and potentially introducing noise and/or fictitious signals in the etv signal . instrumental or astrophysical pulsations on top of the binary signal can change the shape of a single eclipse which can mimic a timing variation . the detrending process attempts to remove these additional signals , but is not perfect , struggles at removing signals that happen during eclipse , and can also introduce spurious signals . also , all polyfits in the current version of the catalog use chains of four second - order polynomials , which does not always result in the ideal fit and can leave slight phase - dependent residuals . for the purpose of pipeline processing , we limit ourselves to second order polynomials , but note that , for special cases and in - depth studies , higher precision timings can be obtained by increasing the order of the fit . until all polyfits are updated to a higher order in the future , we will use the second - order fits and manually run etv signals with a higher order for any individual etv signals that warrant further study . in the cases when a binary has a period that is near - commensurate to _ kepler _ s 29.44 minute cadence , the period and cadence may beat , which results in a separate spurious signal . any combination of these effects can cause issues in determining true and precise eclipse times when dealing with only a few data points . fig . [ anti - phase ] demonstrates how the cost function for the phase shift is affected by the vertical discrepancy in the out - of - eclipse region , creating a fictitious signal in which the etvs of the primary and secondary eclipse are in anti - phase . the left of fig . [ anti - phase ] plots four different eclipses , showing that over time the data in the out - of - eclipse region can be higher on either the right or the left . when measuring timings for the primary and secondary separately , the cost function will artificially be minimized by `` pulling '' the analytic function towards the region with lower flux . since this will affect the primary and secondary in the opposite direction , we can mitigate for this effect by also running the fit over the entire phase . this effectively averages out the anti - phase effect in the primary and secondary eclipses , projecting the real etv signal of the entire system . [ anti - phase - beforeafter ] shows two cases where the anti - phase signal was removed , clearly showing whether there is a presence of any underlying etv signal . these signals that show a `` random walk '' nature are discussed by @xcite . unfortunately , since there is no rigorous way to discriminate between true and fictitious anti - phase signals , this process would also hide a physical etv signal such as apsidal motion . since we are dealing with short - period binaries , most of these orbits will be quite circular so we do not expect to be able to detect any systems with apsidal motion anyway . if a binary has short - cadence data available , they are usually limited to a short time baseline . since we are generally looking for long period trends in the etv signal , we measure timings from the full long cadence dataset . we also run timings on any available short cadence data in case some signal can be detected . a few short period binaries in the catalog that appear to be overcontact or ellipsoidal variables could actually be detached systems whose light curves are convolved by _ kepler _ s 30 minute long cadence exposure . since these systems are less prone to timing noise due to spots or mass transfer than true overcontact systems , the phase smearing and limited data points per eclipse are the main issues preventing us from recovering any etv signal . for this reason , short cadence data were requested and obtained via director s discretionary time for 31 short - period detached ebs in the catalog without previous short - cadence observations in the hope of detecting third - body etv signatures which were not visible in the long - cadence data . unfortunately , none seem to exhibit any significant etvs . eclipse times are computed for all available short - cadence data , but due to the longer baseline , long - cadence timings are used for detection and fitting of potential third - body orbits . eclipse timing variations on the individual eclipses and the entire phase have been run for all objects with a morphology parameter greater ( less detached ) than 0.5 in the latest _ kepler _ eclipsing binary catalog . our method requires at least 3 data points per timing , which allows us to get primary and secondary eclipse timings individually for long cadence data of binaries with periods as short as 3 hours and full phase timings for binaries as short as 1.5 hours . we are able to determine timings for any binary with short cadence data , but since short cadence data availability is sparse and generally not for the whole length of the mission , short cadence etvs are determined separately . plots and data for detrended light curves and eclipse times for the entire sample are available as a part of the _ kepler _ eclipsing binary online catalog at ` http://keplerebs.villanova.edu ` . an excerpt of the eclipse times is shown in table [ tableetvs ] . with the third version of the catalog being released shortly , the database will be updated as new data become available and ephemerides are further refined . this etv code is incorporated into the pipeline : as objects are updated or added , their etvs are recomputed and updated in real - time . as etvs are computed , the ephemerides in the catalog are refined by fitting a linear fit through the entire - phase timings and adjusting the values as necessary to get a `` flat '' trend . for any etv with a long - term sinusoidal trend , this could introduce systematics depending on the part of the sine curve observed and used to fit the linear trend . in particular , for very long etv signals ( of the order of 1000 days and more ) , the measured orbital period of the binary will be anomalous because the variation can not be accounted for from available data . all etv measurements were examined by eye for the presence of any interesting signal , discarding any that seem to be spurious based on their individual primary and secondary eclipse timings . we do not expect to see evidence of apsidal motion in many of our targets due to their short periods and , consequently , circular orbits . we also do not expect to be able to detect any signals due to gravitational quadrupole coupling . this mechanism is able to create period changes with amplitudes on the order of @xmath2 times the period of the binary , meaning a maximum of 3.5 seconds for a binary with a period of 4 days , falling well within our noise limits . etv signals that are sinusoidal in nature or show any sign of curvature are flagged and fit for both a third - body signal and a parabolic mass transfer model . for the cases where we only see a sign of curvature and not a full cycle , we could either be seeing a section in a long period third body signal or mass transfer . to determine whether we consider the signal as a candidate third body or mass transfer , we compare the two models using the bayesian information criterion @xcite : @xmath9 where @xmath10 are the data , @xmath11 the model , @xmath12 the number of data points , and @xmath13 is the number of parameters used in the fit . in the case of the eccentric ltte model @xmath14 , for the circular ltte model @xmath15 , and for the mass transfer model @xmath16 . the better fit , as determined by the lower bic value , then determines whether we consider the signal as a candidate third body or mass transfer . thirty - one etv signals were better fit by a parabola than an ltte orbit , and are possibly caused by mass transfer or the applegate effect instead of the presence of a third body @xcite . a selection of these signals are shown in fig . [ etv_mt ] , all kics are listed in table [ tablemt ] , all of which are available on the online catalog . 236 binaries ( of the sample ) were flagged as candidate third bodies . the results of the model fits are reported in table [ tablerapp ] with a selection plotted in fig . [ etv_ltte ] . based on the fitted period , we then divided these third body candidates into three sections . the first group contains third body signals with periods less than 700 days , such that there are at least two full cycles of the signal present in data through q16 . these systems have the highest confidence and are most likely due to the presence of a tertiary component . the second group contains signals with periods between 700 and 1400 days , such that there is at least one full cycle present . the last group contains signals with periods longer than 1400 days . often these detections merely show some sign of curvature in the etv signal and so a full sinusoidal signal can not yet be confirmed . for this reason the fits generally have large errors and many of these may not even be true triple systems , particularly the signals on the closest binaries which are more likely to be due to mass transfer . @xcite presented an initial study of eclipse timings in 41 _ kepler _ binaries . of their entire sample of 41 binaries , 40 are still in the _ kepler _ eb catalog ( kic 4678873 has since been removed from the catalog as a false positive ) , 32 fall under the scope of this paper ( have a morphology less detached than 0.5 ) , and 9 appear in our list of third - body signals . they identified 14 out of their original 41 as candidate third - body systems , with others being identified as likely caused by starspots , pulsations , and apsidal motion . of their 14 candidate third - body systems , all 14 are still in the _ kepler _ eb catalog , 12 fall under the scope of this paper , and 9 appear in our list of third - body signals . those that appear in both lists are noted in table [ tableresults ] . binaries that they list as candidate third - body systems , but we do not , either show significant noise or would be very long period ltte orbits . @xcite recently reported 39 triple - star _ binaries due to etv signatures . of these 39 , 21 fall under the scope of this paper , 19 of which also appear in our list of third - body signals , with the other two determined to be unlikely caused by a third - body due to their very short periods and notable spot activity . the detections that overlap both of these studies are also noted in table [ tablerapp ] . in most cases , the model fits from both studies are consistent . in general , due to our treatment of the full _ kepler _ dataset now available , tertiary parameters should now be more precise and longer period third body signals are now more apparent . any disagreement is likely due to a slightly differ inner - binary ephemeris or the addition of the physical delay in their models ( discussed further below ) . @xcite presents analytic functions for the light time travel effect ( ltte ) component of the etv residual signal . using the same form as @xcite , the timings can be expressed by @xmath17\ ] ] where @xmath18 p_3^{2/3 } \end{aligned}\ ] ] and @xmath19 is a time offset , @xmath20 is the mass of the third body , @xmath21 is the mass of the entire system , and @xmath22 , @xmath23 , @xmath24 , @xmath25 , @xmath26 , and @xmath27 are the period , inclination , eccentricity , argument of periastron , eccentric anomaly , and mean anomaly of the third body orbit , respectively . this expression was then used to fit all etvs flagged as potential third body signals . the period was first estimated using lomb - scargle periodogram and used as input into a series of levenberg - marquardt fits , each using a different starting guess for eccentricity . the fit with the lowest chi - squared was then kept and the errors estimated from the covariance matrix . if the final fit had an eccentricity consistent with 0 , then @xmath24 was set to 0 , @xmath25 to @xmath28 , and the fitting was redone with circular constraints to get appropriate error estimates on the remaining parameters . this gives values and estimated errors for @xmath22 , @xmath24 , and @xmath29 ( table [ tableresults ] ) . a sample of some of these etv signals and their respective fits can be seen in fig . [ etv_ltte ] , with fits for all candidate third body signals available in the online version of the _ kepler _ eb catalog . we can only provide estimate periods for the sample of etv signals with less than one full cycle in the data . even these periods can be significantly biased based on the section of the cycle that is in the observed baseline and should be treated with reservation . these very long period cases are provided separately at the end of table [ tableresults ] @xcite included physical delays in their models of 39 _ kepler _ binaries with possible third - body etvs , sometimes contributing largely to the overall model . this dynamical effect occurs when the presence of a third body changes the period of the inner binary . [ phys_hist ] shows the distribution in their targets and the overlapping targets of the ratio of the amplitude of the physical delay compared to the total amplitude in the etv signal . 21 of their targets overlap with ours , but due to the short - period inner - binary , the physical delay rarely contributes significantly to these model . from their results , it seems that the ltte effect dominates over the physical delay for binaries with periods less than 3 days , which covers the vast majority of our targets . for some of these binaries with ltte signals , tertiary eclipses have also been found that confirm the presence and third body period , and significantly constrain the inclination of the third body . any binary which was identified to have a possible third body due to its etv signal and also has a detected tertiary eclipse is noted as such in table [ tableresults ] . kic 2856960 , for instance , has an inner - binary period of 0.259 days with an etv signal resulting in a ltte fit with a period of @xmath30 days . this period is consistent with the previously determined period for the tertiary events of 204.25 days ( fig . [ 2856960 ] ) . this is also consistent with the ltte period of @xmath31 days reported by @xcite , and the tertiary event period of @xmath32 days by @xcite . in the case of kic 2835289 ( fig . [ 2835289 ] ) , we have only observed one potential tertiary event in q9 . without at least three consecutive events , we can not rigorously confirm that the eclipse is a third - body as opposed to a blended eclipsing binary . however , the eclipse seems to show the eclipse of both stars in the inner - binary and etv signal shows a possible long - term third body orbit suggesting a period of approximately 800 days . if this proves to be a true third body , then _ kepler _ just missed an event before the beginning of the mission and may have observed another event in q17 , which has yet to be processed . kic 6543674 also shows a single tertiary eclipse in q2 . a second tertiary eclipse was missed during a break in the _ kepler _ data , but we were able to observe an additional tertiary event from the ground , giving a third body period of @xmath33 days @xcite . in this case , we do not have a full orbit of the etv signal and the ltte model period is quite uncertain . nine binaries that show third - body etv signals ( kic 3936357 , 4069063 , 5310387 , 6629588 , 7375612 , 8122124 , 8758716 , 10014830 , and 10855535 ) also show constant changes in their eclipse depths ( fig . [ dv ] ) , which could either be caused by a change in inclination or apsidal motion perhaps induced by the third body . we plan to follow these up later with full photodynamical models . it is also possible that some of these etvs could be composed of multiple signals . kic 5310387 , 6144827 , 8145477 , 11612091 , and 11825204 , for example , may have both an ltte and quadratic component or two ltte signals as is shown in the residuals in fig . in general , the stronger signal is fitted and noted . in this study we find a third body rate of in our sample of close binaries , nearly all of which have inner binary periods shorter than 3 days ( fig . [ ptrip ] ) . this is much lower than the third body rate of 96% found by the previous studies mentioned . however , our identification of tertiary companions is certainly a lower limit for several reasons . first , our ability to detect a third body is very sensitive to both inclination and mass of the third body , such that low - mass tertiaries and/or tertiaries whose orbital planes are highly inclined relative to the inner binary orbital plane do not present detectable ltte effects . of our total sample of 1279 binaries , 3 ( ) show an ltte orbit and visible tertiary eclipses . 111 ( ) have ltte orbits with periods shorter than the span of our photometric data but do not show tertiary eclipses , suggesting that the eclipses fell in a gap in the data or the orbits are not well enough aligned to show eclipses . thus there is evidence from these examples that in a few percent of cases we are indeed missing true third bodies because of inclination non - alignments . 94 ( ) have ltte orbits with periods longer than the photometric baseline . in these cases we do not have well constrained periods and our chances of detecting a tertiary eclipse are slim . a second reason that our determination of the third - body occurrence is likely a lower limit is that the very close binaries that comprise our sample here generally present more noise in the etv signal , which could easily bury a weak ltte signal . we have employed a method that minimizes false positives due to spurious etv signals , and thus necessarily have eliminated some potentially true ltte signals . third , and perhaps most important , the limited timespan of the currently available _ kepler _ data ( @xmath11400 days ) significantly restricts us to detect third bodies with orbital periods comparable to or shorter than 1400 days . relative to the full span of tertiary separations found in previous works @xcite , with separations as large as @xmath11 pc , we are at present sampling only the relatively closest tertiary companions . indeed , @xcite found among tight binaries that the rate of third bodies with orbital periods less than @xmath13 years ( comparable to our limit based on the duration of the available _ kepler _ data ) is 15% @xmath34 3% . thus our finding of a third - body occurrence rate with a period less than 1400 days of is compatible with the expected rate , though it appears we are likely still missing a fraction of some systems for the reasons already mentioned . the distribution of periods of potential third body orbits is also shown in fig . [ ptrip ] . we can clearly see a falloff in detection past the current length of the _ kepler _ mission of @xmath1 1400 days , as expected . however , for third - body periods shorter than @xmath1 1400 days , for which our detectability is relatively good , the occurrence rate does appear to increase toward longer third - body periods , consistent with the period distribution of third bodies among tight binaries found by @xcite . furthermore , we find that the triples on the widest orbits are found around the shortest period binaries , which is consistent with models that tighten the inner binary orbit through the presence , and gradual widening , of a companion . we presented our technique for computing precise eclipse timings for 1279 close eclipsing binaries in the _ kepler _ eclipsing binary catalog . these precise eclipse timings are complemented by the eclipse timings to be reported by @xcite for longer period , detached ebs . for the ebs whose timings are reported here , our method has been developed specifically to deal with the challenge of constantly changing light levels arising from spots and other phenomena that distort the light curves and could cause spurious eclipse timing variation ( etv ) signals . ebs with etv signals suggesting the possible presence of a third body have been identified and have been fit with a ltte orbit model in order to determine the likely parameters of the third bodies . in the current sample of 1279 close ebs , we have identified 236 that likely have tertiary companions . the parameters of these fits are also available online and are updated as new data become available . our measured occurrence rate of of close binaries with tertiary companions with periods up to @xmath35 days ( limited by the current timespan of _ kepler _ data ) , appears to be broadly consistent with the expectation that @xmath36% of close binaries will have tertiaries of such periods @xcite . indeed , we already find in our data that the periods of third bodies rise among the tightest binaries , consistent with previous work that has found a very high rate of third bodies in very wide orbits around the tightest binaries , presumably the result of dynamical tightening of inner binaries through widening of the tertiary . eclipse timings for all ebs are updated in real - time and are freely available as a community resource at ` http://keplerebs.villanova.edu ` . we thank nathan de lee and phil cargile for helpful discussions and darin ragozzine , eric ford , and joshua pepper for their feedback . this project is supported through the _ kepler _ participating scientist award nsr303065 . kec and kgs acknowledge support from nasa adap grant nnx12ae22 g . jao and wfw gratefully acknowledge support from the nsf via grant ast-1109928 , and also nasa via the _ kepler _ psp grant nnx12ad23 g . applegate , j. h. 1992 , , 385 , 621 armstrong , d. , pollacco , d. , watson , c. a. , et al . 2012 , , 545 , l4 batalha , n. m. , borucki , w. j. , koch , d. g. , et al . 2010 , , 713 , l109 bonnell , i. a. 2001 , the formation of binary stars , 200 , 23 borkovits , t. , csizmadia , s. , forgcs - dajka , e. , & hegeds , t. 2011 , , 528 , a53 borucki , w. j. , koch , d. , basri , g. , et al . 2010 , science , 327 , 977 dhital , s. , west , a. a. , stassun , k. g. , & bochanski , j. j. 2010 , , 139 , 2566 doyle , l. r. , carter , j. a. , fabrycky , d. c. , et al . 2011 , science , 333 , 1602 fabrycky , d. c. , ford , e. b. , steffen , j. h. , et al . 2012 , , 750 , 114 fabrycky , d. , & tremaine , s. 2007 , , 669 , 1298 ford , e. b. , ragozzine , d. , rowe , j. f. , et al . 2012 , , 756 , 185 gies , d. r. , williams , s. j. , matson , r. a. , et al . 2012 , , 143 , 137 hilditch , r. w. 2001 , an introduction to close binary stars , by r. w. hilditch , pp . isbn 0521241065 . cambridge , uk : cambridge university press , march 2001 . jenkins , j. m. , caldwell , d. a. , chandrasekaran , h. , et al . 2010 , , 713 , l120 law , n. m. , dhital , s. , kraus , a. , stassun , k. g. , & west , a. a. 2010 , , 720 , 1727 lee , j. w. , kim , s .- l . , lee , c .- u . , et al . 2013 , , 763 , 74 kirk et . al 2013 ( in preparation ) matijevi , g. , pra , a. , orosz , j. a. , et al . 2012 , , 143 , 123 orosz et al . 2013 ( in preparation ) pra , a. , batalha , n. , slawson , r. w. , et al . 2011 , , 141 , 83 pra , a. , guinan , e. f. , devinney , e. j. , et al . 2008 , , 687 , 542 rappaport , s. , deck , k. , levine , a. , et al . 2013 , arxiv:1302.0563 reipurth , b. , & mikkola , s. 2012 , , 492 , 221 schwarz , g. 1978 . the annals of statistics , 6 , 461 schwarz , r. , haghighipour , n. , eggl , s. , pilat - lohinger , e. , & funk , b. 2011 , , 414 , 2763 slawson , r. w. , pra , a. , welsh , w. f. , et al . 2011 , , 142 , 160 steffen , j. h. , ford , e. b. , rowe , j. f. , et al . 2012 , , 756 , 186 thackeray - lacko , b. , hill , m. , orosz , j. a. , et al . 2013 , american astronomical society meeting abstracts , 221 , # 142.09 tokovinin , a. a. 1997 , astronomy letters , 23 , 727 tokovinin , a. , thomas , s. , sterzik , m. , & udry , s. 2006 , , 450 , 681 tran , k. , levine , a. , rappaport , s. , et al . 2013 , , 774 , 81 welsh , w. f. , orosz , j. a. , carter , j. a. , et al . 2012 , , 481 , 475 rrrrr 1433410 & 56107.275 & -6.307 & 39.658 & primary + 1433410 & 56107.275 & -4.493 & 20.390 & entire + 1433410 & 56107.416 & -47.606 & 40.954 & secondary + 1433410 & 56107.558 & -10.541 & 18.835 & entire + 1433410 & 56107.558 & 21.341 & 30.499 & primary + 1433980 & 55740.148 & -443.232 & 143.078 & secondary + 1433980 & 55740.944 & -283.392 & 70.502 & primary + 1433980 & 55740.944 & -170.899 & 69.552 & entire + 1433980 & 55741.741 & -161.482 & 90.634 & secondary + 1433980 & 55742.537 & -248.659 & 61.776 & primary + rrr [ tablemt ] 2305372 & 3104113 & 3765708 + 4074532 & 4851217 & 4853067 + 5020034 & 5471619 & 5770860 + 5792093 & 6044064 & 6044543 + 6066379 & 6213131 & 6314173 + 6464285 & 6677225 & 7696778 + 7938468 & 7938870 & 8758161 + 9087918 & 9402652 & 9840412 + 9934052 & 10030943 & 10292413 + 10736223 & 11097678 & 11144556 + 11924311 & & + rrrrrrr 3228863 & 668.4 & @xmath37 & 0.08\{0.06,0.12 } & @xmath38 & 189\{187,194 } & @xmath39 + 4647652 & 753.5 & @xmath40 & 0.35\{0.10,0.44 } & @xmath41 & 228\{183,274 } & @xmath42 + 4909707 & 505.3 & @xmath43 & 0.54\{0.31,0.66 } & @xmath44 & 493\{378,627 } & @xmath45 + 5128972 & 447.8 & @xmath46 & 0.33\{0.25,0.41 } & @xmath47 & 259\{244,271 } & @xmath48 + 5264818 & 296.3 & @xmath49 & 0.37\{0.13,0.53 } & @xmath50 & 145\{107,196 } & @xmath51 + 5310387 & 214.2 & @xmath52 & 0.53\{0.34,0.61 } & @xmath53 & 31\ { 27 , 37 } & @xmath54 + 5376552 & 334.5 & @xmath55 & 0.40\{0.35,0.45 } & @xmath56 & 94\ { 91 , 98 } & @xmath57 + 6370665 & 285.9 & @xmath58 & 0.22\{0.07,0.33 } & @xmath59 & 67\ { 61 , 74 } & @xmath60 + 6531485 & 48.3 & & 0.44\{0.33,0.63 } & & 72\ { 31,109 } & + 7690843 & 74.3 & @xmath61 & 0.25\{0.08,0.42 } & @xmath62 & 71\ { 51 , 91 } & @xmath63 + 8043961 & 476.7 & @xmath64 & 0.25\{0.14,0.33 } & @xmath65 & 194\{179,213 } & @xmath66 + 8192840 & 803.9 & @xmath67 & 0.63\{0.52,0.70 } & @xmath68 & 208\{187,223 } & @xmath69 + 8386865 & 293 & @xmath70 & 0.38\{0.27,0.48 } & @xmath71 & 171\{156,210 } & @xmath72 + 8394040 & 394.8 & @xmath73 & 0.61\{0.50,0.67 } & @xmath74 & 369\{345,391 } & @xmath75 + 8904448 & 548.1 & @xmath76 & 0.59\{0.50,0.66 } & @xmath77 & 171\{158,192 } & @xmath78 + 9451096 & 106.7 & @xmath79 & 0.24\{0.10,0.36 } & @xmath80 & 90\ { 59,144 } & @xmath81 + 9722737 & 443.9 & @xmath82 & 0.22\{0.16,0.27 } & @xmath83 & 230\{225,236 } & @xmath84 + 9912977 & 753.7 & @xmath85 & 0.31\{0.16,0.39 } & @xmath86 & 105\ { 94,117 } & @xmath87 + 10226388 & 934.9 & @xmath88 & 0.32\{0.24,0.39 } & @xmath89 & 465\{434,493 } & @xmath90 + 10991989 & 554.2 & @xmath91 & 0.30\{0.21,0.37 } & @xmath92 & 256\{239,274 } & @xmath93 + 11042923 & 839 & @xmath94 & 0.17\{0.09,0.25 } & @xmath95 & 223\{213,230 } & @xmath96 + rrrrrr 2856960 & 0.60 & 0.259 & @xmath97 & @xmath98 & @xmath99 + 3228863 & 0.65 & 0.731 & @xmath37 & @xmath38 & @xmath39 + 3245776 & 0.96 & 1.492 & @xmath100 & @xmath101 & @xmath102 + 3641446 & 0.95 & 2.100 & @xmath103 & @xmath104 & @xmath105 + 4037163 & 0.58 & 0.635 & @xmath106 & @xmath107 & @xmath108 + 4909707 & 0.72 & 2.302 & @xmath43 & @xmath44 & @xmath45 + 5128972 & 0.74 & 0.505 & @xmath46 & @xmath47 & @xmath48 + 5264818 & 0.91 & 1.905 & @xmath49 & @xmath50 & @xmath51 + 5310387 & 0.96 & 0.442 & @xmath52 & @xmath53 & @xmath54 + 5376552 & 0.82 & 0.504 & @xmath55 & @xmath56 & @xmath57 + 5459373 & 0.97 & 0.287 & @xmath109 & @xmath110 & @xmath111 + 5560831 & 0.60 & 0.868 & @xmath112 & @xmath113 & @xmath114 + 6302592 & 0.93 & 1.578 & @xmath115 & @xmath116 & @xmath117 + 6370665 & 0.94 & 0.932 & @xmath58 & @xmath59 & @xmath60 + 7362751 & 0.73 & 0.338 & @xmath118 & @xmath119 & @xmath120 + 7657914 & 0.72 & 0.475 & @xmath121 & @xmath122 & @xmath123 + 7685689 & 0.77 & 0.325 & @xmath124 & @xmath125 & @xmath126 + 7690843 & 0.64 & 0.786 & @xmath61 & @xmath62 & @xmath63 + 8043961 & 0.63 & 1.559 & @xmath64 & @xmath65 & @xmath66 + 8145477 & 0.88 & 0.566 & @xmath127 & @xmath128 & @xmath129 + 8190491 & 0.95 & 0.778 & @xmath130 & @xmath38 & @xmath131 + 8211618 & 0.73 & 0.337 & @xmath132 & @xmath133 & @xmath134 + 8330092 & 0.80 & 0.322 & @xmath135 & @xmath136 & @xmath137 + 8386865 & 0.99 & 1.258 & @xmath70 & @xmath71 & @xmath72 + 8394040 & 0.77 & 0.302 & @xmath73 & @xmath74 & @xmath75 + 8904448 & 0.74 & 0.866 & @xmath76 & @xmath77 & @xmath78 + 9075704 & 0.68 & 0.513 & @xmath138 & @xmath139 & @xmath140 + 9451096 & 0.54 & 1.250 & @xmath79 & @xmath80 & @xmath81 + 9706078 & 0.55 & 0.614 & @xmath141 & @xmath142 & @xmath143 + 9722737 & 0.78 & 0.419 & @xmath82 & @xmath83 & @xmath84 + 9994475 & 0.76 & 0.318 & @xmath144 & @xmath145 & @xmath146 + 10014830 & 0.61 & 3.031 & @xmath147 & @xmath148 & @xmath149 + 10855535 & 0.98 & 0.113 & @xmath150 & @xmath151 & @xmath152 + 10991989 & 0.54 & 0.974 & @xmath91 & @xmath92 & @xmath93 + 11247386 & 0.75 & 0.394 & @xmath153 & @xmath154 & @xmath155 + 2302092 & 0.89 & 0.295 & @xmath156 & @xmath157 & @xmath158 + 2450566 & 0.98 & 1.845 & @xmath159 & @xmath160 & @xmath161 + 2835289 & 0.94 & 0.858 & @xmath162 & @xmath163 & @xmath164 + 3839964 & 0.78 & 0.256 & @xmath165 & @xmath166 & @xmath167 + 4069063 & 0.56 & 0.504 & @xmath168 & @xmath169 & @xmath170 + 4138301 & 0.90 & 0.253 & @xmath171 & @xmath172 & @xmath173 + 4244929 & 0.91 & 0.341 & @xmath174 & @xmath175 & @xmath176 + 4451148 & 0.82 & 0.736 & @xmath177 & @xmath178 & @xmath179 + 4547308 & 0.88 & 0.577 & @xmath180 & @xmath38 & @xmath181 + 4647652 & 0.68 & 1.065 & @xmath40 & @xmath41 & @xmath42 + 4670267 & 0.60 & 2.006 & @xmath182 & @xmath183 & @xmath184 + 4681152 & 0.55 & 1.836 & @xmath185 & @xmath186 & @xmath187 + 4762887 & 0.95 & 0.737 & @xmath188 & @xmath65 & @xmath189 + 4859432 & 0.76 & 0.385 & @xmath190 & @xmath191 & @xmath192 + 4937217 & 0.82 & 0.429 & @xmath193 & @xmath194 & @xmath195 + 4945857 & 0.74 & 0.335 & @xmath196 & @xmath56 & @xmath197 + 5269407 & 0.53 & 0.959 & @xmath198 & @xmath38 & @xmath199 + 5478466 & 0.97 & 0.483 & @xmath200 & @xmath201 & @xmath202 + 5611561 & 0.74 & 0.259 & @xmath203 & @xmath65 & @xmath204 + 5790912 & 0.77 & 0.383 & @xmath205 & @xmath206 & @xmath207 + 5791886 & 0.76 & 0.325 & @xmath208 & @xmath209 & @xmath210 + 5975712 & 0.87 & 1.136 & @xmath211 & @xmath212 & @xmath213 + 6050116 & 0.77 & 0.240 & @xmath214 & @xmath215 & @xmath216 + 6118779 & 0.90 & 0.364 & @xmath217 & @xmath218 & @xmath219 + 6281103 & 0.98 & 0.363 & @xmath220 & @xmath221 & @xmath222 + 6469946 & 0.51 & 0.716 & @xmath223 & @xmath224 & @xmath225 + 6516874 & 0.60 & 0.916 & @xmath226 & @xmath227 & @xmath228 + 6543674 & 0.53 & 2.391 & @xmath229 & @xmath230 & @xmath231 + 6615041 & 0.75 & 0.340 & @xmath232 & @xmath233 & @xmath234 + 6629588 & 0.55 & 2.264 & @xmath235 & @xmath236 & @xmath237 + 6671698 & 0.73 & 0.472 & @xmath238 & @xmath239 & @xmath240 + 6766325 & 0.92 & 0.440 & @xmath241 & @xmath242 & @xmath243 + 7035139 & 0.79 & 0.310 & @xmath244 & @xmath245 & @xmath246 + 7119757 & 0.64 & 0.743 & @xmath247 & @xmath248 & @xmath249 + 7272739 & 0.75 & 0.281 & @xmath250 & @xmath251 & @xmath252 + 7385478 & 0.54 & 1.655 & @xmath253 & @xmath254 & @xmath255 + 7518816 & 0.65 & 0.467 & @xmath256 & @xmath257 & @xmath258 + 7877062 & 0.81 & 0.304 & @xmath259 & @xmath260 & @xmath81 + 8045121 & 0.99 & 0.263 & @xmath261 & @xmath262 & @xmath263 + 8192840 & 0.95 & 0.434 & @xmath67 & @xmath68 & @xmath69 + 8242493 & 0.73 & 0.283 & @xmath264 & @xmath38 & @xmath265 + 8563964 & 1.00 & 0.338 & @xmath266 & @xmath262 & @xmath267 + 8690104 & 0.77 & 0.409 & @xmath268 & @xmath269 & @xmath270 + 8739802 & 0.93 & 0.275 & @xmath271 & @xmath272 & @xmath273 + 8957887 & 0.76 & 0.347 & @xmath274 & @xmath275 & @xmath276 + 8982514 & 0.84 & 0.414 & @xmath277 & @xmath262 & @xmath278 + 9091810 & 0.69 & 0.480 & @xmath279 & @xmath38 & @xmath280 + 9101279 & 0.58 & 1.811 & @xmath281 & @xmath282 & @xmath283 + 9272276 & 0.78 & 0.281 & @xmath284 & @xmath285 & @xmath286 + 9283826 & 0.84 & 0.357 & @xmath287 & @xmath56 & @xmath288 + 9353234 & 0.86 & 1.487 & @xmath289 & @xmath290 & @xmath291 + 9412114 & 0.85 & 0.250 & @xmath292 & @xmath293 & @xmath294 + 9532219 & 0.74 & 0.198 & @xmath295 & @xmath110 & @xmath296 + 9592145 & 0.65 & 0.489 & @xmath297 & @xmath298 & @xmath299 + 9612468 & 1.00 & 0.133 & @xmath300 & @xmath301 & @xmath302 + 9665086 & 0.67 & 0.297 & @xmath303 & @xmath304 & @xmath305 + 9821923 & 0.95 & 0.350 & @xmath306 & @xmath307 & @xmath308 + 9838047 & 0.84 & 0.436 & @xmath309 & @xmath310 & @xmath311 + 9912977 & 0.59 & 1.888 & @xmath85 & @xmath86 & @xmath87 + 10226388 & 0.77 & 0.661 & @xmath88 & @xmath89 & @xmath90 + 10275197 & 0.78 & 0.391 & @xmath312 & @xmath313 & @xmath314 + 10322582 & 0.86 & 0.291 & @xmath315 & @xmath316 & @xmath317 + 10383620 & 0.64 & 0.735 & @xmath318 & @xmath319 & @xmath320 + 10388897 & 0.99 & 0.344 & @xmath321 & @xmath322 & @xmath323 + 10724533 & 0.75 & 0.745 & @xmath324 & @xmath325 & @xmath326 + 10727655 & 0.73 & 0.353 & @xmath327 & @xmath262 & @xmath328 + 10848807 & 0.74 & 0.346 & @xmath329 & @xmath330 & @xmath331 + 10905804 & 0.68 & 0.751 & @xmath332 & @xmath333 & @xmath334 + 10916675 & 0.87 & 0.419 & @xmath335 & @xmath336 & @xmath337 + 10934755 & 0.68 & 0.786 & @xmath338 & @xmath339 & @xmath340 + 11042923 & 0.76 & 0.390 & @xmath94 & @xmath95 & @xmath96 + 11246163 & 0.77 & 0.279 & @xmath341 & @xmath342 & @xmath343 + 11347875 & 0.86 & 3.455 & @xmath344 & @xmath345 & @xmath346 + 11604958 & 0.72 & 0.299 & @xmath347 & @xmath348 & @xmath349 + 11716688 & 0.94 & 0.301 & @xmath350 & @xmath351 & @xmath352 + 11825204 & 0.98 & 0.210 & @xmath353 & @xmath354 & @xmath355 + 12019674 & 0.76 & 0.355 & @xmath356 & @xmath357 & @xmath358 + 12055255 & 0.90 & 0.221 & @xmath359 & @xmath360 & @xmath361 + 12071741 & 0.94 & 0.314 & @xmath362 & @xmath363 & @xmath364 + 12458133 & 0.76 & 0.333 & @xmath365 & @xmath366 & @xmath367 + 3114667 & 0.52 & 0.889 & @xmath368 & & + 2983113 & 0.89 & 0.395 & @xmath369 & & + 4066203 & 0.93 & 0.363 & @xmath370 & & + 4074708 & 0.73 & 0.302 & @xmath33 & & + 4904304 & 0.89 & 0.389 & @xmath369 & & + 5282464 & 0.72 & 0.496 & @xmath33 & & + 5956776 & 0.61 & 0.569 & @xmath370 & & + 6153219 & 0.62 & 0.530 & @xmath370 & & + 6265720 & 0.92 & 0.312 & @xmath369 & & + 6794131 & 0.81 & 1.613 & @xmath369 & & + 7506164 & 0.88 & 0.558 & @xmath369 & & + 7590728 & 0.69 & 0.477 & @xmath371 & & + 9544350 & 0.92 & 2.260 & @xmath372 & & + 9776718 & 0.87 & 0.544 & @xmath373 & & + 9788457 & 0.60 & 0.963 & @xmath372 & & + 10095469 & 0.60 & 0.678 & @xmath373 & & + 10796477 & 0.74 & 0.485 & @xmath33 & & + 2159783 & 0.87 & 0.374 & @xmath374 & & + 3127873 & 0.91 & 0.672 & @xmath375 & & + 3221207 & 0.81 & 0.474 & @xmath376 & & + 3342425 & 0.93 & 0.393 & @xmath377 & & + 3766353 & 0.53 & 2.667 & @xmath378 & & + 3848042 & 0.99 & 0.411 & @xmath379 & & + 3935319 & 0.75 & 0.353 & @xmath35 & & + 3936357 & 0.76 & 0.369 & @xmath380 & & + 4077442 & 0.58 & 0.693 & @xmath381 & & + 4464999 & 0.77 & 0.434 & @xmath382 & & + 4563150 & 0.79 & 0.275 & @xmath383 & & + 4758368 & 0.57 & 3.750 & @xmath384 & & + 4945588 & 0.99 & 1.129 & @xmath384 & & + 5008287 & 0.94 & 0.292 & @xmath385 & & + 5015926 & 0.75 & 0.363 & @xmath35 & & + 5097446 & 0.60 & 1.288 & @xmath378 & & + 5296877 & 0.95 & 0.377 & @xmath374 & & + 5353374 & 0.78 & 0.393 & @xmath376 & & + 5389616 & 0.99 & 0.407 & @xmath386 & & + 5513861 & 0.57 & 1.510 & @xmath387 & & + 5770431 & 0.89 & 0.392 & @xmath388 & & + 5820209 & 0.81 & 0.656 & @xmath389 & & + 5951553 & 0.95 & 0.432 & @xmath385 & & + 6144827 & 0.79 & 0.235 & @xmath390 & & + 6187893 & 0.59 & 0.789 & @xmath391 & & + 6287172 & 0.95 & 0.204 & @xmath35 & & + 6370361 & 0.84 & 0.455 & @xmath377 & & + 7137798 & 0.51 & 2.254 & @xmath392 & & + 7269843 & 0.79 & 0.268 & @xmath377 & & + 7339345 & 0.74 & 0.260 & @xmath393 & & + 7367833 & 0.76 & 0.286 & @xmath394 & & + 7375612 & 0.97 & 0.160 & @xmath386 & & + 7680593 & 0.97 & 0.276 & @xmath384 & & + 7697065 & 0.69 & 0.273 & @xmath386 & & + 7709086 & 0.78 & 0.409 & @xmath383 & & + 7773380 & 0.94 & 0.308 & @xmath35 & & + 7871200 & 0.74 & 0.243 & @xmath395 & & + 7878402 & 0.73 & 0.374 & @xmath396 & & + 8016214 & 0.53 & 3.175 & @xmath386 & & + 8122124 & 1.00 & 0.249 & @xmath389 & & + 8189196 & 0.98 & 2.304 & @xmath397 & & + 8222945 & 0.99 & 0.451 & @xmath385 & & + 8231231 & 0.89 & 0.712 & @xmath389 & & + 8257903 & 0.76 & 0.515 & @xmath398 & & + 8265951 & 0.81 & 0.780 & @xmath383 & & + 8703528 & 0.74 & 0.400 & @xmath384 & & + 8715667 & 0.85 & 0.406 & @xmath377 & & + 8758716 & 1.00 & 0.107 & @xmath388 & & + 9026766 & 0.75 & 0.272 & @xmath399 & & + 9083523 & 0.65 & 0.918 & @xmath400 & & + 9097798 & 0.99 & 0.334 & @xmath401 & & + 9181877 & 0.74 & 0.321 & @xmath379 & & + 9347868 & 0.82 & 0.318 & @xmath402 & & + 9657096 & 0.94 & 2.138 & @xmath35 & & + 9724080 & 0.94 & 1.174 & @xmath382 & & + 9882280 & 0.75 & 0.289 & @xmath403 & & + 9956124 & 0.91 & 0.363 & @xmath404 & & + 10007533 & 0.88 & 0.648 & @xmath405 & & + 10135584 & 0.96 & 0.391 & @xmath375 & & + 10216186 & 0.64 & 0.606 & @xmath406 & & + 10228991 & 0.97 & 2.799 & @xmath389 & & + 10229723 & 0.83 & 0.629 & @xmath407 & & + 10481912 & 0.96 & 0.442 & @xmath408 & & + 10485137 & 0.71 & 0.445 & @xmath403 & & + 10557008 & 0.77 & 0.265 & @xmath384 & & + 10711938 & 0.95 & 0.358 & @xmath383 & & + 11091082 & 0.82 & 0.385 & @xmath35 & & + 11151970 & 0.87 & 0.312 & @xmath409 & & + 11305087 & 0.78 & 0.309 & @xmath410 & & + 11496078 & 0.87 & 0.300 & @xmath411 & & + 11509282 & 0.83 & 0.634 & @xmath389 & & + 11566174 & 0.76 & 0.277 & @xmath412 & & + 11717798 & 0.83 & 0.375 & @xmath413 & & + 11805235 & 0.79 & 0.395 & @xmath406 & & + 11910076 & 0.96 & 0.348 & @xmath414 & & + 12055421 & 0.96 & 0.386 & @xmath389 & & + 12267718 & 0.89 & 0.545 & @xmath415 & & + 12554536 & 0.63 & 0.684 & @xmath394 & & +

in this article , we describe a set of algorithms on multiple copies of @xmath0 ( with possibly nonlinear interactions ) , with applications in material science . the physical processes we are interested in are modeled by pdes of the form @xmath1 here @xmath2 is the field describing the physical process of interest , with @xmath3 denoting the spatial variable on some closed , compact subset of @xmath4 , and @xmath5 is a ( typically nonlinear ) function of @xmath6 . one immediately notices that @xmath7 is constant for all time ; this constraint should be respected by the numerical methods used to study such systems . classical numerical integrators fail to preserve this norm , and this motivates our use of geometric integration . the paper is organized as follows : we begin by introducing a problem in micromagnetics . we then describe the lie group framework suitable for the general problem ( [ generalsystem1 ] ) . the new , and fairly geometric , mathematical constructs of partial moving frames and partial connections , which can be used to select generators for use in numerical methods , are introduced in section 3 . readers who are unfamiliar with modern differential geometry , particularly principal bundles , are advised to skip this section at a first reading . in section [ sectionalgo ] , we present some one - step algorithms of different orders . an arbitrary function appears in these algorithms ; different choices of this function yield distinct discrete trajectories . we describe a possible choice that is related both to moving frames ( @xcite ) and , in the case of a geometric version of the forward euler method , to discretization error minimization . work is in progress to identify analogous function choices for higher order methods ( @xcite ) . we present the results of numerical experiments carried out using some of these functions for the symplectically reduced free rigid body , a hamiltonian system on the two - sphere . the conservative nature of the rigid body system facilitates both the derivation of appropriate functions and the assessment of the relative performance of the algorithms . finally , we present some representative numerical results for micromagnetics that indicate that these geometric integration schemes are competitive with conventional numerical algorithms . our numerical experiments were carried out using first , second , and fourth order methods . one of the intrinsic features of the system ( [ generalsystem1 ] ) is that the vector field @xmath8 evolves on a sphere . however , a classical integrator will update @xmath9 at time @xmath10 using the approximation @xmath11 the particular form of @xmath12 depends on the algorithm chosen ; however , it is clear that such updates correspond to _ translations _ of @xmath9 , not _ rotations_. thus , a classical integrator does not account for the fact that @xmath6 evolves on a sphere . this constraint is difficult to efficiently impose in practice . a naive approach is to keep track of changes in the norm @xmath13 during a numerical experiment , and renormalize the iterates after a prescribed tolerance has been exceeded . however , this renormalization is equivalent to the aphysical addition ( or subtraction ) of energy to the system and is therefore an undesirable solution . in addition , this renormalization would also affect @xmath14 . we shall see that in the context of micromagnetics , this change is nontrivial and nonlinear . another observation is that the component of @xmath15 parallel to @xmath6 does not influence the solution curve @xmath16 however , this component does alter the discrete trajectories generated by numerical algorithms . an appropriate selection of the normal component can improve the performance of the scheme . preliminary numerical investigations suggest that such corrections may allow increased accuracy in the capture of key features , e.g. orbits and/or conserved quantities , at a moderate computational cost , without the introduction of significant numerical artifacts , e.g. numerical accelerations . the philosophy that numerical algorithms should respect properties intrinsic to the system momentum conservation , evolution on a manifold , hamiltonian structure is not a novel idea . ( see , e.g. , @xcite , @xcite , @xcite ) . many classes of algorithms have been developed with this guiding principle , especially for energy - conserving or symplectic systems . as far as we know , the application of such methods in the context of numerical micromagnetics is relatively new . encouraged by the success of these algorithms in an industrial context ( @xcite ) , we believe that these techniques will find wider use in the material science community . modern magnetic materials are used in an increasingly large number of applications , including thin film read heads and recording media @xcite , nanocrystalline permanent magnets , and magnetohydrodynamic fluids . in addition , there has been much interest in the use of `` smart materials '' , including magnetostrictive actuators and organic ferromagnets @xcite . these magnetic materials exhibit different responses corresponding to varying magnetic fields . for example , the resistance of a read - head ferromagnetic sensor changes as the device rotates over a recording medium . the magnetization of the material directly interacts with the other physical and chemical characteristics of the material ; micromagnetics theory describes this interaction at a microscopic level . of particularly great interest is the correlation of physical microstructure and magnetization ; the ability to predict the response of one to variations in the other is crucial to the further development of these materials and devices . another physical system whose mathematical model strongly resembles that of micromagnetics arises in the study of liquid crystals . nematic and smectic liquid crystals form the basis of the operation of many every - day devices , such as lcd s , telecommunication devices , thermometers , projection systems even mood rings . the devices operate on the principle that a suitable applied field will change the orientation of the liquid crystals , while conserving the pointwise - norm . therefore , the resulting mathematical model has the same constraints as those in the micromagnetics situation . this system has been studied extensively , yet a norm - conserving algorithm has only been presented for an extremely simple situation ( @xcite ) . the study of long - chain molecules such as those occurring in bio - molecular systems yield another possible application area for the methods we develop . these chemical systems have more complicated constraints on the geometry of possible configurations ; the configuration sought is one that minimizes a free - energy functional . statistical thermodynamics considerations permit the reformulation of these optimization problems as evolutions in time on given manifolds to a steady state ( @xcite,@xcite ) . such studies are used in the development of new drugs . at any point in time , the magnetization @xmath6 is constant in small regions in the material ( termed a domain ) , and the switching of these domains from one state to another is the basis of the functioning of devices built with these magnetic substances @xcite . ideally the entire device would be one large domain , which would switch instantaneously when an applied field was imposed . in practice , there are several domains , and the net magnetization in a desired direction is not optimal . we are interested in tracking the evolution of these domains , which entails following the local behavior of @xmath6 . in industrial applications , a ferromagnetic device is subjected to changing magnetic fields ( corresponding to the various applications a disk moving underneath a read - head , changes in applied voltages for magnetorheological fluids , etc ) . one is interested in the response of the device to the gradual changing of these external fields . a model that is widely used in the industry is the _ landau - lifshitz - gilbert ( llg ) _ model of micromagnetics , which describes the evolution of the state of magnetization @xmath6 in a ferromagnetic sensor , occupying a region @xmath17 in space . the llg equation for the magnetization @xmath18 is given by @xmath19 here @xmath20 is a damping parameter and @xmath21 is an effective magnetic field , described in detail below . the first term on the right hand side of ( [ llg ] ) describes the ( undamped ) _ larmor precession _ of @xmath6 about @xmath21 and is derived from first principles @xcite . it is observed in physical experiments , however , that changes in magnetization decay in finite time . the second term in ( [ llg ] ) is a phenomenological term ( called the _ gilbert damping term _ , see @xcite ) , added to describe this damping behavior ; it can not be derived from first principles . there are situations under which this system is stiff ( see , for example , @xcite ) , and issues of numerical stability of the integration scheme are therefore important . we also thank one of the referees for pointing out @xcite , where it is suggested that poor representations of the exchange energy may lead to the observed stiff behavior . the effective field , which causes the magnetization to change , is derived from energy considerations @xcite and varies nonlinearly with @xmath6 . more precisely , @xmath22 the parameters @xmath23 and @xmath24 are material constants of the permalloy being studied , and @xmath25 is the permeability of free - space . the field @xmath26 is called the _ exchange _ field , preventing rapid spatial variations of @xmath6 and the formation of arbitrarily fine domains ( see @xcite for examples on how this term is computed in general ) . the final contribution in ( [ heff ] ) is due to the nature of ferromagnetic crystals , which causes the magnetic moments to align in preferred directions . this effect is incorporated in the llg model through the _ uniaxial anisotropy _ field @xmath27 . the external _ applied field _ is denoted by @xmath28 . the nonlinear , nonlocal contributions of @xmath6 arise through the _ demagnetizing field _ , @xmath29 , where @xmath30 solves the poisson problem with suitable boundary @xmath31=0,\quad \left [ \frac{\partial \phi}{\partial n}\right ] = { \boldsymbol{\mu}}\cdot { \mathbf n } \,\ , \mbox{on } \,\ , \partial{\mathcal{b}},\ ] ] and radiation conditions@xmath32 here @xmath33 $ ] denotes the jump of the function @xmath34 across @xmath35 . many different methods exist for the calculation of this field , including the use of the full maxwell system , fft techniques , finite element methods , multigrid approaches , finite differences , and recently , fast multipole methods , @xcite , among others . while there are still several unsettled issues in the area of demagnetizing field calculations ( see @xcite for a sharp critique on existing methods ) , it is not our intention in this project to duplicate this work . instead , we shall focus on developing a time - stepping method that is robust , accurate , and requires relatively few field evaluations . theoretical developments in micromagnetics are driven by industrial demands , and the need for accurate algorithms is now imperative . conventional algorithms are still being employed for highly sensitive calculations on large sensors , and are becoming increasingly inadequate . moreover , the time - scales inherent in these problems vary from nano - seconds ( in disk drives ) to tens of seconds ; hence integration techniques that remain effective over long times are required . an ideal integrator would resolve solutions accurately over very small time steps , while allowing large time steps to be taken when the system evolves over a long period . we have already described some of the issues with classical integrators with respect to norm conservation . here we show that renormalizing @xmath6 to conserve the norm while using conventional integrators changes the potential @xmath30 in a nonlinear fashion ; it is easy to construct a simple example in which the renormalization introduces a significant change in the demagnetizing field . assume that the magnetization @xmath6 satisfies @xmath36 for some constant @xmath37 and scalar function @xmath38 ; here @xmath39 denote the usual unit vectors in the @xmath40 and @xmath41 directions . before renormalization , the field @xmath42 is divergence free : @xmath43 . however , the renormalized field is _ not _ divergence free : @xmath44 the potentials obtained by solving ( [ poisson ] ) are clearly not the same for the original and renormalized fields . the effect of renormalization on the demagnetization field is particularly significant near domain boundaries . for example , if the function @xmath38 appearing in ( [ renorm_exam ] ) is a step function , then the divergence of the renormalized magnetic field is a delta function . we thank f. reitich , @xcite , for this illustration of the dangers of normalizing vector fields . we recall that the general system under consideration is @xmath46 the initial condition @xmath25 satisfies @xmath47 for all @xmath3 , i.e. @xmath48 , the unit sphere in @xmath4 . since @xmath49 for all @xmath50 and @xmath10 , @xmath51 for all @xmath10 , the rotation group @xmath52 acts transitively on @xmath0 , i.e. any point on the sphere can be rotated onto any other point on the sphere ; hence there are time dependent curves @xmath53 satisfying @xmath54 we emphasize that these curves are not uniquely determined by ( [ rotategen ] ) . in the spatially discretized version of the system , we choose @xmath55 grid points @xmath56 in @xmath17 and consider only the values of the magnetic field at those grid points . given a curve @xmath57 , we define @xmath58 where there is no confusion , we suppress the argument @xmath10 . the fully discretized version of the system ( [ generalsystem ] ) is @xmath59 here and throughout the paper , vector operations such as cross or inner products on @xmath60 or @xmath61 should be understood as the usual operations in @xmath4 performed on each of the @xmath55 component vectors . the lie group @xmath62 acts transitively , but not freely , on @xmath63 . that is , any point in @xmath64 can be mapped onto any other point in @xmath63 by the action ( by component - wise rotations ) of @xmath65 on @xmath63 , but the group element accomplishing any such transformation is not unique . the isotropy subgroup of a point @xmath42 in @xmath63 ( where @xmath64 is the group of transformations fixing @xmath42 ) is an @xmath55dimensional torus . as in ( [ rotategen ] ) , there are smooth time - dependent curves @xmath66 in the group @xmath65 satisfying @xmath67 we can use the lie algebra @xmath68 of @xmath65 , which is the tangent space to @xmath65 at the identity , to put ( [ rotatedisc ] ) into a more familiar and convenient form . the identification of @xmath69 with @xmath4 is implemented using the map @xmath70 given by @xmath71 } : = \left(\begin{array}{cccc } 0 & -\xi_3 & \xi_2\\ \xi_3&0&-\xi_1\\ -\xi_2&\xi_1&0\end{array}\right),\ ] ] i.e. @xmath72 } { { \mathbf{x}}}= { \boldsymbol{\xi}}\times { { \mathbf{x}}}$ ] for all @xmath73 . the matrix commutator bracket on @xmath69 corresponds to the cross product on @xmath4 under this identification . given any differentiable curve @xmath74 in @xmath65 , there exists a curve @xmath75 satisfying @xmath76 } q(t),\ ] ] where the product of @xmath72}$ ] and @xmath66 is the usual matrix product . thus the system ( [ discretesystem ] ) is equivalent to @xmath77}q(t ) { \mathbf{m}}_0={\boldsymbol{\xi}}(t)\times { \mathbf{m}}(t).\ ] ] comparing ( [ geom_eqs1 ] ) to ( [ discretesystem ] ) , we see that @xmath78 for an arbitrary scalar function @xmath79 . the flexibility in the choice of map @xmath80 arises from the non freeness of the action of @xmath52 on @xmath0 , and thus the action of @xmath65 on @xmath63 ; distinct odes @xmath81 } q \qquad \mbox{and } \qquad \dot q = { \mathrm{skew } \left [ \widetilde { { \boldsymbol{\omega}}}(q \ , m)\right ] } q,\ ] ] where @xmath82 $ ] for all @xmath83 , will typically have distinct solution curves in @xmath52 , but the images in @xmath0 of those solution curves under the map @xmath84 will coincide . when numerically simulating ( [ discretesystem ] ) , we want a time - stepping method that ensures that @xmath85 , i.e. that the norms of the component vectors are identically equal to one . we can regard ( [ geom_eqs1 ] ) and ( [ geom_eqs ] ) as defining a family of odes @xmath86 } q,\ ] ] parametrized by @xmath87 and @xmath79 , on the group @xmath65 and use the techniques developed for geometric integration on lie groups to determine approximate discrete solution curves of these ode ( see @xcite and the references therein ) . when combined with the action of @xmath65 on @xmath63 , these techniques yield geometric integration schemes for ( [ generalsystem ] ) that exactly preserve the constraint @xmath85 , regardless of the step size or the order of the integrator . the key idea is the following . suppose we are given a ( right ) trivialized form @xmath88 of an ode on a lie group @xmath89 and an _ algorithmic exponential _ @xmath90 mapping the lie algebra @xmath91 of @xmath89 into @xmath89 . then an integrator of order @xmath92 corresponds to an update of the form @xmath93 for some map @xmath94 determined by the algorithm and the generator @xmath95 . here we consider only one step methods , with @xmath96 . we emphasize that the algorithmic exponential need not be the true exponential of the lie group , or even a good approximation to the true exponential ; all that we require is that it map algebra elements exactly into the group @xmath89 and that the algorithmic exponential of the zero vector in @xmath91 equal the identity element of @xmath89 . for example , the cayley transform , given by @xmath97})(i - { \mathrm{skew } \left [ { \boldsymbol{\xi}}/2\right]})^{-1},\ ] ] is an algorithmic exponential for the rotation group that has long been used in computational mechanics . the cayley transform has long been used to implement exact rotations in elasticity and plasticity simulations ; see , for example , @xcite ) . more recently , it has been utilized in the geometric integration of a wide variety of mechanical systems , including the llg equations ; see @xcite , and references therein . algorithms of arbitrarily high order can be constructed using the cayley transform , despite the fact that is only a second order approximation of the matrix exponential of @xmath52 . ( note that the cayley transform is , in fact , an algorithmic exponential for any matrix group determined by a quadratic constraint . see , e.g. @xcite . ) for the rotation group @xmath52 on @xmath4 , both the true matrix exponential and the cayley transform can be efficiently evaluated and have frequently been used as algorithmic exponentials . the true exponential takes the form @xmath98 } + \frac{1 - \cos \|{\boldsymbol{\xi}}\|}{\|{\boldsymbol{\xi}}\|^2 } \ , { \mathrm{skew } \left [ { \boldsymbol{\xi}}\right]}^2.\ ] ] the image of a vector @xmath73 under the action of @xmath99 takes the simple , readily evaluated form @xmath100.\ ] ] hence we shall use the cayley transform ( actually , @xmath55 copies of the cayley transform ) as our algorithmic exponential @xmath101 . we observe that the cayley transform has the advantage that the entries of @xmath99 are rational functions of the components of @xmath102 ; in particular , no trigonometric functions need be evaluated . to summarize this section , we have rewritten the discrete system @xmath103 as an ode @xmath104}q(t).\ ] ] on the lie group @xmath105 . we now need to describe choices of infinitesimal update maps @xmath106 and generators @xmath107 that determine one step numerical updates @xmath108 and associated updates @xmath109 with specified properties , e.g. a specified order of overall accuracy . the construction of suitable updates is the subject of the next section . a natural and obvious goal in selection of a numerical scheme is the achievement of the highest possible accuracy working within the given constraints . however , the prioritization of the constraints ( efficiency , stability , developer effort , preservation of key features of the modeled system , etc . ) can lead to significantly different approaches to the achievement of this goal and correspondingly different schemes . for the purposes of this discussion , we shall assume that we are given a family of one step methods of the form @xmath110 for some map @xmath111 . ( recall that @xmath112 . ) we can optimize accuracy within this family of methods by selecting an appropriate isotropy algebra correction ; specifically , we shall first compute the update generator given some ` default ' choice of generator , and then use that generator to determine an element of the isotropy algebra of the current state that minimizes the discretization error for that update . we will derive conditions specifying this choice of isotropy element for algorithms on @xmath64 utilizing the action of @xmath113 using a traditional series expansion approach to the computation of the discretization error . ( see @xcite for the application of this approach and those discussed below to general homogeneous manifolds . ) subsequently , we will discuss some more geometric and , in some cases , less computationally intensive , approaches to this task . the geometric approach most closely related to the naive series expansion treatment is the use of geodesic curvature to characterize the essential information about curves on manifolds , e.g. ; solution curves of differential equations . a less directly related approach , but one that coincides with the direct error minimization approach for the forward euler method on @xmath64 is the use of a partial moving frame and the associated partial connection form to determine a choice of generator . we shall define these constructions in section [ partial_conn ] . to demonstrate the influence of the isotropy algebra on discrete trajectories , we first study the action of the rotation group @xmath52 on a single sphere @xmath0 . the techniques we use in analysing the case of a single sphere immediately generalize to @xmath64 an autonomous vector field @xmath114 on @xmath0 satisfies @xmath115 for all @xmath83 ; hence there exists a ( nonunique ) map @xmath116 , called a _ generator _ of @xmath114 , satisfying @xmath117 on @xmath0 . recall from our earlier discussion that distinct choices of generator @xmath23 typically yield distinct discrete trajectories when used in a numerical algorithm of the form ( [ our_kind ] ) . we will compare the performance of various schemes with two different choices of generator . the first choice is the ` default ' or ` natural ' one , which is not assumed to have any particular geometric properties . for the rigid body system , we will take the body angular velocity as our default generator . for arbitrary vector fields @xmath114 , there need not be a natural choice of generator ; our intent here is to make a plausible choice of generator that one might make if the issue of isotropy were not taken into account . the second choice is the orthogonal generator , i.e. the unique map @xmath118 satisfying @xmath119 for all @xmath83 . a general and direct , but potentially computationally intensive , approach to the choice of a generator that reduces the discretization error is to compute the lowest order nonzero term in the series expansion for the discretization error for the family of algorithms under consideration , leaving the isotropy algebra component of the generator as an undetermined parameter , and thus determine conditions on the isotropy component that minimize the error . see @xcite for a general treatment of this approach and section [ opt ] for the derivation of the optimal generator choice for our forward euler algorithm for the llg system . here we briefly explore some alternatives to this approach that have natural geometric interpretations . we consider an euler update of the form @xmath120 where @xmath80 determines the isotropy ( normal component ) contribution . in the special case that the algorithmic exponential is given by a rescaling of the usual matrix exponential , e.g. by the cayley transform , ( [ simple_up ] ) satisfies @xmath121 for some rescaling @xmath122 of time . hence , in this case , @xmath123 is given by a rigid rotation of @xmath124 about an axis depending only on @xmath124 . hence the curve @xmath125 is a segment of a circle in @xmath0 . our goal is to choose @xmath80 so as to obtain the best circular approximation at @xmath124 to the true orbit segment @xmath126 if @xmath127 , then the optimal circular approximation to @xmath128 at @xmath124 can be characterized using the geodesic curvature @xmath129 of @xmath128 at @xmath124 . the best circular approximation to @xmath128 at @xmath124 is tangent to @xmath130 at @xmath124 and has geodesic curvature equal to that of @xmath128 at @xmath124 . the first condition is clearly satisfied for any consistent update . the geodesic curvature @xmath131 of @xmath132 is easily seen to satisfy @xmath133 , where @xmath30 is the angle between @xmath124 and @xmath134 ( see , e.g. @xcite , p. 249 ) ; thus @xmath135 . hence optimal orbit capture within the class of updates ( [ simple_up ] ) is obtained using @xmath136 if @xmath128 is itself a segment of a circle , then @xmath137 yields @xmath138 . hence any torsion free orbits , e.g. the separatricies of the reduced rigid body equations , are captured exactly by this version of the euler method . note that the choice @xmath139 is suboptimal for the euler method unless @xmath140 along the tractory of interest , i.e. unless the desired trajectory is a great circle . for higher order methods , the axis of rotation used in the update map @xmath141 is typically time dependent and hence the corresponding algorithmic trajectory segment typically is not circular ( i.e. it has nonzero torsion ) . hence the simple argument used in the preceding paragraph can not be applied . however , the strategy of curvature matching can still be followed . since a smooth curve on a two dimensional manifold in @xmath4 is determined up to a time reparametrization by its geodesic curvature , we can determine the conditions on the generator imposed by the restriction that the geodesic curvature of @xmath123 match that of @xmath142 to some order . the higher order derivatives of the curvature can either be determined analytically for a given vector field @xmath114 or numerically approximated using standard difference schemes . more generally , the choice a generator of a vector field on a homogeneous manifold can be viewed as a special case of the choice of a partial connection form , which generalizes to nonfree actions the classical connection form on a principal bundle . a partial connection form is a lie algebra valued one form with appropriate equivariance properties . in section [ partial_conn ] we state the relevant definitions and present a family of partial connection forms on open subsets of @xmath0 that yield a discretization error minimizing algorithm and determine an algorithm that captures orbits to second order for any dynamical system on a single copy of @xmath0 . the interested reader is referred to @xcite for a more detailed treatment of partial connection forms and related constructions . we now briefly discuss a general geometric approach to the selection of generators , using a generalization of the connection form on a principal bundle . for a more detailed treatment of this generalization and proofs of the assertions given below , see lewis et al . [ let @xmath143 be a principle bundle , that is , a manifold @xmath143 acted on by a lie group @xmath89 . let the action of @xmath144 in @xmath143 be denoted by @xmath145 let @xmath146 denote the algebra of @xmath89 and let @xmath147 . recall that a _ connection _ on a principal bundle @xmath148 is a distribution @xmath149 satisfying @xmath150 for all @xmath151 and @xmath152 . specification of a connection @xmath149 is equivalent to specification of an equivariant @xmath91valued one form @xmath153 , called the _ connection form _ , satisfying @xmath154 for all @xmath151 . by equivariance we mean that @xmath155 for all @xmath152 . the connection @xmath149 and connection form @xmath153 are related by the condition @xmath156 = \gamma_p$ ] for all @xmath151 . ( see , e.g. , @xcite for a detailed presentation of the properties of connections and connection forms . ) the equivariance properties of connections and connection forms typically can not be preserved in the context of nonfree actions , hence we relax these conditions , requiring only equivariance with respect to specified representatives of the isotropy equivalence classes . a map @xmath157 is a _ slip map _ if @xmath158 for all @xmath152 and @xmath159 . a ( singular ) distribution @xmath160 assigning a complement @xmath161 to @xmath162 in @xmath163 to each point @xmath159 is a _ partial connection _ if there is a slip map @xmath164 satisfying @xmath165 for all @xmath152 and @xmath159 . a _ partial connection form _ with slip map @xmath164 is a @xmath91valued one form @xmath153 on @xmath63 satisfying @xmath166 and @xmath167 for all @xmath152 and @xmath159 . one natural source of partial connections is a generalization of a _ moving frame _ , in the modern sense introduced by fels and olver ( @xcite , @xcite ) , i.e. a smooth equivariant map @xmath168 on a manifold @xmath148 with a free group action . recall that a principal bundle @xmath148 is trivial if there exists a global section , i.e. a smooth map @xmath169 from the base manifold @xmath63 into@xmath148 such that each group orbit @xmath170 in @xmath148 intersects @xmath171 exactly once and the projection @xmath172 satisfies @xmath173 . this condition corresponds to the existence of a moving frame . specifically , a global section @xmath174 and associated moving frame @xmath175 are related by the equality @xmath176 for all @xmath151 . a global section determines a flat connection , namely the connection that assigns to any point @xmath177 the subspace @xmath178 . if we introduce the notation @xmath179 to denote the right trivialization of the linearization of a map @xmath180 , i.e. @xmath181 for any @xmath151 and @xmath182 , then @xmath183 is the connection form of the connection determined by @xmath174 . the rotation group @xmath52 acts transitively on @xmath0 and freely and transitively on the unit tangent bundle @xmath184 . the map @xmath185 taking @xmath186 to the orthogonal matrix with columns @xmath187 is a ( left ) moving frame with associated connection form @xmath188 where @xmath189 , with @xmath190 and @xmath191 . ( here @xmath192 denotes the canonical projection . ) note that we will regard @xmath34 both as a tangent vector to the sphere at @xmath124 and as a unit vector in @xmath4 . moving frames can be extended to manifolds with nonfree actions as follows : a ( smooth ) map @xmath193 is a ( left ) _ partial moving frame _ if @xmath194 satisfies @xmath195 for all @xmath152 and @xmath159 . a _ partial moving frame _ on a submanifold @xmath196 of a manifold @xmath63 with a @xmath89 action is a map @xmath197 satisfying ( [ partial moving ] ) for any @xmath198 and any @xmath152 such that @xmath199 . the trivialized linearization @xmath200 of a partial moving frame @xmath193 is a partial connection form , with associated slip map @xmath201 . we refer to the trivialized linearization @xmath200 as the partial connection form associated to the partial moving frame @xmath30 . if a group @xmath89 acts transitively on a manifold @xmath63 , then every group orbit is equal to the entire manifold . in this situation , a ( partial ) connection form @xmath153 assigns to each tangent vector a generator of that vector , so that @xmath202 for all @xmath203 and all @xmath159 . in particular , given a vector field @xmath114 on @xmath63 , the map @xmath204 , i.e. @xmath205 , satisfies @xmath206 for all @xmath159 . hence ( partial ) connection forms can be used to construct geometric integration schemes on manifolds with transitive actions . we are currently investigating the role of geometrically motivated choices of partial connection forms in the design of efficient geometric integration algorithms . as can be seen in the example discussed below , simple , natural choices of partial connection forms can lead to significant improvement in numerical performance . to illustrate the somewhat abstract geometric constructions described above , we now present a moving frame associated to the action of the rotation group @xmath52 on the unit tangent bundle @xmath207 of the sphere @xmath0 and an associated family of partial moving frames on @xmath0 . the partial connection form ( which , in this case , is simply a map from the sphere to @xmath4 ) of one of these partial moving frames yields the discretization error minimizing generators used in the versions of the forward euler method described in sections [ opt ] and [ rig_body ] . as we shall see , this partial connection form and the associated generators can be derived without the use of the expansion of the discretization error . any unit vector field @xmath208 on a submanifold @xmath63 of @xmath0 determines a partial moving frame @xmath209 on @xmath63 , with partial connection form @xmath210 the map @xmath211 associated to @xmath212 is @xmath213 , where @xmath214 denotes the angle between @xmath215 and @xmath216 . given an ode @xmath217 , we can set @xmath218 on some set @xmath219 containing no equilibria ( i.e. zeroes of @xmath114 ) ; in this case , ( [ fuzzconn_sphere ] ) takes the form @xmath220 in particular , @xmath221 where @xmath222 denotes the geodesic curvature of the curve @xmath223 in @xmath0 . the partial connection form ( [ fuzzconn_sphere ] ) can be used to select the isotropy correction map @xmath80 used in ( [ simple_up ] ) . following ( [ fuzzconn_sphere3 ] ) , we set @xmath224 in our numerical implementation ( [ sig_choice ] ) , we approximate @xmath222 using the identity @xmath225 for a curve @xmath223 in @xmath0 with nonzero velocity @xmath226 , where @xmath227 is orthogonal to @xmath124 , and replacing @xmath228 and @xmath229 with finite difference approximations . recall that an update of the form @xmath230 where @xmath231 satisfies @xmath232 , determines a one step method on the lie group @xmath65 , where @xmath233 is determined by the generator and the selected scheme , and @xmath234 denotes the time step . given an ode @xmath235 on @xmath63 , we will construct updates of the form @xmath236 it is our objective to identify several classes of infinitesimal update maps @xmath106 , leading to algorithms of first , second , and fourth order . as was previously discussed , the normal component of the generator @xmath107 does not influence the solution curves of the original ode . thus , if we have a numerical algorithm of order @xmath237 , this component does not affect the solution up to order @xmath237 . however , it typically does appear in the higher order terms of the approximation , and theoretically a suitable choice of this component will reduce the discretization error . for the forward euler scheme an optimal choice of @xmath80 , in the sense that this choice minimizes the discretization error , also has a natural geometric interpretation . here we derive this map @xmath80 using a direct discretization error calculation ; in the following section we shall discuss various geometric considerations that can be used in the selection of the generator to be used in a lie group integration scheme . we consider consistent algorithms using standard methods on the tangential component of @xmath238 . the normal component is treated as a function of the tangential one ; we shall see that a component of the local discretization error at second order can be eliminated by a suitable choice of the normal component of the lowest order term in @xmath238 . for the sake of simplicity , we consider here only the lowest order case , in which the discretization error of a first order method is reduced by an appropriate selection of @xmath80 . this is a particular example of a more general result covering a large class of manifolds and higher order algorithms . ( see @xcite . ) work is in progress ( lewis , nigam , olver ) to possibly extend these or related results to an even larger class of systems , including the full discretized llg system . we begin by examining the flow @xmath239 of the ode @xmath240 on @xmath63 . this flow satisfies @xmath241 if the algorithmic update @xmath242 is given by @xmath243 for some map @xmath244 and @xmath245 agrees with the exponential map to second order ( e.g. @xmath246 is the cayley transform ) , then @xmath247 } + { \frac{1}{2}}\delta t^2 { \mathrm{skew } \left [ { \bf a}\right]}^2 + { \mathcal o}(\delta t^3)\right){\mathbf{m}}\\ & = & \left(i+\delta t \ , { \mathrm{skew } \left [ { \boldsymbol{\xi}}_1\right ] } + { \frac{1}{2}}\delta t^2 \left ( { \mathrm{skew } \left [ { \boldsymbol{\xi}}_2\right ] } + { \mathrm{skew } \left [ { \boldsymbol{\xi}}_1\right]}^2 \right ) + { \mathcal o}(\delta t^3 ) \right){\mathbf{m}}\\\ & = & { \mathbf{m}}+\delta t \ , { \boldsymbol{\xi}}_1 \times { \mathbf{m}}+\frac{\delta t^2}{2}\left({\boldsymbol{\xi}}_1\times({\boldsymbol{\xi}}_1\times { \mathbf{m}})+\mathbf{{\boldsymbol{\xi}}}_2\times { \mathbf{m}}\right ) + { \mathcal o}(\delta t^3 ) . \end{aligned}\ ] ] we now derive conditions on the terms @xmath248 , @xmath249, in the series expansion of @xmath106 yielding algorithms of increasingly high order . the consistency condition for @xmath250 is @xmath251 , where @xmath252 denotes component - wise projection onto the orthogonal complements of the component vectors of @xmath42 , i.e. @xmath253 , @xmath254 . if @xmath250 is consistent , then , setting @xmath255 , the local discretization error is @xmath256 the algorithm is thus second order accurate iff @xmath257 in our geometric version of the forward euler method with @xmath258 , @xmath259 for @xmath260 ; thus this method will not be second order . however , we are free to choose @xmath261 so as to satisfy the second equality in ( [ second_order ] ) , e.g. @xmath262 where @xmath263 is some first order approximation to @xmath264 ( e.g. , a discrete difference approximation ) , yielding a discretization error minimizing member of the family of algorithms with @xmath265 analogous expansions can be used to minimize the discretization error of higher order methods . however , the symbolic calculation of such expansions for high order schemes is , at present , relatively laborious and does not seem tractable for systems such as the llg equations , in which the generator @xmath266 is determined in part ( the demagnetization field ) by a nonlinear pde . using ( [ gen_update ] ) , we now define geometric one - step methods that are natural analogs of the standard explicit and implicit euler methods : @xmath267 where @xmath268 denotes the solution of the implicit equation @xmath269 and the scalar functions @xmath80 and @xmath270 are as yet unspecified . the numerical results presented in sections [ gen_crap ] and [ num_results ] illustrate the effect of the parameter @xmath80 on the discrete trajectories determined by the forward euler algorithm when applied to rigid body dynamics and the llg micromagnetism model . we shall see that in the rigid body system , @xmath80 satisfying ( [ movingsigma ] ) yields second order accuracy in energy tracking , and thus second order orbit capture for this conservative system . in the micromagnetics simulations , where damping plays a crucial role in the long term dynamics , large values of @xmath80 cause the trajectories to sharply diverge from those of the ordinary forward euler ; however , the final state is the same . a closer look at the llg equation shows that a larger value of @xmath80 corresponds to the inclusion of more precession in the trajectory . these numerical results clearly show that different choices of the parameter @xmath80 lead to significantly different numerical trajectories and thus motivate the search for an `` optimal '' value of @xmath80 . in section [ gen_crap ] we describe a general geometric approach to selecting values for @xmath80 ; in section [ opt ] we show that when used with the forward euler method , this choice of @xmath80 minimizes the discretization error . ( see @xcite for a description of this approach for more general manifolds . ) we consider four second order methods modeled on the classic heun ( rk2 ) algorithm . in the first , we use the ` default ' generator @xmath266 in the heun method , i.e. @xmath271 the second method , @xmath272 , is entirely analogous , but with @xmath266 replaced with the orthogonal generator @xmath273 , where @xmath274 and hence @xmath275 , @xmath254 . in the third method , the infinitesimal rotation determined by applying the heun method to the default generator @xmath266 is modified by addition of an appropriate isotropy element to yield a higher order of orbit capture ; specifically , @xmath276 the fourth algorithm is analogous , but with @xmath277 replaced by @xmath272 and @xmath278 replaced by an appropriate function @xmath279 . ( the function @xmath279 is a rational function in @xmath124 and the components of @xmath266 , but is significantly more complicated than @xmath278 . ) note that isotropy plays a role both in the choice of the generator and in the selection of a correction term . in this subsection , we describe two families of fourth order lie group integrators on @xmath64 . we emphasize that these algorithms map a point @xmath280 _ exactly _ into @xmath64 ; they are fourth order accurate in the sense that they approximate the true trajectories within @xmath64 to fourth order . a direct implementation of the classic fourth order runge kutta method to a vector field on @xmath64 fails to maps exactly into the manifold @xmath64 , while application of the classical rk4 method to the generator of the flow , followed by application of the exponential map and the group action necessarily yields an update in @xmath64 , but typically does not give a fourth order approximation of the true flow . the generator must be modified to account for the trivialization of the tangent bundle of the group ; this modification can be implemented either before or after the stages of the runge kutta method are computed and averaged . the first method we use is the rkmk4 method , a runge kutta style method due to munthe kaas ( @xcite , @xcite ) in which each stage of a traditional rk4 method is modified so that the resulting generator , followed by ( algorithmic ) exponentiation and application of the group action to the manifold , yields a fourth order method . the second method utilizes a series expansion of the generator along the true flow , followed by a single modification to account for the trivialization of the tangent bundle of the group @xmath65 , again followed by ( algorithmic ) exponentiation and application of the group action . we implemented the rkmk4 method @xcite , using the cayley transform rather than the true matrix exponential . to implement a lie group integrator for ( [ discretesystem ] ) using the cayley transform , we make use of the fact that , for sufficiently small @xmath10 , there is a function @xmath281 satisfying @xmath282 differentiating ( [ use_cay ] ) with respect to @xmath10 , we obtain @xmath283 where the map @xmath284 is the right trivialization of the tangent map of the cayley transform . hence @xmath285 and @xmath286 are related by @xmath287 which is equivalent to @xmath288 for some function @xmath80 . the initial condition for ( [ use_cay ] ) is @xmath289 . the map @xmath290 satisfies @xmath291})\ ] ] and @xmath292 } + \frac{1}{4}{\mathbf{f}}\ , { \mathbf{f}}^t.\ ] ] the cayley version of the rkmk4 method is essentially the conventional rk4 applied to ( [ big ] ) : having found @xmath293 at time @xmath294 , we construct the update @xmath295 where @xmath296 and @xmath297 for more details on this method , see @xcite . the second method involves a series expansion with respect to time of the generator @xmath298 , where @xmath239 denotes the exact flow at time @xmath10 . given the generator @xmath286 , this expansion is computed by iteratively symbolically differentiating @xmath298 and then substituting @xmath299 for @xmath300 . the third order approximation to @xmath298 is then modified to take into account the trivialization of the tangent bundle of @xmath52 and the action of @xmath52 on @xmath0 . the specific expressions for this modification for the rigid body equations on the sphere are given in [ rig_body ] . given the implicit and highly nonlinear nature of the llg equations , symbolic calculation of the derivatives of @xmath286 for this system seemed excessively complicated ; hence we did not implement this algorithm for the llg system . we now apply the results outlined above to a simple and familiar system , the reduced rigid body equations on the sphere . given a positive definite symmetric three by three matrix @xmath301 , define the vector field @xmath302 on @xmath0 . this is a hamiltonian system with respect to the kostant kirillov - souraiu symplectic structure @xmath303 and hamiltonian @xmath304 the system ( [ rigid_body ] ) is the symplectic reduction of the free rigid body equations on @xmath305 ; more concretely , it is the restriction of euler s equation for the body angular momentum to the unit sphere . ( since the norm of the body momentum is preserved by the dynamics of euler s equation , all spheres centered at the origin are invariant submanifolds . ) the conservative nature of this system makes it particularly easy to measure the error in orbit capture ; if the body is triaxial , i.e. the eigenvalues @xmath306 , @xmath307 , @xmath308 of the inertia tensor @xmath301 are distinct , the level sets of the hamiltonian ( [ rb_ham ] ) exactly determine the orbits of the system . thus in this situation the error in the orbit is a function of the fluctuation in the energy . as the numerical results given in tables 14 demonstrate , geometric integration techniques yield efficient , accurate orbit capture for the reduced free rigid body , with good performance even for very large time steps . note that the same randomly generated initial conditions and inertia tensors are used in all of the numerical simulations . if the rigid body is axisymmetric , then all true trajectories consist either of equilibria ( the ` poles ' and the ` equator ' ) or of steady rotations in the plane of symmetry . note that in this situation , even an exactly energy preserving scheme may allow drift across the family of one point orbits along the equator . however , all of the methods considered here detect equilibria as such . thus even in the axisymmetric case , we can use the energy to monitor orbit capture . we shall see that for some of the algorithms considered here , there are significant differences in performance on triaxial and axisymmetric bodies . symmetries play a crucial role in algorithm design and analysis ; see , e.g. @xcite . however , we shall not explore those issues in any detail here . we now consider implementations of the families of algorithms described in [ sectionalgo ] for the rigid body equations . we take as our default generator the body angular velocity @xmath309 . we first consider three first order methods , with infinitesimal updates * @xmath310 * @xmath311 * @xmath312 , where @xmath313 and @xmath314 are given with respect to an eigenbasis of @xmath301 by @xmath315 for any cyclic permutation @xmath316 of @xmath317 . [ plot_errors ] .maximum energy error over the trajectories given in figure [ traj_plots ] . [ cols="<,^,^,^",options="header " , ] we implemented the four different versions of the heun method given in section [ second_ordersection ] for the rigid body system . although the heun methods @xmath318 and @xmath319 described here are only second order accurate , the ( local ) discretization error in the energy is fourth order in the time step . the isotropy corrected versions used here take the form @xmath320 where @xmath321 for any cyclic permutation @xmath322 of @xmath317 and @xmath323 where @xmath279 is another , significantly more complicated , rational function in @xmath124 and the components of the inertia tensor . the isotropy corrections @xmath278 and @xmath279 given above determine algorithms yielding fourth order energy capture . if the body is axisymmetric , @xmath324 preserves the energy to fifth order . as table [ random_errors2 ] shows , some of these algorithms appear to have better _ global _ energy capture than the single step discretization energy error analysis ( which we carried out symbolically using _ mathematica _ ) would suggest . plots of the energy errors in the sample integrations , with randomly generated initial conditions and inertia tensors , show that the energy oscillates about a very slow drift away from the correct value . note that the energy correction term for the fourth order symbolic expansion method using the orthogonal generator is identically zero if the body is axisymmetric ; hence the results generated by @xmath325 and @xmath326 coincide in this case . we consider six fourth order geometric methods . four utilize a series expansion for the generator along a solution curve , while the other two use the rkmk4 algorithm ( with the cayley transform as the algorithmic exponential ) . using the cayley transform , the map @xmath327 determined by the default generator for the rigid body system on @xmath0 is given by @xmath328 where @xmath329 . the corresponding algorithm for the rigid body using the orthogonal generator is @xmath330 the infinitesimal updates @xmath327 and @xmath325 can be modified by the addition of an appropriate multiple of the argument @xmath124 to yield an additional order of energy , and hence orbit , capture . the scalar correction functions , which are rational functions of the components of @xmath124 and the inertia tensor , were determined by symbolic calculation . in this section , the geometric integrators developed in section [ sectionalgo ] are applied to the landau - lifshitz - gilbert equations of micromagnetics . the exact solution of this system is typically not available to us ; indeed , it is the lack of precise analytical results for comparison which makes numerical micromagnetics a challenging field . in our examples , we chose the largest possible time steps for a given method that would lead the system to the solution computed by a higher order method ( within prescribed tolerance ) . as mentioned earlier , numerical micromagnetics has attracted much attention in the mathematical community , for several reasons . in this paper we are focussing on the time - stepping aspect of the problem . the application of geometric integration techniques in this context is relatively new , see for example , @xcite . recently , another technique which modifies existing integrators was developed for numerical micromagnetics , @xcite . this new integrator is of the `` step - and - project '' class , but is stable . and with the geometric forward euler ( @xmath331 ) with optimal @xmath80 are almost identical ; if we assume @xmath332 ( an arbitrary choice ) , the trajectory precesses more before reaching the final point . the figure on the right shows the evolution of the optimal @xmath80 . the applied field is uniform and weak , specifically , @xmath333.__,title="fig:",width=192,height=240 ] and with the geometric forward euler ( @xmath331 ) with optimal @xmath80 are almost identical ; if we assume @xmath332 ( an arbitrary choice ) , the trajectory precesses more before reaching the final point . the figure on the right shows the evolution of the optimal @xmath80 . the applied field is uniform and weak , specifically , @xmath333.__,title="fig:",width=192,height=240 ] and by the geometric forward euler method with optimal @xmath80 and time step @xmath331 are almost identical ; if we set @xmath334 ( an arbitrary choice ) in the geometric forward euler method , the trajectory precesses more before reaching the final point . the figure on the right shows the evolution of the optimal @xmath80 . the applied field is uniform , @xmath335.__,title="fig:",width=192,height=240 ] and by the geometric forward euler method with optimal @xmath80 and time step @xmath331 are almost identical ; if we set @xmath334 ( an arbitrary choice ) in the geometric forward euler method , the trajectory precesses more before reaching the final point . the figure on the right shows the evolution of the optimal @xmath80 . the applied field is uniform , @xmath335.__,title="fig:",width=192,height=240 ] we were particularly interested in the behavior of the free parameter @xmath80 which appears in the geometric time - stepping algorithm ( [ geom_eqs ] ) . in the rigid body case the parameter can be chosen to improve energy conservation . here the system is dissipative and a criterion for the selection of @xmath80 is not immediately obvious . for the forward euler implementation , we can derive a relatively simple expression for a function @xmath80 that minimizes the discretization error . for higher order methods , analogous functions can be described in terms of the series expansions of the true and algorithmic flows , but the cost of computing these expansions , particularly for systems such as the llg equations , rapidly becomes prohibitive . work is in progress to determine computationally tractable criteria for the selection of the isotropy component for higher order methods . we describe a model problem for the llg , for which the analytical solution was particularly simple . recall that the llg for the magnetization @xmath18 is given by ( [ llg ] ) , which we recall here for convenience : @xmath336 @xmath337 in the experiments that follow , we set @xmath338 and vary the applied field . these parameter values are not taken from actual physical data , and were selected solely for purposes of illustration . the saturation magnetization was @xmath339 . we wish to construct a one - dimensional example where the computation of the demagnetizing energy would be simple . to this end , we assume the sample is contained in the infinite slab @xmath340 . we assume the magnetization @xmath341 , ie . , the only variation in the magnetization is along the x - direction . therefore , @xmath342 we assume that there are 100 individual spins uniformly distributed along @xmath343 $ ] . these spins interact with each other through the exchange and demagnetizing fields . this example is admittedly a simplistic one ; the true equilibrium solution for it can easily be found using analytical techniques . therefore , the stopping criterion used was a comparison with the exact final equilibrium point . we see that the geometric integrators take trajectories which respect the point - wise constraints on the magnetization ; conventional integrators do not . thus , the paths traversed by these integrators will be different , as is seen in figures [ fig:1 ] and [ fig:2 ] . as the step - size is shrunk more and more , the trajectories will converge . in the first set of numerical experiments , we implemented the geometrical analog of the forward euler algorithm for the llg equation . we then tracked the evolution of the parameter @xmath80 given by ( [ movingsigma ] ) , an expression derived through arguments of discretization error minimization . we approximated the acceleration @xmath344 of trajectories @xmath345 of ( [ discretesystem ] ) using a one - sided discrete approximation of the derivative of @xmath346 . figures [ fig:1 ] and [ fig:2 ] describe the trajectories followed by the over - damped llg system ( without the larmor precession term ) for two different applied fields.to address the issue of overall computational expense , we ran both a geometric forward euler algorithm and the conventional forward euler algorithm on the example introduced above . for each algorithm , we decreased the time step @xmath234 until the trajectories converged to within @xmath347 . we also required that the final equilibrium point corresponded to that computed by a fourth - order runge - kutta method with time step @xmath348 , to within a relative error of @xmath349 . the geometric forward euler method yielded trajectories which converged , for this example , with time steps of @xmath331 and a cpu time of 1.77 seconds . the usual forward euler required a time step @xmath348 , with a cpu time of 3.88 seconds , to get similar behaviour . in addition , while the pointwise norm of @xmath42 was conserved to machine accuracy by the geometric integrator , the standard forward euler algorithm caused @xmath350 to increase to 1.001183806 times its usual value by the end of the run . as a consequence , the trajectories traversed by the geometric and the usual algorithms differed , though they ended at the nearly same place . as we are only interested in the final equilibrium state of the system , we see the obvious merit of using the geometric integrator we can obtain accurate final states while using much larger time steps . we see the effect of varying the scalar functions @xmath80 on the trajectories is that of changing the amount of precession in the trajectory . we notice certain trends in the optimally chosen function @xmath351 in figures [ fig:1 ] and [ fig:2 ] , and we shall investigate the relationship of these trends to the physical processes occurring at the same times in future work . and damping parameter @xmath352 . the trajectories were computed using the usual forward euler method with time step @xmath348 and the geometric forward euler method with time step @xmath331 . here the trajectory taken by the geometric method differs appreciably from those of the usual method , though the final states appear to be similar . the lefthand plot shows the evolution of one point in the ferromagnetic sample ; the righthand plot shows the evolution of the optimal sigma at that point.__,title="fig:",width=192,height=240 ] and damping parameter @xmath352 . the trajectories were computed using the usual forward euler method with time step @xmath348 and the geometric forward euler method with time step @xmath331 . here the trajectory taken by the geometric method differs appreciably from those of the usual method , though the final states appear to be similar . the lefthand plot shows the evolution of one point in the ferromagnetic sample ; the righthand plot shows the evolution of the optimal sigma at that point.__,title="fig:",width=192,height=240 ] in figure [ fig:4 ] , we implemented the code for the full llg system , including the larmor precession . the applied field is uniform , @xmath353 . the damping parameter @xmath20 was set to a low value , specifically @xmath352 . the trajectories followed by the usual forward euler and @xmath348 and by the geometric forward euler with @xmath331 and optimal @xmath80 diverge appreciably , yet end at the same final state . the drift of the norm is now clearly visible ( see figure ( [ fig:5 ] ) ) . the usual forward euler trajectory moves off the unit sphere in the standard euler integration , while the geometrically integrated one does not . we see that the optimal @xmath80 now varies more ( figure [ fig:4]b ) . in the next set of experiments , we implemented the geometrical versions of the heun algorithm , derived in section [ sectionalgo ] . we did not have an analytical expression for the optimal choice of @xmath80 . therefore , we ran the experiments for several constant values of this parameter , and computed the order of convergence of the algorithm in @xmath234 . the results were interesting , and rather striking . as @xmath80 is varied , the order of convergence changes for the naive choice of generator . what should be noted is that the geometric algorithm appears to converge more rapidly than a conventional heun method ; the order of convergence was @xmath355 , as was borne out in repeated experiments . we now present experiments with a fourth order method derived in section [ sectionalgo ] . lacking an analytical expression for the optimal choice of @xmath80 , we ran the experiments for varying constant values of this parameter , and computed the order of convergence of the algorithm in @xmath234 . as a function of time step , @xmath356 . here we see @xmath357 convergence . on the right : norm of the error of @xmath42 as a function of time step , with @xmath358 . here we see @xmath359 convergence . _ , title="fig:",width=288,height=288 ] as a function of time step , @xmath356 . here we see @xmath357 convergence . on the right : norm of the error of @xmath42 as a function of time step , with @xmath358 . here we see @xmath359 convergence . _ , title="fig:",width=288,height=288 ] in figure [ fig:9 ] we track @xmath350 over @xmath360 $ ] with a time step of @xmath361 . the classical rk4 method without projection shows a drift in the norm ; this drift is of the order of @xmath362 , i.e. , @xmath363 . the lie group integrator , on the other hand , shows no drift ( up to machine precision ) . as we vary @xmath80 , we observe that the rate of convergence of the algorithm varies , see ( figure [ fig:10 ] ) . again , there is clearly some optimal value of this parameter . this behavior is even more pronounced for the rkmk4 method than for the heun method . * acknowledgements : * the authors would like to thank the referees for their insightful and helpful suggestions . we also thank the institute for mathematics and its applications , where the bulk of this research was conducted . d. lewis , conserving and approximately conserving algorithms , in : dynamics of algorithms ( the proceedings of the i m a workshop on dynamics of algorithms ) , i m a volumes in mathematics and its applications , vol . 118 , springer verlag , new york , 2000 .

despite significant advancement over the last two decades , the galactic center magnetosphere is still not well understood ; however , determining the geometry and strength of the magnetic field in this region is crucial to developing an understanding of the dynamics of the galactic center . one reason that the magnetosphere is of interest lies in the galactic center s role as an active galactic nucleus . in the central engines of galaxies , magnetic fields are believed to be important in angular momentum transport and jet dynamics . the most striking evidence for the existence of magnetic fields in the galactic center is the existence of the non - thermal filaments ( ntfs ) of the galactic center radio arc ( gcra ) @xcite . radio polarization measurements have confirmed the synchrotron nature of the emission from these filaments and support the notion that the filaments trace magnetic field lines @xcite . almost all of the confirmed ntfs that have been discovered in the galactic center region are aligned with their long axes within 20@xmath1 of perpendicular to the plane of the galaxy @xcite . the presence of these filaments has led to the idea that these filaments trace the inner part of a dipole or _ poloidal _ magnetic field . in many cases ntfs are observed to be interacting with galactic center molecular clouds . the best example of this is in the center of the gcra where the 25 km s@xmath0 molecular cloud associated with g0.18 - 0.04 ( the sickle ) is coincident with the filaments @xcite . the lack of observed distortion of the filaments allows one to set a lower limit for the strength of the magnetic field . this argument yields @xmath2 @xcite . the association of molecular clouds with ntfs gives additional evidence for the dynamical importance of these fields and has motivated the idea that the generation of the relativistic electrons required to create these ntfs could occur as a result of a molecular cloud / magnetic flux tube interaction . the specific mechanism thought to be responsible for the generation of these relativistic electrons is magnetic reconnection between the fields in the flux tube and those in the cloud ( @xcite ; for a review , see @xcite ) . magnetically aligned dust grains are known to emit polarized radiation in the far - infrared and submillimeter . thus , polarimetry at these wavelengths has proven to be a reliable technique for measuring the projected , line - of - sight integrated magnetic field configuration . previous polarimetry data have shown that the field in the dense molecular clouds is not generally consistent with the poloidal field traced by the filaments . @xcite have shown that in the molecular cloud m-0.13 - 0.08 , the field is nearly parallel to the galactic plane , indicating that gravitational rotation and infall have sheared out the field into a _ toroidal _ configuration . in addition , 60 @xmath3 m polarimetry of the molecular cloud associated with the sickle indicates that the field in this region is parallel to the galactic plane @xcite . @xcite have constructed a model that connects poloidal and toroidal fields . because the magnetic flux is frozen into the matter , differential rotation and infall can shear an initially poloidal field into a toroidal one in sufficiently dense regions of the galactic center . @xcite have shown that on large scales , the dust in the central molecular zone ( cmz ) of the galaxy is permeated by a field that is oriented in a direction parallel to the plane of the galaxy . they compare their observation to the spatial distribution of the direction of faraday rotation in the galactic center and show that these two different techniques for sampling fields yield results that are consistent with such a model . we extend the ideas that underlie the above model to the inner 30 parsecs of the galactic center . when resolved with sub - arcminute resolution in the submillmeter @xcite , this region exhibits a clumpiness resulting from spatial density variations . we find that in less dense regions , the magnetic field that permeates the dust is poloidal , consistent with that found in the filaments . in these regions , the matter density is too low for gravity to shear the field into a toroidal configuration . in the dense molecular clouds , however , the field appears to be generally toroidal , indicating that in such regions , the energy density of gravity dominates that of the permeating magnetic field . section 2 describes the observations . in section 3 , the data are discussed . finally , conclusions are presented in section 4 . the data were obtained using the university of chicago 32-pixel polarimeter , hertz @xcite at the caltech submillimeter observatory on mauna kea , hawaii in april of 2001 . the beam size of hertz is 20fwhm with a pixel spacing of 18 . the central wavelength of the hertz passband is 350 @xmath3 m with a @xmath4 of 0.1 . the observing technique used with hertz involves a combination of high frequency `` chopping '' in which the array footprint is alternately pointed at the source and a reference position 6 away in cross - elevation and low frequency `` nodding '' during which the source is alternately placed in the right and left beams of the telescope . one undesirable complication that this observational scheme introduces is the problem of polarized flux in the reference beam . this problem is discussed in detail in [ sec : refbeam ] . for a full description of the observing and analysis procedures see @xcite and @xcite . during the observations , a total of 158 new polarization measurements were obtained having a polarimetric signal - to - noise greater than 3 . these data are given in tables 1 through 7 . the magnetic field vectors inferred from these data are shown in blue in region iii of figure [ fig : scuba ] . the black vectors in this region are those from @xcite . also shown in this figure are 100 @xmath3 m data from the kuiper airborne observatory @xcite of the arched filaments ( region ii ) and 60 @xmath3 m data @xcite of the sickle ( region i ) . the 100 @xmath3 m and 50 @xmath3 m data have beam sizes of 35@xmath5 and 22@xmath5 , respectively . the contours trace 850 @xmath3 m flux @xcite . major molecular features are labeled . the molecular cloud m-0.13 - 0.08 ( commonly called the 20 km s@xmath0 cloud ) has an elongated shape , and its long axis is oriented at a shallow ( 15@xmath1 ) angle to the galactic plane ( see fig . [ fig : scuba ] ) . the cloud is thought to be located in front of the galactic center since it appears in absorption in the mid - infrared @xcite . in projection , the long axis of this cloud points in the direction of sgr a*. this fact suggests that this cloud is undergoing a gravitational shear as it falls toward sgr a*. this is supported by other observations including a high velocity gradient along the long axis @xcite , increasing line widths and temperatures along the cloud in the direction of sgr a * @xcite , and an extension of the cloud detected in nh@xmath6 emission to within 30@xmath7 of sgr a * in projection @xcite . @xcite have discovered that the magnetic field structure is such that the field is parallel to the long axis of the cloud . they point out that for gravitationally sheared clouds , a consequence of flux - freezing is that regardless of the initial configuration of the field , the field will be forced into a configuration in which it is parallel to the long axis of the cloud . in addition , these authors note that this is not only true for the dense material inside of the cloud , but also for the more diffuse molecular material that belongs to the ambient region . they note that the southern end of this cloud exhibits a flare both morphologically and in magnetic field structure . they suggest that this flare is a connection to the large scale poloidal field traced by the ntfs . our 7 new vectors in this region provide additional support for this interpretation , but a more extensive polarimetric mapping of this cloud is necessary before a complete interpretation is made . the h ii region g0.18 - 0.04 ( commonly called the `` sickle '' ) can be seen in 20 cm thermal emission in figure [ fig:20 cm ] . it is believed to be the ionized surface of a molecular cloud that is interacting with the radio arc @xcite . its molecular counterpart can be seen in figure [ fig : cs]b . it has been suggested that this interaction is responsible for producing the relativistic electrons necessary to light up the radio arc via magnetic reconnection @xcite . the relationship between the geometry of the cloud and its magnetic field structure is similar to that of m-0.13 - 0.08 . the cloud is elongated parallel to the plane of the galaxy and its magnetic field as inferred from far - infrared polarimetry is parallel to the long axis . these similarities to m-0.13 - 0.08 suggest a similar origin of the field geometry . one difference between this cloud and m-0.13 - 0.08 is that in the case of the former there is no observed flaring of the field . in fact , the stark 90@xmath1 difference between the toroidal field observed in the molecular material and the pristine poloidal geometry of the superposed radio arc suggest that there is little connection between the toroidal field and the poloidal field in this vicinity . there are several possibilities for why this might be the case . first , the polarimetry coverage may be too limited in this region . more data taken in surrounding region may reveal a connection to a poloidal field . second , this cloud may be more evolved than m-0.13 - 0.08 , and so we may be observing a magnetic field that has had enough time to shear into a more parallel configuration than that of m-0.13 - 0.08 . finally , because this molecular cloud is farther from the dynamical center of the galaxy than is m-0.13 - 0.08 , the tidal forces on it are weaker , thus allowing for a more uniform shearing along its length . m+0.11 - 0.08 and m+0.07 - 0.08 are two peaks of a molecular cloud complex that , like m-0.13 - 0.08 and the molecular cloud associated with the sickle , is elongated nearly parallel to the plane with its long axis pointed toward sgr a*. the radial velocity of this complex has been measured to be @xmath850 km s@xmath0 ( see fig . [ fig : cs]c ) . the inferred magnetic field configuration for this complex displays a geometry that is quite unique . the field appears to tightly wrap around the southern and eastern edges of this cloud . the implication of this curvature of the magnetic field is that the cloud is moving toward the sgr a * region in projection and is sweeping up magnetic flux from the intercloud medium . the direction of travel inferred from the polarization data indicates that in projection , this cloud is moving toward a region dominated by a poloidal field ( see northeast side of m-0.02 - 0.07 in fig . [ fig : scuba ] ) . as the cloud moves through the less dense intercloud medium , this poloidal flux then is wrapped around the cloud into a toroidal configuration as seen along the eastern edge of the cloud . the result is an observed transition between poloidal and toroidal fields . interior to this cloud complex , there are few measurements ; however , those that exist hint at a field that is parallel to its long axis , similar to that of m-0.13 - 0.08 and the molecular cloud associated with the sickle . the 350 @xmath3 m polarimetric observations of m-0.02 - 0.07 , the cnd , and co+0.02 - 0.02 provide evidence for a magnetic field structure that is contiguous with that derived from the 100 @xmath3 m polarimetric observations of the arched filaments to form an `` x - shaped '' feature that we will refer to as the `` x polarization feature . '' the feature extends ( in the coordinate system of fig . [ fig : scuba ] ) from ( -1 , 13 ) down through ( 5 , 0 ) . the organization of the magnetic field vectors appears to be on scales ( 30 pc ) significantly larger than those of typical galactic center molecular clouds ( 5 - 10 pc ) . this fact reinforces the notion that the vectors trace a single global field that is subject to environmental forces . it is likely that the x polarization feature is a line - of - sight superposition of regions of poloidal and toroidal fields . the orientation of the observed magnetic field at any given location depends on the relative polarized emission from poloidal and toroidal regions that intersect the line of sight . for example , the poloidal field is seen to dominate in the eastern part of the m-0.02 - 0.07 cloud and at the western edge of the arched filaments . the toroidal field dominates at the eastern edge of the arched filaments and around the cnd . the two fields mix in co+0.02 - 0.02 . another example of line - of - sight mixing of poloidal and toroidal fields occurs at the western edge of the arched filaments . here , there exist several magnetic field vectors oriented perpendicular to the plane that are in close proximity to the g0.08 + 0.15 , a ntf that traces the poloidal field . just to the north , the magnetic field vectors return quite abruptly to a toroidal configuration in the vicinity of the molecular features that correspond to the arched filaments . it is possible that the molecular material at @xmath9 km s@xmath0 ( see fig . [ fig : cs]a ) is displaced along the line of sight from g0.08 + 0.15 , and the net polarization observed consists of contributions from dust associated with the respective neighborhoods of these two features . to the south of g0.08 + 0.15 , the vectors again become indicative of a field distorted by gravity as they wrap around the molecular cloud visible in figure [ fig : cs]d . figure [ fig : pvf ] shows the relationship between galactic center polarization measured by hertz versus the 350 @xmath3 m flux measured by sharc , a 350 @xmath3 m photometer with a 15@xmath5 beam @xcite . a linear fit gives a slope of -0.67 . in this plot , we have included all points with a polarization signal - to - noise ( s / n ) greater than 3 and have corrected the polarization by @xmath10 in order to account for the systematic overestimation of polarization @xcite inherent in the conversion from @xmath11 and @xmath12 to @xmath13 and @xmath14 . figure [ fig : pvf ] shows a depolarization effect associated with an increase in 350 @xmath3 m flux . because of a concern over a selection effect that systematically excludes low polarizations at low fluxes , the test was repeated with a s / n cutoff of 1 . for this case , the slope decreases to -0.73 , thus exonerating figure [ fig : pvf ] of this effect . the test was repeated for the 11 individual hertz pointing corresponding to both this work and that of @xcite . a polarization versus flux comparison using the relative fluxes obtained simultaneously with the polarizations gives an average slope of -0.96@xmath150.32 . this number is the mean of the slopes for each of the 11 fields . the error is the standard deviation of these slopes . these numbers are similar to the values found by @xcite for various galactic molecular clouds . @xcite suggest three possibilities for the depolarization effect ( see also @xcite and @xcite ) . first , the lower polarization could be due to poor grain alignment or low polarization efficiency in the cores of these clouds . second , the depolarization could be a result of a geometrical cancellation of the front and back of a three dimensional field that is threaded through the optically thin cloud . finally , it is possible that the spatial structure of the magnetic fields in the cores of clouds is too small to be resolved by hertz s beam . the evidence for depolarization in the galactic center does not provide enough information to differentiate among the three proposed explanations ; however , it does indicate that the depolarization effect is observable in clouds that are a factor of @xmath810 larger than molecular clouds in the disk of the galaxy . historically the lack of information concerning the degree of grain alignment and relative strength of the line - of - sight component of the magnetic field have limited the ability of far - infrared and submillimeter polarimetry to provide information concerning the strengths of magnetic fields . we will show that in the galactic center the additional information concerning magnetic fields ( i.e. the ntfs ) can lead to a model - based estimate of a characteristic global magnetic field strength estimate . in studying the magnetic field in the central 30 pc of the galaxy , we revisit the model of @xcite as it applies to the galactic center . specifically , we keep in mind that the galactic center is clumpy and that some regions seem to contain a field that is poloidal while others contain a field that is toroidal . in the spirit of this model , we adopt a picture in which gravity is assumed to dominate the dynamics in regions of toroidal fields . conversely , in regions of poloidal fields , magnetic fields are assumed to be energetically dominant . figure [ fig : phiflux ] illustrates the dependence of the polarization angle on the flux measured by sharc @xcite . in this figure , the absolute deviation from a poloidal field ( @xmath16 ) is plotted against 350 @xmath3 m flux . this relationship shows that for regions having higher flux ( and thus higher column densities ) , the field is toroidal and the dynamics are gravity - dominated . for lower intensities , poloidal fields are observed indicating that there are regions of low density in which the dynamics are dominated by the energy density associated with the magnetic field . somewhere between these two extremes is a flux for which the energy density of the poloidal field equals that of the gravitational energy density . in order to proceed with the estimate , we will assume that this density corresponds to a measured angle @xmath17 . this choice is somewhat arbitrary . in order to find the flux corresponding to this angle , we perform various binnings as shown in figure [ fig : binned ] . linear fits to each of these four plots give a flux of 125 jy beam@xmath0 . to a crude approximation , this energy balance condition is represented by @xmath18 here , the kinetic energy density of material orbiting in the gravitational potential well of the galaxy is equated to the magnetic energy density . the key to this problem now becomes the estimate of @xmath19 and @xmath20 . the velocity , @xmath20 , can be estimated from typical cloud velocities @xcite and is expected to be between @xmath21 km s@xmath0 . the density can be determined as follows . the brightness one observes from a thermal source having an optical depth @xmath22 is @xmath23 where @xmath24 is the planck function . in the case of 350 @xmath3 m observations , the optical depth is generally small . for @xmath25 , @xmath26 sharc measures @xmath27 , the flux of the incoming radiation where @xmath28 is the solid angle subtended by a sharc array element . from these equations , it is possible to express the optical depth of the dust as a function of temperature and measured flux . @xmath29 alternatively , we can express the optical depth as a function of grain properties along the line of sight . once again we are working in the limit @xmath25 . we can imagine a column of dust along the line of sight that extends through the entire depth of the galactic center . the optical depth is proportional to the number density of dust grains along the line of sight ( @xmath30 ) . it is also proportional to the typical geometrical cross - section of each of the grains ( @xmath31 ) ; however , since the grain sizes are generally much smaller than the wavelenth of the radiation , the efficiencies of the grains for emitting , scattering or absorbing light are much lower than this blackbody approximation indicates . thus we write the optical depth as @xmath32 where @xmath33 is the emissivity of the dust grains and is generally much less than unity for submillimeter radiation . once the @xmath30 is found , the total dust mass observed by a sharc beam is @xmath34 here , @xmath35 is the distance to the source and @xmath36 is simply the physical size of sharc s beam at a distance @xmath35 . @xmath37 and @xmath38 are the density and volume of a dust grain , respectively . if we then make the appropriate substitutions and assume a gas - to - dust ratio , @xmath39 , we get the following expression for the total mass . @xmath40 putting in the appropriate numbers for 350 @xmath3 m radiation yields @xmath41 the density can be calculated by assuming a value for the depth of the dust layer ( @xmath42 ) . @xmath43 we use the following grain properties @xcite for our grain model . these are @xmath44 , @xmath45 , @xmath46 , and @xmath47 . in addition , pierce - price , et al . ( 2000 ) have used scuba to map the central molecular zone ( cmz ) at 450 and 850 @xmath3 m . they have found the dust temperature to be relatively uniform over the cmz and adopt a value of 20 k. with these numbers , one can get an estimate of the magnetic field strength as a function of velocity of the material and the thickness of the dust . @xmath48 based on cs measurements of the cmz @xcite , most molecular material has v @xmath49 150 km s@xmath0 . the cmz has a projected diameter of 200 pc . assuming cylindrical symmetry , this is approximately the scale of the material along the line of sight . because of the clumpy nature of the central 30 pc , this value @xmath42 is most likely overestimated , thereby making the above estimate of @xmath50 conservatively low . in performing differential measurements using chopping techniques , polarized flux in the reference beam positions is always a potential hazard . because of the extended nature of the dust in the galactic center , it is of particular importance to understand the possible effect of reference beam contamination on the data presented here . several attempts have been made to quantify this issue @xcite . the goal of this section is to assess the level of reference beam contamination in our data . note that throughout this analysis quantities pertaining to the `` reference beam '' such as @xmath51 and @xmath52 refer to the average of these quantities over the multiple reference beam positions used in the observations . since the quantitative result of this work centers on the relationship shown in figure [ fig : phiflux ] , we are concerned primarily with the polarization angle ( @xmath14 ) , and so we wish to estimate the maximum effect of polarized flux in the reference beam on @xmath14 . the difference between the source polarization angle ( @xmath53 ) and that measured ( @xmath54 ) is given by @xcite as @xmath55 . \label{eq : rberr}\ ] ] here , @xmath51 is the polarization in the reference beam , @xmath56 is the difference between the measured polarization angle and the polarization angle of the reference beam ( @xmath57 ) , and @xmath58 is the ratio of flux in the reference beam to that measured for a given source ( @xmath59 ) . alternatively , since @xmath60 , @xmath58 can be expressed as @xmath61 . here , @xmath62 is the intrinsic source flux . it is desirable to estimate the maximum effect the reference beam contamination could have on these measurements . specifically , we wish to understand how reference beam contamination could affect the results shown in figures [ fig : phiflux ] and [ fig : binned ] . to do this we need to find @xmath58 , @xmath63 , @xmath64 , and @xmath65 . we find from the scuba survey @xcite , the average ratio of the flux in the main beams of our six sources to that in the reference beams is @xmath66=1.9 . this leads to @xmath67 . the average 350 @xmath3 m polarization measured by hertz in the galactic center is 1.9 % . thus , we assign @xmath68 . finally , we need to make an estimate of @xmath64 and @xmath65 , the quantities describing the polarization properties of the reference beam flux . to do this , we note that @xcite have mapped the central molecular zone(cmz ) in 450 @xmath3 m polarimetry . they find the average polarization into a 6@xmath69 beam is 1.4 % and that the field is uniformly toroidal . because they use a 30@xmath69 chop throw , their reference beam is well off of the cmz and the contamination should be negligible . ignoring any wavelength dependence , we set @xmath70=0.014 and @xmath71=-58.4@xmath1 , the polarization angle that corresponds to a toroidal field . the difference between the measured and actual polarization angle ( @xmath72 ) under the above assumptions is plotted versus measured polarization angle ( @xmath54 ) in figure [ fig : ccurve ] . the maximum error in ( @xmath72 ) is @xmath1520@xmath1 . note that the amplitude of this error is independent of the chosen value of @xmath71 since the curve simply shows the error in the measured polarization angle as a function of the difference between the measured polarization angle and the reference beam polarization angle . ( in this case , we have set the reference beam polarization angle to an `` arbitrary '' value . ) the potential error induced by reference beam contamination is systematically about twice that of our maximum allowable statistical error . ( a signal - to - noise ratio of 3 implies @xmath73 . ) though this is potentially large , it is not large enough to cause poloidal fields to be measured as toroidal and vice - versa . therefore , it is unlikely that the relationship shown in figures [ fig : phiflux ] and [ fig : binned ] is caused by reference beam contamination . we can quantitatively estimate the potential effect of reference beam contamination on the data . this histogram in figure [ fig : ccurve ] shows the number of measurements having a polarization angle in each of the 10@xmath1 bins . as an unfortunate coincidence , the error is not symmetric with respect to our data and thus it is expected that there will be a bias in our field strength estimate . we have applied this correction to the data and the resulting @xmath14 v. @xmath74 plots have the same basic appearance ; however the equilibrium point once the correction is applied is 168@xmath152 jy ( sharc beam)@xmath0 . this implies a 30 % increase in our magnetic field strength estimate . it must be noted that because the large scale field in the galactic center has been found to be toroidal by @xcite , it is possible that a significant amount of dust is present along the line of sight to the central 30 pc that is permeated by this toroidal field and will emit polarized radiation accordingly . in this case , the large scale contribution to the polarized flux can be modeled by a uniform sheet of polarized flux . in this case , chopping can be an advantage in observing magnetic fields in the central 30 pc as it removes the contribution from a uniform field in the foreground and background dust . this model may also explain the lack of polarization measurements by @xcite in the central 30 pc in regions where hertz sees a poloidal field . because of the 6chop , hertz may be sampling a smaller volume along the line of sight than that of @xcite . in the latter case , line - of - sight superposition of the uniform sheet and the polarized emission detected by hertz may cancel , resulting in unpolarized radiation . as described above , submillimeter and far - infrared polarimetry provides an additional method for estimating the magnetic field strength at the galactic center , a quantity that has proven to be controversial . @xcite have argued that the magnetic field strength must exceed @xmath81 mg in order to maintain the linearity of the ntfs . on the other hand , @xcite have derived fields of 10 - 100 @xmath3 g for the arched filaments and the radio arc using faraday rotation measurements . using equipartition arguments , @xcite have found a field of the same order in the plumes that make up the extension of the radio arc on the eastern side of the galactic center lobe . more recently , zeeman splitting measurements have been done in the central 30 pc both using oh masers and h i emission @xcite . these studies show a line - of - sight field of a few mg . our estimate is in agreement with the notion that there is a pervasive mg field in the galactic center , thought obtaining a more accurate number will require a larger sample of polarization measurements with corresponding fluxes . it has been discovered that many of the ntfs in the galactic center are associated with molecular clouds @xcite . this has led to the suggestion that the source of the relativistic electrons in the ntfs is due to magnetic reconnection @xcite . the magnetic reconnection is believed to be precipitated by the collision of the cloud with a magnetic flux tube either by distorting the fields in the flux tube or by forcing these fields into contact with those in the cloud . in the galactic center , relatively diffuse molecular gas is supported by magnetic pressure ( fig . [ fig : model1]a ) ; however , this material is free to collapse along the field lines . as the gas collapses , gravitational energy becomes increasingly important . gravity can accelerate the cloud and the resulting differential motion between the material in the cloud and that of the ambient medium can begin to shear the magnetic field ( fig . [ fig : model1]b ) . this shear continues until the field is more toroidal than poloidal ( fig . [ fig : model1]c ) . note that the field external to the clouds is still quite poloidal because magnetic energy dominates the dynamics of this region . finally , if the gravitational energy is large enough , and the system is given the time to evolve , the oppositely - oriented fields in the center of the cloud may be squeezed together to enable reconnection . this process releases energy that can produce the relativistic electrons required to `` light up '' the filaments . submillimeter and far - infrared polarimetry in concert with radio and submillimeter photometric observations present a picture of molecular material of various sizes at various evolutionary stages according figure [ fig : model1 ] . from the polarimetry data in figure [ fig : scuba ] , we can see examples of each of the four panels in figure [ fig : model1 ] . the area to the north and east of m-0.02 - 0.07 is an example of the situation depicted in figure [ fig : model1]a . here , the molecular material is not very dense and the magnetic field is perpendicular to the galactic plane . an example of figure [ fig : model1]b is the molecular cloud complex containing m0.07 - 0.08 and m0.11 - 0.08 . here the field is seen to be shearing around the front edge of m0.07 - 0.08 , making a transition from poloidal to toroidal at the southern edge of the cloud . the 20 km s@xmath0 cloud ( m-0.13 - 0.08 ) is thought to be shearing out its magnetic field as it falls toward sgr a * @xcite . the flare seen at the southern end may indicate that it has not yet reached the point where magnetic reconnection is occurring and hence best matches figure [ fig : model1]c . the interaction of the sickle with the gcra gives the best example of figure [ fig : model1]d . here , the magnetic field in the molecular cloud is observed to be nearly perfectly aligned with the direction of both the long axis of the cloud and the galactic plane . filaments , namely those of the gcra , are observed , and appear to diffuse into g0.18 - 0.04 , the h ii region associated with this interaction . the clumpiness of the h ii region and the structure of the filaments themselves may stem from the irregularities of the interior of the cloud where reconnection takes place . it is possible that there are similar occurrences in places such as the arched filaments where there is also an observed transition from toroidal to poloidal fields adjacent to an ntf ( in this case , the northern thread ) . in order to test this idea with respect to other filaments and clouds in the galactic center , more complete polarimetric coverage of the central 30 parsecs is required . other scenarios for the explanation of the variety of fields in the central 30 pc can not yet be ruled out . winds produced by large explosions could be responsible for the poloidal fields seen in this region . however , there are several reasons to favor the association of the poloidal fields with a global field . first of all , the spatial proximity of the poloidal fields in the dust to the ntfs lends evidence to this association . second , the @xmath83 mg field strength we derive is consistent with lower limits of the field found in ntfs . third , to date we have not been able to locate any sources of a wind on large enough scales that couple to the magnetic field geometries observed . finally , we see transitions from poloidal to toroidal fields that correspond to dynamics of galactic center molecular clouds , indicating that these two fields have the same origin and that the initial configuration was poloidal . dowell , c. d. , lis , d. c. , serabyn , e. , gardner , m. , kovacs , a. , & yamashita , s. 1999 , in asp conference series 186 : the central parsecs of the galaxy , ed . h. falcke , a. cotera , w. duschl , f. melia , & m. rieke , 453465 pierce - price , d. , richer , j. s. , greaves , j. s. , holland , w. s. , jenness , t. , lasenby , a. n. , white , g. j. , matthews , h. e. , ward - thompson , d. , dent , w. r. f. , zykla , r. , mezger , p. , hasegawa , t. , oka , t. , omont , a. , & gilmore , g. 2000 , apj , 545 , l121 l c c c c c @xmath75 & @xmath76 & @xmath77 & @xmath78 & @xmath79 & @xmath80 + @xmath81 & @xmath76 & @xmath82 & @xmath83 & @xmath84 & @xmath85 + @xmath81 & @xmath86 & @xmath87 & @xmath88 & @xmath89 & @xmath90 + @xmath81 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 + @xmath96 & @xmath75 & @xmath97 & @xmath98 & @xmath99 & @xmath100 + @xmath101 & @xmath75 & @xmath102 & @xmath103 & @xmath104 & @xmath105 + @xmath106 & @xmath81 & @xmath102 & @xmath107 & @xmath108 & @xmath109 + @xmath76 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & @xmath109 + @xmath76 & @xmath75 & @xmath114 & @xmath115 & @xmath116 & @xmath117 + @xmath86 & @xmath75 & @xmath118 & @xmath112 & @xmath119 & @xmath120 + @xmath91 & @xmath75 & @xmath121 & @xmath122 & @xmath123 & @xmath124 + l c c c c c @xmath75 & @xmath125 & @xmath126 & @xmath127 & @xmath128 & @xmath129 + @xmath75 & @xmath96 & @xmath130 & @xmath131 & @xmath132 & @xmath133 + @xmath81 & @xmath75 & @xmath134 & @xmath135 & @xmath136 & @xmath85 + @xmath81 & @xmath81 & @xmath137 & @xmath138 & @xmath139 & @xmath140 + @xmath81 & @xmath125 & @xmath141 & @xmath142 & @xmath89 & @xmath143 + @xmath81 & @xmath144 & @xmath145 & @xmath146 & @xmath104 & @xmath147 + @xmath81 & @xmath96 & @xmath148 & @xmath142 & @xmath149 & @xmath100 + @xmath81 & @xmath101 & @xmath150 & @xmath151 & @xmath152 & @xmath153 + @xmath125 & @xmath75 & @xmath154 & @xmath155 & @xmath156 & @xmath157 + @xmath125 & @xmath81 & @xmath158 & @xmath159 & @xmath160 & @xmath133 + @xmath125 & @xmath125 & @xmath161 & @xmath107 & @xmath162 & @xmath163 + @xmath125 & @xmath144 & @xmath164 & @xmath165 & @xmath166 & @xmath167 + @xmath125 & @xmath96 & @xmath168 & @xmath169 & @xmath170 & @xmath153 + @xmath125 & @xmath101 & @xmath171 & @xmath112 & @xmath172 & @xmath157 + @xmath125 & @xmath106 & @xmath121 & @xmath173 & @xmath174 & @xmath175 + @xmath144 & @xmath81 & @xmath161 & @xmath176 & @xmath177 & @xmath178 + @xmath144 & @xmath125 & @xmath179 & @xmath180 & @xmath181 & @xmath182 + @xmath144 & @xmath144 & @xmath183 & @xmath180 & @xmath184 & @xmath157 + @xmath144 & @xmath96 & @xmath185 & @xmath186 & @xmath187 & @xmath188 + @xmath144 & @xmath101 & @xmath130 & @xmath189 & @xmath190 & @xmath191 + @xmath144 & @xmath106 & @xmath192 & @xmath193 & @xmath194 & @xmath175 + @xmath96 & @xmath81 & @xmath195 & @xmath196 & @xmath197 & @xmath198 + @xmath96 & @xmath125 & @xmath199 & @xmath200 & @xmath104 & @xmath201 + @xmath96 & @xmath144 & @xmath202 & @xmath203 & @xmath204 & @xmath120 + @xmath96 & @xmath96 & @xmath205 & @xmath159 & @xmath206 & @xmath207 + @xmath96 & @xmath101 & @xmath150 & @xmath138 & @xmath208 & @xmath147 + @xmath101 & @xmath81 & @xmath209 & @xmath142 & @xmath210 & @xmath211 + @xmath101 & @xmath125 & @xmath212 & @xmath213 & @xmath214 & @xmath188 + @xmath101 & @xmath144 & @xmath215 & @xmath200 & @xmath216 & @xmath182 + @xmath101 & @xmath96 & @xmath217 & @xmath218 & @xmath219 & @xmath220 + @xmath101 & @xmath101 & @xmath221 & @xmath127 & @xmath222 & @xmath223 + @xmath106 & @xmath125 & @xmath224 & @xmath225 & @xmath226 & @xmath90 + @xmath106 & @xmath144 & @xmath227 & @xmath122 & @xmath228 & @xmath163 + l c c c c c @xmath75 & @xmath144 & @xmath199 & @xmath229 & @xmath230 & @xmath231 + @xmath75 & @xmath101 & @xmath232 & @xmath233 & @xmath234 & @xmath235 + @xmath81 & @xmath96 & @xmath236 & @xmath180 & @xmath237 & @xmath100 + @xmath81 & @xmath101 & @xmath238 & @xmath103 & @xmath194 & @xmath95 + @xmath125 & @xmath81 & @xmath239 & @xmath98 & @xmath240 & @xmath167 + @xmath125 & @xmath144 & @xmath158 & @xmath241 & @xmath242 & @xmath243 + @xmath125 & @xmath101 & @xmath244 & @xmath241 & @xmath245 & @xmath246 + @xmath144 & @xmath101 & @xmath195 & @xmath241 & @xmath204 & @xmath247 + @xmath96 & @xmath81 & @xmath248 & @xmath200 & @xmath249 & @xmath95 + @xmath96 & @xmath144 & @xmath250 & @xmath196 & @xmath251 & @xmath231 + @xmath96 & @xmath96 & @xmath252 & @xmath253 & @xmath254 & @xmath246 + @xmath96 & @xmath101 & @xmath255 & @xmath115 & @xmath256 & @xmath85 + @xmath96 & @xmath106 & @xmath257 & @xmath151 & @xmath258 & @xmath95 + @xmath101 & @xmath81 & @xmath259 & @xmath196 & @xmath260 & @xmath247 + @xmath101 & @xmath125 & @xmath261 & @xmath262 & @xmath263 & @xmath264 + @xmath101 & @xmath144 & @xmath265 & @xmath266 & @xmath267 & @xmath268 + @xmath101 & @xmath96 & @xmath269 & @xmath176 & @xmath270 & @xmath182 + @xmath101 & @xmath101 & @xmath271 & @xmath233 & @xmath272 & @xmath273 + @xmath106 & @xmath81 & @xmath274 & @xmath193 & @xmath275 & @xmath85 + l c c c c c @xmath75 & @xmath125 & @xmath276 & @xmath277 & @xmath278 & @xmath279 + @xmath75 & @xmath144 & @xmath280 & @xmath281 & @xmath282 & @xmath129 + @xmath75 & @xmath96 & @xmath195 & @xmath283 & @xmath284 & @xmath285 + @xmath75 & @xmath101 & @xmath286 & @xmath266 & @xmath287 & @xmath288 + @xmath81 & @xmath125 & @xmath289 & @xmath283 & @xmath290 & @xmath288 + @xmath81 & @xmath96 & @xmath291 & @xmath292 & @xmath293 & @xmath285 + @xmath81 & @xmath101 & @xmath294 & @xmath295 & @xmath296 & @xmath297 + @xmath81 & @xmath106 & @xmath183 & @xmath298 & @xmath299 & @xmath279 + @xmath125 & @xmath81 & @xmath300 & @xmath295 & @xmath301 & @xmath302 + @xmath125 & @xmath125 & @xmath158 & @xmath283 & @xmath303 & @xmath304 + @xmath125 & @xmath144 & @xmath305 & @xmath306 & @xmath307 & @xmath308 + @xmath125 & @xmath96 & @xmath158 & @xmath253 & @xmath309 & @xmath198 + @xmath125 & @xmath101 & @xmath310 & @xmath311 & @xmath312 & @xmath304 + @xmath125 & @xmath106 & @xmath313 & @xmath176 & @xmath314 & @xmath302 + @xmath144 & @xmath81 & @xmath315 & @xmath306 & @xmath316 & @xmath317 + @xmath144 & @xmath125 & @xmath318 & @xmath277 & @xmath319 & @xmath320 + @xmath144 & @xmath144 & @xmath318 & @xmath277 & @xmath321 & @xmath191 + @xmath144 & @xmath96 & @xmath322 & @xmath107 & @xmath323 & @xmath167 + @xmath144 & @xmath101 & @xmath324 & @xmath103 & @xmath325 & @xmath147 + @xmath144 & @xmath106 & @xmath326 & @xmath189 & @xmath327 & @xmath178 + @xmath96 & @xmath75 & @xmath328 & @xmath165 & @xmath329 & @xmath198 + @xmath96 & @xmath81 & @xmath330 & @xmath200 & @xmath331 & @xmath332 + @xmath96 & @xmath125 & @xmath183 & @xmath266 & @xmath333 & @xmath268 + @xmath96 & @xmath144 & @xmath334 & @xmath98 & @xmath335 & @xmath167 + @xmath96 & @xmath96 & @xmath336 & @xmath241 & @xmath337 & @xmath297 + @xmath96 & @xmath101 & @xmath338 & @xmath339 & @xmath340 & @xmath273 + @xmath96 & @xmath106 & @xmath341 & @xmath342 & @xmath343 & @xmath188 + @xmath101 & @xmath81 & @xmath344 & @xmath229 & @xmath345 & @xmath105 + @xmath101 & @xmath125 & @xmath346 & @xmath347 & @xmath348 & @xmath211 + @xmath101 & @xmath96 & @xmath349 & @xmath138 & @xmath350 & @xmath285 + @xmath106 & @xmath144 & @xmath351 & @xmath352 & @xmath353 & @xmath223 + @xmath106 & @xmath96 & @xmath354 & @xmath102 & @xmath355 & @xmath178 + l c c c c c @xmath75 & @xmath81 & @xmath255 & @xmath262 & @xmath356 & @xmath188 + @xmath75 & @xmath125 & @xmath357 & @xmath358 & @xmath359 & @xmath360 + @xmath75 & @xmath144 & @xmath361 & @xmath362 & @xmath363 & @xmath364 + @xmath75 & @xmath96 & @xmath202 & @xmath362 & @xmath365 & @xmath366 + @xmath75 & @xmath101 & @xmath315 & @xmath277 & @xmath367 & @xmath285 + @xmath81 & @xmath75 & @xmath368 & @xmath213 & @xmath369 & @xmath105 + @xmath81 & @xmath81 & @xmath370 & @xmath295 & @xmath371 & @xmath372 + @xmath81 & @xmath125 & @xmath373 & @xmath358 & @xmath327 & @xmath374 + @xmath81 & @xmath144 & @xmath83 & @xmath375 & @xmath287 & @xmath120 + @xmath81 & @xmath96 & @xmath193 & @xmath362 & @xmath376 & @xmath167 + @xmath81 & @xmath101 & @xmath377 & @xmath292 & @xmath378 & @xmath379 + @xmath81 & @xmath106 & @xmath380 & @xmath213 & @xmath381 & @xmath157 + @xmath125 & @xmath75 & @xmath382 & @xmath169 & @xmath383 & @xmath188 + @xmath125 & @xmath125 & @xmath142 & @xmath384 & @xmath385 & @xmath386 + @xmath125 & @xmath96 & @xmath202 & @xmath281 & @xmath387 & @xmath264 + @xmath125 & @xmath101 & @xmath388 & @xmath283 & @xmath389 & @xmath390 + @xmath125 & @xmath106 & @xmath315 & @xmath159 & @xmath391 & @xmath105 + @xmath144 & @xmath75 & @xmath392 & @xmath393 & @xmath394 & @xmath157 + @xmath144 & @xmath81 & @xmath148 & @xmath200 & @xmath395 & @xmath191 + @xmath144 & @xmath125 & @xmath300 & @xmath292 & @xmath396 & @xmath397 + @xmath144 & @xmath144 & @xmath398 & @xmath292 & @xmath399 & @xmath360 + @xmath144 & @xmath96 & @xmath400 & @xmath277 & @xmath401 & @xmath402 + @xmath144 & @xmath101 & @xmath403 & @xmath180 & @xmath404 & @xmath379 + @xmath144 & @xmath106 & @xmath205 & @xmath169 & @xmath405 & @xmath406 + @xmath96 & @xmath125 & @xmath286 & @xmath306 & @xmath407 & @xmath408 + @xmath96 & @xmath144 & @xmath88 & @xmath203 & @xmath409 & @xmath410 + @xmath96 & @xmath96 & @xmath199 & @xmath196 & @xmath411 & @xmath191 + @xmath96 & @xmath101 & @xmath341 & @xmath218 & @xmath412 & @xmath320 + @xmath96 & @xmath106 & @xmath413 & @xmath414 & @xmath415 & @xmath117 + @xmath101 & @xmath125 & @xmath416 & @xmath298 & @xmath417 & @xmath246 + @xmath101 & @xmath144 & @xmath392 & @xmath213 & @xmath418 & @xmath207 + @xmath101 & @xmath96 & @xmath419 & @xmath189 & @xmath376 & @xmath317 + @xmath101 & @xmath101 & @xmath420 & @xmath421 & @xmath319 & @xmath80 + l c c c c c @xmath75 & @xmath125 & @xmath422 & @xmath289 & @xmath363 & @xmath124 + @xmath81 & @xmath125 & @xmath423 & @xmath233 & @xmath190 & @xmath95 + @xmath81 & @xmath144 & @xmath424 & @xmath425 & @xmath426 & @xmath85 + @xmath81 & @xmath101 & @xmath427 & @xmath428 & @xmath429 & @xmath175 + @xmath125 & @xmath75 & @xmath430 & @xmath138 & @xmath431 & @xmath80 + @xmath125 & @xmath81 & @xmath315 & @xmath203 & @xmath345 & @xmath140 + @xmath125 & @xmath125 & @xmath202 & @xmath311 & @xmath432 & @xmath153 + @xmath125 & @xmath144 & @xmath433 & @xmath218 & @xmath434 & @xmath85 + @xmath125 & @xmath96 & @xmath326 & @xmath435 & @xmath436 & @xmath386 + @xmath144 & @xmath75 & @xmath291 & @xmath103 & @xmath437 & @xmath211 + @xmath144 & @xmath81 & @xmath183 & @xmath283 & @xmath302 & @xmath167 + @xmath144 & @xmath144 & @xmath438 & @xmath295 & @xmath439 & @xmath211 + @xmath96 & @xmath75 & @xmath161 & @xmath241 & @xmath440 & @xmath406 + @xmath96 & @xmath125 & @xmath289 & @xmath283 & @xmath441 & @xmath288 + @xmath96 & @xmath106 & @xmath442 & @xmath193 & @xmath443 & @xmath124 + @xmath101 & @xmath81 & @xmath444 & @xmath445 & @xmath441 & @xmath406 + @xmath101 & @xmath125 & @xmath199 & @xmath200 & @xmath446 & @xmath397 + l c c c c c @xmath75 & @xmath96 & @xmath324 & @xmath241 & @xmath447 & @xmath163 + @xmath75 & @xmath101 & @xmath448 & @xmath165 & @xmath449 & @xmath211 + @xmath81 & @xmath81 & @xmath352 & @xmath298 & @xmath450 & @xmath386 + @xmath81 & @xmath106 & @xmath392 & @xmath229 & @xmath451 & @xmath120 + @xmath125 & @xmath101 & @xmath135 & @xmath452 & @xmath453 & @xmath374 + @xmath96 & @xmath75 & @xmath244 & @xmath138 & @xmath454 & @xmath455 + @xmath96 & @xmath81 & @xmath276 & @xmath295 & @xmath456 & @xmath374 + @xmath96 & @xmath125 & @xmath457 & @xmath384 & @xmath458 & @xmath140 + @xmath96 & @xmath144 & @xmath151 & @xmath384 & @xmath459 & @xmath406 + @xmath101 & @xmath81 & @xmath460 & @xmath186 & @xmath461 & @xmath188 + @xmath101 & @xmath125 & @xmath225 & @xmath281 & @xmath462 & @xmath463 + @xmath101 & @xmath96 & @xmath464 & @xmath311 & @xmath465 & @xmath243 + @xmath106 & @xmath144 & @xmath466 & @xmath241 & @xmath467 & @xmath100 +

two - dimensionally extended materials , or membranes , have been paid much attention to in various fields of study , such as lipid bilayers @xcite and surfactants @xcite in soft - matter and biophysics @xcite . alongside , many theoretical studies on those membrane - type objects have been done toward a general understanding of physical properties of such materials @xcite . some of them were simulation approaches @xcite and others were with field theoretical techniques @xcite . for example , one of the authors studied the pore formation in a binary giant vesicle of two - component lipid bilayer , and reproduced its dynamical shapes in the field - theoretical approach with some numerical calculations @xcite . although most of the works contributed successfully to our knowledge of individual membranes , it can be said that we are yet very far from the general understanding . in this work , we aim to give some new insight to the general understanding of the physical properties of soft - material membranes by studying mathematical models on discretized surfaces both analytically and numerically . that is to say , we propose a family of fluid particle models of two - dimensional lattice as simpler models of membranes : tightly pack molecular particles on a given lattice , and give the particles dynamics by some stochastic process , or by a set of mathematical rules . for more simplicity , we consider single - component , single - layered incompressible systems , and assume no explicit viscosity and an upper bound for the flow velocities on the particles . after the formulation , we construct two stochastic models of 2d lattice liquid , as our first steps . one of them is found to be the vicious random walk @xcite in its densest phase and a version of the dynamic lattice liquid model ( dll model ) @xcite in two dimensions . the dynamics of these models is given by completely independent time steps and would not generate various fluid dynamical phenomena such as lasting convective flows , while our formulation does not exclude such possibilities . in fact , the other model is constructed as the flow model with a mathematical rule , which has no corresponding known models . we further restrict ourselves to a finite flat honey - comb lattice and simulate the two models with their highest mobility . comparing with similar lattice models such as various lattice random walks or the loop gas model ( sos model ) @xcite , we show some statistical properties of the models by the simulation results . the results tell that the flow model may exhibit some anomalous diffusion , in terms of time - averaged mean square displacements ( tamsd ) , and the ergodicity with its fairly long relaxation time . relations to other models and some possible extensions of the models will also be discussed in the final section . there are various situations of the membrane systems , and each situation would deserve a single mathematical model in order to simulate it . however , since we aim to construct a basic model to be utilised for those various situations , we focus exclusively on a very simple but basic situation among them as a building block . first , we focus on a single - component and single - layered molecular system on a two - dimensional surface , treating the surrounding medium as a kind of heat bath to the surface . in other words , we introduce some rules or external forces onto the elements of the membrane rather than consider the interactions between the membrane and the medium . secondly , we focus exclusively on a simple classical interaction such as `` collision '' , neglecting any other complicated interaction between the elements of the membrane . let us call this ` no explicit viscosity ' . thirdly , we assume that the density fluctuation can be negligible and two elements would not occupy the same location of the surface at any time . this would correspond to the incompressibility condition of the surface from the two - dimensional fluid point of view . to summarise , we assume and focus on a single - component and single - layered incompressible particle systems with no explicit viscosity on a two - dimensional surface exposed in an external ( random ) field . now , we are ready to define the 2d lattice liquid models . given a two - dimensional lattice @xmath0 , a 2d lattice liquid model can be defined as a mathematical model of the particles which can only reside at the sites of the lattice @xmath0 , covering all or almost all of the sites , and which flow inside @xmath0 as molecular - dynamical units of liquid do inside them . here , the particles represent the elements of the membrane , and are assumed to be of the same physical properties : their sizes , masses , and so on . the word ` liquid ' comes directly from two parts of the above definition , one of which is that the particles cover all or almost all of the sites of @xmath0 , and the other of which is that the particles flow . obviously , there is an ambiguity on the phrase `` all or almost all '' . to avoid this ambiguity in this paper , we force each site to hold only and exactly one particle at a time , covering _ all _ the sites . we call it `` the occupation number is exactly one everywhere '' . although we can make some of the sites empty or put as many particles as possible , the assumption of the density fluctuation eliminates the latter and we simply omit the former in this paper . note that the definition of the lattice liquid would not exclude these cases and would be worth studying elsewhere . one can consider the boundary conditions in the presence of boundary of the surface . there are two types of boundary conditions , one of which is on particles dynamical degrees of freedom , and the other of which is on their internal degrees of freedom , such as particle species . the former type of boundary conditions are naturally embedded in the lattice on which we define the model . on the latter , since we only consider single - component systems , there is no boundary condition on the particle species . to complete the definition of the model , the particles `` flows '' should be defined by imposing a set of mathematical rules on the flows . this consequently determines the dynamics of the system . before going into the details of the rules , let us introduce the lattice conventions and consider the configuration space of the model to begin . a lattice @xmath0 is defined by a set of sites @xmath1 and a set of bonds @xmath2 . a bond @xmath3 is connecting a pair of the sites , @xmath4 , which are called neighbouring sites or in the neighbourhood . the lattice spacing @xmath5 the shortest length of the bonds specifies a fine aspect of the lattice . we consider only the cases with the same bond length , unless otherwise stated . for that reason we set @xmath6 without loss of generality . the total number of the sites and the total number of the bonds are denoted by @xmath7 and @xmath8 , respectively . we abbreviate them by @xmath9 and @xmath10 for simplicity . the degree of the site , @xmath11 , is the number of the bonds connected to the site @xmath12 . if there is only one value for @xmath11 on @xmath0 , we write it by @xmath13 for short , and we consider mainly such cases . in the later sections of some numerical calculations , we focus on the honeycomb lattices of various sizes as a series of examples . on a given lattice @xmath0 , under any set of the rules of the movements , a flow of a particle can be sketched schematically in fig.[fig : worldlines ] below . introducing the minimum time lapse @xmath14 , a particle at the site @xmath12 at the time @xmath15 flew from the site @xmath16 at @xmath17 with the velocity @xmath18 , where @xmath19 is implicitly taken as a unit ( @xmath20 ) . then , the particle at @xmath12 at @xmath15 flows toward @xmath21 at the future time @xmath22 , subject to some fluctuation @xmath23 by external forces . explicitly writing the time dependence of the variables , @xmath24 denotes the flow velocity of the particle from @xmath17 to @xmath15 , and @xmath25 does from @xmath15 to @xmath22 . we denote the latter velocity by a redundant variable @xmath26 for convenience . the solid continuous line in fig.[fig : worldlines ] represents the particle travelling in @xmath27-dimensional space - time , and all the particle movements can be represented by such world - lines . namely , the configuration of the system at a certain time @xmath15 , can be uniquely given by a set of the flow velocities : \{_i(t ) } for all i v ( ) . accordingly , the time evolution is realised by mapping a configuration @xmath28 to another configuration @xmath29 at a successive time , while a particle at @xmath12 travels to @xmath30 with the velocity @xmath31 such that : i^= i+^_i(t ) , _ i^(t+t ) = ^_i(t ) . for consistency , @xmath18 and @xmath32 can not take completely arbitrary values , but are subject to the constraint that @xmath33 and @xmath34 map @xmath35 to itself and they are bijective . in other words , the configuration space is given by the whole set of such bijections . such a space is highly complicated in general , however one assumption actually makes the space more tractable . we assume the following upper bound for the velocities : |_i| 1 for all i v ( ) . this means that the particles are prevented from jumping in between non - nearest neighbouring sites . therefore , the assumption is not too restrictive in the physics context . when the bond lengths are all the same , this leads the absolute value @xmath36 takes either @xmath37 or @xmath38 . accordingly , the velocity @xmath18 takes one of @xmath39 values . since @xmath40 obeys the same , the velocity change @xmath41 takes at most one of @xmath39 values as well . with the upper bound of the velocities and the assumption of the occupation number , a configuration of the particles can be represented by a collection of oriented non - intersecting loops on @xmath0 . in order to illustrate this fact , let us give a set of random velocities , @xmath42 , with the condition @xmath43 . all particles potentially try to move toward the directions they have been kicked by some external forces . however , this does not necessarily give movements to the particles due to the frustration of such momenta ( fig.[fig : frustration - circ](a ) ) . the frustration is an outcome of the excluded volume effect of the particles together with the restriction of the occupation number . in order to generate convective flows , some part of the `` potential '' movements generated by the forces must be enclosed so that the particles can move ( fig.[fig : frustration - circ](b ) ) . + ( a ) + ( b ) hence , the relevant part of the configuration @xmath33 can be represented by a collection of _ oriented _ non - intersecting loop graphs on @xmath0 . we label the space of the oriented non - intersecting loop configurations by @xmath44 . we also label the loop configuration without the orientations by @xmath45 for later use . formal definition of the ( oriented ) loop configurations in graph theory , refer to appendix [ ap : loops ] . note that , if and only if @xmath46 , the above algorithm coincides with a version of the dll model @xcite . in what follows , we focus on the situations where the loop configuration solely plays a crucial role . as was stated before , the set of mathematical rules of the particle - flows give the principal part of the definition of the model . however , regardless of how and what sort of the rules are imposed , they are to be expressed in the form of the velocities and the velocity changes : @xmath33 and @xmath47 , or in the form of the present loop configuration and the resulting one , @xmath48 , representing the velocities . so , we formulate the 2d lattice liquid models as a family of statistical models defined with such variables . one of the most straightforward expressions of the propagator would take the form of : [ eq:1stprop ] & & g ( x_1 , x_2 , ; g , t ; g^ , t^ ) & & or g ( x_1 , x_2 , ; \{_i(t ) } , t ; \ { _ j(t^ ) } , t^ ) , or as its ensemble : [ eq:2ndprop ] & & z^g ( x_1 , x_2 , ; g , t ; t^ ) & & _ g^ ( ) g ( x_1 , x_2 , ; g , t ; g^ , t^ ) , where the @xmath49 s are other unspecified parameters to prescribe the propagator . if all the processes are statistical from the beginning , @xmath50 , even the initial configuration is given by some distribution , the following quantity would be the partition function in question : & & z(x_1 , x_2 , ; t , t^ ) & & _ g ( ) p_init ( g ) z^g ( x_1 , x_2 , ; g , t ; t^ ) , where the function @xmath51 is the statistical weight of the configuration @xmath52 . practically , we will focus on the second expression ( [ eq:2ndprop ] ) of the propagator and the first one ( [ eq:1stprop ] ) for the construction of the second . let us construct a more concrete yet abstract form of the propagator by the probabilities of the configurations . for the time step of @xmath19 , given the present configuration @xmath53 , let us define @xmath54 as the `` potential '' probability of a particle remaining at the site @xmath12 , and @xmath55 is the `` potential '' probability of the particle at @xmath12 being kicked toward one of the nearest neighbouring sites @xmath56 . they satisfy : [ eq : px - const ] p(i)+_j x(i;j ) = 1 , where the summation is over all nearest neighbouring sites to @xmath12 . in general , @xmath54 and @xmath55 may carry the information from the past , but we ignore it in the rest of this manuscript . recall that some of the kicks with the `` potential '' probabilities might not be performed due to the before mentioned frustration . suppose non - frustrated part of the particles are expressed in the configuration @xmath57 . the complementary graph can be given with its projection @xmath58 on @xmath45 by @xmath59 . then , the probability of having @xmath57 in @xmath0 , @xmath60 , can be given by the product of the loop probability @xmath61 of having @xmath57 in @xmath58 and the probability @xmath62 of the frustration in @xmath63 as follows : p(g^ ) = p_l(g^ ) p_f(|g^ ) , where the initial configuration @xmath53 is implicit , and the loop probability can be expressed by multiplying the probabilities @xmath55 s along the oriented loops : p_l(g^ ) _ _ i , j e(g^ ) x(i;j ) . note that the definitions of the above probabilities need some minor extensions to the subgraphs for their consistency . for the detailed definition of the probability functions and their properties , refer to appendix [ ap : pfpl ] . the probability @xmath60 is nothing but the first expression ( [ eq:1stprop ] ) of the propagator for one time step . accordingly , the second expression ( [ eq:2ndprop ] ) of the propagator for one time step is : z^g(g , t ; t+t ) & = & _ g^ ( ) p ( g^ ) . specifying its @xmath53 dependence by @xmath64 , the propagator for a longer time span can be expressed by : & & z^g(g^(0 ) , t ; t^ ) & & = _ g^(1 ) ( ) _ g^(k-1 ) ( ) _ k=1^k p(g^(k ) ; g^(k-1 ) ) , where @xmath65 . hence , the construction of the propagator reduces to a detailed construction of the probability functions , @xmath62 and @xmath61 . note that the above expression actually corresponds to the discretised version of the path integral . given the probabilities , an expectation value of some physical quantity @xmath66 for @xmath19 can be given as usual by the form of : _ g & = & _ g^ ( ) o p(g^ ; g ) . it is obvious that the set of the mathematical rules of the particle - flows give the principal part of the definition of the 2d lattice liquid model as well as of its formulation . in fact , one can easily imagine the following three sorts of the rules for the particle - flows . namely , the particles flow according to some stochastic process , or subject to some external thermodynamic forces from the surrounding medium , or obeying some physics laws with their ( fluid ) dynamical properties such as the langevin - type equation or the navier - stokes equation in the low @xmath67 single - component limit . in order to avoid ambiguities on discretisation and limiting procedures in the equation - based approaches , we deal only with the stochastic cases here . some relations to thermodynamic or equation - based approaches will be discussed later . in this section , we construct and formulate one of the simplest stochastic models , the stochastic markovian model , whose rules are as simple as assigning simple kicks to push the particles and arresting them at the end of each step . the model is turned out to be equivalent to the vicious random walk in its densest phase and a version of the dll model on two - dimensional lattice . the precise definition of the rules are given as follows . first , let @xmath54 be non - negative constant , and @xmath55 be of isotropic kicks such that : [ eq : bounds4p ] p(i ) & = & p , x(i;j ) & = & x(i ) . here , we restrict them to be strictly independent of the past , ensuring that the process is markovian . second , we assign the kicks according to the above probabilities , @xmath68 and @xmath69 , and those kicks are projected onto some oriented loop configuration @xmath53 . perform the moves of @xmath53 . thirdly , as a key part of the rules , we force : [ def : sto ] _ j(t+t ) = _ i(t ) i , where @xmath70 . this means that the configuration of the velocities is set at every step by the fluctuation @xmath71 only . accordingly , one of the two sets , @xmath33 and @xmath72 , becomes redundant . note that @xmath73 inherits the upper bound from @xmath36 : @xmath74 . practically , it follows @xmath75 . the probability of an oriented loop configuration @xmath53 on the sublattice @xmath76 can be written by : p_l ( g ) = _ _ i , je(g ) x(i ) . if @xmath11 s are all the same as of regular lattices , the above probability can be simplified as : p_l ( g ; x ) = x^l(g ) , where we specify @xmath77 as an argument , @xmath78 , and @xmath79 denotes @xmath80 for @xmath81 of @xmath53 , _ i.e. _ , the total length of the loops in @xmath53 . note that @xmath77 plays the role of the fugacity of the bonds in @xmath53 . if we further define the probability of a non - oriented loop configuration @xmath76 on the sublattice @xmath76 , summing up all the orientations of @xmath76 , it follows that : p_l ( g ; x ) = 2^(g ) x^l(g ) , where @xmath82 by definition , @xmath83 counts the number of the loops in @xmath76 , and the factor ` 2 ' indicates the directions of each oriented loop in @xmath53 . accordingly , the propagator , or the partition function , can be written down by : [ eq : part_sto ] g^sto ( x ) = _ g ( ) p_l(g;x ) p_f ( |g ; x ) , which is equivalent to unity by definition . @xmath84 gives the probability of the no - loop configuration . the probabilities @xmath62 and @xmath61 take physical values , @xmath85 and @xmath86 for @xmath87 . it should be mentioned here that if @xmath88 is set to be one for any @xmath76 then the partition function reduces to that of the non - oriented non - intersecting loop configurations on a lattice , which is known as the loop gas model @xcite : @xmath89 with @xmath90 . so , @xmath88 plays a crucial role in our model . as in the definition ( [ def : sto ] ) , the present configuration is always trivial so that the above expression ( [ eq : part_sto ] ) is exact at any @xmath15 . therefore , the propagator can be expressed by : z^sto(x ; t ; t+t ) = g^sto(x ) . hence , it can be summarised and simplified to the propagator of the time lapse @xmath15 as follows : z^sto(x ; t ) = ^t / t . the expectation value of a physical quantity @xmath66 is given in a conventional manner : _ t & = & ^t / t , & = & _ g ( ) op_f ( |g ; x ) p_l(g;x ) . before giving our simulation results , let us show some simple relations and calculations on an isotropic regular lattice . on an isotropic regular lattice , such as a square lattice with periodicities , the one - point function of the velocity can trivially be given by : = 0 , by symmetry . similarly , the expectation value of the absolute value of the velocity at @xmath12 can be given by : = = i , where @xmath91 is the expectation value of the length of the loops : @xmath92 . besides , the variance of the velocity becomes : = = , where the normalised lattice spacing is used . with the above facts , one can state that each particle behaves like a random walker on the given lattice with the constant trapping ratio of @xmath93 , where a particle moves to a certain direction with the probability @xmath94 . namely , for any @xmath95 for large @xmath0 or small @xmath15 : & = & 0 , & = & t , where @xmath96 denotes the position of the @xmath97-th particle at the time @xmath15 . however , it should be noted that the system is not a collection of independent simple random walks , as our particles do not occupy the same site at the same time . in fact , the system is in the densest phase of the vicious random walks mutually avoiding @xmath98 particles random walks on a lattice with @xmath98 sites @xcite . the difference between this system and a collection of the independent simple random walks can be clearly shown by the spacial correlation of the velocities . for example , the spacial correlation of a nearest pair of the system is given by : , whereas those of the simple random walks is given by a simple product of one - point functions : @xmath99 . in the previous sections , we have formulated the model where the streamlines are temporally generated and live only for the duration of @xmath19 . because the probability of the shortest loop in a honeycomb lattice is at most @xmath100 in the bulk , @xmath91 seems to be very small . we infer in the model ( i ) that the stochastic process realises the solid - like nature of the membrane rather than its liquid nature as the dll model targeted supercooled liquids . in this section , we intend to explore another class of stochastic models where the particles keep flowing on the membrane rather than obeying the ` stop and go ' evolution . in order to realise this phenomenon , we take the simplest way : if no fluctuations are added to the system , the particles keep flowing along the oriented loops . this rule can be expressed by the relation in the absence of the fluctuations : [ def : stof ] _ i(t+t ) = _ i(t ) . now , it is evident that the key point of the construction lies on how to add stochastically generated fluctuations onto the existing streamlines . in order to avoid complications for realising the relation ( [ def : stof ] ) and the frustrations in terms of @xmath47 , let us formulate the above addition in terms of the oriented loop configurations . say , @xmath101 is the initial configuration and @xmath102 is the resulting configuration . the arithmetic rule ( addition / subtraction ) for the oriented loop configurations can be illustrated in a multiplicative form by : g g_i = g_f ( ) , where @xmath103 stands for the stochastically generated fluctuations , and vanishing @xmath103 is given by the no - loop configuration @xmath104 naturally satisfying the rule ( [ def : stof ] ) . in @xmath45 , there is a natural definition of such an additive operation as the symmetric difference . we extend and modify it to our case for @xmath105 . let @xmath106 , @xmath107 be elements in @xmath44 . if @xmath107 consists of the same sites and bonds with opposite directions as those of @xmath106 , then we call @xmath107 the inverse of @xmath106 , denoting it by @xmath108 . the additive operation between @xmath106 and @xmath107 is defined in the multiplicative form by : g_1 g_2 = g , such that the directed bonds of the elements satisfy : [ eq : symdiff - directed ] g & = & g_1 g_2 - g_1 g_2 ^ -1 - g_1 ^ -1 g_2 , while the disconnected sites are removed from @xmath53 . the above addition trivially satisfies the followings : g_1 i ( ) & = & i ( ) g_1 = g_1 g_1 ( ) , g_1 g_2 & = & i ( ) g_2 = g_1 ^ -1 , where @xmath104 behaves as the zero in the addition and the identity in its multiplicative form . the problem of this definition is that @xmath53 is in the corresponding ( directed ) cycle space but not necessarily in @xmath44 . to avoid this unwanted situation , we propose the following projected addition for @xmath53 in eq . ( [ eq : symdiff - directed ] ) : [ def : addition ] g_1 g_2 & = & g g ( ) , g_1 g_2 & = & g_2 g ( ) . in the second line , we call @xmath106 is non - additive to @xmath107 , or vice versa . this projection drastically changes the algebraic nature of the operation . in fact , it breaks the commutativity : @xmath109 , and the associativity : @xmath110 . so , the ordering of the operations matters in this case . it should be mentioned here that we can create some different definitions of the addition simply by different projections of @xmath111 onto @xmath44 . as a starting point , we stick to the above definition and will mention another possibility later . note also that , precisely speaking , @xmath103 is either a candidate for a fluctuation or a fluctuation which sometimes does nothing , because of the projection . if the configuration at the time @xmath95 is given by @xmath112 , the configuration @xmath113 at the time @xmath114 can be expressed by a series of successive additions of @xmath103 to @xmath112 : g(t_0+t ) = g(t_0 ) . [ def : flow - part ] the multiplication in the r.h.s . means the multiple application of the loop additions , @xmath115 , and the addition should be done from right to left . the factor expresses nothing but a configurational and kinetic path between two configurations . finally , we complete the definition of the stochastic flow model by stochastically giving @xmath103 in the same manner as in the previous section . namely , assign the `` potential '' probabilities of a particle remaining at the same position and of a particle at @xmath12 being kicked toward one of the nearest site @xmath56 by @xmath68 and @xmath55 , respectively . for the isotropic distribution , @xmath55 can be simplified to @xmath116 . then , project the velocity changes onto the space of the oriented loop configurations @xmath44 , and add it to the current configuration in the end . the degrees of freedom of @xmath41 and @xmath18 are same as @xmath39 , as long as an additional upper bound for @xmath41 is not imposed . so , their spaces differ only by the meanings of @xmath68 and @xmath55 . for simplicity , we neglect this difference and distribute the probabilities isotropically . note that the addition rule implicitly introduced the upper bound for @xmath23 such that @xmath117 . now , we have formulated how the fluctuations are given and added to existing flows . the propagator with the initial configuration @xmath118 can now be written down by : & & z^flow(x ; g_i , t_0 ; t+t_0 ) & & = _ g_t / t ( ) _ g_1 ( ) _ j=1^t / t g_i(t_0 ) , where the probability @xmath119 of having the loop @xmath53 in @xmath0 is the same as in the stochastic markovian model . note the process of this model is also markovian . in this section , we perform numerical calculations on the proposed models and show their statistical properties for further investigation , some of which reveal very interesting features of the models . we restrict the lattice @xmath0 to be as simple as the finite rectangular honeycomb lattice of roughly @xmath120 hexagons for odd @xmath98 , labelling it by @xmath121 . precisely speaking , we define the lattice @xmath121 by piling @xmath122 and @xmath98 aligned hexagons alternatively up to @xmath98 rows so that the lattice becomes left - right symmetric ( fig.[fig : hn ] ) . .,width=188 ] for odd @xmath98 , the lattice becomes actually top - bottom symmetric as well , and therefore symmetric under discrete rotations . in the case of a finite lattice with boundary , there are the boundary sites and bonds which face the outer region of the lattice on its planar embedding . among such sites , some have smaller degrees from those in the bulk . in the case of the honeycomb lattice , there are degree - two boundary sites . accordingly , we can assign two different probabilities : @xmath123 in the bulk and @xmath124 for the boundary sites of degree two . we choose the highest mobility @xmath125 , @xmath126 , and @xmath127 for numerical calculations . as two models share the oriented non - intersecting loop configurations as an input to the system , the input data has been generated in common for a million of time steps for various sizes , @xmath128 . below is the basic information of @xmath121 : dim v(_h_n ) & = & ( n+1)(2n+1)-4 , dim v_b ( _ h_n ) & = & 4n , dim e(_h_n ) & = & 3n^2 + 52 n - 92 , f & = & n^2 - , for odd @xmath129 . @xmath130 is the set of all the boundary sites of degree two . the number of faces is denoted by @xmath131 . we first plot the probability @xmath132 of the no - loop configuration , or of the total frustration , for different sizes in fig.([fig : pf - log ] ) . we have observed that it decays approximately exponentially with respect to the area of the lattice . the @xmath77-axis stands for the area of the lattice counted by the number of hexagonal faces in @xmath0 . there can be obviously seen a scalability . the error bars are given at the 95% confidence level.,width=302 ] here , we propose the approximate functional form of @xmath62 for general shape of the honeycomb lattice with @xmath131 connected faces : p_f ( ) e^- . [ eq : pfapprox ] note the form may not well approximate if @xmath0 takes an extreme shape such as a lattice with small @xmath131 or of aligned faces . fitting the data in fig.[fig : pf - log ] with the above function actually leads to @xmath133 . we have also calculated @xmath62 by hand in the case of the honeycomb lattice of 2 piles of 3 hexagons , @xmath134 , and confirmed that the extrapolation from @xmath134 gives @xmath135 . this result further indicates that the system gradually changes its nature between frustration dominant surface to loop dominant surface as the size changes around @xmath136 . next , we plot the number of events for the total length of the loops @xmath137 , which is equivalent to the number of the moving particles in one time step ( fig.[fig : looplen_dist ] ) . -axis is the length in the unit of the lattice spacing . the @xmath138 shows the number of events in a million trials . periodic oscillations seem to suggest the multiple entries of the same type of loops , e.g. , two length - six loops at once contributes to the length - twelve event.,width=302 ] it is remarkable that the leading contribution is not necessarily from length - six loops , and neither is the sub - leading contribution from length - ten loops . for instance , on @xmath139 , the leading and sub - leading contributions are from length-24 and length-30 loops , resp . we have confirmed that the proposed form of @xmath62 can explain nicely the @xmath98-dependence of the probability of length - six loops . we also report the proportionality of the mean loop length @xmath91 w.r.t . the area of the lattice : |l ~0.01763 f , for @xmath140 . as for the loop number distribution , we find that they obey the poisson distribution in an accurate way with the mean @xmath141 approximately proportional to the number of the faces @xmath131 : not so trivial , suggesting that there is a relatively stable value of the loop number per unit surface as well as that of the total length of the loops per unit surface . together with the average absolute value of the velocity and other basic numbers of @xmath121 , we list them in table [ table : hn ] . . the basic numbers of @xmath121 for various @xmath98 . [ cols=">,>,>,>,^,^,^,^",options="header " , ] the above averages include the data from the beginning , and therefore we neglected the relaxation time @xmath142 from @xmath104 to quasi - critical states . the relaxation time @xmath142 which will be shown shortly is not very long compared to @xmath143 . so , it would not spoil the accuracy of the statistics much . in this flow model , the distribution of the loop number behaves rather gaussian than poisson . in addition , @xmath91 and @xmath141 no longer obey simple proportional relations to the area of the lattice . the former has a rather correlation to the number of the sites @xmath9 , as @xmath144 is almost independent of the size @xmath98 . the value @xmath145 gives the ratio of the moving particles among all , while @xmath146 is of the trapped particles . nearly constant value of @xmath145 might suggest that the value is intrinsically universal for all honeycomb lattices . the average loop number @xmath141 is not so stabilised compared to the average loop length @xmath91 . it should be remarked here that the projection in our addition ( [ def : addition ] ) would certainly change the probability of the fluctuations , but only its characteristic time scale . so , we claim that the above values and qualitative arguments are characteristic in this type of models , regardless of the addition rule defined earlier . it is the dynamical aspect of the model that the addition rule makes different . in this stochastic flow model , the particles are basically trapped in the sites alike in the stochastic markovian model . but , once they are involved in the streamlines , they keep moving until it is stopped by a fluctuation . coarse - graining of the pathway actually make itself look like a brownian motion , but it is actually much more complicated . let us quantify it by the tamsd of eq . ( [ eq : tamsd ] ) . we picked up a particle situated at the centre of the lattice and calculated the tamsd ( fig.[fig : msd11 - 41 ] ) , where the lower limit of the time is set by the relaxation time @xmath147 : @xmath148 , @xmath149 , so that at @xmath150 the particles are randomised enough on @xmath0 from the initial configuration . @xmath151 is set to be the end of the simulation : the million - th step . , for various sizes in the log - log format . the @xmath77-axis is the time lapse @xmath15 , while the @xmath138-axis is the square distance in the unit of the lattice spacing squared . the dashed lines are all exactly calculated asymptotic values for various @xmath98.,width=302 ] as one can see from the left side of fig.[fig : msd11 - 41 ] , the particle starts moving at a certain rate of @xmath145 , and diffuses like liquid , slightly faster than the brownian motion . at around @xmath152 , the exponent @xmath153 in @xmath154 reaches the brownian one , and later decreases to the anomalous diffusion of @xmath155 in a strange way . finally , it approaches its asymptotic value . by the times the particles approach the asymptotic values , we can read off the relaxation times , @xmath142 . their orders of magnitude are roughly the same we set for the initial time @xmath156 for tamsds . it can be said that the system is generally in an anomalous diffusion process where @xmath157 in the short range , and @xmath155 in the distance . there can be found a strange and interesting plateau as well in the middle of the curve for each size of the lattice , but we have not yet obtained any clear explanation of the plateau . we will further mention this in the final section . there is a closest model to compare with our model ( ii ) , which is the non - reversal random walk . in that model , a random walker can bring a little memory for its past path so that it will not come back to the memorised positions . usually , the walker memorises only the previous position in order not to come back at once . the model is sometimes known as a polymer chain model with excluded volume . in our flow model , a particle movement is always associated with a ring of movements and can not reverse immediately . this minimum regulation matches with that of the non - reversal walker . in addition , because the non - reversal walker is not bounded by any loop - like relation to neighbours , we deduce that it gives the upper limit for the tamsd of our model . we define the stochastic flow model ( ii ) and the non - reversal random walk on the lattice @xmath158 , and compare them in terms of their tamsds in fig.[fig : comparisonii_21 ] . notice that the constant trapping ratio is equipped with the non - reversal walkers to keep the pace of his first step with the flows in our model . . the line for the non - reversal random walk is given with the trapping time consistent with the original model , and the asymptotic value for @xmath158 is shown.,width=302 ] as for the ergodicity , we claim that the ergodicity weakly holds in our model ( ii ) of the lattice liquid . in order to illustrate this , we have calculated the asymptotic value of the msd with some assumption as follows . assume that , in the limit of @xmath159 , the time correlation of every particle on @xmath0 vanishes so that a particle situated at any of the sites on @xmath0 at a certain time @xmath95 is to be drifted to every site with equal footing in the infinite future . then , the asymptotic value of @xmath160 can simply be given by the average of the square distances of all the relevant pairs of the sites on @xmath0 . on the rectangular honeycomb lattice @xmath121 , we derived its exact formula and the leading terms are given by : [ eq : asymptotic_approx ] _ t+ = 18 ( 7 n^2 + 10 n ) + o(1 ) . in the case of @xmath161 , the exact values are up to first five digits : [ eq : asymptotic ] 116.75 , 409.04 , 876.45 , 1518.9 , 2336.4 . see appendix [ app : aymptotic ] for the exact form of an arbitrary @xmath98 . the time - averaged msds in fig.[fig : msd11 - 41 ] , [ fig : comparisonii_21 ] approach the calculated asymptotic values ( [ eq : asymptotic ] ) and are stabilised around them as the time grows . therefore , we can state that our models are surely ergodic in this sense . in this paper , we have proposed and formulated a family of models , called the 2d lattice liquid models , as one sort of the simplest models of single - component , single - layered , incompressible fluid particle system on a membrane without explicit viscosity and density fluctuation . the model is formulated as a particle system on a lattice , but the configurations are found to be expressed by the sets of oriented non - intersecting loops , due to the restriction on the occupation number and the upper bound for the velocities . following the formulation , we have constructed two models of the lattice liquid . one of them is the stochastic markovian model ( i ) which is found to be another manifestation of the vicious random walks in its densest phase and the dll model in two dimensions . we have dealt with the model on the isotropic lattice , and derived a few basic quantities of the model . we have also shown that the difference between our model and a collection of non - interacting simple random walks by distinct two - point functions . the other model is the stochastic flow model ( ii ) where the set of rules are given in eqs.([def : stof ] , [ eq : symdiff - directed ] , [ def : addition ] ) . the first rule ( [ def : stof ] ) is on the stability of the existing flow and is written in the euler formulation . we have utilised the notion of the oriented loop configurations @xmath44 and formulated the model . in order to interpret the fluctuation as a direct consequence of the kicks by the environment , it is worth rewriting the rule in the original lagrangian formulation . in this paper , we have focused on the construction of the model itself , and neglected this point . in numerical calculations , we have realised both models on the rectangular honeycomb lattice @xmath121 of various sizes in one million time steps . from the input data of the models on @xmath121 , we have found that the probability of the no - loop configuration , @xmath132 , has a scaling relation with the area of the lattice , and we have proposed the functional form ( [ eq : pfapprox ] ) for a general size of the honeycomb lattice as an approximation . we briefly report here that the probability of one - loop length - six configurations and that of length - ten can also be explained by the proposed form of @xmath62 . this might suggest that the form of @xmath62 with other additional criteria can explain how the distribution of the total length of the loops takes such a peculiar form in fig.[fig : looplen_dist ] . the area laws of the average loop length @xmath91 and the average loop number @xmath141 are also to be explained . we do not know the answers yet . in the numerical calculations of the flow model ( ii ) , we have shown the probability of how many fluctuations of @xmath162 had been added in a million trials . as @xmath98 goes over @xmath163 , the system seems to be stuck against the projected addition . this reduction of the fluctuations is entirely due to the definition of the projected addition . because of the second line of the projection rule ( [ def : addition ] ) , if @xmath164 is huge enough as a set , many elements of @xmath44 become non - additive . even so , practically speaking in quasi - critical states , there are about 30% vacant faces . therefore , the system can not be entirely frozen , changing only the time scale for its dynamics . it is the msd on which the time scale matters . especially , the emergence of the plateaus in fig.[fig : msd11 - 41 ] might be because of this strong restriction of the fluctuations . we then expect that softening the projection in the addition may change the intermediate plateau part of the msds , and this point is to be explored more elsewhere @xcite . there are many related statistical models . particularly , the vicious walk and the dll model are the closest ones to the model ( i ) . the vicious walk is the model of mutually avoiding multiple particle system so that its densest phase inevitably coincides with ours with the simplest stochastic process . the dll model is the model of cooperative particle rearrangement for supercooled liquids and polymer melts , usually equipped with volume fluctuations and their temperature dependence . in both cases , the dynamics is governed by completely independent time steps , and they surely overlap with this sort of the models among the family of our models . especially in the dll model , the dynamics is usually equipped with the temperature dependence of @xmath69 , so it would be intriguing to investigate another manifestation of the temperature dependence and multi - component systems @xcite in both contexts . in addition , the loop gas model is one of the closest models . letting all @xmath62 be one , one automatically obtains the partition function of the loop gas model from that of our stochastic markovian model of the lattice liquid . the loop gas model can be described in the @xmath165 spin model on the lattice by expressing all the loop graphs in terms of @xmath165 spin variables @xcite . to avoid complications , we have not included this algebraic description of our models . also , in the presence of the @xmath62 less than unity , the loop probabilities of both models become different . we can quantify them by the use of number series known as the derangement number and its related series . to invoke the expression , the models must be first written in the form of algebras . we have put aside this procedure , and therefore , to be given elsewhere @xcite . we are afraid that the constructed models are not so applicable and not ready to be compared to experiments , except for a few cases in a very different context . because we used the honeycomb lattice as an explicit example , the current model might be able to be applied directly to a particle system of the same situation : a bigger sort of particles form a most densely packed crystal sitting at the centres of the faces while smaller particles moving in between them . because of the set - up , the msd of the flow model actually resembles that of the super - cooled polymer melt @xcite . since their dimensions are different , we consider this happens by accident though . in order to apply 2d lattice liquid models to soft - material membranes directly , one should remove or relax some of the assumptions we used in this paper for convenience . namely , the assumptions are such that the system must be single - component , single - layered , incompressible , non - explicitly - viscous , less density fluctuating . also , we have not used interactions between particles . note that they are not the requirements nor the definitions of the 2d lattice liquid models . among them , double - layer can easily be embedded just by folding a layer into two , but the interactions between two layers are not obvious . the simplest one is to introduce the interaction between layers only at the common boundary of two layers , possibly as a boundary condition . our construction itself is ready to be extended to the multi - component case where there are more than one particle species and the interaction between different species should be introduced as well . the simplest one will be the ising - type nearest neighbour interaction with which one can observe the phase separation according to some temperature . this might also turn on the effective viscosity in the model . note the concept of the temperature can naturally be installed by letting @xmath166 be a function of @xmath67 , for example , by @xmath167 and @xmath168 . thermodynamic behaviour will depend on the manifestation of the functional forms of @xmath77 and @xmath68 . some of the above proposals will be studied elsewhere @xcite another very interesting yet possibly very hard model to solve is the fluid model with memory effect . as we mentioned in the manuscript , the `` potential '' probabilities can potentially carry the information from the past so that the past partially control the forthcoming event through the probabilities . here , we have only dealt with markov processes in both models , and the problem is beyond our scope of this manuscript . * acknowledgements * the authors acknowledge prof . k. yasuoka , prof . y. masubuchi , prof . s. yasuda , dr . t. akimoto , and all the members of the group _ multiscale simulation for softmatters _ for their stimulating discussions . one of the authors ( yi ) would like to thank prof . k. nagayama for his advice and encouragement , and s. ishimoto for his support . this research was supported by _ core research for evolutional science and technology ( crest ) of japan science and technology agency corporation ( jst)_. to begin with , let us define the ( oriented ) loop configurations on @xmath0 . for a given lattice @xmath0 , a one - loop graph , or configuration , is defined by a sequence of sites and bonds with no repeating sites and no ends . the object is the same as simple cycles in graph theory . an oriented one - loop graph is defined in the same way and is given by a one - loop graph with an orientation of the bonds representing its sequence . the space of the non - oriented non - intersecting loop configurations @xmath45 on @xmath0 is the whole set of the non - intersecting combinations of such one - loop graphs . the space of the oriented non - intersecting loop configuration @xmath44 is those of the oriented one - loop graphs . these spaces can be decomposed into ( oriented ) @xmath169-loop subspaces , or can be defined by them . let @xmath170 be a one - loop graph on @xmath0 for @xmath171 . then , a @xmath169-loop graph @xmath172 is defined by : g_k = _ i=1^k l_i , s.t . l_i _ j = for i j. collecting all such @xmath169-loop graphs in @xmath45 constitutes the @xmath169-loop subspace @xmath173 of @xmath45 : _ k ( ) \ { g_k } , _ k ( ) ( ) . when @xmath0 is finite , there exists the maximum number of co - existing non - intersecting loops on @xmath0 . denoting it by @xmath174 , one can write down the decomposition as : & & l ( ) = _ k=0^k_max l_k ( ) , & & s.t . l_i ( ) _ j ( ) = for ij . in the same manner , one can decompose the oriented space @xmath44 into the oriented @xmath169-loop subspaces : & & _ k ( ) \ { g_k } , _ k ( ) ( ) , & & ( ) = _ k=0^k_max _ k ( ) , & & s.t . _ i ( ) _ j ( ) = for ij , where @xmath175 is an oriented @xmath169-loop graph in @xmath44 . if we introduce the function @xmath176 which counts the number of elements of a set , then the uniqueness of the no - loop configuration in both spaces is trivially given by @xmath177 and @xmath178 . we label this unique configuration by @xmath104 . similarly , @xmath179 due to the orientations . one can further observe the relation : @xmath180 for @xmath181 . the algebraic structure of the space @xmath44 directly corresponds to the physical relations between the states on @xmath0 . however , even for a very basic question of how many configurations are in the space @xmath44 , we have not known any clear answer yet for an arbitrary size of the lattice . in fact , it is shown that counting the number of self - avoiding walks ( saws ) for fixed ends on a given graph is @xmath182-complete @xcite . @xmath182 is a class of the problems for counting and , roughly speaking , they are expected to be as hard as @xmath183-complete problems such as the prime number decomposition of an arbitrary large number . therefore , one can expect that counting the number of the elements of @xmath44 is as hard as @xmath183-complete problems as well , because a non - intersecting loop is nothing but a self - avoiding walk ending at its start @xcite . below , we illustrate the structure of the spaces in terms of the cycle space in graph theory . we restrict ourselves to the planar cases unless otherwise stated . when @xmath0 is planar , _ i.e. _ , @xmath0 can be embedded on a plane with no bonds crossing , the space @xmath45 can be spanned by the set of the smallest loops which are the perimeters of the interior faces of @xmath0 . in graph theory , there is the so - called cycle space @xmath184 which is generated by all the simple cycles and whose basis is given by the fundamental cycles . the addition generates the cycle space from the basis , and makes it the vector space of @xmath185 , where @xmath186 and @xmath187 are the number of the bonds and the number of the sites , respectively . the concepts of @xmath45 and the cycle space @xmath184 are very close to each other , but not always the same because of the presence of the intersecting loops in the cycle space . in fact , there is a necessary and sufficient condition for their equivalence that the degrees of the sites must be less than four everywhere : @xmath188 for all @xmath189 . otherwise , the addition allows intersecting loops . therefore , ( ) cycle ( ) , in general , and @xmath45 is not necessarily a vector space but can be spanned by the set of the smallest loops . when @xmath0 is a planar honeycomb lattice of a finite size , ( ) = cycle ( ) _ 2^e - v+1 , and ( l ( ) ) = 2^e - v+1 = 2^f , where @xmath131 denotes the number of the interior faces and the second equality is given by euler s formula on a planar graph . its basis can be given by the set of the smallest loops . note that , in general , ( l ( ) ) 2^e - v+1 . in the case of the oriented loop configurations @xmath44 , the relation is not as simple as above . we can only provide an estimate for the number of the elements to close . in the case of @xmath190 , & & 2^f+1 -1 < ( ( _ h_n,3 m ) ) < & & 2^ f - 2^(f+1 ) + 2f + 1 . a more refined relation will be given elsewhere @xcite . in this section , we define the probability functions of the no - loop configuration and of having ( oriented ) loops on the given graph , @xmath62 and @xmath61 , respectively . that is to say , the complementary functions satisfying : @xmath191 . upon their definitions , we will show some of the basic properties of the probability functions . note that the word ` loop graph ' indicates the same object as the loop configuration in the original context , but the latter emphasises an outcome of some stochastic process . on the loop graphs as events of the stochastic process , we define the following probability functions . let @xmath192 be the probability of having the loop graph @xmath172 in @xmath0 . let @xmath193 be the probability of having any loop graph on the graph @xmath194 . let @xmath88 be the probability of the graph @xmath195 being completely frustrated , _ i.e. _ , having no loops . here , the graph @xmath196 is an extended graph of @xmath76 such that it additionally includes all the bonds adjacent to the sites in @xmath76 , in order to precisely explain and estimate the stochastic nature of our model . introducing the complementary graph @xmath197 to a loop graph @xmath198 , the probability function @xmath199 of having the loop graph @xmath198 in @xmath0 is given by the product of the probability @xmath200 of the loop configuration @xmath198 on the graph @xmath198 , and the probability @xmath201 of all the other sites being frustrated : p(l ) = p_l(l ) p_f(|l ) . by definition , the following equalities and inequalities hold : [ eq : ineq ] p_l ( ) & = & _ g_k>0 ( ) p(g_k ) , p_f(g ) & = & 1 - p_l(g ) , p_f ( ) = 1- p_l ( ) = 1 , 0 & & p_l(g ) 1 , 0p_f(g ) 1 . in the third line , for finite @xmath76 and @xmath0 , the equalities @xmath202 and @xmath203 only hold when @xmath76 or @xmath0 is linear or the assignment of the `` potential '' probabilities of the particles moving are unnaturally given in the non - stochastic limit . therefore , we assume @xmath204 and @xmath205 for finite size @xmath76 and @xmath0 . letting @xmath198 be the naive projection of the oriented loop graph @xmath206 by removing the orientations , and assuming that the different oriented loop graphs which share the same projected loop graph @xmath198 are equiprobable , then the mapping of the oriented loop graphs by the probability functions become apparent as : p(g_k ) & = & p_l(g_k ) p_f(|g_k ) , p(g_k ) & = & 2^k p(g_k ) , p_l(g_k ) & = & 2^k p(g_k ) , p_f(g_k ) & = & p_f(g_k ) . @xmath175 is not a subgraph of @xmath0 so that the complementary graph is defined by the projected graph @xmath172 : @xmath207 . one can interpret that the same projection is done for the @xmath62 as well . because the frustrations in disconnected graphs are independent , the following relation holds : for two disconnected graphs , @xmath208 : p_f(g_1 g_2 ) = p_f(g_1 ) p_f(g_2 ) g_1 g_2 = . note that , using the above definitions and successive application of the relations , one can obtain @xmath192 in a nested expression of the loop probabilities only in this section , we calculate the asymptotic value of the mean square displacements , @xmath209 , which is the average of @xmath210 over @xmath97 , in the limit of @xmath211 . before going into the actual calculation , we fix the notations and introduce some functions to facilitate it . number the position on @xmath0 starting from 0 to @xmath212 , and assign a pair of integers to the @xmath68-th site , @xmath213 . the position of the @xmath68-th site , @xmath214 , on the two - dimensional @xmath215 plane can be given by , for the @xmath216 honeycomb lattice : & & p_x = p mod col , p_y = [ p / col ] , & & x_p = p_x , y_q = 32 p_y + 14 ( -1)^p_x+p_y+1 , where @xmath217 , and @xmath218 $ ] denotes the greatest integer which does not exceed @xmath219 . @xmath220 takes the value from @xmath37 to @xmath221 , while @xmath222 . @xmath223 are all integers . the lattice spacing is taken to be a unit length . the square distance between @xmath68-th and @xmath224-th positions are : & & r_q , p - q^2 = ( x_p - x_q ) ^2 + ( y_p - y_q ) ^2 & & = 34 \ { ( p_x - q_x)^2 + 3 ( p_y - q_y)^2 } & & + 18 \{1 - 6(-1)^p_x+p_y ( p_y - q_y ) } ( 1 - ( -1)^(p_x - q_x ) + ( p_y - q_y ) ) . note that we used the following identity : @xmath225 for @xmath226 . replacing @xmath227 by @xmath228 , one can rewrite this as @xmath229 : [ c3 ] r_q,(k , l)^2 & = & 34 k^2 + 94 l^2 + 18 c(q , l ) - 8 c(q , l ) ( -1)^k+l , where @xmath230 from this expression , one can verify that @xmath231 is integer . assume that , in the limit of @xmath159 , all the pairs of the sites are independent to each other so that the asymptotic value of @xmath209 can simply be given by the average of the square distances of all the relevant pairs in @xmath232 combinations . namely , the value leads to the expression : [ eq : r_inf ] [ eq : r^2 ] & & _ t & & = _ relevant . here , the word `` relevant '' means that some sites at the boundary can not contribute to the above , because our flow dynamics can only involve those points which have more than one bond . for instance , regardless of the value of @xmath98 , the following two sites should always be excluded : q = \ { ( 0,0 ) , ( 2n,0 ) } . besides , when @xmath98 is odd , the following two sites are to be taken into account in the same manner : q = \ { ( 0,n ) , ( 2n , n ) } . they are connected only one nearest neighbouring site , so that the particle can not reach these four points via our flow dynamics . therefore , the square distances related to these points do not contribute to the above calculation . substituting ( [ c3 ] ) into ( [ eq : r^2 ] ) and subtracting such unnecessary terms from the sum , we obtain for odd @xmath98 : & & & & = & & & & 18 ( 7 n^2 + 10 n ) , up to @xmath233 . in the cases of @xmath234 , the exact values are given in the manuscript up to first five digits . h. noguchi and g. gompper , phys . e * 73 * , 021903 ( 2006 ) ; t. akimoto , e. yamamoto , k. yasuoka , y. hirano , and m. yasui , phys . lett . , * 107 * , 178103 ( 2011 ) ; h. noguchi , j. chem . phys . * 134 * , 055101 ( 2011 ) ;

ultraluminous infrared galaxies ( ulirgs ) output more energy in the infrared than at all other wavelengths combined , and have @xmath8 . in the local universe , these relatively rare objects are frequently associated with peculiar galaxies , thought to be merging systems @xcite . their high infrared emission comes from warm dust , heated by a combination of starbursts and active galactic nuclei ( agn ) which are thought to be fueled by huge quantities of molecular gas that fall inward toward the nucleus during a major merger of two gas - rich spiral galaxies @xcite . while at the present epoch ulirgs are rare in number , numerical simulations by @xcite suggest that they are not merely anomalies resulting from special " merger conditions , but rather their rarity is due to the fact that they represent a short - lived yet common phase of galactic evolution , namely when spiral galaxies collide to form massive elliptical galaxies . detections by the scuba telescope of hundreds of high - redshift sub - mm galaxies ( smgs ) lend further support to this theory ( e.g. * ? ? ? the properties of smgs are similar to those of local ulirgs @xcite , and the number density of smgs is similar to the number density of massive elliptical galaxies at the present epoch . thus , assuming that structure in our universe forms by hierarchical assembly , ulirgs offer us the unique opportunity to glimpse a brief but important stage in a galaxy s life , when massive spirals transform into ellipticals . though ngc 6240 ( @xmath9 , @xmath10 mpc , @xmath11 pc ; @xmath12 km @xmath13 mpc@xmath14 , @xmath15 , @xmath16 ) technically lies at the high - luminosity edge of the lirg class with @xmath17 , in many ways it is a prototypical ulirg , and is commonly referred to as such . ngc 6240 is a late - stage merger of two massive disk galaxies @xcite ; it has extended tidal tails and a close nuclear pair separated by @xmath01.5 , or a projected distance of 715 pc ( e.g. * ? ? ? * ; * ? ? ? its k band light is dominated by red supergiants in @xmath020 myr - old starbursts @xcite , and at the double nucleus two agn have been discovered in hard x - rays @xcite as well as at 5 ghz @xcite . as in other merging systems , most notably ngc 4038/9 ( the antennae galaxy ) @xcite , star clusters have previously been observed in ngc 6240 s main body and tidal tails with the wfpc2 f450w , f547w , and f814w filters @xcite . but recently @xcite detected previously unseen circumnuclear star clusters at j , h and k bands using natural guide star ( ngs ) adaptive optics ( ao ) with the nirc2 wide - field camera on keck ii . these clusters , and other similar clusters in ulirgs and galaxy mergers , may hold clues regarding a variety of open questions , from the distribution of dust and star formation in ulirgs , to the formation and evolution of globular clusters ( gcs ) . historically , measuring star formation rates ( sfrs ) in galaxy mergers has been difficult , and until recently the contribution of smgs to the total luminous energy density in the high - redshift universe has been significantly underestimated . a variety of heterogeneous techniques have been used to measure sfrs @xcite , but in starburst galaxies these are complicated by the presence of dust @xcite , including the usual methods of measuring sfrs through h@xmath18 line fluxes or uv continuum luminosities . in these cases astronomers have turned to far - infrared ( fir ) wavelengths to measure the dust - reprocessed light . but a well - known problem arises when using fir luminosities to estimate sfrs in ulirgs : in addition to young ob stars , agn also help heat the dust . while studies have shown that the luminosities of most ulirgs are dominated by star formation even when agn are present @xcite , accurately measuring the fractional agn contribution to the bolometric luminosity ( @xmath19 ) has proved difficult . for example , @xcite reported on their detection of the [ ne v ] 14.3 @xmath20 m emission line feature , with a flux of @xmath21 erg @xmath22 s@xmath14 , which provides direct evidence of the buried agn in ngc 6240 . they note that the [ ne v]/[ne ii ] ratio is often incorrectly treated as a reddening independent quantity , even though [ ne v ] comes directly from gas heated by the agn , while [ ne ii ] can be dominated by a starburst . without correcting for dust , @xcite state that the large equivalent width of the @xmath23 polycyclic aromatic hydrocarbon ( pah ) emission feature , and the small [ ne v]/[ne ii ] and [ ne v]/ir flux ratios , are consistent with an apparent agn contribution of only @xmath24% of @xmath19 . but after correcting the measured [ ne v ] flux for extinction using values implied by their sed fitting , @xcite estimate that the intrinsic fractional agn contribution to the bolometric luminosity of ngc 6240 is much higher , between 20 and 24% . by detecting individual , massive , star clusters in merging galaxies we can trace the most violent star formation activity , while circumventing some of these hurdles which make accurately measuring sfrs a difficult task . for example , the circumnuclear locations of clusters makes them easily distinguishable from agn . in addition , while the exact size distribution of dust grains in both star clusters and diffuse stellar populations is impossible to know , the complicated dust mixing models required for analyzing diffuse stellar fields reduce to simple dust screens toward clusters , if we assume that all intracluster dust has been evacuated . even when some intracluster dust is present , the dust screen approximation is still reasonable assuming that the amount of dust internal to the cluster is far less than the intervening dust along the line of sight through the entire merging galaxy . furthermore , individual clusters are well approximated by single stellar populations , making stellar population analysis of clusters more straightforward than for diffuse stellar fields . finally , near - infrared ( nir ) cluster observations provide a more localized measure of the sfr compared to estimates based on longer wavelength observations . but unfortunately , while star clusters seem to represent a common mode of star formation in spirals and lirgs alike ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , detecting dust - reddened clusters is challenging because longer wavelength observations are innately of lower spatial resolution . however , with the improvement of ao systems on large , ground based telescopes , the possibility of detecting individual infrared star clusters , even at @xmath25 , is now a reality . in order to measure _ localized _ sfrs with fewer complications due to agn and dust , we have obtained @xmath26 resolution images of roughly 10 kpc surrounding the merging galaxies nuclei using the nirc2 narrow - field camera with ngs ao at keck . we combine our data with hst images taken with the nicmos nic2 camera ( using the f110w , f160w , and f222 m filters ) and the wfpc2 planetary camera ( using the f450w and f814w filters ) . using the combined data set we study a total of 32 clusters with inhomogeneous wavelength coverage , partially due to variations in the fields of view , and partially due to non - detections in some wavebands . in order to estimate the sfr associated with these clusters we estimate masses , ages , and extinctions toward these 32 clusters by comparing the clusters spectral energy distributions ( seds ) to model seds from stellar population synthesis codes @xcite , employing simple , screen extinction models by @xcite . given the data currently available , we do not attempt to quantify the cluster s properties with the precision of previous studies of nearby mergers . this work is meant to be a preliminary analysis of ngc 6240 s recently discovered cluster population one of the most distant such populations ever detected ; future high - resolution spectral observations will provide a much clearer understanding of the cluster s properties . in [ observations ] we describe our observations , in [ data processing ] we describe our data reduction , including the techniques we used to identify clusters and perform photometric calibrations , and in [ results ] and [ discussion ] we present and discuss our results . we observed ngc 6240 with the ngs ao system on the keck ii telescope using the narrow - field camera ( 0.009942 pixel@xmath14 ; 4.74 pc pixel@xmath14 ) of the nirc2 infrared instrument at h ( @xmath27 m ) and k ( @xmath28 m ) bands . the h and k observations were obtained on 24 april 2005 and 17 august 2003 , respectively . the natural guide star ( b = 13.5 ) is usno - b 1.0 0924 - 0386013 , located 35.8 east - northeast of ngc 6240 s dual nuclei . due to the faint nature of the star clusters superposed on the diffuse background light from the galaxy , it is problematic to measure the resolution and strehl ratio ( s ) at the location of the clusters , i.e. off - axis from the guide star . we did not observe psf star pairs and therefore can not reasonably guess what the strehl may be at the clusters locations . if there were no tip - tilt errors then the core of the off - axis point spread function ( psf ) would be diffraction - limited , and at 2.1@xmath20 m the full - width - half - maximum ( fwhm ) of this diffraction - limited , off - axis psf core is @xmath29 ( @xmath30 pc at @xmath31 ) , making our observations slightly oversampled . throughout this paper we assume that the off - axis psf and strehl whatever they may be are constant across the off - axis fov . this assumption is well justified when a small fov ( @xmath32 ) is far off - axis ( @xmath3330 ) , and when the areas of interest in the fov are oriented perpendicular to the direction of the guide star . most of the clusters in our images are located along the north - south orientated double nuclei , and therefore are oriented approximately perpendicular to the direction of the guide star . the assumed constancy of the resolution and strehl ratio across the fov implies that our detection limits and photometric measurements should be accurate in the relative sense ; in [ caseii ] ( case ii ) we describe the steps we have taken to ensure that our photometric measurements are accurate in the absolute sense as well , by sidestepping the need to know the absolute strehl ratio to perform photometry . in table [ tableobskeck ] we outline the details of these keck observations , including the on - axis strehls achieved for the guide star on each night . while the off - axis strehl is unknown , the strehl measured on the guide star is an indicator of the overall performance of the ao system , including the effects of wind turbulence , sampling rate , and weather . we have analyzed archival wfpc2 images taken as part of the go proposal 6430 ( pi : r. van der marel ) . the observations were taken with the planetary camera ( .0455 pixel@xmath14 ; 21.7 pc pixel@xmath14 ) using the f450w and f814w filters on 28 march 1999 . the theoretical fwhm of the diffraction - limited psfs for these two wfpc2 images are @xmath34 and @xmath35 , respectively . at ngc 6240 s distance , these angular resolutions correspond to 18.1 and 32.7 pc , respectively . the f450w and f814w filters are roughly equivalent to the johnson - cousins b and i filters , therefore throughout this paper we use the shorter johnson - cousins _ notation _ interchangeably with the true wfpc2 filter names . however , when performing all calculations and measurements we use the exact filter transmission functions for the wfpc2 filters . we have also made use of archival nicmos images obtained on 1998 february 02 with the hubble space telescope using the near infrared camera and multi - object spectrometer ( nicmos ) as part of the guaranteed time observing ( gto ) program 7219 ( pi : scoville ; @xcite ) . the imaging program employed three spectral filters ( f110w : @xmath36 , fwhm = 0.592 @xmath37 ; f160w : @xmath38 , fwhm = 0.403 @xmath37 ; f222 m : @xmath39 , fwhm = 0.143 @xmath37 ) in nicmos camera 2 ( pixel scale : @xmath40 pixel@xmath14 , @xmath41 pixel@xmath14 ; @xmath42 36.3 pc pixel@xmath14 , @xmath43 36.0 pc pixel@xmath14 ) . four exposures in each filter , using step8/nsamp = 10 , 11 , and 12 multiaccum sampling @xcite for f110w , f160w and f222 m , respectively , were taken with corresponding integration times of 40s , 48s , and 56s in each exposure . nicmos camera 2 critically samples its point spread function at @xmath44 ( see * ? ? ? all images were acquired under a four - point dither " pattern with pointing offsets of 1.9215 , thus enabling critical sampling at @xmath45 ( after image combination ) , as well as improved sampling and rejection of data from defective pixels at all wavelengths . because the f160w and f222 m filters provide wavelength coverage roughly similar to the h and k mauna kea filters ( fig . [ filters ] ) , these two nicmos images are used primarily as a tool to help perform photometric calibrations on the nirc2 , ao images . ( the details of this photometric calibration technique are described in [ fluxcalibrations ] . ) however where the fields of view of the nicmos f160w and f222 m images extend beyond the nirc2 fields of view , the nicmos images provide additional spectral information and thus we analyze these images . note that of all the data discussed in this paper the nicmos images have the lowest spatial resolutions about 4 times lower than the resolution provided by the keck ao system at similar wavebands . the theoretical fwhm of the diffraction - limited psfs for the three nicmos images are 0.093 , 0.13 , and 0.18 , corresponding to linear scales of 44.2 , 64.3 , and 85.0 pc . details regarding both the wfpc2 and nicmos hst observations are shown in table [ tableobshst ] . the complete set of data is composed of 7 images from 3 different instruments at the following bandpasses , in order of increasing central wavelength : b , i , f110w , f160w , h , k , and f222 m . the inhomogeneous nature of these images , including the manner in which they were taken , the morphology that they display , and their resolutions which span more than a factor of five , requires that we employ non - uniform processing techniques . we shall discuss the pitfalls inherent to treating data in a non - uniform fashion . the near - infrared images taken with nirc2 were reduced in the standard way , by masking bad pixels , flat fielding each frame using twilight images , and dark subtracting and sky subtracting frames of the appropriate exposure times . before combining nirc2 narrow camera frames , we removed the geometric distortions using the iraf drizzle package with the distortion coefficients published in the nirc2 manual , and the kernel parameter set to `` lanczos3 . '' the nicmos data analyzed in this paper were obtained by recalibrating and reprocessing the archival data . instrumentally calibrated count - rate images were created from the raw archival multiaccum frames after applying linearity corrections , dark - frame subtraction , dc bias compensation , flat - field correction , and cosmic - ray rejection . darks , made from contemporaneous on - orbit calibration observations , calibration reference files for linearity corrections , obtained from stsci , and flats , obtained from the nicmos instrument definition team ( idt ) , were applied in an idl - based process closely following the iraf / stsdas calnica task . the image data were photometrically calibrated based upon the absolute instrumental calibration established by the idt ( f110w : @xmath46 jy count@xmath14 s ; f160w : @xmath47 jy count@xmath14 s ; f222 m : @xmath48 jy count@xmath14 s ) . known bad " or defective pixels were replaced by two - dimensional gaussian weighted interpolation of good neighbor pixels with wavelength - dependent weighting radii of 3 , 5 , and 7 pixels for the f110w , f160w , and f222 m frames respectively . the fully calibrated images were geometrically rectified to correct for the @xmath49% linear geometrical distortion in nicmos camera 2 , and for each filter , those four images were median combined into a single image on a 2-times finer resampled grid ( @xmath50 mas pix@xmath14 ) . all image resampling ( including re - mapping for registration , and rectification ) was done with sinc - function apodized gaussian weighted interpolation using the idp3 package developed by the nicmos idt @xcite , conserving the target flux density while resampling . the three resulting images ( one for each filter ) are those that are discussed further in this paper . we obtained the archival wfpc2 data from the wfpc2 associations at mast ( multimission archive at stsci ) , and these data are corrected for bias , dark current , flat fielding , and cosmic rays . we performed no additional processing on these wfpc2 associations . we attempted to identify unresolved star clusters in the 7 images in a systematic way by using the find routine adapted for idl from daophot . however , this method and variations on this method , including using the find algorithm on unsharp - masked images , were all unsuccessful , especially in regions where the diffuse background light from the galaxy is highly variable or bright . therefore in this paper we have visually identified clusters , and we define a cluster to be any source that appears unresolved in an unsharp masked image . we neglect clusters that are visible only at b and/or i bands because these have been previously analyzed by @xcite , and our analysis would provide no new spectral coverage . figure [ grayimage ] shows the cluster positions superimposed on images at b , i , and k bands . in total we identified 32 clusters that are visible in at least one of the infrared passbands ( f110w , f160w , h , k , and f222 m ) . of these , 27 lie within the intersection of the 7 fields of view , lending themselves to the most complete spectral analysis . of these 27 clusters , only 7 have low enough extinction to be detected at i band , and this number decreases to 5 at b band . figure [ fovfig ] shows the locations of all 32 clusters and the relative sizes and orientations of the 7 fields of view . in figure [ clusterids ] we label each cluster with a number from 1 through 32 with 1 being the northernmost cluster . note that cluster # 15 is coincident with the site of the northern black hole , as determined by @xcite . for reference , in the publicly available wfpc2 i band image , the brightest pixel in the northern nucleus is 0.27 east and 0.099 north of the northern black hole ( i.e. cluster # 15 ) . in figures [ fovfig ] and [ clusterids ] we mark the position of the northern black hole with an orange cross . the positions of each cluster relative to cluster # 15 can be found in table [ clustertable ] . the complicated morphology of ngc 6240 makes it impossible to specify a single cluster detection limit across the entire fov . the morphology of the h and k images is relatively smooth , with background luminosity increasing sharply toward the double nuclei . therefore the cluster detection limits at h and k worsen monotonically towards the double nuclei . while it is generally true that the detection limits worsen toward the double nuclei in the b and i images as well , the situation here is more complicated because these images are far more patchy due to dust obscuration . we estimate our detection limits by adding simulated clusters to various locations in each image , and decreasing the fluxes of the simulated clusters until the clusters become undetectable by eye . ( see [ fluxcalibrations ] cases i and ii for a detailed description of how we create the simulated clusters . ) using this method , we estimate that clusters brighter than @xmath51 mag , @xmath52 mag , @xmath53 mag , and @xmath54 mag should be detectable throughout much of the images . in order to align all 7 images to the same spatial grid we first visually determined which star clusters are common to pairs of images . ( no stars are visible in the small nirc2 fov . ) we fed the centroids of the matched cluster pairs into the iraf geomap routine , and in this way computed second - order polynomial spatial transformation functions . we applied the transformations with the iraf geotran package using bilinear interpolation to rebin the images . due to ngc 6240 s drastically different morphology between b and k bands , finding enough clusters in common at these two wavelengths was challenging . therefore , instead of aligning the b image directly to the k image , the b image was first aligned to the i image , and then to the k image using the transformation computed for the i - k alignment . this two step process also ensures that @xmath55 colors are as accurate as possible , with the fewest errors introduced from poor image alignment . typical rms fitting errors computed by geomap were @xmath56 when aligning keck images , @xmath57 when aligning hst images , and @xmath58 when aligning hst images to keck images . because the star clusters , although unresolved , can not be conveniently modeled as many point sources on a smooth sky background , and because the strehl ratio for the ao images is uncertain , we performed aperture photometry rather than using the standard psf - fitting photometry packages like daophot . we explored two aperture photometry techniques . first , we simply performed aperture photometry on all images using an aperture radius of 0.07 and a 0.02-wide sky annulus centered at a radius of 0.1 from the cluster centroid . this annulus is near enough to the cluster that we can accurately measure the local background in a complicated environment , while the aperture is large enough to lessen the effects of poor image alignment . second , we tried to minimize the color errors by creating matched - resolution image pairs before performing aperture photometry . we found that this second technique produced poorer results because we were forced to degrade the image quality of the nirc2 images , making many of the clusters disappear into the noise . therefore , the results presented in this paper were calculated by performing simple aperture photometry on all the clusters , using the same aperture and annulus size for each filter . note that the wfpc2 and nirc2 images have very similar resolutions , with the best resolution ( from the h band ao image ) being a factor of 2 better than the worst resolution ( from the wfpc2 i band image ) . however the nicmos images have poorer resolutions than the wfpc2 and keck ao images , so that performing aperture photometry on these images with the same size aperture and annulus introduces color errors in the case where different colored background flux from the galaxy falls within the aperture . thus caution must be used when interpreting the nicmos data , especially in regions near the nuclei where the background flux from the galaxy is bright . in figures [ seds_7_18 ] - [ seds_1_32 ] we present the results of our aperture photometry measurements in the form of seds for each cluster , and in the following paragraphs we outline the procedure we used to create these seds for two key cases . other cases can be deconstructed into variations on these two cases . in order to populate the sed of a cluster that is detected in only the b , i , f110w , f160w , and f222 m images , first we perform aperture photometry on the cluster in each image ( using aperture parameters just described ) . we calculate appropriate aperture corrections by repeating this procedure on simulated point sources extracted from the tinytim ( ver 6.3 ) software package . finally , we convert the ( aperture corrected ) counts to calibrated flux ( in janskys ) using the photflam and photplam keywords in the fits header of the wfpc2 images , and the instrumental calibration values determined by the nicmos idt ( see [ datareduction ] ) . on an sed graph we represent these measured fluxes in the form of a single point at the position of the central wavelength for each filter . to calculate the photometry errors shown in figures [ seds_7_18 ] - [ seds_1_32 ] , for each cluster we simulated 100 additional clusters using tinytim psfs . the simulated clusters were made to have the same sed as that which was measured for the real cluster , and they were placed in approximately similar areas of the image as the real cluster . ( we defined 3 regions of the images far , close , and intermediate distances from the nuclei . ) next , we performed the same aperture photometry procedure on the simulated clusters as we did on the real clusters . the resulting distribution of 100 measurements is sensitive to the exact placement of the simulated clusters and is not gaussian . therefore we discard the 10 smallest and largest measured values , and quote errors as 80% confidence limits from the pared distribution . the error bars on the single data point at which all models are normalized represent something slightly different from the error bars on the other points in the sed . the former are 80% confidence limits on the flux at that frequency , while the latter are 80% confidence limits on the color . thus when analyzing these normalized seds , which are plotted in log - linear space , realize that the seds may shift vertically by an amount governed by the error bar on the point at which the seds are normalized , while the slopes of the seds may change only by the size of the error bars on the _ other _ points . since the slope of an sed is governed by the stellar population of that cluster and any intervening dust , the color error bars indicate how confidently we know the cluster s stellar population . likewise , since the total mass of the cluster corresponds to an overall multiplicative factor , i.e. a vertical shift in log - linear space , the flux error bars indicate how confidently we know the cluster s mass , assuming no errors on the stellar population and extinction values . in the error analysis procedure described above , if the 80% confidence limits indicate that the cluster may have negative flux , as is sometimes the case when performing aperture photometry on barely detected clusters , then we replace the lower error bar with a downward arrow , signifying an upper limit . ( we still plot the upper error bar . ) for clusters that we deem undetected we calculate upper limits using a different method . we create a simulated cluster using a tinytim psf , and place it exactly at the location of the known cluster . we decrease the flux of the simulated cluster until it is undetectable ( by eye ) and call the final flux an upper limit . because absolute flux calibration of ngs ao images is very difficult when the off - axis strehl is unknown , the procedure we have adopted for populating a cluster s sed with photometric information taken by the keck ao system is quite different from the procedure outlined for hst in case i. as an example , we outline the procedure used to populate the sed of a cluster detected only in the h and k images . cluster 18 is an example of such a cluster for which the following procedure was employed . we begin as before by first performing aperture photometry on the cluster in the k image . next , because we have no way of performing an accurate aperture correction or converting counts into janskys for this image taken using ao , we make use of the nicmos f222 m image . though the particular cluster in question is not visible in the lower resolution nicmos image , one cluster , namely cluster # 2 , _ is _ visible in both images , and is relatively far from the bright nuclei . we calculate the total flux of cluster # 2 in the f222 m image using the standard procedure for nicmos data , namely performing aperture photometry and applying an aperture correction calculated from a tinytim psf , and converting aperture corrected counts into calibrated janskys by using the appropriate instrumental calibration value determined by the nicmos idt . next we perform aperture photometry on cluster # 2 in the k image . the ratio of the total flux of cluster # 2 to the aperture photometry result can be used to convert from counts in the k image ( calculated using a particular aperture and annulus ) to _ total _ , _ calibrated _ flux for any cluster . we use this ratio to convert the counts in the particular aperture and annulus for cluster # 18 in the k image to _ total _ , _ calibrated _ flux . on an sed graph , this flux is represented in the form of a single point at the position of the central wavelength for the k filter . because the k and f222 m filters are not identical ( as can be seen in fig . [ filters ] ) , the bootstrapping procedure described above produces accurate results only when the slopes of the sed in the nir are flat , or the same , for clusters 2 and 18 . indeed , the nir sed is very flat for the case of young , dusty clusters . to estimate the errors introduced by this bootstrapping procedure we convolved various @xcite stellar population models with the transmission functions for the f222 m and k filters . assuming a salpeter imf , solar metallicity , and padova evolutionary tracks , we determine that observations taken with the two different filters should differ by less than 0.1 magnitudes for a cluster between the ages of 0 and 10 gyr . if an intervening dust screen , corresponding to extinctions ranging from @xmath59 to 12 , reddens the cluster and flattens the nir sed , then observations taken with the two different filters should differ by less than @xmath60 for a cluster that is less than 1 gyr old . a completely analogous procedure , which makes use of cluster # 29 , is used to bypass the aperture correction and flux calibration of the h band ao images ; the errors introduced by calibrating h band data with f160w images are comparable to the 0.1 mag errors quoted above . the photometry error calculations and the upper limit calculations in the case where the cluster is only detected in the ao images are completely analogous to the procedures described in case i. however , instead of using tinytim psfs to simulate clusters , we used a theoretical ao psf . we model the ao psfs as the sum of two gaussians . one gaussian has a fwhm equal to @xmath61 where @xmath62 is the observing wavelength and d is the 10 m diameter of the keck telescope . the other gaussian has fwhm equal to @xmath63 , where @xmath64 is the fried parameter , which we assume to be 30 cm . we let the on - axis strehl ratio govern the ratio of the total amounts of energy in each gaussian . we recognize that the strehl used for the simulated clusters is certainly over estimated . one of the consequences of these assumptions is evident in figure [ seds_1_32 ] , where upper limits for cluster # 31 are clearly far too low . ( larger strehls make fainter clusters easier to detect . ) therefore when interpreting infrared upper limits shown in figures [ seds_7_18 ] - [ seds_1_32 ] , the upper limits computed from nicmos data for which the psf is well modeled by tinytim should be trusted more than the upper limits computed from the ao data . in this complicated galaxy merger the values we have calculated , namely a cluster s magnitude in _ at most _ 5 distinct wavebands , are outnumbered by the unknown variables , which include , but are not limited to , the following list : 1 ) the metallicity of gas that formed each cluster ; 2 ) the duration and temporal profile of the starburst for each cluster ; 3 ) the initial mass function ( imf ) of stars in each cluster ; 4 ) the stellar evolutionary track followed by every star in each cluster ; 5 ) the extinction , scattering , and absorption properties of the dust in and in front of each cluster ; and 6 ) the mass and age of each cluster . in principle we would like to find the stellar population that most closely matches the observed photometry for each cluster , but because it is impossible to fully constrain this problem , rather than perform minimization techniques on a many - dimensional parameter space we adopt a different approach which requires making the following simplifying assumptions . we assume that all clusters : 1 ) have solar metallicity ; 2 ) were formed in an instantaneous burst ; 3 ) are composed of stars with stellar masses governed by a salpeter imf ; 4 ) are composed of stars that evolve on padova ( 1994 ) evolutionary tracks ; and 5 ) have evacuated all intracluster dust before an age of @xmath03 myr , and are obscured behind a simple dust screen that behaves according to @xmath65 . the fact that all 32 of these clusters are concentrated within @xmath66 kpc of the double nuclei ( in projection ) suggests that their formation was triggered by the merger and that they were formed roughly contemporaneously . to simplify our analysis we initially assume that all clusters were formed simultaneously , although this is obviously false . later we relax this assumption to allow for clusters of three possible young ( or intermediate ) ages . it is possible that some subset of the clusters are old gcs from the progenitor galaxies that happen to be passing through the center of the merger , but this is surely not the case for the majority of the clusters especially those closest to the nuclei . assuming the clusters are not old gcs , it is reasonable to choose a metallicity close to solar . if the triggering event was short , then the assumption that the clusters stellar populations are well modeled by an instantaneous burst is roughly accurate . however , those clusters that are located very near to the double nuclei may be experiencing a continual accretion of gas , and therefore it is possible that they would be better modeled by a stellar population with a constant sfr . we leave the clusters ages and the amount of intervening dust along the line of sight to each cluster as free variables . after making the aforementioned simplifications , our approach entails visually comparing the seds that resulted from performing aperture photometry at b through k bands , to model seds created by the isochrone synthesis code of @xcite . in figures [ seds_7_18 ] - [ seds_1_32 ] we present the results of this approach for the case where we assume that all clusters are 14.5 myr old . ( we justify this choice of age shortly . ) we organize the results into 3 groups ; figure [ seds_7_18 ] shows the 12 clusters closest to the northern nucleus , figure [ seds_3_28 ] shows the 12 next closest clusters , that lie at an intermediate distance from the norther nucleus , and figure [ seds_1_32 ] shows the remaining 8 clusters that are farthest from the northern nucleus . all model seds shown in these figures embody the former 6 simplifying assumptions , and for the purpose of visual comparison , each sed has been scaled so that it passes through either the k or the i data points . implies an insignificant k - correction . ] the different colors of the sed curves correspond to varying column densities of interstellar dust applied using the @xcite dust prescription with @xmath67 . ( grain size distribution comes from @xcite ; pah optical properties come from @xcite . ) for the 12 clusters nearest the northern nucleus , column densities corresponding to between 3.5 and 11 magnitudes of visual extinction were required to match the observations with the models . similarly , when comparing the intermediate - distance clusters to the model seds , column densities corresponding to @xmath5 between 4.5 and 14 mag were applied to the models . but for the clusters farthest from the northern nucleus , much smaller column densities , corresponding to @xmath68 mag , were needed to produce seds spanning the range of observations . immediately we see that if our simplifying assumptions are correct , and if all clusters are assumed for the time being to be the same age of 14.5 myr , then on average the outermost clusters are obscured by a thinner dust screen than that which obscures the inner clusters . we repeat this procedure for two additional cases , where all clusters are 2.88 myr old , and where all clusters are 180 myr old . these ages were chosen for the following two reasons . 1 ) the ages span a range that is consistent with our previous assumptions . for example , clusters younger than 3 myr could have considerable intracluster dust , and clusters older than several hundred million years would have been formed before the onset of the merger , thus it is unlikely that they would have formed contemporaneously or be situated coincidentally near the nuclei , the masses required to match the observations to the models for clusters older than @xmath0180 myr are too large to be credible . ] . 2 ) the ages bracket the results published in the literature for other star clusters in ngc 6240 . using photometry at the f450w , f547w , and f814w wavebands , @xcite found that the clusters in ngc 6240 s main body and tidal tails have probable ages ranging from 5 to 13 myr . by analyzing the co 2 - 0 absorption bandhead equivalent width , @xcite show that ngc 6240 s k - band light is dominated by a @xmath020 myr - old , short - duration starburst . for each of the three ages analyzed , we choose as the best - fitting " model the sed with the smallest value of @xmath5 that matches the observations for each cluster . for example , the observed sed for cluster # 13 in figure [ seds_7_18 ] is consistent with model seds having @xmath5 ranging from 5.5 to 16 mag ; therefore we assume that cluster # 13 , _ if it is 15 myr old _ , is extinguished by @xmath69 mag . note that larger extinctions translate to more massive clusters , so our method finds the least massive cluster that matches the data for a given age . we do not show graphs of the 2.88 and 180 myr - old model seds for every cluster ; instead we tabulate the results of this exercise in table [ clustertable ] . the values of @xmath5 listed for each cluster and for each assumed cluster age correspond to the best - fitting " model just described . the masses of each cluster are directly proportional to their un - extincted fluxes ; for two clusters with equal ages and equal apparent magnitudes , the more extinguished cluster is more massive . in figure [ histograms ] we depict these same results in the form of histograms . the blue , red , and yellow histograms show the best - fitting " @xmath5 , and corresponding mass and ( un - extinguished ) @xmath70 , of all 32 clusters for the cases where all clusters are assumed to be 2.88 , 14.5 , and 180 myr old , respectively . while the values listed in table [ clustertable ] indicate the stellar population with the best - fitting " @xmath5 assuming 3 different cluster ages , we have not yet specified which age stellar population best matches the observations . we address this next step using figure [ examplaryplots ] , in which we use 3 typical clusters to directly compare the model seds with the best - fitting @xmath5 for 4 different ages . each of the measured seds for clusters # 1 , 28 , and 21 are shown over - plotted with model seds corresponding to stellar populations that are 2.88 , 14.5 , 180 , and 2000 myr old . cluster # 1 is typical of a cluster for which it is difficult to distinguish between the 14.5 and 180 myr - old populations , but it is easy to rule out the youngest and oldest ages with the help of spectral information at wavelengths beyond 1@xmath37 . in cases like these , to be consistent with our previous method of quoting the least massive clusters that match the data , we choose as the most likely cluster age that which yields the smallest cluster mass . for cluster # 1 , the 14.5 and 180 myr - old populations correspond to @xmath71 and @xmath72 , respectively , and so we designate this cluster as having aged 14.5 myr . in table [ clustertable ] we indicate this final designation with bold text . the center panel in figure [ examplaryplots ] uses cluster # 28 to demonstrate the case where the 2.88 myr - old model sed visibly fits the observations better than stellar populations of the other three ages . in cases like these , although the cluster s young age requires it to be very massive , we nevertheless designate it as having aged 2.88 myr by using bold text in table [ clustertable ] . finally , cluster # 21 is representative of the majority of the clusters , for which it is not possible to rule out any age stellar population through visual inspection of the various model seds . in cases like these , we again resort to choosing the cluster age which yields the smallest cluster mass . for cluster # 21 , the 2.88 , 14.5 , 180 , and 2000 myr - old populations correspond to clusters that have approximate masses of @xmath73 , @xmath74 , @xmath75 , and @xmath76 , respectively , and therefore we designate this cluster ( and clusters like this one ) as having aged 14.5 myr . note that in order for the younger models to be consistent with the observations we must assume more intervening dust ; the 2.88 myr - old model shown for cluster # 21 corresponds to @xmath77 mag , while the 2000 myr - old model corresponds to @xmath78 mag . this , of course , demonstrates the typical astronomer s dilemma of distinguishing between young , dusty populations and older stellar populations . finally , having just described the method we used to estimate the properties of each cluster , we reiterate our motivation to _ not _ perform a detailed @xmath79 minimization when matching model seds to our observations . given the number of assumptions we are required to make in this work , a detailed minimization of the sed fits would provide very little additional insight , and could be misleading . in particular , a minimization would only describe which of the three chosen ages best fits the models ; a fourth ( intermediate ) age could fit the models better than any of the three . therefore it is more instructive to examine the cases of young and old clusters and to discuss how our results would differ if we had chosen different ages . we do this in [ clusterproperties ] , and indeed , we discover not unexpectedly that a different intermediate age may best describe some of the clusters . as shown with bold text in table [ clustertable ] , the best - fitting model stellar populations and dust screens imply that the clusters are consistent with being young ( @xmath80 myr ) and having masses that range from approximately @xmath2 to @xmath81 . the summed mass of all 32 clusters is @xmath82 , with the single most massive cluster making up @xmath83% of the total mass . the 8 next most massive clusters with @xmath84 contribute @xmath85% of the total mass . cluster # 15 , which is coincident with the site of the northern black hole , is the only cluster which we expect may have a non - stellar sed , or may be poorly modeled by an instantaneous burst . this cluster is the second most massive cluster and accounts for approximately 10% of the total mass . with the exception of the most massive cluster , the clusters analyzed in this paper have masses similar to , albeit slightly larger than , those found in other well - studied merging galaxies . @xcite estimate that the young , optically bright clusters in the antennae galaxies have masses ranging from @xmath86 to @xmath87 . @xcite find that the brightest near - ir star cluster in the antennae is @xmath88 myr old and has @xmath89 . similarly , many of the young ( @xmath90 myr ) clusters in arp 220 have masses @xmath91 , or even as much as @xmath92 @xcite . ( note that all three of these aforementioned works assume a salpeter imf . ) @xcite found kinematic masses of @xmath93 and @xmath94 for two clusters in m82 . recall that the masses stated in this paper are underestimated in the sense that we adopted stellar population models with the smallest value of @xmath5 possible . in addition , for all cases where the observed seds fit the different aged models equally well , we designated the correct model to be that which yielded the least massive cluster . but by comparing the resulting cluster masses to those in the literature , we judge that these two assumptions were probably necessary for the majority of the clusters . figure [ histograms ] shows that if we had assumed the clusters were much younger or older than @xmath95 myr , many of the resulting masses would be far larger than those measured in any other merging system . in fact , the clusters masses put stringent constraints on their maximum ages , assuming we have correctly modeled other properties of the stellar populations . as clusters age past 180 myr , they become dimmer and dimmer so that the masses required for us to detect them become unreasonably large ( @xmath96 ) . the star formation rates measured for ngc 6240 using a variety of methods put additional stringent constraints on the cluster s ages . @xcite use the 1.4-ghz luminosity to show that @xmath97 in the nuclear region of ngc 6240 . @xcite show that the hcn luminosity , which traces dense molecular gas , is a good indicator of the star formation rate in spirals and ulirgs alike . using the relationship described by @xcite and the value of @xmath98 measured by @xcite , we calculate that the total star formation rate in ngc 6240 is approximately @xmath99 . this value is slightly larger than that derived from ngc 6240 s integrated ir luminosity , @xmath100 @xcite . the total star formation rate due to the 32 clusters analyzed in this paper is @xmath101 . considering that our measurement exceeds the sfr measured by @xcite for the nuclear region of ngc 6240 , and that the unresolved clusters do not include the majority of the infrared nuclear light ( see fig . [ grayimage ] ) , our measured sfr may be implausibly high . however , @xmath102% of the total sfr from the clusters comes from just 4 clusters which each contribute sfr @xmath103 . these 4 clusters are the only clusters for which we determined that the 3 myr - old model seds fit the observations better than models of other ages . also have @xmath104 . ] we have already conceded that our method of visual inspection does not search the entire parameter space for a best fitting model . while a salpeter imf or a solar metallicity may be to blame , it is unlikely that all but 4 clusters can be reasonably modeled by our assumptions . thus we suggest that these 4 clusters are neither 3 , nor 15 , nor 180 myr old , but rather some other intermediate age . the k band brightness implied by the instantaneous burst stellar population models @xcite monotonically decreases for population ages greater than @xmath105 myr , and increases for population ages between @xmath7 and 10 myr . thus , these 4 clusters may be better described by a stellar population with an age of @xmath105 myr , implying that they are less massive and contribute less to the total sfr . the total sfr measured for the remaining 28 clusters , for which the observations are consistent with 15 myr - old stellar populations , is @xmath106 . this agrees well with the literature , and implies that the newly detected clusters account for @xmath107% of the nuclear sfr , or @xmath108% of the total sfr in the ulirg . finally , if we assumed that the same 28 clusters were all 3 myr old and summed their appropriate star formation rates in table [ clustertable ] , we would conclude that they contribute a whopping @xmath109 . thus given our assumptions ( outlined in detail in paragraph 1 in [ results ] ) the total star formation rate for ngc 6240 puts rather stringent constraints on the clusters ages . to summarize , the masses of star clusters observed in other mergers like arp 220 and the antennae strongly suggest that the clusters in ngc 6240 are younger than 180 myr , while the sfrs measured for ngc 6240 using @xmath110 , @xmath98 , and @xmath111 require that the majority of the clusters be older than 3 myr . 28 of the 32 detected clusters are consistent with being 15 myr old , while the large sfrs of 4 clusters ( clusters # 6 , 28 , 29 , and 32 , their high sfrs do not directly contradict @xmath112 limit set by @xcite . ] ) suggest that these 4 may be more consistent with another intermediate age . the large mass of cluster # 15 and the fact that cluster # 15 is coincident with the northern black hole suggests that the measured sed may have a non - stellar component . ngc 6240 s cluster population excluding the 5 aforementioned clusters is very similar to other cluster populations in the literature . the cluster masses range from approximately @xmath2 to @xmath3 , and the total contribution to the sfr from these 27 clusters is @xmath113 , or @xmath105% of the total sfr in the nuclear region . the cluster population discussed in this paper is considerably more massive than that which was discussed in @xcite . the latter consists of 41 clusters located in ngc 6240 s main body " and 13 clusters located in the galactic tails . @xcite estimate that these clusters have probable masses of @xmath114 . because @xcite do not give the locations of the clusters they analyze it is difficult to know whether this is a discrepancy or whether we are simply noticing an increase in the cluster mass toward the double nuclei , as was noted by @xcite . while it is obvious that @xcite did not analyze the majority of the clusters discussed in this paper , for which we measure only upper limits on the b and i photometry , it is intriguiging that clusters # 2 and 4 , which are located in what appears to be the main body " of ngc 6240 , and for which b and i measurements were made , are rather massive @xmath115 compared to estimates made by @xcite . it is possible that this apparent discrepancy is due to the fact that @xcite used starburst99 stellar population models with a different reddening law and sub - solar metallicity . this apparent discrepancy can not , however , be attributed to a difference in imf , since @xcite used the same imf as we did in this paper , namely a salpeter imf with lower and upper mass cutoffs of 0.1 and 100 m@xmath116 , respecitvely . finally , we note that while using a salpeter imf is helpful when comparing our results to the literature , the @xcite imf is more realistic and results in lower cluster mass estimates . since the infrared luminosity of clusters is dominated by light from high mass stars that quickly turn off the main sequence , and since the salpeter and @xcite imfs differ only for masses below 0.5 m@xmath116 , the cluster masses one derives when employing these imfs differ by a simple scaling factor for clusters with ages less than the main sequence lifetime of a 0.5 m@xmath116 star . the exact mass scaling can be easily derived by piecewise - integrating the two imfs from 0.1 to 100 m@xmath116 . the result is that two young clusters with equal infrared luminosities but different imfs have masses that differ by a factor of @xmath117 , with the larger mass corresponding to the cluster formed with a salpeter imf and the smaller mass corresponding to the cluster formed with a @xcite imf . thus if we had used a more realistic imf throughout our analysis in this paper , we would have estimated cluster masses to be smaller by a factor of about 0.77 . correspondingly , our sfr estimates would decrease by this same factor . ngc 6240 s cluster population , if in fact obscured by a simple dust screen with @xmath118 as described by @xcite , is extinguished by column densities of dust corresponding to @xmath5 between 0.5 and 14 mag . mag and is negligible given the uncertainty on our measurements . ] there is no significant difference in the values of @xmath5 measured for the 24 clusters closest to the northern nucleus , but the 8 clusters farthest from the nucleus exhibit significantly less extinction . this is to be expected for an active merger , in which gas is funneled onto the nuclei , in this case powering the agn that @xcite detected . excluding clusters # 6 , 28 , 29 , 32 , and 15 as previously discussed , on average the 6 clusters farthest from the northern nucleus are extinguished by @xmath119 mag , and the remaining 21 clusters closest to the nucleus are extinguished by @xmath120 mag . @xcite and @xcite both estimate extinctions toward the northern and southern nuclei assuming foreground dust screens . for the northern nucleus we can directly compare our results to theirs using clusters # 12 and 14 as two sight lines through this region . ( @xcite and @xcite both classify the brightest northern knot as the northern nucleus . ) @xcite and @xcite estimate that @xmath121 mag and @xmath122 mag , respectively . both of these values are considerably less than the measurements made in this paper : @xmath123 mag and @xmath124 mag toward clusters # 12 and 14 , respectively . in addition , @xcite measure a peak extinction of @xmath125 mag , while we find that many clusters discussed in this paper have @xmath126 mag . for these reasons we suggest that @xcite and @xcite have both underestimated the amount of extinction in this ulirg . our values of extinction agree better with the values in the literature , measured using a variety of methods . @xcite used x - ray observations and found @xmath127 mag , @xcite used line ratios and found @xmath128 mag , and @xcite used continuum k and l spectrophotometry and inferred @xmath129 mag . the cause of the previous underestimates may be due in part to the different spatial scales being probed , and in part to the fact that the measurements can only probe the optically thin layer surrounding the nuclei . the observations presented in this paper are the first infrared observations to resolve the northern nucleus , and thus it is sensible that we would infer higher values of extinction than were found by @xcite . finally , note that while @xcite may have underestimated the overall extinction in this ulirg , the morphology of their extinction map is consistent with our findings . they find that the extinction is greatest between the nuclei and least toward the northern nucleus . we detect a cluster coincident with the northern nucleus and we find a large concentration of clusters just northwest of the northern nucleus , yet we can not resolve the southern nucleus , nor do we detect many clusters directly between the nuclei . thus the locations of the clusters in our sample strongly suggest that the region between the nuclei is the most obscured , and that the southern nucleus is more highly obscured than the northern nucleus . this evidence suggests that deeper observations would result in additional cluster detections , particularly between the two nuclei . the circumnuclear locations of the clusters detected in these new , high - resolution observations strongly suggest that their formation was triggered by the merger event that created the ulirg ngc 6240 . this motivates our decision to model the clusters as solar metallicity , instantaneous burst stellar populations . if we further assume that the stellar populations were created with a salpeter imf , and that the stars in the clusters follow padova evolutionary tracks , then we find that 27 of the 32 newly discovered clusters are consistent with being @xmath95 myr old , and 4 of the 32 clusters are probably between 3 and 15 myr old . cluster ages much younger or much older than 15 myr can be ruled out by the large star formation rates and masses inferred , respectively . assuming the clusters are 15 myr old , then the majority of the clusters have masses ranging from @xmath2 to @xmath3 , and the total contribution to the sfr from these 27 clusters is @xmath130 , or roughly 10% of the total sfr in the nuclear region . the extinctions calculated toward these clusters range from @xmath131 14 mag , with the least extinguished clusters being located farthest from the double nuclei . while the range of extinctions quoted in the literature for ngc 6240 varies greatly , the values measured in this paper agree well with some of the larger published values . the locations of the clusters detected here , along with the extinction map published by @xcite , suggest that there may be as - yet - unobserved clusters located between the nuclei that are still obscured . the complicated morphology of this ulirg makes it very difficult to determine our completeness limits empirically . we would like to thank peter bodenheimer , seran gibbard , raja guhathakurta , david koo , jason melbourne , lynne raschke , constance rockosi , and jay strader , for many helpful discussions . the authors would also like to thank gabriela canalizo for reducing the nirc2 k data , and we would like to acknowledge the helpful support staff at keck observatory . 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 and the keck ii adaptive optics system were both made possible by the generous financial support of the w.m . keck foundation . the authors wish to extend special thanks to those of hawaiian ancestry on whose sacred mountain we are privileged to be guests . this work was supported in part by the national science foundation science and technology center for adaptive optics , managed by the university of california at santa cruz under cooperative agreement no . this work was based , in part , 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 . gs acknowledges support provided by nasa through a grant from the space telescope science institute , under nasa contract nas 5 - 26555 . this work was also supported in part by nasa grant nag 5 - 3042 . armus , l. , et al . 2006 , , 640 , 204 beswick , r. j. , pedlar , a. , mundell , c. g. , gallimore , j. f. 2001 , , 325 , 151 borys , c. , chapman , s. , halpern , m. , & scott , d. 2003 , , 344 , 385 bruzual , g. , & charlot , s. 2003 , , 344 , 1000 calzetti , d. 2001 , pasp , 113 , 1449 conselice , c.j . , chapman , s.c . , & windhorst , r.a . 2003 , , 596 , l5 draine , b. t. 2003 , , 598 , 1017 fall , s. m. , chandar , r. , & whitmore , b. c. 2005 , , 631 , l133 gao , y. & solomon , p. m. 2004 , , 606 , 271 gallimore , j.f . & beswick , r. 2004 , , 127 , 239 genzel , r. et al . 1998 , , 498 , 579 gerssen , j. et al . 2004 , , 127 , 75 gilbert , a. m. , et al . 2000 , , 533 , l57 heckman , m.t . , armus , l. , & miley , g.k . 1990 , , 74 , 833 kennicutt , r. c. , jr . 1998 , ara&a , 36 , 189 kewley , l.j . , heisler , c.a . , dopita , m.a . , & lumsden , s. 2001 , , 132 , 37 komossa , s. , et al . 2003 , , 582 , l15 kroupa , p. 2001 , asp conf . ser . 228 : dynamics of star clusters and the milky way , 228 , 187 larsen , s.s . 2002 , , 124 , 1393 li , a. & draine , b.t . 2001 , , 554 , 778 mackenty , j. , skinner , c. , calzetti , d. , & axon , d. 1997 , nicmos instrument handbook " , version 2.0 ( baltimore : stsci ) max , c. e. , et al . 2005 , , 621 , 738 max , c. e. , canalizo , g. , & de vries , w. 2006 , submitted to science mccrady , n. , gilbert , a. m. , & graham , j. r. 2003 , , 596 , 240 mihos , j. c. & hernquist , l. 1994 , , 431 , l9 pasquali , a. , de grijs , r. , & gallagher , j s. 2003 , , 345 , 161 rieke , g.h . , et al . 1985 , , 290 , 116 sanders , d. b. , et al . 1988 , , 325 , 74 sanders , d. b. & mirabel , i. f. 1996 , ara&a , 34 , 749 schneider , g. , & stobie , e. 2002 , asp conf . ser . 281 : astronomical data analysis software and systems xi , 281 , 382 scoville , n.z . , et al . 2000 , , 119 , 991 solomon , p. m. , downes , d. & radford , s. j. e. 1992 , , 387 , l55 springel , v. , et al . 2005 , , 620 , l79 tacconi , l. j. , et al . 1999 , , 524 , 732 tecza , m. , et al . 2000 , , 537 , 178 vignati , p. et al . 1999 , a&a , 349 , l57 weingartner , j.c . , & draine , b.t . 2001 , , 548 , 296 whitmore , b. c. & schweizer , f. 1995 , , 109 , 960 whitmore , b.c . & zhang , q. 2002 , , 124 , 1418 wilson , c. d. , harris , w. e. , longden , r. , & scoville , n. z. 2006 , , 641 , 763 lcclc go 6430 ( pi : van der marel ) & wfpc2 & pc & f814w ( b ) & @xmath132 + & wfpc2 & pc & f450w ( i ) & @xmath133 + gto 7219 ( pi : scoville ) & nicmos & nic2 & f110w & @xmath134 + & nicmos & nic2 & f160w & @xmath135 + & nicmos & nic2 & f222 m & @xmath136 cllcccc 1 & 1.43 & 9.95 & -14.9 , * -13.0 * , -13.6 & 2.5 , * 0.5 * , 1.0 & 2.8 , * 0.9 * , 7.9 & 0.97 , * 0.06 * , 0.04 + 2 & -0.96 & 4.89 & -18.0 , * -15.8 * , -16.4 & 4.5 , * 2.0 * , 2.5 & 49.0 , * 11.4 * , 105.7 & 17.00 , * 0.79 * , 0.59 + 3 & -0.06 & 2.23 & -17.6 , * -14.1 * , -14.8 & 12.0 , * 6.5 * , 7.0 & 34.0 , * 2.4 * , 23.9 & 11.81 , * 0.17 * , 0.13 + 4 & -7.31 & 1.87 & -16.7 , * -14.5 * , -15.1 & 6.0 , * 3.5 * , 4.0 & 14.8 , * 3.4 * , 31.9 & 5.13 , * 0.24 * , 0.18 + 5 & 0.01 & 1.34 & -19.1 , * -15.8 * , -16.1 & 18.0 , * 13.5 * , 11.0 & 137.5 , * 11.1 * , 77.3 & 47.73 , * 0.77 * , 0.43 + 6 & 0.41 & 1.12 & * -17.7 * , -14.4 , -15.1 & * 9.0 * , 5.0 , 6.0 & * 35.8 * , 3.0 , 31.4 & * 12.42 * , 0.21 , 0.17 + 7 & -0.80 & 0.63 & -17.2 , * -13.8 * , -14.3 & 10.5 , * 6.0 * , 5.0 & 22.9 , * 1.8 * , 15.2 & 7.95 , * 0.13 * , 0.08 + 8 & -0.15 & 0.44 & -15.7 , * -12.4 * , -13.1 & 7.0 , * 3.5 * , 4.0 & 5.6 , * 0.5 * , 4.9 & 1.93 , * 0.03 * , 0.03 + 9 & -0.46 & 0.42 & -18.0 , * -14.6 * , -15.4 & 10.5 , * 6.0 * , 7.5 & 48.8 , * 3.9 * , 42.8 & 16.93 , * 0.27 * , 0.24 + 10 & -0.67 & 0.41 & -16.7 , * -13.3 * , -14.0 & 9.5 , * 5.0 * , 5.5 & 13.8 , * 1.1 * , 10.8 & 4.80 , * 0.08 * , 0.06 + 11 & -0.82 & 0.32 & -18.3 , * -15.0 * , -15.4 & 15.0 , * 11.0 * , 9.0 & 63.5 , * 5.4 * , 39.9 & 22.05 , * 0.37 * , 0.22 + 12 & 0.26 & 0.17 & -20.4 , * -17.1 * , -17.8 & 10.0 , * 6.5 * , 6.5 & 433.5 , * 39.0 * , 359.6 & 150.53 , * 2.70 * , 1.99 + 13 & -0.25 & 0.15 & -17.6 , * -14.3 * , -14.9 & 10.0 , * 5.5 * , 6.0 & 34.6 , * 2.8 * , 27.2 & 12.02 , * 0.19 * , 0.15 + 14 & 0.28 & 0.11 & -20.1 , * -16.7 * , -17.4 & 8.0 , * 3.5 * , 4.0 & 325.2 , * 26.2 * , 255.2 & 112.93 , * 1.81 * , 1.41 + 15 & 0.00 & 0.00 & -21.0 , * -17.5 * , -18.2 & 14.5 , * 9.0 * , 9.5 & 782.1 , * 56.3 * , 549.1 & 271.57 , * 3.90 * , 3.04 + 16 & -0.41 & -0.06 & -19.1 , * -15.5 * , -16.2 & 14.0 , * 8.0 * , 8.5 & 129.7 , * 8.8 * , 86.2 & 45.05 , * 0.61 * , 0.48 + 17 & -0.27 & -0.13 & -18.6 , * -15.1 * , -15.9 & 12.0 , * 6.5 * , 7.5 & 86.2 , * 6.2 * , 64.0 & 29.93 , * 0.43 * , 0.35 + 18 & -0.53 & -0.37 & -18.9 , * -15.2 * , -16.0 & 15.0 , * 8.5 * , 10.0 & 105.9 , * 6.8 * , 74.3 & 36.77 , * 0.47 * , 0.41 + 19 & 0.06 & -0.94 & -16.8 , * -13.4 * , -14.1 & 11.0 , * 6.0 * , 6.5 & 16.4 , * 1.3 * , 12.2 & 5.71 , * 0.09 * , 0.07 + 20 & 0.19 & -0.97 & -17.3 , * -13.9 * , -14.6 & 12.5 , * 7.5 * , 8.0 & 26.1 , * 2.0 * , 19.4 & 9.06 , * 0.14 * , 0.11 + 21 & -0.45 & -1.15 & -18.4 , * -15.1 * , -15.8 & 11.0 , * 6.5 * , 7.5 & 71.6 , * 5.8 * , 59.4 & 24.85 , * 0.40 * , 0.33 + 22 & -0.62 & -1.19 & -16.2 , * -13.0 * , -13.7 & 8.5 , * 5.0 * , 6.0 & 9.3 , * 0.8 * , 8.7 & 3.24 , * 0.06 * , 0.05 + 23 & -1.07 & -1.29 & -17.6 , * -14.3 * , -15.0 & 11.0 , * 7.0 * , 7.0 & 34.7 , * 2.9 * , 27.2 & 12.04 , * 0.20 * , 0.15 + 24 & 1.94 & -1.60 & -14.4 , * -12.5 * , -12.8 & 4.0 , * 2.0 * , 2.0 & 1.8 , * 0.5 * , 3.8 & 0.61 , * 0.04 * , 0.02 + 25 & 0.13 & -2.21 & -16.7 , * -13.5 * , -14.2 & 8.0 , * 5.0 * , 5.5 & 14.5 , * 1.4 * , 13.4 & 5.03 , * 0.10 * , 0.07 + 26 & -0.22 & -2.46 & -15.9 , * -12.7 * , -13.2 & 8.5 , * 5.5 * , 4.5 & 7.2 , * 0.7 * , 5.7 & 2.51 , * 0.05 * , 0.03 + 27 & 0.89 & -2.62 & -18.0 , * -14.6 * , -15.3 & 19.0 , * 14.0 * , 15.0 & 50.0 , * 3.8 * , 39.2 & 17.35 , * 0.26 * , 0.22 + 28 & -0.19 & -3.10 & * -17.9 * , -14.6 , -15.3 & * 8.0 * , 4.5 , 5.5 & * 42.3 * , 3.8 , 39.2 & * 14.68 * , 0.26 , 0.22 + 29 & 6.58 & -4.13 & * -17.0 * , -14.5 , -15.1 & * 6.5 * , 3.5 , 4.0 & * 19.3 * , 3.4 , 31.9 & * 6.71 * , 0.24 , 0.18 + 30 & 0.95 & -4.25 & -16.0 , * -13.2 * , -13.8 & 7.5 , * 4.0 * , 4.5 & 7.3 , * 1.0 * , 9.2 & 2.53 , * 0.07 * , 0.05 + 31 & -1.02 & -5.30 & -14.4 , * -12.8 * , -13.4 & 3.5 , * 2.0 * , 2.5 & 1.8 , * 0.7 * , 6.6 & 0.62 , * 0.05 * , 0.04 + 32 & -3.54 & -7.71 & * -19.0 * , -16.2 , -16.9 & * 7.5 * , 4.0 , 4.5 & * 124.7 * , 17.0 , 157.6 & * 43.31 * , 1.18 , 0.87 +

the lma msw solution @xcite has been identified @xcite@xmath6@xcite as the correct solution of the solar neutrino problem . the @xmath7 conversion probability of this solution gives a very good description of all available data : no statistically significant deviation has been found so far . new physics effects beyond lma , if exist , are below few per cent . the program of future solar neutrino studies includes 1 ) . further tests of the lma solution , in particular , searches for signatures of this solution such as the day - night asymmetry and the distortion ( upturn " ) of the boron @xmath0 spectrum at low energies ; 2 ) . precise determination of the oscillation parameters , especially the 1 - 2 mixing angle ; 3 ) . searches for the sub - leading effects which originate from - 1 - 3 mixing , - sterile neutrino mixing , - non - standard neutrino interactions , - spin - flavor flip in the magnetic fields of the sun , - violation of the fundamental symmetries ( cpt , equivalence principle , _ etc_. ) . + already the present solar neutrino measurements have sensitivity at the level of few per cent . for instance , the predicted day - night asymmetry of the superkamiokande signal is about @xmath8 which is comparable with the existing @xmath9 experimental error @xcite . at sno one expects the @xmath10 asymmetry , consistent with the experimental result @xcite at the @xmath9 level . future experiments will have substantially higher sensitivity @xcite . the solar neutrino studies enter a phase of precision measurements . + in this connection it is important * to give precise description of the lma conversion , both in the sun and in the earth , taking into account various corrections ; * to estimate accuracy of the approximations made ; * to find the precise _ analytic _ expressions for probabilities and observables as functions of the oscillation parameters ( @xmath11 , @xmath12 ) . this will help to test the lma solution and to search for physics beyond lma . we address these issues in the present paper . in section [ sec2 ] we consider the non - adiabatic corrections to the lma conversion probability . we calculate these corrections for propagation inside the sun and the earth . in section [ sec3 ] we obtain the analytical formula for the probability averaged over the distribution of neutrino sources . in section [ sec4 ] we derive the analytic formula for the @xmath0 regeneration effect in the earth . we present our conclusions in section [ sec5 ] . in the appendices a and b , alternative derivations of formulas for the regeneration factor are given . according to the lma msw solution of the solar neutrino problem , a conversion of the solar electron neutrinos is driven by mixing of the two active neutrinos , @xmath13 : _ f = u ( ) _ mass , where , in general , the mixing matrix is determined as u ( ) = , [ matr ] and @xmath14 is the vector of mass states . the main feature of the lma solution is the adiabaticity of conversion . according to lma the averaged @xmath15 survival probability of the electron neutrinos is given by the adiabatic formula @xcite : p_ee = ( 1 + 2 ^ 0_m 2 ) . [ adprob ] here @xmath16 is the vacuum mixing angle , @xmath17 is the mixing angle in matter in the neutrino production point , @xmath18 , and the mixing angle in matter is determined by 2_m(v ) = , v= g_f n_e(x ) . [ cost ] here @xmath11 is the mass squared difference , @xmath19 is the neutrino energy , @xmath20 is the potential , @xmath21 is the fermi coupling constant and @xmath22 is the number density of electrons in the point @xmath23 . how precise the expression ( [ adprob ] ) is , and what are the non - adiabatic corrections ? to answer these questions , we will elaborate on the adiabatic perturbation theory . dynamics of neutrino conversion is described in terms of the instantaneous eigenstates of the hamiltonian in matter , @xmath24 , @xmath25 . representing an arbitrary neutrino state as @xmath26 , we can write the evolution equation in the base @xmath27 , @xmath28 as @xcite i = , [ evoleigen ] where ( x ) m^2 [ massdiff ] and _ m(x ) = . [ tdot ] the adiabatic approximation corresponds to a situation when 1 , [ adcond ] and the off - diagonal terms in the hamiltonian ( [ evoleigen ] ) can be neglected . in this case there are no transitions between the eigenstates , and therefore the eigenstates propagate independently . the solution of ( [ evoleigen ] ) is straightforward : _ m^ad(x ) = s^ad ( x , x_0 ) _ m ( x_0 ) , where _ m^ad(x ) = , _ m ( x_0 ) = , [ defpsi ] and the adiabatic evolution matrix is s^ad(x , x_0 ) = s^ad ( ) = . [ evad ] the adiabatic phase @xmath29 equals ( x ) = _ x_0^x dx ( x ) . a state initially produced as the electron neutrino , @xmath30 , @xmath31 , evolves as ( x ) = _ m^0 e^i ( x)_1 m + _ m^0 e^-i(x ) _ 2 m . the incoherent survival probability ( [ adprob ] ) can be immediately obtained by averaging the matrix element squared , @xmath32 , over the oscillations . + the non - adiabatic corrections correspond to transitions between the instantaneous eigenstates . we calculate these corrections by solving the equation ( [ evoleigen ] ) . we will implement a perturbation theory using the fact that for lma the adiabaticity parameter ( [ adcond ] ) is very small . let us search for the solution of the equation ( [ evoleigen ] ) in the form & & = + & & = , [ evolnew ] where @xmath33 is supposed to hold everywhere along the neutrino trajectory . ( we check this _ a posteriori _ ) . the expression ( [ evolnew ] ) can be rewritten as _ m ( x ) = s(x , x_0 ) _ m ( x_0 ) , where the evolution matrix equals s(x , x_0 ) c s^ad . [ ev ] inserting ( [ evolnew ] ) into ( [ evoleigen ] ) we find the differential equation for @xmath34 from the condition that the off - diagonal elements of the evolution equation for @xmath35 are zero : i c(x)= - c(x)-i _ m(x ) . [ evolc ] here the first order terms in @xmath34 and @xmath36 are kept only . in this approximation the energy gap between the states coincides with the adiabatic split @xmath37 given in eq . ( [ massdiff ] ) . the solution of equation ( [ evolc ] ) can be written in the following form c(x ) = - ^x_x_0 dx . [ newsolutb ] the integration constant is fixed by the condition : @xmath38 as @xmath39 , so that @xmath40 . since for the lma solution the phase @xmath29 is a fast oscillating function , the integral in ( [ newsolutb ] ) can be calculated using the following formula ( essentially the integration by parts ) ^b_a f(x ) e^i g(x ) dx = e^i g(x)|^b_a + o(1/g^3 ) [ approxm ] which is valid for smooth functions of @xmath41 and @xmath42 . here @xmath43 and @xmath44 . the formula gives very good approximation if @xmath45 . in the case of integral ( [ newsolutb ] ) this condition coincides with the adiabaticity condition ( [ adcond ] ) which is well satisfied . + according to ( [ approxm ] ) and ( [ newsolutb ] ) we find c(x_f ) & & = -i & & = -i sign(_m ) |^x = x_f_x = x_0 , [ csol1 ] or explicitly c(x_f)= - i sign@xmath46 . we will apply this formula for neutrino propagation inside the sun in section [ sec2.3 ] and inside the earth in section [ sec4 ] . the survival probability with the adiabaticity violation effect included can be written as p_ee = , [ surprob ] where @xmath47 is the jump probability @xmath6 the probability of transition @xmath48 on the way from @xmath18 to @xmath49 . let us calculate @xmath50 . notice that for the lma solution , one can not use the landau - zener formula @xcite for @xmath50 for the following reasons . the mixing angle in the final point of evolution , @xmath49 , is large . the adiabaticity parameter @xmath51 is of the same order for all points inside @xmath52 of the solar radius @xcite . the point of maximal adiabaticity violation is not the resonant point , though not far from it . moreover , the resonance layer defined as @xmath53 , is broad since the vacuum mixing angle is large . futhermore , significant part of the neutrino flux is produced inside the resonance region or does not cross the resonance region at all . the double exponential formula @xcite is not valid too . it requires production of neutrinos far above the resonance region in the density scale . this formula is not applied in the range 2~ ( 1.6 - 8.0)10 ^ -6 , [ range ] for which the density in the production point turns out to be close to the resonance density . for the best fit values of the lma oscillation parameters the range ( [ range ] ) corresponds to @xmath54 mev , that is , to the region of interest . let us apply the results of section 2.2 for calculation of the non - adiabatic corrections . notice that at the surface of the sun the effective potential @xmath20 is negligible and @xmath55 can be taken zero . using eq . ( [ csol2 ] ) we find the transition amplitude @xmath56 on the way from the production point to the surface of the sun in the leading order approximation as c(x_f)= i |_x = x_0 . [ nonadinsunb ] then the probability of non - adiabatic transition is given by p_c = |c(x_f)|^2 = = ^2 , [ nonadinsun ] where h(x ) ^-1 , l_osc(x ) are the density height and the oscillation length in matter . the transition probability @xmath50 ( [ nonadinsun ] ) depends only on parameters of the initial ( production ) point . one can understand this by noting that @xmath57 . therefore many oscillation lengths are obtained on the distance in which the potential changes sizably . the non - adiabatic corrections are averaged out being negligible along the trajectory of the neutrino except for the boundaries of trajectory , _ i.e. _ , around the production point or the point at the surface of the sun . at the surface of the sun the contribution can be neglected because the potential is zero . the probability is determined basically by square of the ratio of the oscillation length and the density height . second factor in ( [ nonadinsun ] ) is of the order one . so , essentially @xmath58 ^ 2 $ ] . using the best fit values of the lma oscillation parameters we find from ( [ nonadinsun ] ) p_c = ( 10 ^ -9 - 10 ^ -7)()^2 . [ pc ] here the numerical prefactor depends on the production point . as a function of @xmath18 , the probability @xmath50 reaches maximum at around @xmath59 , where the potential does nt drop down substantially and @xmath60 reaches its almost minimal value due to increase of the density gradient . the corrections are negligible in the whole relevant range of neutrino energies and production points . the probability ( [ pc ] ) strongly differs from what one would get using double - exponential formula @xcite : @xmath61 . notice that the jump probability ( [ nonadinsun ] ) equals ( up to factor 4 ) the adiabaticity parameter in the production point squared , as is expected in the adiabatic perturbation theory this contrasts the landau - zener probability , @xmath62 , which is essentially non - perturbative effect . .__the average value of potential @xmath63 and the corresponding value of @xmath64 for different component of the solar neutrino spectrum . _ _ [ cols="^,^,^,^,^,^,^,^,^",options="header " , ] in the adiabatic approximation the survival probability depends on the potential ( density ) in neutrino production point @xmath65 : p_ee = p_ee(v_0 ) , v_0=v(x_0=r ) . observables at the earth are determined by the survival probability averaged over the neutrino production region : p_k = , k = pp , pep , be , n , o , f , b , hep , [ aveprob1 ] where @xmath66 is the distribution of sources of the @xmath67 component of neutrino spectrum . the distributions are different for different components . let us introduce the average value of the potential in the production region for the type @xmath67 neutrinos : |v_k . [ avepot ] we will use the fact that in the effective production region , @xmath68 deviates weakly from @xmath63 . therefore the survival probability can be expanded in series around @xmath63 : p_ee = p_ee(|v_k ) + ( ) _ v = |v_k(v - |v_k ) + ( ) _ v = |v_k(v - |v_k)^2 + inserting this expression into ( [ aveprob1 ] ) and using the definition ( [ avepot ] ) we find p_k = p_ee(|v_k ) - v^2_k , [ aveprobb ] where v^2_k . [ delpot2 ] notice that the correction to @xmath69 appears in the second order of the deviation of potential from the average value . the expression for probability ( [ aveprobb ] ) can be rewritten as p_k = + ( 1-_k ) 2_m(|v_k ) 2 , [ aveprob ] where the correction @xmath70 equals _ k = . [ deltan ] in the table [ tab : para ] we present the average values of potentials and the corresponding second order deviations from the average values for eight types of solar neutrinos . we use the distributions of neutrino sources from the bp2000 model @xcite . the expansion parameters @xmath71 are all small , especially for the boron neutrinos which have the narrowest distribution of sources . in fig . [ figdiff ] we compare the probability @xmath72 obtained from the approximate analytic formula ( [ aveprob ] ) with results of numerical calculations , @xmath73 . @xmath74 is shown as a function of @xmath75 . the lines have been cut for @xmath76 because of their lower energies in comparison with the energies of @xmath77 and boron neutrinos . the plot shows that the analytic formula is rather precise . in particular , the deviations are extremely small ( @xmath78 ) for small and large values of @xmath79 . relatively large deviations can be seen in the intermediate region of @xmath79 . for example , for @xmath80 the magnitude of @xmath81 reaches maximum @xmath82 at around @xmath83 ev@xmath84 . the corrections @xmath70 are important : _ e.g. _ , for the @xmath77 neutrinos the deviation would be up to @xmath85 without @xmath70 . the solar neutrinos arrive at the surface of the earth as incoherent fluxes of the mass states . the mass states oscillate in the matter of the earth producing partial regeneration of the electron neutrino flux @xcite . previously the effect has been described in one or two layers approximation . in the later case , interference effects of contributions from the core and the mantle have been discussed @xcite . in this section we will study effects for the realistic density profile of the earth . the probability of @xmath86 transition can be written as p(_2 _ e ) ^2 + f_reg , [ freg ] where @xmath87 is the regeneration factor which describes the earth matter effect . in the absence of matter ( _ i.e. _ , during the day ) @xmath88 . using the definition ( [ freg ] ) we find the @xmath0 survival probability with the regeneration effect included as p_ee = ( 1 + 2_m^0 2 ) - 2_m^0 f_reg . notice that the mixing angle @xmath89 in the neutrino production point in the sun determines the mass ( @xmath90 , @xmath91 ) composition of the neutrino flux which arrives at the earth . the essential feature of the lma solution is that the earth matter effect is small . this smallness is characterized by the ratio = 0.024 ( ) ( ) , v_a= g_f n_a , where @xmath92 is the avogadro number . we will use @xmath93 as the expansion parameter . the @xmath86 transition probability can be written as p(_2 _ e ) & & = |_e|u(_mr ) s(x_f , x_0 ) u^(_mr)u()|_2|^2 + & & = |_e|u(_mr ) s(x_f , x_0 ) u^(_mr - ) |_2|^2 , [ p2e ] where @xmath94 is the mixing angle in matter at the surface of the earth and the matrix @xmath95 given in ( [ ev ] ) describes evolution of the neutrino eigenstates in matter . noting that in matter of the earth ( _ mr - ) 21 , we find from ( [ evolnew ] ) , ( [ freg ] ) and ( [ p2e ] ) an expression for the regeneration factor in the lowest order in @xmath34 and @xmath96 as f_reg = ^2 2 ^ 2 ( x_f ) + 2 re\{c(x_f ) } , [ regcor ] where @xmath97 is the potential at the surface of the earth and @xmath98 is the total phase acquired along the trajectory in the earth . for the profile with slowly changing density ( lowest adiabatic approximation ) , @xmath99 , we obtain f^ad_reg= ^2 2 ^ 2(x_f ) . [ regold ] the prefactor ( the depth of oscillations ) is determined by the potential at the surface of the earth , whereas the phase is given by the integral along whole trajectory . for one layer with constant potential ( density ) , and therefore @xmath100 , the regeneration factor ( [ regcor ] ) or ( [ regold ] ) is reduced to the well known expression : f_reg= ^2 2 ^2 . [ regold1 ] here @xmath101 is the distance traveled by neutrino in the earth and @xmath102 is the oscillation length in matter . let us consider a neutrino propagation in realistic density profile of the earth . the profile can be described by @xmath103 nearly spherical shells of matter with sharp ( step - like ) density changes at the borders of shells and slow variation of density in layers between the borders . according to the prem model @xmath104 @xcite . so , in @xmath105th shell ( @xmath106 ) , the potential @xmath107 is a smooth function of the radial distance @xmath65 . crossing @xmath108 shells corresponds to crossing @xmath109 layers ( see fig . [ figearth ] ) . we denote by @xmath110 the outer radius of @xmath105th shell , so that @xmath111 corresponds to the radius of the earth : @xmath112 . the trajectory of the neutrino is characterized by the zenith angle @xmath113 . we determine a position of neutrino along trajectory by the coordinate @xmath23 with origin in the center of trajectory , so that x , x^2=r^2-r_e^2 ^ 2_z . [ trajcoor ] here @xmath101 is the total length of the trajectory in the earth . the length of the part of trajectory inside border @xmath114 is given by l_i = , [ length ] and @xmath115 by definition . we introduce the adiabatic phase @xmath116 acquired by neutrinos in the interval @xmath117 along the trajectory , that is , inside the outer border of the @xmath118th shell : _ i & & = ^l_i/2_-l_i/2 dx + & & ^l_i/2_-l_i/2 dx , i=0,,n-1 . [ phasei ] @xmath37 is given in eq . ( [ massdiff ] ) . here we keep the order @xmath119 term since due to integration its contribution to the phase is not negligible . at the borders of shells there are jumps of the potential and hence the discontinuities of the mixing angle in matter . we denote them as v_i & & v_i+1(r_i)-v_i(r_i ) , i=0,,n-1 , [ jump1 ] + _ mi & & _ m(v_i+1(r_i))- _ m(v_i(r_i ) ) , i=0 , , n-1 . [ jump ] at the surface of the earth , we obtain v_0= v_r , _ m0 = _ mr- . [ jump2 ] furthermore , we find _ mi 2 , _ mi 1 , i=0,,n-1 . [ approxth ] corrections to ( [ approxth ] ) are of the order @xmath120 , and hence negligible . smooth variation of the potential in each shell of the earth can be approximated by the analytic formula @xcite : v_i = v_a , i=1,,n . [ poteni ] let us find an analytic expression for the regeneration factor using the density profile described above . the problem can be solved in two steps : ( 1 ) computation of the non - adiabatic corrections to propagation within a given layer , @xmath121 ; ( 2 ) computation of effect of the borders between layers , @xmath122 . so that f_reg = f^ad_reg + _ i f_i+_i f_i^jump . the virtue of the lma solution is that it enables us to study both effects using the same formalism of the adiabatic perturbation theory . let us compute the non - adiabatic corrections for one layer . suppose a neutrino trajectory crosses the @xmath105th layer with the borders at @xmath123 and @xmath124 . using ( [ trajcoor ] ) and ( [ poteni ] ) , the potential in this layer can be expressed in terms of the trajectory coordinate as v_i = v_a . [ potenib ] here _i=_i+ _ i ^2_z+_i ^4_z , _i=_i+2 _ i ^2_z , _i=_i . [ parametb ] according to ( [ length ] ) and ( [ parametb ] ) the gradients of potential at the borders equal |_x = l_i 2 & & = ( _ i+ 2 _ i y_i^2 ) , [ gradi ] + @xmath125 . then for this layer eq . ( [ csol2 ] ) gives the amplitude of non - adiabatic transition c_i & & = -2i e^2 i _ 0 e^-2 i ( x ) |^x = l_i-1 2_x = l_i 2 + & & = - v_a e^i _ 0 , [ ampli ] where phases @xmath116 are defined in ( [ phasei ] ) . inserting this expression into ( [ regcor ] ) we obtain the non - adiabatic correction from this layer , @xmath126 to the regeneration factor as f_i & & = v_a . the ratio of the absolute value of correction and the adiabatic term equals ~ ~ , [ ratio ] where @xmath127 plays the role of typical scale of the density change . as an example , let us consider the layer between @xmath128 and @xmath129 . in this layer @xmath130 , @xmath131 and @xmath132 @xcite . from ( [ ratio ] ) we obtain that the non - adiabatic correction to the regeneration factor is about @xmath133 at @xmath134 mev . notice that for some particular values of energies and @xmath113 , the contributions @xmath121 from different layers @xmath105 may sum up constructively " producing larger effect . in this connection let us notice the following . \1 ) the enhancement effect may occur for exceptional values of @xmath19 and @xmath113 and therefore any realistic averaging over @xmath19 and integration over @xmath113 will wash it out . \2 ) the enhancement can not be large ( proportional to the number of layers , @xmath103 ) since ( i ) only few layers give significant contribution and for the rest , the effect is below @xmath135 ; ( ii ) there is a systematic cancellation of contributions from the upper limit of integration in @xmath121 and the lower limit of integration in @xmath136 ( the adiabatic phases are the same for both ) ; ( iii ) typically , contributions from two layers of the same shell have opposite signs . so , we conclude that the non - adiabaticity within layers of the earth can be safely neglected . the jumps of potential between the layers strongly violate the adiabaticity and on the first glance , the adiabatic perturbation theory can not be applied . we show , however , that the results for non - adiabatic case obtained in section [ sec2.2 ] can be also used here . the key point is that for the lma parameters the earth matter effects are small , whatever the density profile in the earth is . as a consequence , variations of the mixing angle in matter are small : @xmath137 , and essentially the expansion parameter here is @xmath93 . consider a neutrino trajectory which crosses @xmath138 layers ( @xmath103 shells ) . in the points of the trajectory @xmath139 neutrinos cross the borders of shell with @xmath140 , as illustrated in fig . [ figearth ] . the corresponding potential jumps equal @xmath141 for @xmath139 . using the expression ( [ tdot ] ) we obtain in the lowest approximation _ m(x ) = ^n-1_i=1 v_i . [ theder ] as it has been shown in section [ sec4.2 ] to a good approximation one can take @xmath142 everywhere outside the borders . the evolution equation ( [ evoleigen ] ) can be averaged in small intervals @xmath143 to eliminate @xmath144-functions which originate from @xmath145 . however , this is not necessary since in the expression for @xmath34 in ( [ evolc ] ) @xmath145 is integrated anyway . plugging expression ( [ theder ] ) into eq . ( [ newsolutb ] ) we obtain c(l/2 ) = e^i _ 0 ^n-1_i=1 v_i ( e^-i _ i - e^i _ i ) , where @xmath116 are defined in ( [ phasei ] ) . inserting the real part of @xmath146 into ( [ regcor ] ) we find the regeneration factor for the case of @xmath103 shells crossing : f_reg = ^2 2 ^2 _ 0 + ^n-1_i=1 ^2 2_i _ 0 . [ regnew ] in the appendix a we present the rigorous derivation of this factor considering evolution in the sequent layers explicitly . the results of two approaches coincide exactly in the first order in @xmath147 . noting that @xmath148 is the jump of potential at the surface of the earth , we can rewrite the expression ( [ regnew ] ) in the following compact form : f_reg = _ 0 ^n-1_i=0 v_i _ i. [ regnewb ] so , @xmath87 is proportional to the sum of similar terms which correspond to the borders of the shells . each term is the product of the potential jump at a given border and sine of the total adiabatic phase acquired on the part of trajectory inside a given border ( that is , from @xmath149 to @xmath150 for the border @xmath105 ) . the sum runs over all borders including the surface of the earth . the expression ( [ regnewb ] ) corresponds to the symmetric density profile . the zenith angle dependence of the regeneration factor appears via the phases : @xmath151 . the formula ( [ regnewb ] ) ( which is the main result of our study ) allows us to get complete understanding of the earth matter effects including effects of complicated shell structure . apparently , this is not possible using the one layer approximation ( [ regold ] ) , where the interference terms induced by different shells are absent . in fig . [ figreg ] we compare the zenith angle dependence of the regeneration factor computed using the analytic formula ( [ regnew ] ) with the one obtained by the exact numerical integration for the prem profile . two results coincide extremely well . one can see that the analytic formula reproduces quite precisely the magnitude and the phase structure of the regeneration factor . let us mark some features . the change of the oscillatory behaviors for @xmath152 is induced by the sharp density jump at the border between the mantle and the core of the earth , at @xmath153 . notice that at @xmath152 the amplitude of oscillations for some periods increases , however , the frequency of large peaks becomes lower . so that the average value of @xmath87 does not increase in comparison with the value for @xmath154 . for small @xmath155 the dependence of @xmath87 on @xmath113 is a result of interference of terms in ( [ regnewb ] ) which correspond to the outer shells of the earth . to understand this dependence , it is convenient to introduce the phase @xmath156 : _ i _ 0 - _ i , [ phased ] so that @xmath157 is the phase acquired by neutrino on the way from the surface of the earth to @xmath140 . using @xmath156 we can rewrite the expression for regeneration factor ( [ regnewb ] ) as f_reg = ^n-1_i=0 v_i . [ newreg1 ] apparently , if averaging over @xmath156 occurs only the term with @xmath158 survives ( @xmath159 ) in @xmath87 which is reduced to the adiabatic expression for one layer . the increase of regeneration factor with @xmath155 in the range @xmath160 is related to the effect of three density jumps near the surface of the earth : according to the prem profile they are situated at depths 10 km , 22 km and 31 km correspondingly . the distance , @xmath161 , on which @xmath156 is acquired depends on the zenith angle as l - l_i 2 , _ z > ( @xmath162 is the depth from the surface of the earth to the borders of the @xmath105th shell ) . in the case of small @xmath155 ( @xmath163 ) , the distance @xmath161 can be of several hundreds kilometers which is comparable to or larger than the oscillation length . furthermore , @xmath161 and @xmath156 are fast changing functions of @xmath113 . so , @xmath156 are large and different for different @xmath105 . therefore , the terms @xmath164 for different @xmath105 ( @xmath165 ) are quite different and therefore partially cancel each other ( `` interfere destructively '' ) . in constrast , for @xmath166 , the distances @xmath161 ( @xmath165 ) for the outer shells become much smaller than the oscillation length and they slowly change with @xmath155 . in this case the phases @xmath156 ( @xmath165 ) are all small and @xmath167 . so , for @xmath168 the effects of outer shells `` interfere constructively '' producing larger regeneration factor . it is then possible to account these close layers effectively as a single layer , as it was done in ref . increase of the regeneration factor in the transition region @xmath169 corresponds to convergence of the term @xmath170 ( @xmath165 ) to @xmath171 . for the lma solution the oscillation length in the earth is small : @xmath172 . since the time of the neutrino detection is well known , averaging over the zenith angle can be avoided , and in fact , in the unbinned analysis of the data developed recently @xcite one needs to know the zenith angle dependence without averaging . at the same time since the recoil electron ( and not neutrino ) energy is measured and a detector has finite energy resolution , averaging over the neutrino energy occurs . in the leading approximation the phase equals @xmath173 . therefore the energy resolution @xmath174 corresponds to averaging over the interval of phases : _ i _ i . if @xmath175 , so that @xmath176 , the interval @xmath177 can be large , thus leading to strong averaging of terms @xmath170 in ( [ regnewb ] ) . this happens to the contributions from structures situated far from the surface of the earth . in fig . [ figave ] we show the result of averaging of the regeneration factor folded with the cross section of the neutrino - electron elastic scattering over two different energy intervals . comparing with fig . [ figreg ] , one sees that the complicated oscillatory pattern produced by the density jumps in the central regions of the earth is strongly averaged when @xmath155 is large ( for general analysis of this effect see @xcite ) . according to fig . [ figave ] for @xmath178 the regeneration factor oscillates with small depth around @xmath179 . this happens because the main term @xmath180 is strongly averaged too . in contrast , for @xmath163 only the outer structures of the earth can contribute and the averaging is not as efficient as for large @xmath155 . indeed , for the borders of outer shells ( @xmath165 ) the distance where @xmath156 is acquired , @xmath181 , can be about several hundreds kilometers , that is , comparable with the oscillation length in matter . in this case the phase interval @xmath182 is still not large enough to give sufficient averaging . for @xmath183 , @xmath184 is also small and averaging is weak . furthermore , in the interval @xmath185 the regeneration factor increases with @xmath155 . the reason is that in this interval the main term , @xmath171 , starts to `` interfere constructively '' with the terms produced by the outer shells , @xmath170 ( @xmath165 ) , as it has been discussed in section [ sec4.3 ] . in fig . [ snocc1 ] and [ snocc2 ] we show dependence of the regeneration factor for the charged current events at sno . here we have taken into account the energy resolution of the sno detector and also performed integration over various energy bins of the observed kinetic energy . on the basis of our analytic formulas and discussion , the interpretation of results of fig . [ snocc1 ] and [ snocc2 ] is straightforward . there are small scale structures in the outer mantle of the earth of depth @xmath186 km in which matter has quite different densities ( _ e.g. _ , ocean , rock and soil ) . in contrast to the ideal prem model , these structures are not isotropically distributed and can be quite complicated . in section [ sec4.3 ] we have shown for the ideal prem model that contributions produced by structures close the surface of the earth interfere destructively for small @xmath155 . furthermore , averaging over the energy does nt smooth the dependence of these contributions on @xmath155 completely . this produces an uncertainty for the future high statistics solar neutrino experiments ( see also comments in @xcite ) unless the local density distribution is well known @xcite . in view of this we will consider general ( not spherically symmetric ) density profile . suppose neutrinos cross @xmath187 layers of matter . the density jumps occur in the points @xmath188 ( @xmath189 ) ; @xmath190 and @xmath191 are the points where neutrinos enter and leave the matter correspondingly . similarly to ( [ theder ] ) we parameterize @xmath55 as _ m= ^k-1_i=1 v_i ( x - x_i ) , [ thederasy ] where the jump of potential at the @xmath105th border between layers equals v_i = v(x_i+)-v(x_i- ) , i=1 , , k-1 . [ jumpnewden ] @xmath192 is the infinitesimally small distance . noting that the potential is zero for neutrinos before entering the earth and after leaving the earth , we define also v_0 = v(x_0 ) , v_k = -v(x_k ) . [ jumpnewden1 ] plugging the potential jumps into ( [ newsolutb ] ) gives c(x_k)= - ^k-1_i=1 v_i e^2 i _ i , where _ i = ^x_k_x_i dx , i=0 , , k [ phasex ] is the phase acquired from a given border @xmath105 to the final point of the trajectory ( detector ) . now it is straightforward to compute the regeneration factor , and in the leading order in @xmath147 we obtain f_reg & & = | _ e| u(_m(x_k ) ) s(x_k , x_0 ) u^(_m(x_0))u ( ) | _ 2 |^2 -^2 + & & = - ^k_i=0 v_i 2_i . [ regasy ] in the appendix b , a direct computation of @xmath87 for this case is given . its result coincides with ( [ regasy ] ) in the leading order in @xmath147 . using this formula one can easily reproduce ( [ regnewb ] ) assuming a symmetric density profile and taking into account that @xmath193 for @xmath194 and @xmath195 . using ( [ regasy ] ) it is easy to study averaging effects following the discussion in section [ sec4.4 ] . structures situated far from the detector have @xmath196 . so that , after averaging , remote structures do not produce significant effect . however , if the energy resolution is improved , effects of these remote structures can be large . this agrees with general consideration in @xcite . thus , we arrive at the following conclusion . if @xmath155 is large , small scale structures near the entering point can be taken effectively as a single layer . furthermore , averaging over the energy makes contributions of these small structures to be unimportant . if @xmath155 is small , we can not consider small scale structures near the entering point as a single layer . however uncertainties produced by these structures can be significantly reduced if averaging is performed over broad energy interval , _ i.e. _ @xmath197 . after averaging , the regeneration factor still shows an oscillatory behavior in the region of small @xmath155 , and this effect is produced by contributions of the shells close to the detector @xcite . we have performed detailed analytic study of the lma msw conversion of the solar neutrinos . our main result is the precise analytic formula for the survival probability which includes non - adiabatic corrections , averaging over the neutrino production region and the earth regeneration effect . for the @xmath67 component of the solar neutrino spectrum ( @xmath198 ) it can be written as p_k = + ( 1-_k ) 2_m(|v_k ) 2-(1-_k ) 2_m(|v_k ) f_reg . here the correction due to averaging effect , @xmath70 , is given in eq . ( [ deltan ] ) ; the average values of matter potential in the production regions of @xmath67 components , @xmath63 , are defined in ( [ avepot ] ) and their numerical values are presented in the table 1 . the regeneration factor @xmath87 is given in ( [ regnew ] ) for the symmetric density profile and in ( [ regasy ] ) for general asymmetric density profile . effect of averaging over the neutrino production region in the sun is reduced to specific value of the initial mixing angle in matter which should be taken for the average value of the potential , @xmath199 , and to the appearance of the correction @xmath70 . we have compared the analytic results with the results of numerical computation and found that maximal deviation @xmath200 happens for the @xmath77 neutrinos . for the boron neutrinos the precision is better than 0.2% . we have obtained precise analytic formula for the regeneration effect in the earth using the realistic density profile . we present simple derivation of this formula which uses the adiabatic perturbation theory . performing also explicit calculations of the evolution in sequent layers we show that this derivation is correct . the analytic formula reproduces results of numerical computations with accuracy determined by @xmath201 . essentially the regeneration effect is the sum of contributions from different shells which are determined by jumps of the potential at the borders and by the adiabatic phase acquired inside the outer borders of the corresponding shells . the dependence of regeneration factor on the zenith angle can be understood in terms of interference of contributions from different borders . the derived analytical formula allows us to understand the effect of averaging over the neutrino energy . using the analytical formula we have considered effects of small scale structures ( @xmath202 km ) of the earth profile . these effects can be important for small values of @xmath203 . we stress that local `` perturbations '' of the density profile can produce sizable uncertainties in @xmath87 . + * acknowledgment * one of the authors ( p.c.h . ) would like to thank fapesp for financial support . a.y.s . thanks tokyo metropolitan university where this work has been accomplished for hospitality . let us derive the expression for the regeneration factor by considering neutrino evolution in sequent layers of the earth explicitly . first , we find the complete evolution matrix , @xmath204 , in the basis of the mass eigenstates @xmath205 : ( x_f)=(x_f , x_0 ) ( x_0 ) . as discussed in section [ sec4.2 ] , the adiabaticity violation effect within layers is suppressed by @xmath206 in comparison with the leading order earth matter effect ( @xmath207 ) . therefore , we neglect the adiabaticity violation within layers . 1 ) . in the case of neutrino propagation in one shell ( one layer ) we can simply project the adiabatic evolution matrix ( [ evad ] ) obtained for the matter eigenstates on to the basis of the mass states . in the leading order in @xmath147 we find & & = _ 1 ( , - ) + & & = u^ ( ) u(_mr ) s^ad(,- ) u^(_mr ) u ( ) = u(_m0 ) s^ad(_0 ) u^(_m0 ) + & & = s^ad(,- ) + _ m0 ( e^-i _ 0-e^i _ 0 ) + & & = + ( e^-i _ 0-e^i _ 0 ) . [ matr1 ] here @xmath208 is the jump of the mixing angle at the surface of the earth . we have used eq . ( [ approxth ] ) , and @xmath209 is defined in eq . ( [ phasei ] ) . 2 ) . in the case of two shells crossing , the neutrino encounters three layers ( the outer shell is crossed twice ) . the evolution matrix can be similarly obtained by using the adiabatic evolution matrix ( [ evad ] ) in each layer and by rotation from the matter eigenstates basis in the layer before the border to the basis after the border . as a result , we find = u(_m0 ) s^ad ( , ) s_1 ( , - ) s^ad(-,- ) u^(_m0 ) , where @xmath210 is the evolution matrix in the inner shell which has a form similar to eq . ( [ matr1 ] ) and it can be written as s_1 & & = u(_m1 ) s^ad(,- ) u^(_m1 ) + & & = s^ad(,- ) + 2(e^-i _ 1-e^i _ 1 ) . [ matr2b ] here @xmath211 is the jump of mixing angle on the border between the first and the second shells , and @xmath212 is defined in eq . ( [ phasei ] ) . combining the last two formulas we find to the order @xmath147 & & = s_2(,- ) + & & = s^ad(,- ) + _ i=0 ^ 1 ( e^-i _ i - e^i _ i ) . [ matr2 ] 3 ) . suppose the evolution matrix for @xmath108 shells crossing ( @xmath109 layers ) equals & & = s_j ( , - ) + & & = s^ad(,- ) + _ i=0^j-1 ( e^-i _ i - e^i _ i ) . [ matrj ] consider now the trajectory with @xmath213 shells crossings . the evolution matrix is & & = s_j+1 + & & = ^j-1_i=0 s_j ( , - ) ^0_i = j-1 , [ sss ] where @xmath214 defined in ( [ jump ] ) , is the jump of the mixing angle in matter at the border @xmath114 . @xmath215 is the evolution matrix in the central shell which can be written similarly to eq . ( [ matr2b ] ) as s_j = s^ad(,- ) + 2(e^-i _ j - e^i _ j ) [ matrjb ] . @xmath216 is given in eq . ( [ phasei ] ) . after insertion into ( [ sss ] ) , the first term of ( [ matrjb ] ) leads to @xmath217 . the second term in ( [ matrjb ] ) is already of the order @xmath147 . note that @xmath214 is small , as is shown in ( [ approxth ] ) . so , we can approximate @xmath218 by the unit matrix when plugging the second term in ( [ matrjb ] ) into ( [ sss ] ) . as a result , we find = s_j + ( e^-i _ j - e^i _ j ) . using then expression ( [ matrj ] ) for @xmath217 , the formula ( [ matrj ] ) is immediately extended to the case of crossing @xmath213 shells , thus accomplishing the proof . the result for the case of @xmath103 shells crossing is = s^ad(,- ) + _ i=0^n-1 ( e^-i _ i - e^i _ i ) . [ matrn ] using ( [ matrn ] ) we obtain the expression for the regeneration factor in the leading order in @xmath147 as f_reg & & =| e^-i _ 0+_i=0^n-1 ( e^-i _ i - e^i _ i ) |^2 -^2 + & & = _ 0 _ i=0^n-1v_i _ i , [ regen ] where @xmath116 is given in ( [ phasei ] ) and @xmath219 is defined in ( [ jump1 ] ) . this expression coincides with ( [ regnew ] ) or ( [ regnewb ] ) which have been obtained in section [ sec4.3 ] using the adiabatic perturbation theory . as in the section [ sec4.5 ] , we define _ i _ m(x_i+)-_m(x_i- ) , i=0 , , k , where @xmath220 and @xmath221 , and @xmath188 are the points of density jumps . @xmath18 and @xmath222 are the initial and final points of neutrino trajectory in matter . we will use the following expression : _ i 2 , _ i 1 , [ approxb ] which is a good approximation in the leading order in @xmath223 . @xmath219 is given in ( [ jumpnewden ] ) . neglecting the adiabaticity violation within each layer , we obtain the evolution matrix @xmath224 as = ( ^1_i = k u^(_i ) s^ad(x_i , x_i-1 ) ) u^(_0 ) . [ matrasy ] @xmath225 is the adiabatic evolution matrix given in ( [ evad ] ) . we approximate @xmath226 as u^(_i ) = + q(_i ) , where q(_i ) . straightforward computation gives the following result in the leading order in @xmath147 & & = s^ad(x_k , x_0)+q(_k ) s^ad(x_k , x_0 ) + s^ad(x_k , x_0 ) q(_0 ) + & & + ^k-1_i=1 s^ad(x_k , x_i ) q(_i ) s^ad(x_i , x_0 ) + & & = s^ad(x_k , x_0)+ ^k_i=0 . @xmath227 is defined in ( [ phasex ] ) . then , using ( [ approxb ] ) , the regeneration factor can be directly computed in the first order in @xmath147 as f_reg & & = | e^-i _ 0-^k_i=0_i e^i(2 _ i-_0 ) |^2 -^2 + & & = - ^k_i=0 v_i 2 _ i. it coincides with ( [ regasy ] ) . + * note added * + this note has been added on request of the referee . 1 ) . after the present paper had appeared in the hep - ph archive [ hep - ph/0404042 ] , the preprint by akhmedov _ et . , [ hep - ph/0404083 ] , has been published in which the analytic integral formula is given for the the regeneration effect in the earth in the three neutrino framework . in the first version of [ hep - ph/0404083 ] the correct oscillation phase in this integral formula has been introduced on the `` heuristic '' basis : it does not follow from their perturbation theory . correct integral formula ( with the correct phase ) has been derived for the first time in the paper by ioannisian and smirnov , [ hep - ph/0404060 ] . later in the journal version jhep 0405 ( 2004 ) 057 , akhmedov _ et . _ , have also presented derivation of correct phase . in ioannisian and smirnov paper hep - ph/0404060 and akhmedov _ _ paper jhep 0405 ( 2004 ) 057 the integral formula has been obtained using the improved perturbation theory in the small parameter @xmath228 . in the present paper we use the adiabatic perturbation theory . it can be shown that in the lowest order in @xmath93 both approaches coincide . indeed , inserting expression for @xmath55 from ( [ tdot ] ) into ( [ newsolutb ] ) and performing integration by parts in eq . ( [ newsolutb ] ) of the present paper one can derive the integral formula . this formula is valid for arbitrary density profile provided that the condition @xmath229 is satisfied . inserting the potential of prem model ( eqs . ( [ jump1 ] , [ potenib ] ) ) into the integral formula one can reproduce the result ( [ regnewb ] ) . however technically the use of formula ( [ newsolutb ] ) of the present paper is more convenient for the derivation of ( [ regnewb ] ) since the derivative @xmath230 gives @xmath144- functions at the borders of layers and further integration becomes trivial . 2 ) . the kamland collaboration has published recently results of improved measurements of oscillations on the basis of 766.3 ton - year exposure [ t. araki _ et . al . _ , hep - ex/0406035 ] . in assumption of the cpt conservation , the global analysis of the solar neutrino data and kamland gives slightly ( @xmath231 ) higher best fit value @xmath232 ev@xmath233 than it was before . the increase of @xmath11 leads to the corresponding small decrease of the adiabaticity parameter @xmath234 and the expansion parameter @xmath93 for a given energy of neutrinos . therefore increase in @xmath11 ( i ) further improves the adiabatic perturbation theory and implies that the non - adiabatic correction for probability in the sun is smaller ; ( ii ) diminishes the earth matter regeneration effect ; ( iii ) makes our analytic formula for @xmath87 preciser . notice that the analytic study of this paper has a general character and does not rely on particular values of @xmath11 . we use specific value of @xmath11 for illustration only . s. p. mikheyev and a. yu . smirnov , yad . 42 * , 1441 ( 1985 ) [ sov . . phys . * 42 * , 913 ( 1985 ) ] ; nuovo cim . * c9 * , 17 ( 1986 ) ; s. p. mikheyev and a. yu . smirnov , zhetf , * 91 * , ( 1986 ) , [ sov . jetp , * 64 * , 4 ( 1986 ) ] ( reprinted in `` solar neutrinos : the first thirty years '' , eds . j.n.bahcall _ et . al . _ ) . next - generation cherenkov detector hyper - kamiokande , k. nakamura , talk at neutrinos and implications for physics beyond the standard model 2002 , http://insti.physics.sunysb.edu/itp/conf/neutrino/talks/nakamura.pdf . s. p. mikheyev and a. yu . smirnov , _ 86 massive neutrinos in astrophysics and in particle physics _ , proceedings of the sixth moriond workshop , edited by o. fackler and j. tr@xmath236n thanh v@xmath236n ( editions frontires , gif - sur - yvette , 1986 ) , pp . 355 . baltz and j. weneser , phys . rev . * d35 * , 528 ( 1987 ) ; a. dar _ et . _ , phys . rev . * d 35 * ( 1987 ) 3607 ; s. p. mikheyev and a. yu . smirnov , sov . 30 ( 1987 ) 759 - 790 ; l. cherry and k. lande , phys . rev d * 36 * 3571 ( 1987 ) ; s. hiroi , h. sakuma , t. yanagida , m. yoshimura , phys . lett . b*198 * 403 , ( 1987 ) and prog . theor . phys . * 78 * 1428 , ( 1987 ) ; a. j. baltz and j. weneser , phys . rev . * d37 * , 3364 ( 1988 ) . m. spiro and d. vignaud , phys . b * 242 * 297 ( 1990 ) . a. j. baltz and j. weneser , phys . rev . * d50 * , 5971 ( 1994 ) ; _ ibid _ * d51 * , 3960 ( 1994 ) ; m. maris and s.t . petcov , phys . d*62*(2000)093006 ; n. hata and p. langacker , phys . d * 48*(1993)2937 and phys . d * 50*(1994)632 ; e. lisi , d. montanino , phys . rev . d*56*(1997)1792 ; j. m. gelb , wai - kwok kwong , s. p. rosen , phys . lett . * 78*(1997)2296 . q. y. liu , m. maris and s. t. petcov phys . d * 56 * 5991 ( 1997 ) ; m. maris , s.t . petcov , phys . d * 56 * 7444 , 1997 . m. narayan , g. rajasekaran , r. sinha ; mod . a13 : 1915 ( 1998 ) ; a. de gouvea , a. friedland , h. murayama , hep - ph/9910286 ; g.l . fogli , e. lisi , d. montanino , phys . rev . d * 61 * 073009 ( 2000 ) ; g.l . fogli , e. lisi , d. montanino , a. palazzo , hep - ph/0008012 . j. n. bahcall , p. i. krastev , a. yu . smirnov , phys . * d60*(1999 ) 093001;p . c. de holanda , c. pena - garay , m.c . gonzalez - garcia , j. w. f. valle , phys . d*60*(1999)093010 ; a. h. guth , l. randall , m. serna , jhep * 9908*(1999)018 . p. i. krastev and a. yu . smirnov , phys . b * 226*(1989)341 ; j.m . losecco , phys . d*47*(1993)2032 ; q. y. liu and a. yu . smirnov , nucl . b * 524*(1998)505 ; q. y. liu , s. p. mikheyev , and a. yu . smirnov , phys . b*440*(1998)319 ; s.t . petcov , phys . b * 434*(1998)321 ; kh . akhmedov , a. dighe , p. lipari , and a. yu . smirnov , nucl . b*542*(1999)3 ; e. kh . akhmedov , hep - ph/9903302 ; e.k . akhmedov , nucl . b*538*(1999)25 ; m.v . chizhov , s.t . petcov , phys . rev . lett . * 83*(1999)1096 ; e.k . akhmedov , a.yu . smirnov , phys . * 85*(2000)3978 . e. lisi and d. montanino , phys . a. ioannissian , a. smirnov , hep - ph/0404060 . g. l. fogli , e. lisi , a. marrone , d. montanino , a. palazzo , phys . a. ioannissian , a. smirnov , in preparation .

systems whose thermodynamic parameters are drastically changed give rise to highly nonlinear and far from equilibrium processes that have been intensively studied for decades in various contexts @xcite . these may range from binary fluids to alloys and spin systems exhibiting a disordered phase at high temperatures , whilst having two or more ordered phases below a critical point @xcite . there is already a vast body of research studying such nonequilibrium problems in terms of kinetic ising models @xcite under both glauber and kawasaki dynamics @xcite , respectively associated with the so called models @xmath0 and @xmath1 in the terminology of critical dynamic theories @xcite . after a quench to a subcritical temperature these dynamics attempt to minimize the interfacial energy between different equilibrium domains which therefore grow and coarsen with time . at late stages the spreading of domains is such that if their typical lengths are rescaled by a factor @xmath2 the domain patterns at different times will be statistically similar . here , the scaling or dynamic exponent @xmath3 is characteristic of the universality class to which the dynamic belongs , and is usually independent of the spatial dimensionality but sensitive to conservation laws ( see , e.g. @xcite and references therein ) . for one - dimensional ( 1d ) systems these dynamics are also amenable to experimental probe @xcite . in particular , the magnetic relaxation of synthesized molecular chains with strong ising anisotropy @xcite was considered in the framework of a glauber dynamics with both first- and second - neighbor interactions @xmath4 @xcite . in that regard , recently it was suggested that weak competing regimes in the low but nonzero - temperature limit ( @xmath5 ) give rise to an almost ballistic dynamic exponent ( @xmath6 ) @xcite rather than the usual diffusive value of @xmath7 @xcite . irrespective of how small the frustration might be , note that for @xmath8 such discontinuous scaling behavior is also accompanied by the sudden appearance of a large basin of metastable states @xcite . when it comes to the ferromagnetic kawasaki dynamics , these latter already exist for @xmath9 and are characterized by kinks or domain walls separated by two or more lattice spacings @xcite . in that situation the mean density of kinks reaches a finite value @xcite , and so the average size of metastable domains can not but remain bounded . at zero temperature the dynamics rapidly gets stuck in these states -thus preventing the system to reach equilibrium- but for @xmath5 their structure is at the origin of the @xmath10 growth of domain length scales @xcite . despite metastability , note that as long as temperature is held finite , no matter how small , the dynamics is still ergodic and eventually the equilibrium state is accessible . continuing the development initiated in ref.@xcite , in this work we further investigate the metastable effects brought about by second neighbor couplings on the scaling regimes of this typical phase separation dynamics . as we shall see , for @xmath11 there are new metastable scenarios and activation barriers that come into play , ultimately affecting the large time kinetics in the low temperature limit . following the methodology of ref.@xcite , first we will construct and diagonalize numerically the kink evolution operator associated to the master equation @xcite of this dynamics in finite chains . this will enable us to determine the relaxation time ( @xmath12 ) of these processes from the spectral gaps of that liouvillian operator . as is known , in nearing a critical point those time scales diverge with the equilibrium correlation length as @xmath13 @xcite , which in the present context also grows unbounded in the limit of @xmath14 ( except for @xmath15 where the ground state is highly degenerate and @xmath16 remains finite @xcite ) . on the other hand , in a system of typical length @xmath17 evidently @xmath16 can not grow beyond that scale ; hence in practice it is customary to trade that scaling relation by the finite - size behavior @xmath18 provided @xmath17 is taken sufficiently large @xcite . thus , once armed with the above referred spectral gaps we will aim at obtaining dynamic exponents from this finite - size scaling relation across a host of metastable situations . at low but nonzero - temperatures , however notice that owing to the presence of arrhenius activation energies the time scales involved are unbounded even for finite systems ( cf . sec.iv ) . the structure of this work is organized as follows . in sec.ii we recast the master equation of these stochastic processes in terms of a quantum spin analogy that readily lends itself to evaluate the low lying levels of the associated evolution operator . owing to detailed balance @xcite this latter can be brought to a symmetric representation by means of simple nonunitary spin transformations , thus simplifying the subsequent numerical analysis . sec.iii describes various metastable regimes while attempting to identify their decay patterns and activation barriers . calculational details regarding the proliferation rates of these states with the system size are relegated to appendix a. in sec.iv we evaluate numerically the spectrum gaps of finite chains using standard recursive techniques @xcite which yield clear arrhenius trends for relaxation times at low temperature regimes . this provides a sequence of finite - size approximants to dynamic exponents which are then combined with extrapolations to the thermodynamic limit @xcite . finally , sec.v contains a summarizing discussion along with some remarks on open issues and prospects of future work . let us consider the kawasaki dynamics in a chain of ising spins ( @xmath19 ) coupled with both first- and second - neighbor interactions @xmath4 , thus setting energy configurations @xmath20 while in contact with a heat bath at temperature @xmath21 . here , frustration arises when combining antiferromagnetic ( af ) @xmath22 couplings ( @xmath23 ) with @xmath24 exchanges of any sign . in particular , for @xmath25 the ground state consists of consecutive pairs of oppositely oriented spins ( @xmath26 ) , while for @xmath27 the ordering is f or af depending on the respective sign of @xmath24 . for @xmath28 there is no frustration and the order type is also set by @xmath24 . unless otherwise stated , periodic boundary conditions ( pbc ) and a vanishing magnetization will be assumed throughout . the bath is represented as causing the ising states @xmath29 to fluctuate by exchanges of nearest neighbor ( nn ) spin pairs chosen randomly from @xmath17 locations . to enforce the system to relax towards the boltzmann distribution @xmath30 , the transition probability rates per unit time @xmath31 between two configurations @xmath32 ( here differing just in an exchanged pair of nn spins ) , are chosen to satisfy the detailed balance condition @xcite @xmath33 ( also , see its role by the end of sec.iia ) . however , detailed balance itself can not determine entirely the form of such rates , thus in what follows we take up the common choice used in the context of kinetic ising models , namely @xcite @xmath34\,\big\}\,,\ ] ] where @xmath35 just sets the time scale of the microscopic process , and is hereafter set to 1 . also , from now on temperatures are measured in energy units , or , equivalently , the boltzmann constant in @xmath36 is taken equal to unity . in the specific case of spin exchanges , say at locations @xmath37 , clearly from eq.([energy ] ) the above energy differences reduce to @xmath38 $ ] . thus , after introducing the parameters @xmath39 ( @xmath40 ) and using basic properties of hyperbolic functions , it is a straightforward matter to verify that the exchange rates @xmath41 associated to those locations actually deploy multi - spin terms of the form @xmath42\,,\end{aligned}\ ] ] which already anticipate the type of many - body interactions to be found later on in the evolution operator of sec.iia . here , the factors @xmath43 and @xmath44 ensure vanishing rates for parallel spins while contributing to reproduce eq.([choice ] ) for antiparallel ones also , note that although the standard case of @xmath9 is left with terms of just two - spin interactions , its dynamics is _ not _ yet amenable for exact analytic treatments at @xmath45 @xcite . the range of the sites involved in eq.([rates ] ) , or equivalently , in the energy differences of eq.([choice ] ) , basically distinguishes between eight situations of spin exchanges . for later convenience we now regroup them in two sets of dual events in which kinks or ferromagnetic domain walls are thought of as hard core particles @xmath0 undergoing pairing @xmath46 , and diffusion @xmath47 processes , as schematized in table [ tab1 ] . its columns also summarize the information needed to construct the operational form of the dynamics , while allowing to infer the variety of metastable structures alluded to in sec.i . in what follows we turn to the first of these issues , and defer the discussion of the second to sec.iii . 0.5 cm c c c c c pairing @xmath48 & 0.75 cm rate @xmath49 & 0.7 cm @xmath50 & 0.7 cm s element & 0.5 cm projector + + @xmath51 & 0.75 cm @xmath52 & 0.75 cm @xmath53 & 0.75 cm @xmath54 & 0.5 cm @xmath55 + @xmath56 & 0.75 cm @xmath57 & 0.75 cm @xmath58 & 0.75 cm @xmath59 & 0.5 cm @xmath60 + @xmath61 & 0.75 cm @xmath62 & 0.75 cm @xmath63 & 0.75 cm @xmath59 & 0.5 cm @xmath64 + @xmath65 & 0.75 cm @xmath66 & 0.75 cm @xmath67 & 0.75 cm @xmath68 & 0.5 cm @xmath69 + diffusion @xmath70 & 0.75 cm rate @xmath49 & 0.7 cm @xmath71 & 0.7 cm s element & 0.5 cm projector + + @xmath72 & 0.75 cm @xmath73 & 0.75 cm @xmath74 & 0.75 cm @xmath73 & 0.5 cm @xmath55 + @xmath75 & 0.75 cm @xmath76 & 0.75 cm @xmath77 & 0.75 cm @xmath78 & 0.5 cm @xmath60 + @xmath79 & 0.75 cm @xmath80 & 0.75 cm @xmath81 & 0.75 cm @xmath78 & 0.5 cm @xmath64 + @xmath82 & 0.75 cm @xmath73 & 0.75 cm @xmath74 & 0.75 cm @xmath73 & 0.5 cm @xmath69 + 0.3 cm as is known , in a continuous time description of these markovian processes the stochastic dynamics is controlled by a gain - loss relation customarily termed as the master equation @xcite @xmath83\,,\ ] ] which governs the time development of the probability distribution @xmath84 . conveniently , this relation can also be reinterpreted as a schrdinger equation in imaginary time , i.e. @xmath85 under a pseudo - hamiltonian or evolution operator @xmath86 . this is readily set up by defining diagonal and nondiagonal matrix elements @xcite @xmath87 thus formally enabling to derive the state of the system @xmath88 at subsequent times from the action of @xmath86 on a given initial condition , that is @xmath89 . in particular the relaxation time @xmath12 of any observable with nonzero matrix element between the steady state and the first excitation mode of @xmath86 , is singled out by the eigenvalue @xmath90 corresponding to that latter , i.e. @xmath91 , whereas by construction the former merely yields an eigenvalue @xmath92 @xcite . note that the numerical analysis of these spectral gaps ( or inverse relaxation times ) will first require to obtain an operational analog of eqs.([diag ] ) and ( [ non - diag ] ) , as the phase - space dimension of these processes grows exponentially with the system size . that will allow us to implement the recursive diagonalization techniques of sec.iv , where the matrix representation of @xmath86 is not actually stored in memory @xcite . on the other hand , to halve the number of machine operations it is convenient here to turn to a dual description in which new ising variables @xmath93 standing on dual chain locations denote the presence ( @xmath94 ) or absence ( @xmath95 ) of the kinks referred to above . thus , if we think of the states @xmath96 as representing configurations of @xmath97-spinors ( say in the @xmath3 direction ) , we can readily construct the counterpart of the above matrix elements by means of usual raising and lowering operators @xmath98 . clearly , the nondiagonal parts @xmath99 accounting for the kink pairing and diffusion processes depicted in table [ tab1 ] must involve respectively terms of the form @xmath100 and @xmath101 , say for events occurring at locations @xmath102 under the presence of a central kink . however , due to the @xmath22-couplings , note that these terms should also comprise the kink occupation @xmath103 and vacancy @xmath104 numbers of second neighbor sites surrounding that central kink , as these also matter in the rate values of table [ tab1 ] . thus , to weight such correlated processes here we classify them according to projectors defined as @xmath105 to which in turn we assign the variables @xmath106 , and @xmath107 . therefore , with the aid of these latter , the contributions of the pairing and diffusion parts to the operational analog of eq.([non - diag ] ) can now be written down as @xmath108\,,\\ \label{diffusion } h_{nd}^{^{(di\!f\!f ) } } & = & - \sum_{i,\,j}\,{\cal \hat p}_i^{(j ) } \big[\,f ( y_j)\ , \sigma^+_{i-1}\,\sigma^-_{i+1 } + f ( -y_j)\,\sigma^+_{i+1}\,\sigma^-_{i-1}\,\big]\,,\end{aligned}\ ] ] where @xmath109 , and the @xmath110 index runs over the four types of projectors specified in eq.([proj ] ) . when it comes to the diagonal terms associated with eq.([diag ] ) , in turn needed for conservation of probability , notice that these basically count the number of manners in which a given configuration can evolve to different ones in a single step . in the kink representation this amounts to the summation of all pairing and diffusion attempts that a given state is capable of . as before , these attempts also can be probed and weighted by means of the above projectors , vacancy , and number operators , in terms of which those diagonal contributions are expressed here as @xmath111\,,\\ h_d^{^{(diff)}}\!\!\!\ ! & = & \sum_{i,\,j}\,{\cal \hat p}_i^{(j ) } \big[\,f ( -y_j)\ , \hat{\rm v}_{i-1}\ , \hat n_{i+1 } + f ( y_j)\,\hat n_{i-1}\ , \hat{\rm v}_{i+1}\,\big]\,.\end{aligned}\ ] ] thus , after simple algebraic steps and using the @xmath112 parameters defined in eqs.([cpm ] ) and ( [ dpm ] ) , the net contribution @xmath113 of these diagonal terms is found to involve two- , three- , and four - body interactions of the form @xmath114\,,\end{aligned}\ ] ] some of which had already appeared at the level of the original spin rates mentioned in eq.([rates ] ) . _ detailed balance_. further to the correlated pairing and diffusion terms of eqs.([pairing]),([diffusion ] ) which would leave us with a non - symmetric representation of the evolution operator , we can make some progress here by exploiting detailed balance [ eq.([db ] ) ] . this latter warrants the existence of representations in which @xmath86 is symmetric and thereby fully diagonalizable @xcite . for our purposes , it suffices to consider the _ diagonal _ nonunitary similarity transformation @xmath115\,,\ ] ] stemming from the original spin energies of eq.([energy ] ) but re - expressed in terms of kinks , i.e. @xmath116 . hence @xmath117 , implying that the nondiagonal matrix elements of @xmath118 will transform as @xmath119 but since in the kink representation @xmath120 also comply with the detailed balance condition ( [ db ] ) , then clearly these elements become symmetric under @xmath121 ( see the symmetrized elements of table [ tab1 ] ) . equivalently , under eq.([transf ] ) the pairing and diffusion operators involved in eqs.([pairing ] ) and ( [ diffusion ] ) will transform respectively as @xmath122\ ; \sigma^{\pm}_{i-1 } \,\sigma^{\pm}_{i+1}\;,\\ \phantom{\big [ } \sigma^{\pm}_{i-1 } \,\sigma^{\mp}_{i+1 } & \to & \exp \big [ \pm k_2 \left ( \sigma^z_{i-2 } - \sigma^z_{i+2}\right ) \big]\ ; \sigma^{\pm}_{i-1 } \,\sigma^{\mp}_{i+1}\;,\end{aligned}\ ] ] while leaving @xmath123 and all projectors of eq.([proj ] ) unchanged . thus , after introducing the @xmath124 and @xmath125 coefficients @xmath126 it is straightforward to check that the symmetric counterparts of @xmath127 and @xmath128 are then given by @xmath129 \ , \left(\,\sigma^+_{i-1}\,\sigma^+_{i+1 } + { \rm h.c.}\,\right)\,,\\ \label{s - diffusion } { \cal h}_{nd}^{^{(di\!f\!f ) } } & = & - \frac{1}{8}\,\left(\,1 + { \rm sech}\ , 2k_2 \ , \right)\,\sum_i \,\left(1+\sigma^z_i\right ) \left(\,1+\tanh^2\!k_2\ ; \sigma^z_{i-2}\ , \sigma^z_{i+2}\,\right ) \,\left(\,\sigma^+_{i-1}\,\sigma^-_{i+1 } + { \rm h.c.}\,\right)\,.\end{aligned}\ ] ] together with eq.([diagonal ] ) this completes the construction of the operational analog of eqs.([diag ] ) and ( [ non - diag ] ) in an hermitian representation . in passing , it is worth to point out that all above non - diagonal operators not only preserve the parity of kinks @xmath130 ( being even for pbc ) , but that they also commute , by construction , with @xmath131 which simply re - expresses the conservation of the total spin magnetization in the original system . in practice , for the numerical evaluation of spectral gaps ( sec.iv ) we will just build up the adequate basis of kink states from the corresponding spin ones . after an instantaneous quench down to a low but non - zero temperature often this stochastic dynamics rapidly reaches a state in which further energy - lowering processes are unlikely . this is because the configuration space contains ` basins ' of local energy minima from which the chances to access lower energy states must first find their way through a typical ` energy barrier ' @xmath132 . in the limit of @xmath14 the average time spent in these configuration , or metastable ( m ) states , then diverges with an arrhenius factor @xmath133 . in common with the standard ferromagnetic dynamics , here the decay from these m - configurations is mediated by diffusion of kink pairs . in particular , for @xmath9 their release requires activation energies of @xmath134 which at low - temperatures involves time scales @xmath135 ( see pairing rates of table [ tab1 ] ) . as a result of the diffusion of these pairs , entire ferromagnetic domains can move rigidly by one lattice spacing @xcite . following an argumentation given in refs.@xcite , the repeated effect of that rather long process ultimately leads to coarsening of domains , and is at the root of their @xmath136 growth @xcite . for @xmath137 however there are other energy barriers that also come into play , so the identification of a net arrhenius factor in the actual relaxation time is less straightforward . in addition , due to the discontinuities already appearing at the level of transition rates ( specifically at @xmath138 in the limit of @xmath14 ) , note that there are several coupling regimes where that identification must be carried out . as we shall see , that will prove much helpful in the finite - size scaling analysis of sec.iv thus , here we focus attention on the variety of m - structures arising in the coupling sectors _ ( a ) _ to _ ( h ) _ listed in table [ tab2 ] . reasoning with table [ tab1 ] and guided by simulated quenches down to @xmath139 , we turn to the characterization of these structures while trying to identify their decay patterns . also , a measure of the basin of these states , such as the rate at which they proliferate with the system size , is provided with the aid of appendix a. the results of the arguments and observations that follow in this section are summarized in table [ tab2 ] . 0.2 cm c c c c coupling regime & typical m - state & rates ( @xmath140 ) & barrier ( @xmath141 ) + + @xmath142 & 0.7 cm @xmath143{\,0\ , \cdots\,}_{\rm v\,\ge 1}\ , { \blue\bf 1}\ ; 0\ , \cdots$ ] & -0.8 cm 1.6180 & @xmath144 + @xmath145 & 0.7 cm @xmath146{\underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v\ , \ge 1}\ , { \blue\bf 1}\ , \underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v'\,\ge 1}}^{\rm v\,\,\,or\,\,\ , v'\,>\ , 1}\ , { \blue\bf 1}\ , \cdots$ ] & -0.8 cm 1.5701 & @xmath144 + -1.5 cm _ ( c ) _ @xmath147 & 0.6 cm @xmath148{\,0\ , \cdots\,}_{\rm v\ , \ge 2}\,{\blue\bf 1}\;0\ , \cdots$ ] & -0.8 cm 1.4655 & @xmath144 + -1.5 cm _ ( d ) _ @xmath149 & 0.54 cm @xmath150 { \,{\blue\bf 1}\,\cdots\,}_{k\ , \ne 2,3}\;\underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v\ , \ge 2}\ , { \blue\bf 1}\,\cdots$ ] & -0.8 cm 1.6180 & @xmath144 + -1 cm _ ( e ) _ @xmath151 & 0.55 cm @xmath150 { \,{\blue\bf 1}\,\cdots\,}_{k\ , \ge 3}\;\underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v\ , \ge 2 } \,{\blue\bf 1}\,\cdots$ ] & -0.8 cm 1.5289 & @xmath152 + _ ( f ) _ @xmath153 & 0.54 cm @xmath150{\,{\blue\bf 1}\,\cdots\,}_{k\,=1,2}\,0\ ; { \blue\bf 1}\,\cdots$ ] & -0.8 cm 1.3247 & -0.2 cm @xmath154 + -1.6 cm _ ( g ) _ @xmath155 & 1.1 cm @xmath156 { \underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v\ , \ge 1}\ ; \underbracket[0.5pt ] { \,{\blue\bf 1}\,\cdots\,}_{k\,=1,2 } \ ; \underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v'\ , \ge 1}}^ { { \rm v \,\,\ , or \,\,v ' = \ , 1,\,\,\,if } \,\,\ , k\,=1}\,{\blue\bf 1}\ , \cdots$ ] & -0.8 cm 1.7437 & @xmath157 + -1.08 cm _ ( h ) _ @xmath158 & 0.57 cm @xmath150{\,{\blue\bf 1}\,\cdots\,}_{k\,=1,2}\ , \underbracket[0.5pt]{\,0\ , \cdots\,}_{\rm v\ , \ge 1}\,{\blue\bf 1}\,\cdots$ ] & -0.8 cm 1.8392 & @xmath159 + 0.35 cm 0.25 cm _ ( a ) _ and _ ( b)_. in these two first coupling sectors the kinks of all m - states must be separated by at least one vacancy because in the pairing rates of table [ tab1 ] both @xmath160 and @xmath161 are positive . however , note that in sector _ ( b ) _ sequences of the form @xmath162 could not show up because there @xmath163 ( see fourth pairing process of table [ tab1 ] ) . as is indicated in appendix a this further constraint ( schematized in table [ tab2 ] ) , significantly reduces the proliferation of m - states with respect to sector _ ( a)_. for this latter the number of m - configurations turns out to grow as @xmath164 with golden mean @xmath165 , whereas for sector for _ ( b ) _ it grows only as @xmath166 . as mentioned above , on par with the usual dynamics of @xmath9 the low - temperature decay from either of these structures involves the diffusion of kink pairs between otherwise isolated domain walls @xcite . however , notice that for @xmath167 this requires the occurrence of two successive and raising energy events ( cf . table [ tab1 ] ) namely , pairing around an existing wall ( @xmath168 ) followed by detachment of neighboring pairs from the triplets so formed ( @xmath169 ) . this allows the kink pair to diffuse at no further energy cost until eventually a new wall is encountered and the energy excess is rapidly released @xcite . the net thermal barrier of this composite process is therefore @xmath170 , thus setting arbitrarily large relaxation time scales in the low - temperature limit , even for _ finite _ lattices . we will find back these time scales later on in the exact diagonalizations of sec.iva . yet , it remains to be determined whether the above reduction of metastability ( as well as that of the following case ) , is of any consequence for dynamic exponents . 0.25 cm _ ( c)_. here @xmath171 but now @xmath172 , implying that kinks must at least be separated by two vacancies . otherwise , the isolated vacancies involved both in the second and third processes of table [ tab1 ] would alternately originate a random walk of pairings at no energy cost until encountering another kink . then the walk could no longer advance , and eventually after few but energy - decreasing processes the isolated vacancy would be finally canceled out . this constraint brings about even further reductions in the number of m - states which actually now grows as @xmath173 ( appendix a ) . however , this new minimum separation of kinks has no effect in the decay pattern mentioned in the previous two cases , so here the arrhenius factors can also be expected to diverge as @xmath174 . as before , this will be corroborated in sec.iva . 0.25 cm _ ( d)_. in this coupling regime @xmath160 and @xmath175 are both negative , meaning that groups of four or more consecutive kinks , i.e. af spin domains , may well show up in these new m - states ( see table [ tab1 ] ) . also , since @xmath176 isolated kinks can still appear scattered throughout . in turn , these latter as well as all af domains must be separated by at least two vacancies otherwise , as indicated in table [ tab1 ] , the energy would decrease further . on the other hand , it turns out that similarly to sector _ ( a ) _ the number of m - states here also proliferates with the golden mean as @xmath164 ( see appendix a ) . when it comes to energy barriers , at a first stage on time scales @xmath177 , and so long as @xmath178 , most af domains can disgregate by pair annihilations followed by energy - decreasing processes . the steps involved may be schematically represented as ( say , starting from the leftmost triplet ) @xmath179 where @xmath180 , and @xmath181 . ( note that annihilations around innermost kinks would require even larger time scales of order @xmath182 because @xmath183 throughout this regime ) . since at low temperatures @xmath184 the process can recur and eventually the entire af domain can be disaggregated on times @xmath185 . however , for @xmath186 that process could no longer advance within those time scales . instead , disgregation simply proceeds via successive detachment of outer pairs , each one requiring scales @xmath187 . in either case , these pairs can then realize a free random walk ( @xmath188 ) , until finally annihilating with an isolated kink ( @xmath189 ) . in a much later second stage , when most consecutive kinks disappear , large vacancy regions then proceed to coarse - grain as in previous cases @xcite . as before , this ultimately requires time scales of order @xmath190 ( cf . with spectral gaps of sec.iva ) . 0.25 cm _ ( e)_. much alike the previous situation , here groups of consecutive kinks separated by two or more vacancies may also appear in these m - sequences because @xmath191 and @xmath192 are still negative . however , since @xmath193 there can be no isolated kinks , and consequently af groups may now range from three kinks onwards ( see table [ tab1 ] ) . as is shown in appendix a , the proliferation of m - states given rise by these new constraints turns out to grow as @xmath194 . like in the initial stages of case _ ( d ) _ , the decay to equilibrium ( now fully af ) also proceeds through a random walk of kink pairs which successively detach from af domains on time scales @xmath195 @xcite . as the random walk goes on , these pairs may coalesce with others , or stick back to the same or to a new domain ( in all cases with @xmath196 ) , until eventually some groups get fragmented into triplets @xcite . yet , within times @xmath195 these may further decompose as @xmath197 say , detaching a rightmost pair from one of those triplets . thus , in the low - temperature limit the above process is unlikely to reverse on these time scales , and so the process can recur in a progressively denser medium until the full af state is reached . as we shall see in sec.ivb this decay pattern , now activated by @xmath198 energies , actually brings about a faster coarsening than those obtained in previous cases . 0.25 cm _ ( f)_. as suggested by simulated quenches down to @xmath139 , for vanishing magnetization there are no consecutive vacancies in the m - states of this regime @xcite . since here @xmath199 and @xmath200 , then following table [ tab1 ] these configurations should be consistent with single and , at most , four contiguous kinks scattered throughout . however , for @xmath201 it turns out that just single and double kinks actually show up @xcite . the recursions of appendix a then show that these tighter constraints give rise to a number of m - states which proliferates only as @xmath202 . as for the manner in which such configurations proceed to equilibrium , contrary to the previous situations it is difficult to identify here a specific pattern of decay . rather , we content ourselves with mentioning that the largest energy barrier of these states ( i.e. @xmath203 actually corresponding to the activation of diffusing pairs ) , turns out to be associated with the actual relaxation times @xmath204 evaluated by the exact diagonalizations of sec.ivc . 0.25 cm _ ( g ) _ and _ ( h)_. for these two last sectors the signs of @xmath175 and @xmath205 remain as in the previous case , though now @xmath160 is positive . thus , following table [ tab1 ] , as before only single and double kinks may appear distributed throughout but now they can be separated by one or more vacancies , even for @xmath206 . thereby , kink pairs are able to diffuse freely ( @xmath188 ) across consecutive vacancies while keeping at least one space from each other and other kinks . so , the typical m - state of these sectors actually results in an alternating sequence of mobile and jammed blocks . in these latter there can be just one vacancy aside each kink pair , whereas single kinks may appear separated by one or more spaces . however , note that in sector _ ( g ) _ sequences of the form @xmath207 would be unstable because there @xmath193 ( see first pairing process of table [ tab1 ] ) . on par with what occurs in cases _ ( a ) _ and _ ( b ) _ , this further constraint ( schematized in table [ tab2 ] ) then reduces the proliferation of m - configurations with respect to sector _ ( h)_. in fact , the recursions of appendix a show that for that latter sector the number of m - states grows as @xmath208 , while for _ ( g ) _ it grows just as @xmath209 . with regard to the decay of these states , as in case _ ( f ) _ here we limit ourselves to mention that in nearing low - temperature regimes it turns out that it is the largest activation barrier the one that takes over ( fourth paring process of table [ tab1 ] ) , as the arrhenius factors of these sectors actually will turn out to diverge as @xmath210 ( see sec.ivc ) . 0.25 cm finally , notice that in the region @xmath211 with @xmath212 there are _ no _ m - basins hindering the access to the af ground state , i.e. @xmath213 . here , the paths to this latter proceed much as in sector _ ( e ) _ except that now @xmath23 , so the dynamics can further decrease the energy either by triplet splittings or pair creations ( @xmath214 ) . therefore as temperature is lowered in this region , the relaxation time towards the af ordering of _ finite _ chains remains bounded ( see beginning of sec.ivc ) . having built up the kink evolution operator in a symmetric representation [ eqs.([diagonal ] ) , ( [ s - pairing ] ) , and ( [ s - diffusion ] ) ] , next we proceed to evaluate numerically its spectral gap in finite chains via a recursion type lanczos algorithm @xcite . as mentioned in sec.ii we focus on the case of vanishing magnetization in the original spin model , thus corresponding to a subspace of @xmath215 kink states . as a preliminary test first we verified that the transformed boltzmann distribution @xmath216 resulting from eq.([transf ] ) , actually yields the ground state of our quantum ` hamiltonian ' with eigenvalue @xmath217 . thereafter , as usual the lanczos recursion was started with a random linear combination of kink states , but here chosen orthogonal to that boltzmann - like direction . in turn , all subsequent vectors generated by the lanczos algorithm were also reorthogonalized to @xmath218 . this allowed us to obtain the first excited eigenmodes of the evolution operator in periodic chains of up to @xmath219 sites , the main limitation for this being the exponential growth of the space dimensionality . another restrictive issue we are confronted with is that as temperature decreases the spectral gaps ( @xmath90 ) get arbitrarily small due to the energy barriers mentioned in sec.iii , i.e. @xmath220 . on the other hand , to ensure that these finite - size quantities are actually scaled within the arrhenius regime , in this context it is more appropriate to put forward a ` normalized ' version of the scaling hypothesis ( [ scaling ] ) , namely @xmath221 where the amplitude @xmath0 would involve at most a @xmath4 dependent quantity ( assuming @xmath17 is large enough ) . moreover , alike dynamic exponents such proportionality factors will also come out as sector - wise universal constants . in practice , below @xmath222 the evaluation of ( [ normalization ] ) requires the use of at least quadruple precision but as the spacing between low - lying levels gets progressively narrow , for @xmath223 it turns out that the pace of the lanczos convergence becomes impractical in most of the coupling sectors of table [ tab2 ] . nevertheless , as we shall see in the following subsections , already when temperature is lowered within the ranges in hand the normalized gaps @xmath224 exhibit clear saturation trends , thus constituting accurate estimations of @xmath225 . let us start by considering the first four cases of table [ tab2 ] , all sharing at large times the decay pattern of the standard dynamics of @xmath9 @xcite , and the energy barriers @xmath227 alluded to in the previous section . in fig.[one ] we display the above normalized gaps for several coupling ratios -0.75 cm of the evolution operator for @xmath228 on approaching low - temperature regimes in sectors _ ( a ) _ and _ ( b ) _ of table [ tab2 ] . from top to bottom , alternating solid and dashed lines stand for coupling ratios @xmath229 . upper- and lowermost dotted curves denoting respectively the cases @xmath230 [ sector _ ( c ) _ ] and @xmath231 , are shown for comparison . details of that latter standard case are displayed in fig.[two](a).,title="fig:",scaledwidth=56.0% ] 0.25 cm @xmath232 in sectors _ ( a ) _ , _ ( b _ ) , and _ ( c ) _ in a chain of 20 sites . as temperature decreases , the saturated behavior of most of the @xmath233 values considered clearly signals the emergence of the expected arrhenius regime . note that even a slight deviation from the conjectured barriers would result in strong departures from this behavior . also , the saturation values of @xmath234 , i.e. the amplitudes involved in eq.([normalization ] ) , come out to be @xmath233-independent so long as @xmath235 . apart from finite - size corrections ( fig.[two ] ) , this independence also holds for all other accessible lengths ( a general feature applying also to other sectors of table [ tab2 ] ) . however in approaching @xmath236 or @xmath237 , where discontinuities already appear at the level of transition rates ( see table [ tab1 ] ) , the arrhenius trend is only incipient and in some limiting cases it remains beyond our reach . when it comes to dynamic exponents ( @xmath3 ) , in the main panels of fig.[two ] we show the finite - size behavior of these normalized gaps comparing the case of @xmath9 with others in sectors _ ( a ) _ and _ ( b)_. 0.1 cm -13 cm for ( a ) @xmath238 ( usual ferromagnetic dynamics ) , ( b ) @xmath239 , and ( c ) @xmath240 . solid and dashed lines show alternately the cases of @xmath241 . in ( a ) sizes increase in upward order , while in ( b ) and ( c ) they do so from top to bottom . the data collapse of larger sizes was attained using dynamic exponents read off from the slope of the insets . these latter estimate the finite - size decay of spectral gaps close to @xmath14 [ ( eq.([normalization ] ) ] . in that limit , the decay of ( b ) ( not shown ) is indistinguishable from that of ( c ) due to the common saturation trends of fig.[one ] . for displaying convenience vertical scales of main panels were normalized by a factor @xmath242.,title="fig:",scaledwidth=37.0% ] -6.8 cm -1 cm for ( a ) @xmath238 ( usual ferromagnetic dynamics ) , ( b ) @xmath239 , and ( c ) @xmath240 . solid and dashed lines show alternately the cases of @xmath241 . in ( a ) sizes increase in upward order , while in ( b ) and ( c ) they do so from top to bottom . the data collapse of larger sizes was attained using dynamic exponents read off from the slope of the insets . these latter estimate the finite - size decay of spectral gaps close to @xmath14 [ ( eq.([normalization ] ) ] . in that limit , the decay of ( b ) ( not shown ) is indistinguishable from that of ( c ) due to the common saturation trends of fig.[one ] . for displaying convenience vertical scales of main panels were normalized by a factor @xmath242.,title="fig:",scaledwidth=37.0% ] -6.82 cm 11.2 cm for ( a ) @xmath238 ( usual ferromagnetic dynamics ) , ( b ) @xmath239 , and ( c ) @xmath240 . solid and dashed lines show alternately the cases of @xmath241 . in ( a ) sizes increase in upward order , while in ( b ) and ( c ) they do so from top to bottom . the data collapse of larger sizes was attained using dynamic exponents read off from the slope of the insets . these latter estimate the finite - size decay of spectral gaps close to @xmath14 [ ( eq.([normalization ] ) ] . in that limit , the decay of ( b ) ( not shown ) is indistinguishable from that of ( c ) due to the common saturation trends of fig.[one ] . for displaying convenience vertical scales of main panels were normalized by a factor @xmath242.,title="fig:",scaledwidth=37.0% ] 0.25 cm clearly , the data collapse onto larger sizes is better for the standard dynamics , although in all cases the dynamic exponents producing these scaling plots are close to @xmath243 ( check later on the extrapolations given in sec.ivd ) . in turn , their values were estimated from the slopes fitting the finite - size decay of @xmath234 at the saturation limit ( insets of fig.[two ] ) , this being almost identical in _ ( a ) _ and _ ( b ) _ as it is evidenced in fig.[one ] . the above observations can also be extended to fig.([three ] ) , where the normalized gaps of sectors _ ( c ) _ and _ ( d ) _ are exhibited . in parallel with the difficulties occurring in fig.([one ] ) as @xmath244 , here these also appear in approaching the 0.1 cm -0.75 cm for @xmath228 within regime _ ( d ) _ of table [ tab2 ] . from top to bottom solid and dashed lines refer in turn to coupling ratios @xmath245 . for comparison , also the case @xmath230 [ regime _ ( c ) _ ] is shown by the dotted line . the inset estimates the typical finite - size decay of these gaps in the low - temperature limit [ ( eq.([normalization ] ) ] . filled triangles are representative of @xmath246 in sector _ ( d ) _ , while open circles stand for @xmath247 . however , see extrapolations of sec.ivd.,title="fig:",scaledwidth=56.0% ] 0.25 cm limit of @xmath248 . but otherwise , as before , the arrhenius regime can be reached already within our low temperature ranges . in that respect , the inset shows that the common finite - size decay of @xmath249 in sector _ ( d ) _ closely follows that of _ ( c ) _ ( @xmath250 ) , both regimes being characterized by a slope ( dynamic exponent ) very close to that obtained for sectors _ ( a ) _ and _ ( b)_. thus , an asymptotic scaling regime similar to the standard one of @xmath9 might be expected in these first four cases ( despite the different proliferation rates of their corresponding m - states ) . but for the moment we defer that discussion to sec.ivd . next we turn to regime _ ( e ) _ where , as referred to in sec.iii , the dynamics follows a rather different decay pattern into af states on time scales @xmath195 . as temperature is lowered , the saturation trends observed in fig.[four](a ) already disclose the emergence of these arrhenius factors for several @xmath233-coupling ratios . as in the previous subsection , the agreement with the former is very precise given the persistence of the @xmath234 plateaus . also , the amplitudes concerning eq.([normalization ] ) here turn out to be @xmath233-independent although , likewise with what occurs in fig.([one ] ) , as @xmath252 decreases the arrhenius regime barely shows up for @xmath253 . on the other hand , the trend of decreasing minima approaching the standard non - metastable gaps of @xmath9 is disrupted in the low temperature limit . as discussed below , this already signals an abrupt crossover of scaling regimes . note that regardless of how small @xmath233 might be , the metastability of this sector does not disappear so long as @xmath28 . thus , in the limit of @xmath254 this poses a situation reminiscent to that mentioned in sec.i for the 1d glauber dynamics under weak competing interactions . irrespective of the weakness of the frustration , in the limit of @xmath14 metastability takes over and changes the dynamic exponents of that non - conserving dynamics from diffusive ( @xmath7 ) to almost ballistic ( @xmath255 ) @xcite . in that regard , here the analogy goes deeper as a similar discontinuity in scaling regimes also appears in this ( non - frustrated ) sector . this is exemplified for @xmath256 in the scaling plot of fig.[four](b ) where the data collapse towards larger sizes is attained on choosing a dynamic exponent @xmath257 , in turn read off from the slope of the inset . as in sec.iva , this latter depicts the finite - size behavior of normalized gaps within the common saturation regime of fig.[four](a ) , thus showing a @xmath249-decay which presumably is also representative of all @xmath258 _ ( e)_. let us anticipate that the dynamic exponent arising from the finite - size extrapolations of these data ( see sec.ivd ) also tends to a nearly ballistic value , far apart from the standard diffusive case of @xmath259 ( @xmath260 , see fig.[five ] below ) as well as from the subdiffusive one with @xmath261 [ @xmath262 , fig.[two](a ) ] . 0.15 cm -8 cm in regime _ ( e ) _ of table [ tab2 ] using @xmath228 . from left to right solid and dashed lines denote in turn the cases of @xmath263 . cusps stem from level crossings in the spectrum of the evolution operator . the dotted line shows for comparison the standard case of @xmath231 with @xmath211 . ( b ) finite - size scaling of normalized gaps for @xmath264 upon identifying @xmath3 with the slope of the inset . alternating solid and dashed lines in downward direction indicate sizes @xmath265 . for displaying purposes the vertical scale was normalized by a factor @xmath242 . the inset estimates the typical finite - size decay of these gaps within sector _ ( e ) _ in the limit of @xmath14 . see however extrapolations of sec.ivd.,title="fig:",scaledwidth=40.0% ] -8.1 cm in regime _ ( e ) _ of table [ tab2 ] using @xmath228 . from left to right solid and dashed lines denote in turn the cases of @xmath263 . cusps stem from level crossings in the spectrum of the evolution operator . the dotted line shows for comparison the standard case of @xmath231 with @xmath211 . ( b ) finite - size scaling of normalized gaps for @xmath264 upon identifying @xmath3 with the slope of the inset . alternating solid and dashed lines in downward direction indicate sizes @xmath265 . for displaying purposes the vertical scale was normalized by a factor @xmath242 . the inset estimates the typical finite - size decay of these gaps within sector _ ( e ) _ in the limit of @xmath14 . see however extrapolations of sec.ivd.,title="fig:",scaledwidth=40.0% ] 0.25 cm before moving on to other sectors of table [ tab2 ] , first we consider the non - metastable regime mentioned by the end of sec.iii , namely the situation of @xmath211 with @xmath267 . as in sector _ ( e ) _ , here the phase ordering is still af . in fig.[five](a ) we display the plain spectral gaps for several coupling ratios in this region [ no need of normalization as in eq.([normalization ] ) ] . -8.2 cm and @xmath268 in the non - metastable region @xmath269 . from left to right solid and dashed lines refer alternately to coupling ratios @xmath270 . the case @xmath238 is also shown for comparison ( doted line ) . ( b ) scaling plot of these levels for @xmath239 and @xmath271 ( solid and dashed lines from top to bottom ) using @xmath272 . for convenience the vertical scale was normalized by a factor @xmath273 . the slope of the inset corroborates a common diffusive decay in the low - temperature limit of all ratios considered in panel ( a).,title="fig:",scaledwidth=40.0% ] -7.87 cm and @xmath268 in the non - metastable region @xmath269 . from left to right solid and dashed lines refer alternately to coupling ratios @xmath270 . the case @xmath238 is also shown for comparison ( doted line ) . ( b ) scaling plot of these levels for @xmath239 and @xmath271 ( solid and dashed lines from top to bottom ) using @xmath272 . for convenience the vertical scale was normalized by a factor @xmath273 . the slope of the inset corroborates a common diffusive decay in the low - temperature limit of all ratios considered in panel ( a).,title="fig:",scaledwidth=40.0% ] 0.25 cm contrariwise to all other sectors , in this case the relaxation times ( @xmath274 ) of finite chains are kept bounded in the low - temperature limit , and so the lanczos convergence is now faster . unlike the glauber case briefly touched upon in sec.ivb , here the presence of frustration does not bring about changes in scaling regimes . taking for instance @xmath239 , this is checked in fig.[five](b ) where at low - temperatures all finite - size data can be made to collapse into a single curve by choosing the same diffusive exponent of the standard af dynamics . in turn , the inset also corroborates this by estimating the slope with which these gaps decay with the system size as @xmath275 . since in that limit @xmath90 becomes @xmath233-independent ( just as do the amplitudes accompanying the arrhenius factors in the above subsections ) , clearly this scaling behavior persists through the entire non - metastable region . turning to sectors _ ( f ) _ and _ ( g ) _ , there are various thermal barriers affecting the decay of their respective m - structures , namely ( in increasing order ) @xmath276 for @xmath258 _ ( f ) _ , and @xmath277 for @xmath258 _ ( g)_. among these barriers , it turns out that actually the largest one of each sector is comprised in the normalized gaps exhibited in fig.[six ] . as before , the precision of the corresponding arrhenius -7.5 cm for @xmath228 in sectors _ ( f ) _ and _ ( g ) _ of table [ tab2 ] [ panels ( a ) and ( b ) respectively , exhibiting results of several coupling ratios ] . from top to bottom solid and dashed lines stand in turn for ( a ) , @xmath278 , and ( b ) , @xmath279 . the inset shows the case of @xmath247 for which these gaps become size - independent in nearing the arrhenius regime . here , lengths @xmath280 ( in upward direction ) match the four - fold periodicity of the ground state alluded to in the text.,title="fig:",scaledwidth=37.0% ] -7.98 cm 7.2 cm for @xmath228 in sectors _ ( f ) _ and _ ( g ) _ of table [ tab2 ] [ panels ( a ) and ( b ) respectively , exhibiting results of several coupling ratios ] . from top to bottom solid and dashed lines stand in turn for ( a ) , @xmath278 , and ( b ) , @xmath279 . the inset shows the case of @xmath247 for which these gaps become size - independent in nearing the arrhenius regime . here , lengths @xmath280 ( in upward direction ) match the four - fold periodicity of the ground state alluded to in the text.,title="fig:",scaledwidth=39.5% ] 0.25 cm factors is reflected in the clear saturation behavior obtained in the low - temperature regime . however just as in sec.iva , in approaching @xmath281 or @xmath282 [ where the trend of increasing maxima continues for heights larger than the range of fig.[six](a ) ] , the discontinuities arising in transition rates carry that saturation limit beyond our reach . but surprisingly , as is shown by the inset of fig.[six](b ) , for @xmath230 that limit becomes _ size_-independent . this suggests an exponentially fast relaxation even in the thermodynamic limit , though as @xmath14 the time scales involved get arbitrarily large , i.e. @xmath283 . in considering the finite - size behavior of @xmath284 for other coupling ratios in sectors _ ( f ) _ and _ ( g ) _ , note that there the four - fold periodicity of the ground state @xmath285 mentioned by the beginning of sec.ii leaves us with few sizes to draw conclusions about dynamic exponents . however , it is worth mentioning that the rather small logarithmic slopes resulting from the gaps of @xmath286 and 24 ( namely , @xmath287 and 0.16 ) , are consistent with the size - independent gaps obtained for @xmath230 . 0.25 cm _ sector ( h)_. as before , there are several thermal barriers affecting the m - states of this sector ( @xmath288 ) , though now the largest one ( @xmath289 ) ends up imposing even more severe restrictions on the lanczos procedure as temperature is lowered . in fact , for @xmath290 the relenting convergence pace precluded us to obtain further results within the arrhenius regime . in part , this also stems from level crossings in the spectrum of the evolution operator , on the other hand responsible for the pointed cusps observed in fig.[seven ] . there , we just content with evidencing the presence of a common activation factor characterizing the decay towards either the ferromagnetic or four - fold ground state ( @xmath291 , or @xmath292 respectively ) . 0.35 cm in regime _ ( h ) _ of table [ tab2 ] , using @xmath293 for @xmath294 ( solid and dashed lines from left to right).,title="fig:",scaledwidth=41.0% ] 0.25 cm armed with the finite - size estimations of the normalized gaps evaluated in secs.iva and ivb , next we turn to the issue of going a step further than the scaling plots considered so far . in that respect , an improved estimation of dynamic exponents can be made by introducing the sequence of approximants or effective exponents @xmath295 } { \ln[\,(l\!-\!2)/l\,]}\,,\ ] ] each of which simply derives a measure of @xmath3 from the gaps of successive chain lengths . similarly , it is worth introducing a sequence of approximants to the amplitudes involved in eq.([normalization ] ) , as their common saturation values strongly suggest that these quantities are robust within each coupling sector of table [ tab2 ] . thus , concurrently with eq.([extra - z ] ) we shall also consider the accompanying set of effective amplitudes @xmath296 given by @xmath297}\,.\ ] ] in general , the elements @xmath298 of a finite - size sequence obtained close to a critical point ( here @xmath299 ) , are assumed to converge logarithmically @xcite as @xmath300 , with @xmath301-constants and @xmath302-exponents such that @xmath303 . to minimize the number of fitting parameters here we keep only the leading - order term of that expansion which just leave us with a nonlinear least - squares fit of three quantities . the results of those regressions are depicted in fig.[eight ] which summarizes the trends of sequences ( [ extra - z ] ) and ( [ extra - a ] ) across sectors _ ( a ) _ to _ ( e ) _ , together with those found in the non - metastable region . specifically , the extrapolated exponents and amplitudes of each case turn out to be @xmath304 the pace of convergence of effective exponents in sectors _ ( a ) _ and _ ( b ) _ comes out slightly slower than that arising in _ ( c ) _ and _ ( d ) _ ( @xmath305 and 2.07 respectively ) , though in the case of effective amplitudes that pace is inverted ( @xmath306 and 1.58 ) . in sector _ ( e ) _ the convergence is still a bit slower [ @xmath307 in fig.[eight](a ) and @xmath308 in fig.[eight](b ) ] , but as anticipated in sec.ivb the extrapolated dynamic exponent is close to that resulting from the 1d glauber dynamics under weak competing interactions @xcite . thus , both scenarios are characterized by a discontinuous crossover from a non - metastable diffusive regime to a metastable one with nearly ballistic exponents . in the former case ( @xmath309 ) the convergence is somewhat faster ( @xmath310 for exponents , and @xmath311 for 0.1 cm -9 cm ) and ( [ extra - a ] ) . data of regimes _ ( a ) _ and _ ( b ) _ ( listed in table [ tab2 ] ) , are represented by circles , regimes _ ( c ) _ and _ ( d ) _ by triangles , and regime _ ( e ) _ by rhomboids . rightmost horizontal lines depict the confidence intervals arising from non - linear fittings in those regimes ( see text for details ) . leftmost dashed lines indicate the limiting values of the rapidly converging approximants for the usual case of @xmath312 . small dots stand for results in the non - metastable sector @xmath313 with @xmath211.,title="fig:",scaledwidth=44.0% ] -9.65 cm 8.5 cm ) and ( [ extra - a ] ) . data of regimes _ ( a ) _ and _ ( b ) _ ( listed in table [ tab2 ] ) , are represented by circles , regimes _ ( c ) _ and _ ( d ) _ by triangles , and regime _ ( e ) _ by rhomboids . rightmost horizontal lines depict the confidence intervals arising from non - linear fittings in those regimes ( see text for details ) . leftmost dashed lines indicate the limiting values of the rapidly converging approximants for the usual case of @xmath312 . small dots stand for results in the non - metastable sector @xmath313 with @xmath211.,title="fig:",scaledwidth=44.0% ] 0.25 cm amplitudes ) and the estimated errors become smaller than 0.3% . by contrast , in sectors _ ( a ) _ to _ ( d ) _ the errors are such that the resulting confidence intervals superpose each other [ see eq.([extrapol ] ) ] , though this is also due to the slight differences between the extrapolated values obtained in those sectors . since in practice it is never really clear whether the assumed asymptotic behavior is sufficiently well realized by the data available @xcite , those differences might well be ascribed to our finite - size limitations . in that sense , the merging of confidence intervals [ rightmost center of panels [ eight](a ) and [ eight](b ) ] suggests a common characterization of these four sectors ( as anticipated by the end of sec.iva ) , within an error margin of less than @xmath314 . on this particular , it is also worth pointing out that for the standard ferromagnetic case of @xmath9 the differences between our higher approximants , namely @xmath315 and @xmath316 , are both less than 0.02% [ also see fig.[two](a ) ] . the corresponding sequences approach swiftly towards @xmath317 ( thus suggesting a slightly slower kinetics than the lifschitz - slyozov type @xcite ) , and @xmath318 , both values being consistent with eq.([extrapol ] ) and included within the merged intervals of fig.[eight ] . finally , we should add that the seemingly fast convergence of approximants ( [ extra - z ] ) and ( [ extra - a ] ) in fig.[eight ] only occurs within a small region of our scaled sizes ( @xmath319 , @xmath320 ) . this is due to the big @xmath301-slopes stemming from our nonlinear least - squares fits , so that the measure of successive errors is actually @xmath321 . nonetheless , the larger extrapolated errors of eq.([extrapol ] ) resulted in less than 4% . to summarize , we have studied a 1d kawasaki dynamics considering up to second neighbor interactions thus uncovering a range of metastable situations [ sectors _ ( a ) _ to _ ( h ) _ specified in table [ tab2 ] ] . following the thread of arguments given in the glauber counterpart @xcite , we have constructed a quantum spin analogy whose ` hamiltonian ' [ eqs.([diagonal ] ) , ( [ s - pairing ] ) , and ( [ s - diffusion ] ) ] played the role of the evolution operator of these processes in the kink representation ( see table [ tab1 ] ) . the relaxation times of these former were then evaluated numerically in finite chains by analyzing the spectral gaps associated with those hamiltonians using standard recursive methods @xcite . we focused attention on the low but nonzero - temperature regimes where magnetic domains tend to coarsen and relaxation times can grow arbitrarily large -even for finite chains- due to the activation barriers discussed through secs.iii and iv . the usual finite - size scaling hypothesis ( [ scaling ] ) was then ` normalized ' as in eq.([normalization ] ) so as to actually scale the spectral gaps of each sector within their corresponding arrhenius regimes . at time scales of order @xmath322 the decay patterns of sectors _ ( a ) _ to _ ( d ) _ were argued to be those of the standard ferromagnetic case @xcite , although the proliferation of metastable states in sectors _ ( b ) _ and _ ( c ) _ turns out to be smaller . however , those differences appear to have no effect on the dynamic exponents , at least within the confidence intervals estimated in the extrapolations of sec.ivd [ eq.([extrapol ] ) and fig.[eight](a ) ] . by contrast , those extrapolations yielded nearly ballistic values for the exponents of sector _ , on the other hand conjectured to decay through times @xmath323 in a rather different form . note here then the abrupt crossover of scaling regimes in passing from sector _ ( d ) _ to _ ( e)_. also , in moving from this latter to the non - metastable region @xmath324 with @xmath211 , another discontinuous change of dynamic exponents occurs . in the absence of activation barriers now these former become diffusive [ figs.[five](b ) and [ eight](a ) ] whilst the relaxation time of finite systems remains bounded even at @xmath139 . this situation is highly reminiscent of that of the glauber dynamics studied in ref.@xcite where the sudden emergence of metastable states under a small @xmath23 also changes these exponents from diffusive to nearly ballistic . when it comes to sectors _ ( f ) _ and _ ( g ) _ the four - fold periodicity of the ground state mentioned in sec.ii left us with few sizes to consider in eq.([normalization ] ) , thus restricting our ability to extrapolate dynamic exponents . however for @xmath325 , just where the activation barriers of these sectors coincide ( see table [ tab2 ] ) , surprisingly the normalized gaps become independent of the system size in the arrhenius regime [ inset of fig.[six](b ) ] . clearly , this suggests an exponential relaxation to equilibrium through time scales @xmath326 that would persist up to the thermodynamic limit . in turn , this would be consistent with the small exponents preliminarily obtained for other coupling ratios in these sectors , but that is an issue requiring further investigation . similarly , the study of sector _ ( h ) _ remains quite open given the convergence difficulties encountered in larger chains as temperature is decreased . nonetheless , _ all _ sectors indicate that the amplitudes involved in eq.([normalization ] ) possibly stand for piecewise - universal quantities . except at @xmath327 , where the original transition rates get discontinuous in the limit of @xmath14 , this is evidenced by the common saturation values of normalized gaps observed throughout figs.[one ] , [ three ] , [ four](a ) , [ five](a ) , [ six](a ) , [ six](b ) , and [ seven ] . as with dynamic exponents , those values were extrapolated to their thermodynamic limit in sectors _ ( a ) _ to _ ( e ) _ as well as in the non - metastable region [ eq.([extrapol ] ) and fig.[eight](b ) ] . in common to a variety of finite - size scaling studies ( see e.g. ref.@xcite and references therein ) , ultimately small sized systems have been analyzed . often , as is the case here , the dimensionality of the operators involved ( transfer matrices , liouvillians , quantum hamiltonians ) grows exponentially with the system size thus severely limiting the manageable length scales , even for optimized algorithms . in an attempt to avoid those limitations we also considered the scaling of relaxation times in larger chains via monte carlo simulations however , due to the arrhenius barriers the difficulties introduced by small temperatures in such simulations are by far more restrictive than those associated to the system size ( recall that @xmath328 ) . in fact , starting from a disordered phase and quenching down to @xmath329 within the range of @xmath330 , it turned out that a significant fraction of the evolutions considered gets stuck in the typical metastable states of table [ tab2 ] , even at large times . finally , and with regard to a possible extension of this study , it would be interesting to derive the activation barriers @xmath132 quoted in that latter table directly from the evolution operator @xmath331 constructed in sec.iia . however irrespective of the sector considered , note that for @xmath14 the leading - order of its diagonal terms [ eq.([diagonal ] ) ] is different from that of their non - diagonal counterparts [ eqs.([s - pairing ] ) and([s - diffusion ] ) ] . thereby , the identification of an overall arrhenius factor in the low - temperature limit of @xmath331 is not evident in the kink representation . but in view of the universal amplitudes obtained above , one can further ask whether there might be a uniform spin rotation @xmath332 around a sector dependent axis such that @xmath333 , for some sector - wise but constant operator @xmath334 . that would not only single out activation barriers but would also allow computational access to the strict limit of @xmath14 via the low - lying eigenvalues of @xmath334 . further work along that line is under consideration . we thank e. v. albano , t. s. grigera , and f. a. schaposnik for helpful discussions . support from conicet and anpcyt , argentina , under grants no . pip 20120747 and no . pict 20121724 , is acknowledged . as schematized in table [ tab2 ] , these metastable ( m ) structures are characterized by the restrictions imposed on the number of consecutive kinks ( @xmath335 ) and vacancies ( @xmath336 ) scattered throughout the chain . in turn , for each coupling sector these constraints affect the rate at which these configurations proliferate with the system size . in order to evaluate such specific rates , in what follows we will construct a set of recursive relations for the number @xmath337 of those states on chains of generic length @xmath17 . to ease the analysis , open boundary conditions ( obc ) will be assumed throughout . below , we address each case in turn . 0.25 cm _ ( a)_. since for this sector @xmath338 and @xmath339 , it is helpful to consider the relation between the quantities @xmath340 and @xmath341 defined as the number of m - configurations of length @xmath17 having respectively 1 or 0 as kink occupations on their first site . clearly , under obc these latter quantities must then be recursively related as @xmath342 therefore , either of these quantities as well as the total number @xmath343 of m - states follow a fibonacci recursion @xmath344 , from which an exponential growth @xmath345 with golden mean @xmath346 is obtained for large sizes ( this growth also coincides with that of the ground state degeneracy at @xmath347 @xcite ) . 0.25 cm _ ( b)_. in addition to the kink restrictions of the previous case , here there is also a ban on sequence parts of the form @xmath162 as indicated in table [ tab2 ] . to take into account that further constraint it is now convenient to introduce the number @xmath348 of m - sequences of length @xmath17 having @xmath349 and @xmath350 as their first and second characters respectively ( @xmath351 0 or 1 ) . under obc it is then a simple matter to check that these quantities must be related as while clearly @xmath353 . in eq.([b - b ] ) , @xmath354 cancels out just all extra sequences from @xmath355 which would not form part of @xmath356 . thereby , it can be readily verified that all @xmath357 s , along with the total number of m - states , i.e. @xmath358 , will then follow the recurrence @xmath359 the general solution of this latter @xcite is associated to the roots of the polynomial @xmath360 , thus for long chains , where the largest root dominates , the m - configurations of this sector finally turn out to grow as @xmath166 . 0.25 cm _ ( c)_. further to @xmath361 , in this coupling sector every kink must appear separated by at least two vacancies , i.e. @xmath362 , so now there are even more reductions in the number of m - states . on considering for instance the @xmath341 and @xmath340 quantities referred to in case _ ( a ) _ , it is clear that under obc here these should verify @xmath363 from where the total number of m - configurations is obtained recursively as @xmath364 thus , for @xmath365 the largest root of the associated polynomial @xmath366 implies that @xmath367 . 0.25 cm _ ( d)_. in this case not only @xmath368 and @xmath338 , but also there may be consecutive kinks now appearing in groups of @xmath369 . to evaluate the proliferation of the corresponding m - states it is convenient to reintroduce here the @xmath348 quantities referred to in case _ ( b)_. for these latter , we readily obtain the recursive relations @xmath370 ( obc throughout ) , evidently now with @xmath371 as there can be no isolated vacancies . thus , after a small amount of algebra it turns out that the total number of m - configurations as well as all @xmath372 satisfy the recursive form @xmath373 from where the golden mean is recovered in the largest root of the associated polynomial @xmath374 . hence , analogously to sector _ ( a ) _ , in the thermodynamic limit @xmath375 proliferates as @xmath164 . 0.25 cm _ ( e)_. as it was referred to in table [ tab2 ] for this coupling regime @xmath376 and @xmath368 . thus , resorting back to the @xmath377 quantities considered above we readily find that in this sector these must be related as @xmath378 whereas @xmath379 , as neither vacancies nor kinks may appear isolated in this case ( obc assumed ) . thereby , it can be checked that the total number of m - states is given recursively by @xmath380 from the largest root of the polynomial @xmath381 linked to this recurrence , it then follows that for large sizes @xmath337 finally grows as @xmath194 . 0.25 cm _ ( f)_. in this sector kinks and vacancy constraints are respectively specified by @xmath382 , and @xmath383 . therefore , in terms of the @xmath357-quantities introduced above this means that their recursion relations should now read @xmath384 while clearly @xmath385 . hence , after simple substitutions it is found that each of these @xmath357 s , and correspondingly the total number of m - states , all follow the recursive form @xmath386 which for large sizes is taken over by the largest root of the polynomial @xmath387 . thereby , it turns out that for this coupling regime @xmath337 grows only as fast as @xmath388 . 0.25 cm _ ( g)_. following table [ tab2 ] , in this coupling regime @xmath382 ( as before ) , but now @xmath389 . in addition , there is also the constraint impeding the appearance of sequence parts of the form @xmath207 hence , assuming as usual obc , the four @xmath357-quantities of this case must be linked recursively as due to the above restriction , and on par with case _ ( b ) _ , here @xmath391 appears subtracting unwanted sequences which otherwise would overestimate @xmath392 in eq.([g - a ] ) . it is then a straightforward matter to verify that all of the above @xmath357 s ( and therefore also @xmath337 ) , ought to comply with the relation @xmath393 so , the characteristic polynomial associated to this latter recurrence is @xmath394 , from where it follows that at large sizes @xmath337 should proliferate as @xmath209 . 0.25 cm _ ( h)_. finally , in this sector kinks and vacancy restrictions remain as in the previous case except that the ban on the sequences referred to above is now lifted . thus , recursions ( [ g ] ) still hold provided eq.([g - a ] ) is modified as @xmath395 i.e. the cancellation of sequences contained in @xmath396 is no longer required here . after that modification it can be readily checked that the recursions arising in this coupling regime are all of the form @xmath397 from the largest root of @xmath398 , we thus find that in the limit of large @xmath17 here @xmath337 grows as @xmath208 . for reviews , consult _ kinetics of phase transitions _ , edited by s. puri and v. wadhawan ( crc press , boca raton , fl , 2009 ) ; a. j. bray , adv . 43 * , 357 ( 1994 ) ; j. d. gunton , m. san miguel , and p. s. sahni , in _ phase transitions and critical phenomena _ , edited by c. domb and j. l. lebowitz ( academic press , london , 1983 ) , vol . 8 . l. bogani , c. sangregorio , r. sessoli , and d. gatteschi , angew . * 44 * , 5817 ( 2005 ) ; k. bernot _ et al . _ , j. am . * 128 * , 7947 ( 2006 ) ; a. caneschi _ et al . _ , europhys . lett . * 58 * , 771 ( 2002 ) . see e.g. k. tanaka , t. morita , and k. hiroike , prog . . phys . * 77 * , 68 ( 1987 ) ; r. m. hornreich , r. liebmann , h. g. schuster , and w. selke , z. phys . b * 35 * , 91 ( 1979 ) ; j. stephenson , phys . b * 1 * , 4405 ( 1970 ) ; j. marro and r. dickman , _ nonequilibirum phase transitions in lattice models _ , ( cambridge university press , 1999 ) , section 8.4 . see for example , g. h. golub and c. f. van loan , _ matrix computations _ , 3rd . ed . ( johns hopkins university press , baltimore , 1996 ) ; g. meurant , _ the lanczos and conjugate gradient algorithms _ , ( siam , philadelphia , 2006 ) . however , for exact results at large times of the zero - temperature dynamics , see g. de smedt , c. godrche , and j. m. luck , eur . j. * 32 * , 215 ( 2003 ) ; v. privman , phys . lett . * 69 * , 3686 ( 1992 ) . for @xmath399 these times are associated with the smallest activation barrier of this sector . otherwise , the lead is taken by the first pairing process of table [ tab1 ] , but its effects are inconsequential on times @xmath195 .

it is widely known that many classical results in model theory fail or become meaningless when considered at the finite level . an important class of such results are preservation theorems , which relate the form of syntactic expressions to the closure properties of classes they define . classical preservation theorems are almost uniformly false when restricted to the class of all finite relational structures , but also tend to fail when restricted to the class all of finite algebras . a key example is the os - tarski preservation theorem , which says that a first - order sentence is preserved under taking extensions if and only if it is equivalent to an existential sentence . see tait @xcite or gurevich and shelah @xcite for a counterexample in the finite setting for relational structures , and clark , davey , jackson and pitkethly ( * ? ? ? * example @xmath0 ) for a counterexample in the finite setting for algebras . for some preservation theorems only one side of the story is explained ; for example the @xmath1-preservation theorem , which asserts that a first - order formula is preserved under taking substructures and direct products if and only if it is equivalent to a universal - horn sentence , remains to be settled at the finite - level in the relational setting . in contrast , example 4.3 of @xcite also gives a counterexample for algebraic structures at the finite level . on the other hand , lyndon s positivity theorem , which says that a first - order sentence is preserved under taking surjective homomorphisms if and only if it is equivalent to a positive sentence , fails in the finite setting for relational structures ( see ajtai and gurevich @xcite ) , while its status remains unknown in the algebraic setting . a notable exception is rossman s celebrated finite homomorphism preservation theorem ( * ? ? ? * theorem 5.16 ) : a first - order sentence is preserved under homomorphisms on finite relational structures if and only if it is equivalent , amongst finite relational structures , to an existential positive sentence . the question whether the classical homomorphism preservation theorem would continue to hold on finite relational structures was a long - standing one in finite model theory . the analogous question for algebraic signatures remains unresolved . the proof for the relational case is highly non - trivial and , as rossman suggests , the task of adjusting for function symbols is conceivably more difficult . preservation theorems that hold for the class of all finite relational structures do not necessarily relativise to subclasses of this class , and moreover , while many classical preservation theorems fail amongst all finite relational structures , some preservation theorems are recovered in restricted classes of finite relational structures . in @xcite , atserias , dawar , and grohe show that the os - tarski preservation theorem is recovered in a number of classes of finite relational structures that are of interest from a computational point of view , including classes of finite acyclic graphs , classes of structures of bounded degree , and classes of structures of treewidth at most @xmath2 , for any positive integer @xmath2 . in @xcite , atserias , dawar and kolaitis show that the homomorphism preservation property holds for a number of classes of finite relational structures , including classes of structures of bounded degree , classes of bounded treewidth , and more generally , classes of finite structures whose cores exclude at least one minor . however these results are not implied by rossman s finite homomorphism preservation theorem . furthermore , preservation theorems need not behave consistently when we restrict our attention to a particular subclass of the class of all finite relational structures : for example , the homomorphism preservation property holds for the class of all planar graphs , but in contrast , the extension preservation property fails in this setting . relativising preservation theorems to restricted classes of finite structures appears to be completely independent of whether or not the corresponding preservation property holds amongst all finite relational structures . on the other hand , relativisation of preservation theorems to restricted classes of finite algebras is not as well studied . in this paper , we are able to give some insight into this topic . 1 . [ item:1 ] we show that within the class of finite bounded lattices , there are many classes that demonstrate the failure of the homomorphism preservation property . more specifically , we show that for a fixed finite bounded lattice @xmath3 , the class of finite bounded lattices admitting a homomorphism into @xmath3 is definable by a single first - order sentence , relative to the class of finite bounded lattices , but is not definable by universal sentences . see theorem [ thm : algfail ] in section [ section : sectionalg ] . [ item:2 ] in contrast to ( [ item:1 ] ) , we establish a relativised homomorphism preservation theorem ( see theorem [ thm : sausage ] in section [ section : sectrelational ] ) for a wide range of classes of finite relational structures . theorem [ thm : sausage ] gives us a complete characterisation of first - order definable classes of relational structures admitting a homomorphism into a fixed finite relational structure , relative to classes @xmath4 definable by sentences of some general forms . in theorem [ thm : colourfam ] we extend this characterisation to classes of relational structures admitting a homomorphism into one of a finite number of relational structures . we will see that theorem [ thm : algfail ] mentioned in ( [ item:1 ] ) not only demonstrates the failure of the homomorphism preservation property for the class of finite bounded lattices , but also the failure of the extension preservation property ( os - tarski preservation theorem ) . so the extension preservation property , which fails for the class of all finite algebras and also the class of all finite relational structures , continues to fail when we restrict to the class of finite bounded lattices . this result could be contrasted with the positive results of atserias , dawar and grohe @xcite mentioned previously , which show that the extension preservation property is recovered in a number of restricted classes of finite relational structures . when the class @xmath4 in theorem [ thm : sausage ] is relaxed to include all finite relational structures , our result coincides with a well known one due to atserias , which says that first - order definable constraint satisfaction problems are precisely those with finite tree duality ( or equivalently , those definable by existential positive sentences ) , see theorem @xmath5 of @xcite . it is also a consequence rossman s finite homomorphism preservation theorem , which holds for classes of relational structures admitting a homomorphism into one of _ any _ number of finite relational structures . our proof , however , is independent of both atserias and rossman and relies mostly on some constructions borrowed from larose , loten and tardif given in @xcite . we remark that while it remains unclear if the homomorphism preservation property in its full generality holds in these classes , the restricted preservation property , nevertheless , covers many commonly encountered homomorphism - closed classes . this article is comprised of three main sections . we begin with some background and definitions that will be required for the remainder of the paper , we then investigate the status of relativised homomorphism preservation theorem [ thm : relhompresthm ] at the finite level for both algebraic and relational signatures . in section [ section : sectionalg ] , we give counterexamples to the finite - level version of relativised homomorphism preservation theorem [ thm : relhompresthm ] for algebraic structures . moreover , we completely characterise those classes of finite lattices admitting a homomorphism into a fixed finite bounded lattice . in section [ section : sectrelational ] , we establish a restricted form of the finite - level version of relativised homomorphism theorem [ thm : relhompresthm ] for relational structures . the proof follows an ehrenfeucht - frass game argument on structures obtained by adjusting some constructions borrowed from larose , loten and tardif given in @xcite . a _ signature _ ( or _ language _ or _ type _ ) is a pair @xmath6 consisting of a set of relation symbols @xmath7 and a set of function symbols @xmath8 . each relation and function symbol has an associated _ arity _ @xmath9 ; we do not allow for nullary relation symbols so @xmath9 is quantified over @xmath10 in this case ; we do however include the possibility of @xmath11-ary function symbols so @xmath9 is quantified over @xmath12 in this case . an @xmath13-_structure _ @xmath14 consists of a universe @xmath15 together with an interpretation of * each @xmath9-ary relation symbol @xmath16 as a subset @xmath17 of @xmath18 , * each @xmath9-ary function symbol @xmath19 as a function @xmath20 . if @xmath21 , then @xmath22 is an _ algebra _ , if @xmath23 then @xmath22 is a _ relational structure_. let @xmath24 be a structure in some signature @xmath13 consisting of possibly both relation and function symbols . we let @xmath25 be the language obtained by adding a nullary function symbol @xmath26 to @xmath13 , for each @xmath27 . let @xmath28 , let @xmath29 be a purely relational signature , and let @xmath30 . weak substructure _ of @xmath31 is a structure @xmath32 , where @xmath33 and @xmath34 for each @xmath35 . the weak substructure @xmath36 is _ proper _ if either @xmath37 or @xmath38 and there exists @xmath39 such that @xmath40 . let @xmath13 be some signature consisting of possibly both relation and function symbols , and let @xmath41 be a class of @xmath13-structures . we define @xmath42 to be the class of all @xmath13-structures admitting a homomorphism into some member of the class @xmath41 . let @xmath43 be a set and let @xmath44 be a set of structures in some signature @xmath13 . we say that @xmath45 is a _ homomorphism - independent _ set if for all @xmath46 with @xmath47 , there is no homomorphism from @xmath48 to @xmath49 . let @xmath50 and let @xmath51 be a finite set of finite structures in the purely relational signature @xmath52 . we denote the finite models in @xmath53 by @xmath54 . obstruction _ for the class @xmath55 is any finite @xmath7-structure not belonging to @xmath54 . an obstruction @xmath56 for the class @xmath55 is _ critical _ if every proper weak substructure of @xmath56 belongs to @xmath54 . we say that @xmath54 satisfies the _ finite duality property _ if there exists a finite set @xmath57 of finite structures with the following property : a finite @xmath7-structure @xmath58 belongs to the class @xmath54 if and only if @xmath58 avoids a homomorphism from every member of @xmath57 . let @xmath24 be a finite structure in the purely relational signature @xmath7 . the class of all finite @xmath7-structures admitting a homomorphism into @xmath24 is denoted @xmath59 . the membership problem for @xmath59 is known as the _ constraint satisfaction problem _ over template @xmath24 . note that @xmath60 . let @xmath61 , let @xmath24 be a finite relational structure in the purely relational signature @xmath52 with @xmath62 of arity @xmath63 , for all @xmath64 . the _ incidence multigraph _ of @xmath24 , denoted @xmath65 , is defined as the bipartite multigraph with parts @xmath15 and @xmath66 , where @xmath66 consists of all pairs @xmath67 such that @xmath68 and @xmath69 , and with edges @xmath70 joining @xmath27 to @xmath71 , when @xmath72 for some @xmath73 . let @xmath26 and @xmath74 be two elements in @xmath15 . then the _ distance between @xmath26 and @xmath74 in _ @xmath75 is defined to be half the number of edges in a shortest path connecting them in @xmath65 . the _ girth _ of @xmath24 is defined to be half the shortest length of a cycle in @xmath65 . the _ diameter _ of @xmath24 is defined to be @xmath76 . let @xmath4 be a class of @xmath13-structures and let @xmath41 be a subclass of @xmath4 . we define the following class : @xmath77 in the case where @xmath4 is the class of all @xmath13-structures we abbreviate @xmath78 to @xmath79 . in general , @xmath80 . let @xmath81 , let @xmath82 , and let @xmath83 be a finite set of structures in @xmath4 . we follow gorbunov @xcite and refer to the class @xmath84 as a _ principal colour - family in @xmath4 _ , and we refer to the class @xmath85 as a _ colour - family in _ @xmath4 . let @xmath86 and let @xmath61 . a sentence @xmath87 is called an _ anti - identity _ if it is of the form @xmath88,\ ] ] where @xmath89 is an atomic formula for all @xmath90 . let @xmath91 be a set of first - order sentences in some signature @xmath92 . we let @xmath93 denote the class of all models of @xmath91 . . a subclass @xmath94 of a class @xmath95 is called _ strictly elementary relative to _ @xmath4 if there is a first - order sentence @xmath96 such that @xmath97 . a subclass @xmath94 of a class @xmath95 is called an _ anti - variety relative to _ @xmath4 if @xmath98 where @xmath99 is a set of anti - identities . the proof of the relativised homomorphism preservation theorem [ thm : relhompresthm ] requires only standard applications of some model - theoretic fundamentals . we have included its proof , however , for completeness sake . we remark that the results of this paper are more easily formulated for classes that are complements of homomorphism closed classes . such classes are closed under taking homomorphic pre - images : classes @xmath41 satisfying @xmath100 whenever @xmath58 admits a homomorphism to @xmath24 and @xmath101 . for this reason , we work with the homomorphism preservation theorem in its complemented form . [ thm : relhompresthm ] let @xmath91 be a set of sentences in the signature @xmath13 , let @xmath102 , and let @xmath41 be a subclass of @xmath4 . then the class @xmath103 is an anti - variety relative to @xmath4 . moreover , if @xmath103 is strictly elementary with respect to @xmath4 , then there is a sentence consisting of a finite conjunction anti - identities that is logically equivalent to @xmath104 modulo @xmath91 . we first show that @xmath103 is axiomatisable by a set of anti - identities with respect to the class @xmath4 . the second part of the theorem will follow by compactness . let @xmath105 and let @xmath106 be the set of anti - identities true in @xmath41 . clearly , if @xmath107 , then @xmath108 , since anti - identities are preserved under taking homomorphic pre - images . now let @xmath82 and suppose that @xmath108 . let @xmath109 be the set of atomic sentences true in @xmath24 in the expanded signature @xmath25 , and let @xmath110 be the class of all @xmath25-structures whose @xmath13-reduct belongs to @xmath111 . for any finite subset @xmath112 of @xmath109 , we have @xmath113 so , @xmath24 fails the sentence @xmath114.\ ] ] hence there is some structure @xmath115 that fails @xmath116 . let @xmath117 be the elements of @xmath118 that witness the failure of the sentence @xmath116 . now expand @xmath58 in the signature @xmath25 under the following interpretation : let @xmath74 be any element of @xmath118 , and define @xmath119 , if @xmath120 , and @xmath121 , otherwise . then @xmath122 , and satisfies @xmath123 since @xmath4 is elementary and @xmath111 is first order definable relative to @xmath4 it follows , by compactness , that there is some @xmath124 such that @xmath125 . now let @xmath126 be the map that sends each element @xmath127 of @xmath24 to its corresponding interpretation in @xmath128 . then , for every atomic formula @xmath129 in @xmath13 and every @xmath9-tuple @xmath130 , we have @xmath131 thus , the map @xmath126 is a homomorphism from @xmath24 to @xmath132 . since @xmath133 , there exists a homomorphism @xmath134 from @xmath132 to some member @xmath135 of @xmath136 , and so @xmath137 is a homomorphism from @xmath24 to @xmath135 , thus @xmath138 . hence we have shown that there is a set @xmath139 of anti - identities such that @xmath140 , namely @xmath141 . now assume there is a single first - order sentence @xmath96 such that @xmath142 . then every model of @xmath143 is also a model of @xmath144 . by the completeness theorem of first - order logic , we get that @xmath145 . it then follows , by compactness , that there is a finite ( possibly empty ) set of anti - identities @xmath146 such @xmath147 . thus , the sentence @xmath96 is logically equivalent to the conjunction of anti - identities @xmath148 modulo @xmath91 . we now show that the finite - level version of the relativised homomorphism preservation theorem [ thm : relhompresthm ] fails for algebraic signatures . if we restrict the class @xmath4 to all finite bounded lattices , then we can find many classes closed under taking homomorphic pre - images that are first - order definable relative to @xmath4 , but not definable by universal sentences . [ thm : algfail ] let @xmath4 be the class of finite bounded lattices in the signature @xmath149 and let @xmath3 be any lattice in @xmath4 with @xmath150 . the class @xmath151 of all finite bounded lattices admitting a homomorphism into @xmath3 is strictly first - order definable relative to @xmath4 but not definable by any universal sentence . the class @xmath151 of all finite bounded lattices admitting a homomorphism into @xmath3 is definable by a @xmath152 sentence @xmath96 by ( * ? ? ? * theorem 4.2 ) . we will show that the class @xmath151 is not definable by any universal sentence . specifically , we show that , for each @xmath153 , there exists a finite bounded lattice @xmath154 satisfying the following two conditions . 1 . @xmath155 ; 2 . every @xmath9-generated sublattice of @xmath154 belongs to @xmath151 . for each @xmath153 , let @xmath154 be the lattice depicted in figure [ figure : a_n ] . the lattice @xmath154 is a `` stack '' of @xmath9 copies of @xmath156 , where @xmath2 is a natural number strictly greater than @xmath157 . ( @xmath158 ) : let @xmath153 . the smallest lattice that @xmath154 collapses to , without identifying @xmath11 and @xmath158 , is isomorphic to the lattice @xmath156 . since @xmath156 is simple ( as @xmath159 ) , every congruence on @xmath156 collapses everything or collapses nothing . thus if @xmath154 is to map homomorphically into @xmath3 , then @xmath3 must contain a sublattice isomorphic to @xmath156 , but this is not possible since @xmath156 contains more elements than @xmath3 . hence @xmath160 . ( @xmath161 ) : we will show that every @xmath9-generated sublattice of @xmath154 has a homomorphism onto the two - element lattice @xmath162 , and therefore has a homomorphism into @xmath3 . assume that we are selecting @xmath9 generators from the lattice @xmath163 . in order to generate a sublattice of @xmath154 that does not admit a homomorphism onto @xmath162 ( or equivalently , a sublattice that does not contain a prime ideal ) , we would need to select _ at least _ three elements from each of the @xmath9 maximal anti - chains in @xmath154 . thus , the lattice @xmath164 given in figure [ figure : p ] is the smallest sublattice of @xmath154 that does not admit a homomorphism onto @xmath162 . but this lattice is @xmath165-generated and not generated by fewer than @xmath165 elements . hence every @xmath9-generated a sublattice of @xmath154 admits a homomorphism onto @xmath162 , and therefore into @xmath3 . notice that within the class of finite bounded lattices , the ( first - order definable ) property of admitting a homomorphism onto @xmath166 is preserved by taking sublattices , and so the above examples also give us a counterexample to the finite - level relativised os - tarski preservation theorem for algebraic signatures . also notice that the proof of theorem [ thm : algfail ] works for the class @xmath167 , where @xmath168 is any class of finite bounded lattices containing at least one non - trivial lattice , and for which there exists @xmath159 such that @xmath169 . the proof of theorem [ thm : algfail ] also works with @xmath4 replaced by any class of finite lattices containing the lattices in figure [ figure : a_n ] ( for each @xmath170 ) . as the lattice @xmath154 is modular , the class @xmath4 in theorem [ thm : algfail ] could be taken to be the class of all finite bounded modular lattices . alternatively , @xmath4 could to be the finite part of the variety generated by the bounded lattices @xmath154 . this variety can be shown to coincide with the variety generated by @xmath171 . [ unshaded/.style = draw , shape = circle , fill = white , inner sep=1.5pt , order/.style = very thin ] label at ( 0,0 ) ; ( bot ) at ( 0,0 ) ; ( 2 ) at ( -2,1 ) ; ( 8) at ( -4,1 ) ; at ( 2.8,1 ) @xmath172 ; at ( -2.8,1 ) @xmath172 ; ( 3 ) at ( 2,1 ) ; ( 9 ) at ( 4,1 ) ; ( 5 ) at ( -1,1 ) ; ( 6 ) at ( 0,1 ) ; ( 7 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( half1 ) at ( -1.9,1.35 ) ; ( half2 ) at ( 1.9,1.35 ) ; ( half3 ) at ( -1.9,.65 ) ; ( half4 ) at ( 1.9,.65 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( bot ) to ( 5 ) ; ( bot ) to ( 6 ) ; ( bot ) to ( 7 ) ; ( bot ) to ( 8) ; ( bot ) to ( 9 ) ; ( bot ) to ( half3 ) ; ( bot ) to ( half4 ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; ( 5 ) to ( top ) ; ( 6 ) to ( top ) ; ( 7 ) to ( top ) ; ( 8) to ( top ) ; ( 9 ) to ( top ) ; ( half1 ) to ( top ) ; ( half2 ) to ( top ) ; ( bot ) at ( 0,0 ) ; ( 2 ) at ( -2,1 ) ; ( 8) at ( -4,1 ) ; at ( 2.8,1 ) @xmath172 ; at ( -2.8,1 ) @xmath172 ; ( 3 ) at ( 2,1 ) ; ( 9 ) at ( 4,1 ) ; ( 5 ) at ( -1,1 ) ; ( 6 ) at ( 0,1 ) ; ( 7 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( half1 ) at ( -1.9,1.35 ) ; ( half2 ) at ( 1.9,1.35 ) ; ( half3 ) at ( -1.9,.65 ) ; ( half4 ) at ( 1.9,.65 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( bot ) to ( 5 ) ; ( bot ) to ( 6 ) ; ( bot ) to ( 7 ) ; ( bot ) to ( 8) ; ( bot ) to ( 9 ) ; ( bot ) to ( half3 ) ; ( bot ) to ( half4 ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; ( 5 ) to ( top ) ; ( 6 ) to ( top ) ; ( 7 ) to ( top ) ; ( 8) to ( top ) ; ( 9 ) to ( top ) ; ( half1 ) to ( top ) ; ( half2 ) to ( top ) ; at ( 0,5 ) @xmath173 ; ( bot ) at ( 0,0 ) ; ( 2 ) at ( -2,1 ) ; ( 8) at ( -4,1 ) ; at ( 2.8,1 ) @xmath172 ; at ( -2.8,1 ) @xmath172 ; ( 3 ) at ( 2,1 ) ; ( 9 ) at ( 4,1 ) ; ( 5 ) at ( -1,1 ) ; ( 6 ) at ( 0,1 ) ; ( 7 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( half1 ) at ( -1.9,1.35 ) ; ( half2 ) at ( 1.9,1.35 ) ; ( half3 ) at ( -1.9,.65 ) ; ( half4 ) at ( 1.9,.65 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( bot ) to ( 5 ) ; ( bot ) to ( 6 ) ; ( bot ) to ( 7 ) ; ( bot ) to ( 8) ; ( bot ) to ( 9 ) ; ( bot ) to ( half3 ) ; ( bot ) to ( half4 ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; ( 5 ) to ( top ) ; ( 6 ) to ( top ) ; ( 7 ) to ( top ) ; ( 8) to ( top ) ; ( 9 ) to ( top ) ; ( half1 ) to ( top ) ; ( half2 ) to ( top ) ; ( 0,7.8)(0,0 ) node[midway , right , xshift=10pt ] * n copies * ; [ unshaded/.style = draw , shape = circle , fill = white , inner sep=1.5pt , order/.style = very thin ] ( bot ) at ( 0,0 ) ; ( 1 ) at ( -1,1 ) ; ( 2 ) at ( 0,1 ) ; ( 3 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( bot ) to ( 1 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( 1 ) to ( top ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; ( bot ) at ( 0,0 ) ; ( 1 ) at ( -1,1 ) ; ( 2 ) at ( 0,1 ) ; ( 3 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( bot ) to ( 1 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( 1 ) to ( top ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; at ( 0,5 ) @xmath173 ; ( bot ) at ( 0,0 ) ; ( 1 ) at ( -1,1 ) ; ( 2 ) at ( 0,1 ) ; ( 3 ) at ( 1,1 ) ; ( top ) at ( 0,2 ) ; ( bot ) to ( 1 ) ; ( bot ) to ( 2 ) ; ( bot ) to ( 3 ) ; ( 1 ) to ( top ) ; ( 2 ) to ( top ) ; ( 3 ) to ( top ) ; ( 0,7.8)(0,0 ) node[midway , right , xshift=10pt ] * n copies * ; we now turn to the purely relational side of things . in this section , we establish a finite - level version of the relativised homomorphism preservation theorem for relational signatures . we present this result as theorem [ thm : sausage ] below . we first prove the non - relativised version of theorem [ thm : sausage ] , namely theorem [ thm : atserias ] , from which theorem [ thm : sausage ] will follow almost immediately . the proof requires three constructions given in definition [ defn : nlink ] , definition [ defn : unpinch ] and definition [ defn : npinch ] below , borrowed from @xcite . the quotient structure given in definition [ defn : npinch ] , will play a key role in establishing theorem [ thm : sausage ] . we follow jackson and trotta @xcite and refer to this quotient structure as the _ @xmath9-pinch _ over @xmath24 . in @xcite , larose , loten and tardif show that , for a relational structure @xmath24 , demonstrating membership of the @xmath9-pinch in @xmath59 , for some natural number @xmath9 , is equivalent to @xmath59 admitting the finite duality property . this fact will be crucial in the proof of theorem [ thm : sausage ] . we conclude the section by extending theorem [ thm : sausage ] . more specifically , we obtain a complete characterisation of not only first - order definable principal colour - families but also first - order definable colour - families in classes @xmath4 closed under forming the @xmath9-pinch construction and taking disjoint unions ; see theorem [ thm : colourfam ] . [ defn : unpinch ] let @xmath153 and let @xmath24 be a structure in the signature @xmath52 . consider the product structure @xmath179 . we define an equivalence relation @xmath180 on @xmath179 as follows : @xmath181 observe that the first coordinate of the elements in @xmath179 is constant on each of the @xmath183-equivalence classes . in other words , the first coordinate is unchanged by the formation of the quotient structure @xmath184 . we therefore use the first coordinate as a means for specifying the position of an element in a copy of @xmath24 or @xmath185 in the @xmath9-pinch : notationally , we define a function @xmath186 by @xmath187 . we refer to the copy of @xmath24 in @xmath184 at position @xmath188 as the _ left pinch _ of @xmath184 , and we refer to the copy of @xmath24 at position @xmath189 as the _ right pinch _ of @xmath184 . we denote the substructures of @xmath184 induced by the sets @xmath190 and @xmath191 as @xmath192 and @xmath193 , respectively . we refer to the copy of @xmath185 in @xmath192 at position @xmath194 as the _ right slice _ of @xmath192 , and we refer to the copy of @xmath185 in @xmath193 at position @xmath195 as the _ left slice _ of @xmath193 . 1 . ( @xcite , lemma @xmath199 ) for each natural number @xmath9 , the substructures @xmath192 and @xmath193 of @xmath184 admit homomorphisms into @xmath24.[itm : part1 ] 2 . ( @xcite theorem @xmath200 and theorem @xmath201)[itm : pinchequiv ] the following are equivalent : * @xmath59 satisfies the finite duality property ; * there exists some natural number @xmath9 such that @xmath184 admits a homomorphism to @xmath24 . [ itm : pinchsurjhom](@xcite lemma @xmath202 ) for each @xmath153 , there exists a surjective homomorphism from @xmath203 to @xmath204 . for ( @xmath158 ) , it is routine to check that the map @xmath205 defined by @xmath206 is a homomorphism from @xmath192 to @xmath24 and similarly , that the map @xmath207 defined by @xmath208 is a homomorphism from @xmath192 to @xmath24 , see the proof of lemma @xmath199 in @xcite . the proof of ( @xmath161 ) depends mostly on the observation that there are only finitely many critical obstructions of a given diameter ( see lemma @xmath209 of @xcite . for ( @xmath210 ) , the map defined by @xmath211 is easily shown to be a surjective homomorphism , see the proof of lemma @xmath202 in @xcite . let @xmath4 be a class of @xmath7-structures and let @xmath55 be a subclass of @xmath4 . we say that @xmath4 is _ closed under forming the @xmath9-pinch construction over @xmath55 _ , if for every @xmath28 and every @xmath212 , we have that @xmath213 and its substructures @xmath192 and @xmath193 belong to the class @xmath4 . [ thm : sausage ] let @xmath4 be a class of finite @xmath214-structures and let @xmath82 . if @xmath4 is closed under forming the @xmath9-pinch construction over @xmath24 and taking disjoint unions , then @xmath215 is equal to @xmath216 for some first order sentence @xmath96 if and only if there is a finite conjunction of anti - identities @xmath87 such that @xmath217 . we now prove ( @xmath210 ) @xmath218 ( @xmath158 ) by establishing the contrapositive . we show that for a finite relational structure @xmath24 , if the class @xmath59 fails the finite duality property , then the property of admitting a homomorphism to @xmath24 is not expressible by a first - order sentence . the proof follows an ehrenfeucht - frass game arguement ; we show that , for each non - negative integer @xmath2 , there are two structures @xmath219 and @xmath220 such that : if @xmath59 does not satisfy the finite duality property , then theorem [ thm : sausagefacts ] tells us that for each natural number @xmath9 , there exists no homomorphism from @xmath184 to @xmath24 , but the substructures @xmath192 and @xmath193 of @xmath184 admit homomorphisms to @xmath24 . this suggests that we construct our structures @xmath219 and @xmath220 using combinations of @xmath184 and its substructures @xmath192 and @xmath193 , where @xmath9 is some number defined in terms of @xmath2 . let @xmath2 be a non - negative integer and assume that @xmath221 is an integer strictly greater than @xmath222 . now consider @xmath223 and the disjoint union @xmath224 . we build @xmath219 and @xmath220 in the following way . let @xmath220 be the structure obtained by taking the disjoint union of @xmath225 copies of @xmath224 , and let @xmath219 be @xmath226 . see figure [ figure : sausies ] for an abstract depiction of the structures @xmath219 and @xmath220 . node[coordinate , label = below:@xmath15 ] ( v16 ) at ( -2,0.5 ) node[coordinate ] ( v17 ) at ( -3,3.5 ) ; ( v16)(v17 ) ; node[coordinate , label = below:@xmath228 ( v0 ) at ( 0,0 ) node[coordinate ] ( v1 ) at ( 1,3 ) node[coordinate ] ( v2 ) at ( 0,6 ) node[coordinate ] ( v3 ) at ( -1,3 ) ; ( v0)(v1)(v2)(v3)(v0 ) ; node[coordinate ] ( v4 ) at ( 1,0.5 ) node[coordinate ] ( v5 ) at ( 2,3.5 ) node[coordinate ] ( v6 ) at ( 1,6.5 ) node[coordinate ] ( v7 ) at ( 0,3.5 ) ; ( v4)(v5)(v6)(v7)(v4 ) ; node[coordinate ] ( v8 ) at ( 4,2 ) node[coordinate ] ( v9 ) at ( 5,5 ) node[coordinate ] ( v10 ) at ( 4,8 ) node[coordinate ] ( v11 ) at ( 3,5 ) ; ( v8)(v9)(v10)(v11)(v8 ) ; node[coordinate , label = below:@xmath228 ( v12 ) at ( 5,2.5 ) node[coordinate ] ( v13 ) at ( 6,5.5 ) node[coordinate ] ( v14 ) at ( 5,8.5 ) node[coordinate ] ( v15 ) at ( 4,5.5 ) ; ( v12)(v13)(v14)(v15)(v12 ) ; node[coordinate , label = below:@xmath15 ] ( v18 ) at ( 7,4.5 ) node[coordinate ] ( v19 ) at ( 8,7.5 ) ; ( v18)(v19 ) ; ( v17)(v2)(v6)(v10)(v14)(v19 ) ; ( v17)(v3)(v7)(v11)(v15)(v18 ) ; ( v16)(v0)(v4)(v8)(v12)(v18 ) ; ( v16)(v1)(v5)(v9)(v13)(v19 ) ; node[coordinate ] ( v16 ) at ( -2,0.5 ) node[coordinate ] ( v17 ) at ( -3,3.5 ) ; ( v16)(v17 ) ; node[coordinate ] ( v0 ) at ( 0,0 ) node[coordinate ] ( v1 ) at ( 1,3 ) node[coordinate ] ( v2 ) at ( 0,6 ) node[coordinate ] ( v3 ) at ( -1,3 ) ; ( v0)(v1)(v2)(v3)(v0 ) ; node[coordinate ] ( v4 ) at ( 0.6,0.3 ) node[coordinate ] ( v5 ) at ( 1.6,3.3 ) node[coordinate ] ( v6 ) at ( 0.6,6.3 ) node[coordinate ] ( v7 ) at ( -0.4,3.3 ) ; ( v4)(v5)(v6)(v7)(v4 ) ; node[coordinate ] ( v8 ) at ( 4.4,2.2 ) node[coordinate ] ( v9 ) at ( 5.4,5.2 ) node[coordinate ] ( v10 ) at ( 4.4,8.2 ) node[coordinate ] ( v11 ) at ( 3.4,5.2 ) ; ( v8)(v9)(v10)(v11)(v8 ) ; node[coordinate ] ( v12 ) at ( 5,2.5 ) node[coordinate ] ( v13 ) at ( 6,5.5 ) node[coordinate ] ( v14 ) at ( 5,8.5 ) node[coordinate ] ( v15 ) at ( 4,5.5 ) ; ( v12)(v13)(v14)(v15)(v12 ) ; node[coordinate ] ( v20 ) at ( 3.4,1.7 ) node[coordinate ] ( v21 ) at ( 4.4,4.7 ) node[coordinate ] ( v22 ) at ( 3.4,7.7 ) node[coordinate ] ( v23 ) at ( 2.4,4.7 ) ; ( v20)(v21)(v22)(v23)(v20 ) ; node[coordinate ] ( v24 ) at ( 1.6,0.8 ) node[coordinate ] ( v25 ) at ( 2.6,3.8 ) node[coordinate ] ( v26 ) at ( 1.6,6.8 ) node[coordinate ] ( v27 ) at ( 0.6,3.8 ) ; ( v24)(v25)(v26)(v27)(v24 ) ; node[coordinate ] ( v18 ) at ( 7,4.5 ) node[coordinate ] ( v19 ) at ( 8,7.5 ) ; ( v18)(v19 ) ; ( v17)(v2)(v6)(v26 ) ; ( v22)(v10)(v14)(v19 ) ; ( v17)(v3)(v7)(v27 ) ; ( v23)(v11)(v15)(v18 ) ; ( v16)(v0)(v4)(v24 ) ; ( v20)(v8)(v12)(v18 ) ; ( v16)(v1)(v5)(v25 ) ; ( v21)(v9)(v13)(v19 ) ; node[coordinate ] ( v16 ) at ( -2,0.5 ) node[coordinate ] ( v17 ) at ( -3,3.5 ) ; ( v16)(v17 ) ; node[coordinate ] ( v0 ) at ( 0,0 ) node[coordinate ] ( v1 ) at ( 1,3 ) node[coordinate ] ( v2 ) at ( 0,6 ) node[coordinate ] ( v3 ) at ( -1,3 ) ; ( v0)(v1)(v2)(v3)(v0 ) ; node[coordinate ] ( v4 ) at ( 0.6,0.3 ) node[coordinate ] ( v5 ) at ( 1.6,3.3 ) node[coordinate ] ( v6 ) at ( 0.6,6.3 ) node[coordinate ] ( v7 ) at ( -0.4,3.3 ) ; ( v4)(v5)(v6)(v7)(v4 ) ; node[coordinate ] ( v8 ) at ( 4.4,2.2 ) node[coordinate ] ( v9 ) at ( 5.4,5.2 ) node[coordinate ] ( v10 ) at ( 4.4,8.2 ) node[coordinate ] ( v11 ) at ( 3.4,5.2 ) ; ( v8)(v9)(v10)(v11)(v8 ) ; node[coordinate ] ( v12 ) at ( 5,2.5 ) node[coordinate ] ( v13 ) at ( 6,5.5 ) node[coordinate ] ( v14 ) at ( 5,8.5 ) node[coordinate ] ( v15 ) at ( 4,5.5 ) ; ( v12)(v13)(v14)(v15)(v12 ) ; node[coordinate ] ( v20 ) at ( 3.4,1.7 ) node[coordinate ] ( v21 ) at ( 4.4,4.7 ) node[coordinate ] ( v22 ) at ( 3.4,7.7 ) node[coordinate ] ( v23 ) at ( 2.4,4.7 ) ; ( v20)(v21)(v22)(v23)(v20 ) ; node[coordinate ] ( v24 ) at ( 1.6,0.8 ) node[coordinate ] ( v25 ) at ( 2.6,3.8 ) node[coordinate ] ( v26 ) at ( 1.6,6.8 ) node[coordinate ] ( v27 ) at ( 0.6,3.8 ) ; ( v24)(v25)(v26)(v27)(v24 ) ; node[coordinate ] ( v18 ) at ( 7,4.5 ) node[coordinate ] ( v19 ) at ( 8,7.5 ) ; ( v18)(v19 ) ; ( v17)(v2)(v6)(v26 ) ; ( v22)(v10)(v14)(v19 ) ; ( v17)(v3)(v7)(v27 ) ; ( v23)(v11)(v15)(v18 ) ; ( v16)(v0)(v4)(v24 ) ; ( v20)(v8)(v12)(v18 ) ; ( v16)(v1)(v5)(v25 ) ; ( v21)(v9)(v13)(v19 ) ; node[coordinate ] ( v16 ) at ( -2,0.5 ) node[coordinate ] ( v17 ) at ( -3,3.5 ) ; ( v16)(v17 ) ; node[coordinate ] ( v0 ) at ( 0,0 ) node[coordinate ] ( v1 ) at ( 1,3 ) node[coordinate ] ( v2 ) at ( 0,6 ) node[coordinate ] ( v3 ) at ( -1,3 ) ; ( v0)(v1)(v2)(v3)(v0 ) ; node[coordinate ] ( v4 ) at ( 0.6,0.3 ) node[coordinate ] ( v5 ) at ( 1.6,3.3 ) node[coordinate ] ( v6 ) at ( 0.6,6.3 ) node[coordinate ] ( v7 ) at ( -0.4,3.3 ) ; ( v4)(v5)(v6)(v7)(v4 ) ; node[coordinate ] ( v8 ) at ( 4.4,2.2 ) node[coordinate ] ( v9 ) at ( 5.4,5.2 ) node[coordinate ] ( v10 ) at ( 4.4,8.2 ) node[coordinate ] ( v11 ) at ( 3.4,5.2 ) ; ( v8)(v9)(v10)(v11)(v8 ) ; node[coordinate ] ( v12 ) at ( 5,2.5 ) node[coordinate ] ( v13 ) at ( 6,5.5 ) node[coordinate ] ( v14 ) at ( 5,8.5 ) node[coordinate ] ( v15 ) at ( 4,5.5 ) ; ( v12)(v13)(v14)(v15)(v12 ) ; node[coordinate ] ( v20 ) at ( 3.4,1.7 ) node[coordinate ] ( v21 ) at ( 4.4,4.7 ) node[coordinate ] ( v22 ) at ( 3.4,7.7 ) node[coordinate ] ( v23 ) at ( 2.4,4.7 ) ; ( v20)(v21)(v22)(v23)(v20 ) ; node[coordinate ] ( v24 ) at ( 1.6,0.8 ) node[coordinate ] ( v25 ) at ( 2.6,3.8 ) node[coordinate ] ( v26 ) at ( 1.6,6.8 ) node[coordinate ] ( v27 ) at ( 0.6,3.8 ) ; ( v24)(v25)(v26)(v27)(v24 ) ; node[coordinate ] ( v18 ) at ( 7,4.5 ) node[coordinate ] ( v19 ) at ( 8,7.5 ) ; ( v18)(v19 ) ; ( v17)(v2)(v6)(v26 ) ; ( v22)(v10)(v14)(v19 ) ; ( v17)(v3)(v7)(v27 ) ; ( v23)(v11)(v15)(v18 ) ; ( v16)(v0)(v4)(v24 ) ; ( v20)(v8)(v12)(v18 ) ; ( v16)(v1)(v5)(v25 ) ; ( v21)(v9)(v13)(v19 ) ; node[coordinate ] ( v16 ) at ( -2,0.5 ) node[coordinate ] ( v17 ) at ( -3,3.5 ) ; ( v16)(v17 ) ; node[coordinate ] ( v0 ) at ( 0,0 ) node[coordinate ] ( v1 ) at ( 1,3 ) node[coordinate ] ( v2 ) at ( 0,6 ) node[coordinate ] ( v3 ) at ( -1,3 ) ; ( v0)(v1)(v2)(v3)(v0 ) ; node[coordinate ] ( v4 ) at ( 0.6,0.3 ) node[coordinate ] ( v5 ) at ( 1.6,3.3 ) node[coordinate ] ( v6 ) at ( 0.6,6.3 ) node[coordinate ] ( v7 ) at ( -0.4,3.3 ) ; ( v4)(v5)(v6)(v7)(v4 ) ; node[coordinate ] ( v8 ) at ( 4.4,2.2 ) node[coordinate ] ( v9 ) at ( 5.4,5.2 ) node[coordinate ] ( v10 ) at ( 4.4,8.2 ) node[coordinate ] ( v11 ) at ( 3.4,5.2 ) ; ( v8)(v9)(v10)(v11)(v8 ) ; node[coordinate ] ( v12 ) at ( 5,2.5 ) node[coordinate ] ( v13 ) at ( 6,5.5 ) node[coordinate ] ( v14 ) at ( 5,8.5 ) node[coordinate ] ( v15 ) at ( 4,5.5 ) ; ( v12)(v13)(v14)(v15)(v12 ) ; node[coordinate ] ( v20 ) at ( 3.4,1.7 ) node[coordinate ] ( v21 ) at ( 4.4,4.7 ) node[coordinate ] ( v22 ) at ( 3.4,7.7 ) node[coordinate ] ( v23 ) at ( 2.4,4.7 ) ; ( v20)(v21)(v22)(v23)(v20 ) ; node[coordinate ] ( v24 ) at ( 1.6,0.8 ) node[coordinate ] ( v25 ) at ( 2.6,3.8 ) node[coordinate ] ( v26 ) at ( 1.6,6.8 ) node[coordinate ] ( v27 ) at ( 0.6,3.8 ) ; ( v24)(v25)(v26)(v27)(v24 ) ; node[coordinate ] ( v18 ) at ( 7,4.5 ) node[coordinate ] ( v19 ) at ( 8,7.5 ) ; ( v18)(v19 ) ; ( v17)(v2)(v6)(v26 ) ; ( v22)(v10)(v14)(v19 ) ; ( v17)(v3)(v7)(v27 ) ; ( v23)(v11)(v15)(v18 ) ; ( v16)(v0)(v4)(v24 ) ; ( v20)(v8)(v12)(v18 ) ; ( v16)(v1)(v5)(v25 ) ; ( v21)(v9)(v13)(v19 ) ; we begin by defining a notion of what it means for two elements in @xmath219 or @xmath220 to be close to " or far away " from each other ; but first we need to give a concrete definition of distance between two elements in @xmath219 or @xmath220 . we will refer to a copy of @xmath223 or @xmath230 or @xmath231 in @xmath219 or @xmath220 as a _ block - component_. every element @xmath232 in @xmath219 or @xmath220 is a copy of one of the form @xmath233 , thus we may unambiguously extend the function @xmath234 to both @xmath219 and @xmath220 . now define the distance between two elements within @xmath219 by @xmath235 with @xmath236 defined identically on @xmath220 . the subscripts @xmath219 or @xmath220 are dropped when no ambiguity arises . note that these `` distances '' are not metrics in the strict sense , as distinct points may have distance @xmath11 ( if they lie in the same block component and have the same @xmath237 value ) . at each round @xmath238 , we will say that the distance between two elements @xmath232 and @xmath239 in @xmath219 or @xmath220 is @xmath240 if it at least @xmath241 . note that this also includes the case where the distance between @xmath232 and @xmath239 is infinite . assume that spoiler and duplicator have played @xmath242 rounds of the @xmath2 round game and have selected some elements @xmath243 and @xmath244 . let @xmath245 . if the distance between @xmath232 and @xmath239 is greater than or equal to @xmath246 and there are no previously played elements between them , then @xmath247 is large enough so that spoiler can not connect @xmath232 and @xmath239 . more precisely : at each round of play , spoiler may halve the distance between @xmath232 and @xmath239 ; the value of @xmath240 is chosen such that all further plays ( to the @xmath248 round ) will result in a gap of distance at least @xmath249 somewhere `` between '' @xmath232 and @xmath239 . see libkin @xcite for very similar arguments . the board " on which spoiler and duplicator will play the @xmath2-round ehrenfeucht - frass game essentially consists of lots of copies of @xmath24 and @xmath250 ( some of which are connected to one another ) . of course , the orientation of @xmath250 is interchangeable , because @xmath250 admits a natural automorphism given by @xmath251 . however it is convenient for our discussion to fix the ordering on these coordinates , so that we make think of pinching " the second coordinate on the left side and the first coordinate on the right side . in the same way , @xmath230 is in fact isomorphic to @xmath231 , but it is convenient for us to visualize them with their pinches in dual orientation . we can use the following assumption to shed some light on duplicator s strategy to win the @xmath2-round game . we can assume that spoiler s first move in the @xmath2-round ehrenfeucht - frass game will be in @xmath223 in @xmath219 , since @xmath252 is isomorphic to @xmath220 and so a first move made by spoiler outside of @xmath253 is a wasted one . assume that spoiler chooses an element @xmath254 that is distance @xmath255 strictly less than @xmath256 to the left pinch in @xmath253 ( the case for the right pinch is symmetric ) . since @xmath255 is small enough so that , after round @xmath2 , it is possible for spoiler to `` connect '' two elements and because the orientations at the left and right pinches differ , duplicator must choose an element @xmath257 from @xmath220 that is a copy of the element @xmath254 and in the exact corresponding position , that is , exactly distance @xmath255 from a left pinch . now assume that spoiler chooses an element @xmath254 that is @xmath256 from both the left and right pinch in @xmath253 . again , duplicator is forced to choose an element from a copy of @xmath192 or @xmath193 in @xmath220 and she must select an element @xmath257 that is also distance @xmath256 from any right or left pinch in @xmath220 . she can achieve this by selecting an exact corresponding element to @xmath254 that is somewhere around the middle of a copy of @xmath192 or @xmath193 ( her choice of @xmath192 or @xmath193 does not matter since they both have diameter greater than or equal to @xmath240 ) . 1 . if @xmath260 , then @xmath261 ; 2 . if @xmath262 , then @xmath263 ; 3 . if @xmath264 is of distance @xmath265 from a left pinch , then @xmath266 is of distance @xmath255 from left pinch ; 4 . if @xmath264 is of distance @xmath265 from a right pinch , then @xmath266 is of distance @xmath255 from right pinch ; 5 . if @xmath264 is of distance @xmath265 from a left slice , then @xmath266 is of distance @xmath255 from a left slice ; 6 . if @xmath264 is of distance @xmath265 from a right slice , then @xmath266 is of distance @xmath255 from a right slice ; 7 . the dual conditions to the above ( in other words , switching the roles of @xmath134 and @xmath267 ) . the base case holds vacuously . for the induction step , suppose that duplicator has survived @xmath268 rounds of the game by maintaining the above conditions to the completion of round @xmath268 . assume that spoiler is making his @xmath269st move in @xmath219 ( the case for @xmath220 is symmetric ) . if spoiler s selection for @xmath270 is equal to some previously played element @xmath271 , where @xmath272 , then duplicator s response should be @xmath273 . case @xmath158 : spoiler chooses @xmath270 so that it has distance greater than or equal to @xmath274 from any other previously played element in @xmath219 . + case @xmath158(a ) : spoiler chooses @xmath270 to be distance greater than or equal to @xmath274 from any pinch or slice in @xmath219 . duplicator must select @xmath275 so that it is also distance greater than or equal to @xmath274 from any other previously played element and from any pinch or slice in @xmath220 . duplicator can always find a copy of @xmath230 or @xmath231 in @xmath220 that contains no previously played points : after round @xmath268 there are at least @xmath276 unplayed copies ( each of length @xmath277 ) remaining in @xmath220 . if duplicator selects @xmath275 to be as close as possible to the middle of one of these intervals , then we ensure that the @xmath278 property and all other conditions of the hypothesis are carried through to the ( @xmath279)st round . case @xmath158(b ) : spoiler chooses @xmath270 so that it is distance @xmath280 from a pinch or slice in @xmath219 . without loss of generality , suppose spoiler chooses @xmath270 so that it is distance @xmath280 from a left pinch in @xmath219 . duplicator must select an element @xmath275 in @xmath220 that is exactly distance @xmath255 from a left pinch in @xmath220 and such that @xmath275 is distance greater than or equal to @xmath274 from any other previously played element . duplicator can always find a copy of @xmath231 in @xmath220 that contains no previously played elements : after round @xmath268 there are at least @xmath281 copies remaining in @xmath220 . now duplicator should select an element @xmath275 in one of these copies corresponding to the element that spoiler chose in his copy of @xmath231 ( or @xmath253 ) relative to the natural embedding of @xmath231 into @xmath253 , and so that it has distance exactly @xmath280 from the left pinch . then all the conditions of hypothesis are preserved to the ( @xmath282)st round in this case . case @xmath161(a ) : spoiler chooses @xmath270 to be distance greater than or equal to @xmath274 from any pinch or slice in @xmath219 . without loss of generality assume that spoiler has selected @xmath270 to be to the right of @xmath271 . there are two sub - cases to consider , however duplicator s strategy is the same in each case : she should select @xmath275 in @xmath220 to be an exact copy of the element @xmath270 , and exactly distance @xmath283 to the right of @xmath273 . we only need to check that conditions @xmath284 and @xmath285 are maintained in this case . + case @xmath161(a)(i ) : if there exists a previously selected element @xmath264 to the right of @xmath270 that is distance strictly less than @xmath274 , then @xmath260 , and so condition ( @xmath161 ) hypothesis tells us that @xmath261 . hence duplicator s selection of @xmath275 at exactly distance @xmath283 to the right of @xmath273 ensures that @xmath286 and @xmath287 . hence all conditions of the hypothesis are maintained in this case . + case @xmath161(a)(ii ) : if the distance of any other previously played element to the right of @xmath270 is greater than or equal @xmath274 . then , at round @xmath268 , the distance of any previously played element to the right of @xmath271 was greater than or equal to @xmath241 , so condition ( @xmath158 ) of the hypothesis tells us that the distance of any previously played element to the right of @xmath273 in @xmath220 is @xmath241 . hence duplicator s selection of @xmath275 at exactly distance @xmath283 to the right of @xmath273 ensures that @xmath286 and the distance of any other previously played element to the right of @xmath270 is greater than or equal @xmath274 . hence all conditions of the hypothesis are maintained in this case . + case @xmath161(b ) spoiler chooses @xmath270 to be distance strictly less than @xmath274 from any pinch or slice in @xmath219 . without loss of generality assume that spoiler chooses @xmath270 to be distance strictly less than @xmath274 from a left pinch . + case @xmath161 ( b)(i ) : if spoiler chooses @xmath270 to be to the right of @xmath271 at distance @xmath288 , then at round @xmath268 , the distance @xmath289 between @xmath271 and the left pinch must have been strictly less than @xmath241 . so condition ( @xmath210 ) of the hypothesis ensures that @xmath273 is also distance @xmath289 from a left pinch . if duplicator selects @xmath275 to be distance @xmath288 to the right of @xmath273 , then @xmath275 is exactly the same distance ( @xmath290 ) as @xmath270 from a left pinch , so condition ( @xmath210 ) of the hypothesis is maintained to this round . it remains to check conditions ( @xmath158 ) and ( @xmath161 ) : applying the arguments given in case @xmath161(a)(i ) and case @xmath161(a)(ii ) , respectively guarantees that all the conditions of the hypothesis are maintained in this case . + case @xmath161 ( b)(ii ) : if spoiler chooses @xmath270 to be to the left of @xmath271 at distance @xmath291 strictly less than @xmath274 , then @xmath271 must be distance @xmath292 strictly less than @xmath241 from the left pinch , and so the hypothesis ensures that @xmath273 is also distance @xmath292 from a left pinch . duplicator should select @xmath275 to be distance @xmath291 to the right of @xmath273 , then @xmath275 is exactly the same distance ( @xmath293 ) as @xmath270 from a left pinch , and so condition ( @xmath210 ) of the hypothesis is maintained to this round . this selection also guarantees that condition ( @xmath158 ) is maintained in this case . if there is no previously played element further to the left of @xmath270 , then we are done . if on the other hand , there exists a previously played element @xmath264 further to the left of @xmath270 , we know that @xmath294 , and so we need to check condition ( @xmath158 ) of the hypothesis . at round @xmath268 , we had @xmath295 , and so the hypothesis tells us that @xmath296 , and hence duplicator s selection ensures that @xmath297 . we prove the contrapositive . assume that @xmath298 is not definable by a finite conjunction of anti - identities modulo @xmath299 . then @xmath298 fails the finite duality property relative to the class @xmath4 ( clearly , ( @xmath158 ) @xmath218 ( @xmath161 ) of theorem [ thm : atserias ] relativises to any class ) . hence @xmath59 fails the finite duality property and so for every natural number @xmath9 , the structure @xmath184 does not admit a homomorphism to @xmath24 . now since the class @xmath4 is closed under forming the @xmath9-pinch construction over @xmath24 and taking disjoint unions , the proof of ( @xmath210 ) @xmath218 ( @xmath158 ) of theorem [ thm : atserias ] relativises to @xmath4 : we have found , for each non - negative integer @xmath2 , two structures @xmath219 and @xmath220 in @xmath4 such that : we now complete this section by extending theorem [ thm : sausage ] . the proof follows a similar structure : we first prove a non - relativised version , namely theorem [ thm : genrlsdsausagefacts ] , from which the extended result will follow almost immediately . the following three results given in lemma [ lem : erdos ] , corollary [ cor : extendederdos ] and lemma [ lem : finiteduality ] are required . the first lemma appears as lemma @xmath301 in @xcite , and is proved by a probabilistic argument based on that originally given by erds to show the existence , for each non - negative @xmath2 , of a graph with chromatic number greater than @xmath2 and no cycles of length less than or equal to @xmath2 . see the proof given in theorem @xmath302 of @xcite . [ lem : erdos ] let @xmath24 and @xmath58 be @xmath7-structures . if there exists no homomorphism from @xmath58 to @xmath24 , then for every natural number @xmath9 there exists an @xmath7-structure @xmath303 of girth at least @xmath9 such that the following conditions hold : [ cor : extendederdos ] let @xmath61 , let @xmath304 be a finite set of @xmath7-structures and let @xmath58 be an @xmath7-structure . if @xmath58 does not admit a homomorphism to @xmath48 , for each @xmath178 , then for any natural number @xmath9 , there exists an @xmath7-structure @xmath303 of girth at least @xmath9 such that the following conditions hold : let @xmath90 . since @xmath58 does not admit a homomorphism to @xmath48 , we can use lemma [ lem : erdos ] , to find a structure @xmath305 of girth at least @xmath9 that admits a homomorphism to @xmath58 but does not admit one to @xmath48 . define @xmath303 to be the disjoint union of the @xmath305 . clearly , the structure @xmath303 has girth at least @xmath9 and does not admit a homomorphism to @xmath48 , for all @xmath178 . the proof of the next lemma is identical , up to replacing lemma @xmath301 of @xcite with its extended version corollary [ cor : extendederdos ] , to the proof of lemma @xmath209 in @xcite . we refer the reader to the proof there . let @xmath55 be a class of finite @xmath214-structures , let @xmath56 be a critical obstruction for the class @xmath307 and let @xmath308 . we let @xmath309 denote the substructure of @xmath56 induced by the set @xmath310 . [ lem : hom2npinch ] let @xmath311 , let @xmath304 be a finite set of homomorphism - independent @xmath7-structures and let @xmath56 be a critical obstruction for the class @xmath54 with diameter at least @xmath312 . assume that there are @xmath313 and @xmath314 with @xmath232 and @xmath315 distance @xmath312 apart and with both @xmath309 and @xmath316 admitting a homomorphism to @xmath317 . then there exists a homomorphism from @xmath56 to @xmath318 . the lemma is essentially borrowed from @xcite , so we give only a proof sketch . let @xmath232 and @xmath315 be elements in @xmath56 at distance @xmath312 and let @xmath319 and @xmath320 be homomorphisms . the maps @xmath321 and @xmath96 defined below are shown to be homomorphisms in the proof of proposition @xmath306 in @xcite . define @xmath322 by @xmath323 1 . @xmath54 is definable by a first - order sentence ; 2 . for each @xmath178 , the class @xmath329 satisfies the finite duality property ; 3 . @xmath54 satisfies the finite duality property ; 4 . @xmath54 is definable by a finite conjunction of anti - identities . ( @xmath158 ) @xmath218 ( @xmath161 ) : we prove the contrapositive . assume that @xmath330 fails to satisfy the finite duality property . for each @xmath331 , define @xmath332 and let @xmath333 be the @xmath334-pinch over @xmath317 . let @xmath192 and @xmath193 be the substructures of @xmath333 induced by the sets @xmath335 and @xmath336 , respectively . from part ( [ itm : part1 ] and [ itm : pinchequiv ] ) of theorem [ thm : sausagefacts ] , we have that , for all @xmath337 , the structure @xmath338 does not admit a homomorphism to @xmath317 , but the substructures @xmath192 and @xmath193 both admit homomorphisms to @xmath317 . now consider the disjoint union @xmath339 . since @xmath304 is a homomorphism independent set , it follows that , for all @xmath340 , the structure @xmath341 does not belong to the class @xmath54 . for each @xmath340 let @xmath220 be the structure obtained by taking the disjoint union of @xmath49 with @xmath225 copies of @xmath224 , and let @xmath342 . then , for each @xmath340 , we have @xmath343 , while @xmath344 . now apply the argument given in theorem [ thm : sausage ] with the extra condition that whenever spoiler selects an element @xmath134 from @xmath219 in the extra copy of @xmath49 , then duplicator selects the corresponding element @xmath267 from the extra copy of @xmath49 in @xmath220 , and vice - versa . hence , for each non - negative integer @xmath2 , duplicator has a winning strategy in the @xmath2-round ehrenfeucht - frass game played on @xmath219 and @xmath220 , and so the property of admitting a homomorphism to @xmath55 is not definable by a first - order sentence . ( @xmath161 ) @xmath218 ( @xmath210 ) : we will show that there is a bound on the the diameter of the critical obstructions for @xmath54 . the required result will then follow by lemma [ lem : finiteduality ] . assume that , for each @xmath178 , the class @xmath345 satisfies the finite duality property . then , for each @xmath346 , there exists a natural number @xmath347 with @xmath348 admitting a homomorphism to @xmath349 , by part ( [ itm : pinchequiv ] ) of theorem [ thm : sausagefacts ] . let @xmath56 be any critical obstruction for @xmath54 . of course , for any element @xmath350 , the substructure @xmath351 admits a homomorphism to at least one of the @xmath349 . now let @xmath2 be the diameter of @xmath56 , and let @xmath232 and @xmath315 be elements in @xmath324 at distance @xmath2 . we need to consider two cases . _ case @xmath158 _ : there exists @xmath314 with @xmath352 and @xmath353 admitting homomorphisms to @xmath317 . lemma [ lem : hom2npinch ] tells us that @xmath56 admits a homomorphism to @xmath354 . but since @xmath355 admits a homomorphism to @xmath317 , and therefore @xmath356 admits a homomorphism to @xmath49 , for all @xmath357 ( by part [ itm : pinchsurjhom ] of theorem [ thm : sausagefacts ] ) , we must have that @xmath2 is strictly less than @xmath358 . _ case @xmath161 _ : there exists @xmath359 with @xmath360 and with @xmath352 admitting a homomorphism to @xmath361 and @xmath353 admittng a homomorphism to @xmath362 . we will show that @xmath2 is less than or equal to @xmath363 . if there exists @xmath364 and @xmath314 with @xmath365 and @xmath366 admitting homomorphisms to @xmath367 , then from case @xmath158 , the maximum distance between @xmath239 and @xmath368 in @xmath56 is strictly less than @xmath358 . thus , since @xmath352 admits a homomorphism to @xmath361 , any element @xmath350 with @xmath351 admitting a homomorphism to @xmath361 must be at most distance @xmath369 from @xmath232 . therefore any element @xmath370 with distance exactly @xmath371 from @xmath232 must have @xmath372 admitting a homomorphism to @xmath373 , where @xmath374 and @xmath375 . then any element @xmath376 with distance @xmath377 from @xmath378 ( and distance @xmath379 from @xmath232 ) must have @xmath380 admitting a homomorphism to @xmath381 , for some @xmath382 . we can continue this process until we run out of @xmath383 s . it follows that the diameter @xmath2 of @xmath56 is at most @xmath384 . [ thm : colourfam ] let @xmath386 . let @xmath4 be a class of finite relational structures in the signature @xmath7 and let @xmath304 be a finite set of homomorphism - independent structures in @xmath4 . if @xmath4 is closed under forming the @xmath9-pinch construction over @xmath55 and taking disjoint unions , then @xmath387 is equal to @xmath216 for some first order sentence @xmath96 if and only if there is a finite conjunction of anti - identities @xmath87 such that @xmath388 . we prove the contrapositive . assume that @xmath389 is not definable by a finite conjunction of anti - identities modulo @xmath299 . then @xmath389 fails the finite duality property relative to the class @xmath4 , and therefore @xmath390 fails the finite duality property . so there is some @xmath314 with @xmath330 failing to satisfy the finite duality property , using ( @xmath161 ) @xmath218 ( @xmath210 ) of theorem [ thm : genrlsdsausagefacts ] . now since the class @xmath4 is closed under forming the @xmath9-pinch construction over @xmath55 and taking disjoint unions , the proof of ( @xmath158 ) @xmath218 ( @xmath161 ) of theorem [ thm : genrlsdsausagefacts ] relativises to @xmath4 . hence the property of admitting a homomorphism into @xmath55 relative to the class @xmath4 is not definable by a first - order sentence . we have made multiple references to closure under formation of the @xmath9-pinch and taking disjoint unions . we conclude this section with some examples of sentences that are closed under these constructions . [ eg : examplelaws ] let @xmath391 be a relational signature . then a class of @xmath214-structures @xmath55 is closed under forming the @xmath9-pinch ( for sufficiently large @xmath9 ) and disjoint unions if it is definable by sentences of any of the following forms : closure under disjoint unions is obvious in each case ; in case ( 3 ) this uses the fact that the conclusion of the implication involves variables only appearing in the premise . we now discuss the formation of @xmath9-pinches ( for sufficiently large @xmath9 ) . case ( 1 ) is trivial , while case ( 2 ) follows from the fact that for @xmath396 , the @xmath289-vertex substructures of the @xmath9-pinch over @xmath397 admit a homomorphism into @xmath397 ( by theorem [ thm : sausagefacts](1 ) ) , and hence the @xmath9-pinch will satisfy any anti - identity in at most @xmath289 variables that is true on @xmath397 . for case ( 3 ) , let @xmath398 . we have @xmath399 . if @xmath400 , then from the definition of @xmath401 , for each @xmath402 , there exists @xmath403 satisfying @xmath404 . it follows that @xmath69 and @xmath405 . note that here we are implicitly using the condition that @xmath395 ; if @xmath406 contained repeated elements then we could not guarantee repeated elements in @xmath407 . now interpreting in @xmath24 , gives @xmath408 and @xmath409 , where @xmath410 . it follows that @xmath411 . a similar argument works for @xmath412 and even easier argument works for @xmath413 .

this article is a review of the main experimental applications of fluctuation theorems ( fts ) and summarizes the plenary talk given at statphys24 . in order to define the main contents let us consider several simple examples . the simplest and most basic out of equilibrium system is a thermal conductor whose extremities are connected to two heat baths at different temperatures , as sketched in fig.[fig : conduction ] . the second law of thermodynamics imposes that in average the heat flows from the hot to the cold reservoir ( from h to c in fig.[fig : conduction ] ) . however the second law does not say anything about fluctuations and in principle one can observe for a short time a heat current in the opposite direction . what is the probability of observing these rare events ? as a general rule when the size of the system decreases the role of fluctuations increases . thus from an experimental point of view it is reasonable to think that such rare events can be observed in systems that are small . a good candidate could be for example the thermal conduction in a nanotube whose extremities are connected to two heat baths @xcite , exactly in the spirit of fig.[fig : conduction ] . in reality in this kind of experiments the measure of the mean quantities @xcite is already difficult and of course the analysis of fluctuations is even more complicated . however there is an electrical analogy , shown in fig.[fig : conduction]b ) , of the thermal model of fig.[fig : conduction]a ) . let us consider an electrical conductor connected to a potential difference @xmath0 and kept at temperature @xmath1 by a heat bath . if the mean current @xmath2 ( @xmath3 being the electrical resistance of the conductor ) is of the order of @xmath4 a and the injected power is about @xmath5 j ( @xmath6 is the boltzmann constant ) then the instantaneous current inside the resistance has fluctuations whose amplitude is comparable to the mean , as shown in fig.[fig : conduction]c ) . the variance of these fluctuations is @xmath7 where @xmath8 is the characteristic time constant of the electrical circuit . in the specific case of fig . [ fig : conduction]c ) the current reverses with respect to the mean value . the probability of having those negative currents have been studied both theoretically and experimentally in ref . @xcite within the context of fluctuation theorems , that we will present in sec.[section_ft ] . and @xmath9 with @xmath10 . b ) electrical analogy . a conductor of electrical resistance @xmath3 and kept at a temperature @xmath1 is submitted to a potential difference @xmath11 . c ) instantaneous current @xmath12 flowing into the resistance using @xmath13 m@xmath14 , @xmath15 k and @xmath16 ms . [ fig : conduction ] ] we discuss a second example where the source of fluctuation is not the coupling with a thermal bath , as in the case of the electrical conductor , but it is either a chaotic flows or a chaotic force produced by the non - linear interaction of many degrees of freedom of a dissipative system sustained by an external driving . we will refer to them as chaotic systems . let us consider a turbulent wind flowing around an object as sketched in fig.[fig : turbulence]a),b ) . the wind exerts a mean force @xmath17 on the object but the instantaneous force , plotted in fig.[fig : turbulence]c ) , is a strongly fluctuating quantity which presents negative values @xcite , i.e. the object moves against the wind , fig.[fig : turbulence]b ) . in such a case the mean work done by the wind on the object is about @xmath18 j@xmath19 and obviously thermal fluctuations do not play any role but so does the chaotic flow , which produces the fluctuations . other similar examples can be found for example in shaken granular media @xcite , discussed in sec.[section : dynamical_system ] . ] these examples stress that in the two experiments the electrical conductor and the turbulent wind we may observe the counterintuitive effect that the instantaneous response of the system is opposite to the mean value , in other words the system has an instantaneous negative entropy production rate . this effect is induced by the thermal fluctuations in the first case and by the chaotic flow in the second case . the question that we want analyze in this article is whether the fluctuations theorem ( defined in section 3 ) is able to predict the probability of these rare events in both cases , i.e. for the stochastic and the chaotic systems . we will take an experimentalist approach and we will use experimental results in order to introduce the main concepts . the largest part of the article concerns stochastic systems described by a langevin dynamics . for chaotic systems we will mainly discuss the difficulty of comparing the experimental results with the theoretical predictions . the article is organized as follows . in section 2 we present the experimental results on the energy fluctuations measured in a harmonic oscillator driven out of equilibrium by an external force . in section 3 the experimental results on the harmonic oscillator are used to introduce the property of fluctuation theorems ( fts ) . in section 4 the non linear case of a brownian particle confined in a time dependent double well potential is presented . in section 5 we introduce the applications of the ft , and as a more specific example we describe the measure of the torque of a molecular motor . finally in section 6 we discuss the chaotic systems and we conclude in section 7 . the choice of discussing the dynamics of the harmonic oscillator is dictated by the fact that it is relevant for many practical applications such as the measure of the elasticity of nanotubes@xcite , the dynamics of the tip of an afm @xcite , the mems and the thermal rheometer that we developed several years ago to study the rheology of complex fluids @xcite . this device is a very sensitive torsion pendulum as sketched in fig.[fig : pendulum]a ) . it is composed by a brass wire ( length @xmath20 @xmath21 , width @xmath22 @xmath21 , thickness @xmath23 @xmath24 ) and a glass mirror with a golden surface , fig.[fig : pendulum]c ) . the mirror ( length @xmath25 @xmath21 , width @xmath26 @xmath21 , thickness @xmath27 @xmath21 ) is glued in the middle of the brass wire . the elastic torsional stiffness of the wire is @xmath28 @xmath29 . it is enclosed in a cell , fig.[fig : pendulum]d ) , which is filled by a fluid . we used either air or a water - glycerol mixture at @xmath30 concentration . the system is a harmonic oscillator with resonant frequency @xmath31 and a relaxation time @xmath32 . @xmath33 is the total moment of inertia of the displaced masses ( _ i.e. _ the mirror and the mass of displaced fluid ) @xcite . the damping has two contributions : the viscous damping @xmath34 of the surrounding fluid and the viscoelasticity of the brass wire . the angular displacement of the pendulum @xmath35 is measured by a differential interferometer @xcite which uses the two laser beams reflected by the mirror fig.[fig : pendulum]a ) . the measurement noise is two orders of magnitude smaller than thermal fluctuations of the pendulum . @xmath36 is acquired with a resolution of @xmath37 bits at a sampling rate of @xmath38 @xmath39 , which is about 40 times @xmath40 . we drive the system out - of - equilibrium by forcing it with an external torque @xmath41 by means of a small electric current @xmath42 flowing in a coil glued behind the mirror ( fig . [ fig : pendulum]b ) . the coil is inside a static magnetic field . the displacements of the coil and therefore the angular displacements of the mirror are much smaller than the spatial scale of inhomogeneity of the magnetic field . so the torque is proportional to the injected current : @xmath43 ; the slope @xmath44 depends on the geometry of the system . the practical realization of the montage is shown in figs . [ fig : pendulum]c ) , [ fig : pendulum]d ) . in equilibrium the variance @xmath45 of the thermal fluctuations of @xmath35 can be obtained from equipartition , i.e. @xmath46 nrad for our pendulum , where @xmath1 is the temperature of the surrounding fluid . the dynamics of the torsion pendulum can be assimilated to that of a harmonic oscillator damped by the viscoelasticity of the torsion wire and the viscosity of the surrounding fluid , whose equation of motion reads in the temporal domain @xmath47 where @xmath48 is the memory kernel and @xmath49 the thermal noise . in fourier space ( in the frequency range of our interest ) this equation takes the simple form @xmath50\ , \hat{\theta } = \hat{m}+^\eta , \end{aligned}\ ] ] where @xmath51 denotes the fourier transform and @xmath52 $ ] is the complex frequency - dependent elastic stiffness of the system . @xmath53 and @xmath54 are the viscoelastic and viscous components of the damping term . . a ) square root of the power spectral density of @xmath35 . @xmath55 directly measured spectrum , black dotted line is the spectrum estimated from the measure of @xmath56 and using eq.[fdt ] the red dashed and dotted lines show the viscous and viscoelastic component of the damping respectively . b ) probability density function of @xmath57 . the continuous line is a gaussian fit ] at equilibrium,_i.e . _ @xmath58 , the fluctuation dissipation theorem ( fdt ) gives a relation between the amplitude of the thermal angular fluctuations of the oscillator and its response function . the response function of the system @xmath59 can be measured by applying a torque with a white spectrum . when @xmath60 , the amplitude of the thermal vibrations of the oscillator is related to its response function via the fluctuation - dissipation theorem ( fdt ) . therefore , the thermal fluctuation power spectral density ( psd ) of the torsion pendulum reads for positive frequencies @xmath61 ^ 2}. \label{fdt}\ ] ] the brackets are ensemble averages . as an example , the spectrum of @xmath35 measured in the glycerol - water mixture is shown in fig.[fig : spectrum]a ) . in this case the resonance frequency is @xmath62 @xmath39 and the relaxation time @xmath63 @xmath64 the measured spectrum is compared with that obtained from eq.[fdt ] using the measured @xmath56 . the viscoelastic component at low frequencies correspond to a constant @xmath65 . indeed if @xmath66 then from eq.[fdt ] @xmath67 as seen in fig.[fig : spectrum]a ) . instead if @xmath68 then for @xmath69 from eq.[fdt ] the spectrum is constant as a function of @xmath70 . it is important to stress that in the viscoelastic case the noise @xmath49 is correlated and the process is not markovian , whereas in the viscous case the process is markovian . thus by changing the quality of the fluid surrounding the pendulum one can tune the markovian nature of the process . in the following we will consider only the experiment in the glycerol - water mixture where the viscoelastic contribution is visible only at very low frequencies and is therefore negligible . this allows a more precise comparison with theoretical predictions often obtained for markovian processes . the probability density function ( pdf ) of @xmath35 , plotted in fig.[fig : spectrum]b ) , is a gaussian . we now consider a periodic forcing of amplitude @xmath71 and frequency @xmath72 , i.e. @xmath73 @xcite-@xcite . this is a very common kind of forcing which has been already studied in the case of the first order langevin equation @xcite and of the two level system @xcite and in a different context for the second order langevin equation @xcite . furthermore this is a very general case because using fourier transform , any periodical forcing can be decomposed in a sum of sinusoidal forcing . we explain here the behavior of a single mode . experiments have been performed at various @xmath71 and @xmath72 . we present here the results for a particular amplitude and frequency : @xmath74 @xmath75 and @xmath76 @xmath39 . this torque is plotted in fig . [ fig : sinustorque]a . the mean of the response to this torque is sinusoidal , with the same frequency , as can be seen in fig . [ fig : sinustorque]b . the system is clearly in a non - equilibrium steady state ( ness ) . . ] the work done by the torque @xmath77 on a time @xmath78 is @xmath79 as @xmath35 fluctuates also @xmath80 is a fluctuating quantity whose probability density function ( pdf ) is plotted in fig . [ fig : pdf_w_u]a ) for various @xmath81 . this plot has interesting features . specifically , work fluctuations are gaussian for all values of @xmath81 and @xmath82 takes negative values as long as @xmath83 is not too large . the probability of having negative values of @xmath82 decreases when @xmath83 is increased . there is a finite probability of having negative values of the work , in other words the system may have an instantaneous negative entropy production rate although the average of the work @xmath84 is of course positive ( @xmath85 stands for ensemble average ) . in this specific example is @xmath86 . we now consider the energy balance for the system . ; @xmath87 ( @xmath88 ) , @xmath89 ( @xmath90 ) , @xmath91 ( @xmath92 ) and @xmath93 ( @xmath94 ) . b ) pdf of @xmath95,title="fig : " ] as the fluid is rather viscous we will take into account only the standard viscosity that is @xmath68 and @xmath96 . in such a case eq.[eqofmotion ] simplifies @xmath97 where @xmath49 is the thermal noise , which in this case is delta - correlated in time : @xmath98 . when the system is driven out of equilibrium using a deterministic torque , it receives some work and a fraction of this energy is dissipated into the heat bath . multiplying eq . ( [ eq : langevin_oscillator ] ) by @xmath99 and integrating between @xmath100 and @xmath101 , one obtains a formulation of the first law of thermodynamics between the two states at time @xmath100 and @xmath101 ( eq . ( [ eq : energy_conservation ] ) ) . this formulation has been first proposed in ref.@xcite and used in other theoretical and experimental works @xcite . the change in internal energy @xmath95 of the oscillator over a time @xmath102 , starting at a time @xmath100 , is written as : @xmath103 where @xmath82 is the work done on the system over a time @xmath102 : @xmath104 and @xmath105 is the heat dissipated by the system . the internal energy is the sum of the potential energy and the kinetic energy : @xmath106^{2 } + \frac{1}{2 } c \theta(t)^2 \right\}. \label{eq : udef}\ ] ] the heat transfer @xmath105 is deduced from equation ( [ eq : energy_conservation ] ) ; it has two contributions : @xmath107^{2}{\mathrm{d}}t ' - \int_{t_i}^{t_i+\tau } \eta(t ' ) \frac{{\mathrm{d}}\theta}{{\mathrm{d}}t}(t ' ) { \mathrm{d}}t ' . \label{eq : qdef}\end{aligned}\ ] ] the first term corresponds to the viscous dissipation and is always positive , whereas the second term can be interpreted as the work of the thermal noise which has a fluctuating sign . the second law of thermodynamics imposes @xmath108 to be positive . b ) pdf of @xmath109 for various @xmath81 : @xmath87 ( @xmath88 ) , @xmath89 ( @xmath90 ) , @xmath91 ( @xmath92 ) and @xmath93 ( @xmath94 ) . the continuous lines in this figures are not fits but are analytical predictions obtained from the lnagevin dynamics as discussed in sect.[sec : theoretical_results],title="fig : " ] the dissipated heat @xmath105 can not be directly measured because we have seen that eq.[eq : qdef ] contains the work of the noise ( the heat bath ) that experimentally is impossible to measure , because @xmath49 is unknown . however @xmath105 can be obtained indirectly from the measure of @xmath82 and @xmath95 , whose pdf measured during the periodic forcing are exponential for any @xmath102 , as shown in fig.[fig : pdf_w_u]b . we first do some comments on the average values . the average of @xmath95 is obviously vanishing because the time @xmath102 is a multiple of the period of the forcing . therefore @xmath110 and @xmath111 are equal . we rescale the work @xmath82 ( the heat @xmath105 ) by the average work @xmath112 ( the average heat @xmath108 ) and define : @xmath113 ( @xmath114 ) . in the present article , @xmath115 , respectively @xmath116 , stands for either @xmath117 or @xmath118 , respectively @xmath82 or @xmath105 . we compare now the pdf of @xmath117 and @xmath118 in fig . [ fig : compare_pdf_w_q ] . the pdfs of heat fluctuations @xmath119 have exponential tails ( fig . [ fig : compare_pdf_w_q]b ) . it is interesting to stress that although the two variables @xmath82 and @xmath105 have the same mean values they have a very different pdf . the pdf of @xmath117 are gaussian whereas those of @xmath118 are exponential . on a first approximation the pdf of @xmath118 are the convolution of a gaussian ( the pdf of @xmath82 ) and exponential ( the pdf of @xmath95 ) . in figs . [ fig : compare_pdf_w_q ] the continuous lines are analytical predictions obtained from the langevin dynamics with no adjustable parameter ( see sect.[sec : theoretical_results ] ) . in the previous section we have seen that both @xmath82 and @xmath105 present negative values , _ i.e. _ the second law is verified only on average but the entropy production can have instantaneously negative values . the probabilities of getting positive and negative entropy production are quantitatively related in non - equilibrium systems by the fluctuation theorem ( fts ) . there are two classes of fts . the _ stationary state fluctuation theorem _ ( ssft ) considers a non - equilibrium steady state . the _ transient fluctuation theorem _ ( tft ) describes transient non - equilibrium states where @xmath102 measures the time since the system left the equilibrium state . a fluctuation relation ( fr ) examines the symmetry around @xmath120 of the probability density function ( pdf ) @xmath121 of a quantity @xmath115 , as defined in the previous section . it compares the probability to have a positive event ( @xmath122 ) versus the probability to have a negative event ( @xmath123 ) . we quantify the ft using a function @xmath124 ( symmetry function ) : @xmath125 the _ transient fluctuation theorem _ ( tft ) states that the symmetry function is linear with @xmath115 for any values of the time integration @xmath102 and the proportionality coefficient is equal to @xmath27 for any value of @xmath102 . @xmath126 contrary to tft , the _ stationary state fluctuation theorem _ ( ssft ) holds only in the limit of infinite time ( @xmath102 ) . @xmath127 in the following we will assume linearity at finite time @xmath102 @xcite and use the following general expression : @xmath128 where for ssft @xmath129 takes into account the finite time corrections and @xmath130 whereas @xmath131 for tft . however these claims are not universal because they depend on the kind of @xmath115 which is used . specifically we will see in the next sections that the results are not exactly the same if @xmath116 is replaced by any one of @xmath82 , @xmath105 and @xmath132 , defined in sect.[section_ft_harmonic ] . furthermore the definitions given in this section are appropriate for stochastic systems and in sect.[section : dynamical_system ] we will discuss the differences between stochastic and chaotic systems . the first numerical evidence of relations of this kind has been given by evans et al . in ref.@xcite whereas the tft was proved in ref.@xcite . in 1995 gallavotti and cohen @xcite proved ssft for dynamical systems although in such a case @xmath115 takes a different meaning that we will discuss in sect.[section : dynamical_system ] . the proof of ssft has been extended to stochastic dynamics in ref . furthermore van zon and cohen proved that there is an important difference between the fts for the injected power and those for the dissipated power @xcite . the ssft has been proved also for other quantities such as the dissipation function @xcite and the total entropy @xcite . other theoretical papers studied ft and the reader may find a review in ref.@xcite . experiments searching for fts have been performed in dynamical systems @xcite , but interpretations are very difficult because a quantitative comparison with theoretical prediction can be doubtful . other experiments have been performed in stochastic systems described by a first order langevin equation : a brownian particle in a moving optical trap @xcite and an out - of - equilibrium electrical circuit @xcite in which existing theoretical predictions @xcite were verified . other experimental tests for fts have been performed on driven two level systems @xcite and on colloids @xcite . let us suppose the the variable @xmath116 has a gaussian distribution of mean @xmath133 and variance @xmath134 . it is easy to show that in order to satisfy fts , the variable @xmath116 must have the following statistical property : @xmath135 this is an interesting relation because it imposes that the relative fluctuations of @xmath116 are @xmath136 this means that the probability of having negative events reduces by increasing @xmath116 , specifically from eq.[eq : relative_fluct ] it follows that @xmath137 where @xmath138 is the complementary error function . it is now possible to estimate the length of the time interval @xmath139 needed to observe at least one negative event , which is : @xmath140 where we used the fact that all the values @xmath82 computed on different intervals of length @xmath102 are independent , which is certainly true if @xmath102 is larger than the correlation time . let us consider the specific example of section [ sec : sinusexp ] , i.e. @xmath141 at @xmath142 hz , @xmath143 @xmath75 and @xmath144 . the pdf of @xmath82 are gaussian in this case ( fig.[fig : pdf_w_u ] ) and , as we will see in the next section , they satisfy ssft for large @xmath102 . therefore eq.[eq : relative_fluct ] holds for @xmath145 and we may estimate @xmath139 in the asymptotic limit @xmath146 . for example at @xmath147 , one obtains from the above mentioned experimental values @xmath148 and @xmath149 . inserting these experimental values in eq.[eq : relative_fluct ] one gets roughly a negative event over an observational time @xmath150 , which is already a rather long time for the distance between two events . for larger @xmath81 and larger @xmath151 this time becomes exponentially large . this justifies the fact that millions of data are necessary in order to have a reliable measure of ssft . the questions we ask are whether for finite time fts are satisfied for either @xmath152 or @xmath153 and what are the finite time corrections . in a first time , we test the correction to the proportionality between the symmetry function @xmath154 and @xmath115 . in the region where the symmetry function is linear with @xmath115 , we define the slope @xmath129 , i.e. @xmath155 . in a second time we measure finite time corrections to the value @xmath156 which is the asymptotic value expected from fts . in this review article we will focus on the ssft applied to the experimental results of sect . [ sec : sinusexp ] and to other examples . the tft will be not discussed here and the interested readers may look at ref.@xcite . plotted as a function of @xmath117 for various @xmath81 : @xmath87 ( @xmath88 ) , @xmath89 ( @xmath90 ) , @xmath91 ( @xmath92 ) and @xmath93 ( @xmath94 ) . for all @xmath81 the dependence of @xmath157 on @xmath117 is linear , with slope @xmath158 . b ) symmetry functions @xmath159 plotted as a function of @xmath118 for various @xmath81 . the dependence of @xmath159 on @xmath118 is linear only for @xmath160 . continuous lines are is theoretical predictions.,title="fig : " ] as a function of n obtained from the slopes of the straight lines of fig.[fig : s_w_s_q]a ) ( @xmath161 ) . the circles correspond to another measurement performed at a different frequency . the finite time corrections depend on the driving frequency . the slope @xmath162 measured for @xmath160 ( fig.[fig : s_w_s_q]b ) have exactly the same values of @xmath163 as a function of @xmath81 . b ) linear forcing . @xmath158 measured as a function of @xmath102 with the driving torque @xmath41 has a linear dependence on time . the finite time corrections depend on form of the driving . ] from the pdfs of @xmath117 and @xmath118 plotted in figs.[fig : compare_pdf_w_q ] , we compute the symmetry functions defined in eq.[eq : ft ] . the symmetry function @xmath164 are plotted in fig . [ fig : s_w_s_q]a ) as a function of @xmath165 . they are linear in @xmath165 . the slope @xmath166 is not equal to @xmath27 for all @xmath81 but there is a correction at finite time ( fig . [ fig : compare_sigma]a ) . nevertheless , @xmath166 tends to @xmath27 for large @xmath81 . thus ssft is satisfied for @xmath82 and for a sinusoidal forcing . the convergence is very slow and we have to wait a large number of periods of forcing for the slope to be @xmath27 ( after @xmath167 periods , the slope is still @xmath168 ) . this behavior is independent of the amplitude of the forcing @xmath71 and consequently of the mean value of the work @xmath169 , which , as explained in sec.[sec : gaussian_ft ] , changes only the time needed to observe a negative event . the system satisfies the ssft for all forcing frequencies @xmath72 but finite time corrections depend on @xmath72 , as can be seen in fig . [ fig : compare_sigma]a ) . we now analyze the pdf of @xmath118 ( fig.[fig : compare_pdf_w_q]b ) ) and we compute the symmetry functions @xmath170 of @xmath119 plotted in fig . [ fig : s_w_s_q]b ) for different values of @xmath81 . they are clearly very different from those of @xmath165 plotted in fig . [ fig : s_w_s_q]a ) . for @xmath170 three different regions appear : \(i ) for large fluctuations @xmath119 , @xmath170 equals @xmath25 . when @xmath102 tends to infinity , this region spans from @xmath171 to infinity . \(ii ) for small fluctuations @xmath119 , @xmath170 is a linear function of @xmath119 . we then define @xmath172 as the slope of the function @xmath170 , _ i.e. _ @xmath173 . we have measured @xcite that @xmath174 for all the values of @xmath81 , i.e. finite time corrections are the same for heat and work . thus @xmath172 tends to @xmath27 when @xmath102 is increased and ssft holds in this region ii which spans from @xmath175 up to @xmath176 for large @xmath102 . this effect has been discussed for the first time in refs.@xcite . \(iii ) a smooth connection between the two behaviors . these regions define the fluctuation relation from the heat dissipated by the oscillator . the limit for large @xmath102 of the symmetry function @xmath159 is rather delicate and it has been discussed in ref.@xcite . the conclusions of this experimental analysis is that ssft holds for work for any value of @xmath117 whereas for heat it holds only for @xmath160 . the finite time correction to fts , described by @xmath177 are not universal . they are the same both for @xmath117 and @xmath118 but they depend on the driving frequency as shown in fig.[fig : compare_sigma]a ) . furthermore they depend on the kind of driving force . in fig . [ fig : compare_sigma]b ) we plot @xmath158 measured when the harmonic oscillator is driven out of equilibrium by a linear ramp . the difference with respect fig . [ fig : compare_sigma]a ) is quite evident . this experimental analysis allows a very precise comparison with theoretical predictions using the langevin equation ( eq.[eq : langevin_oscillator ] ) and using two experimental observations : a ) the properties of heat bath are not modified by the driving and b ) the fluctuations of the @xmath82 are gaussian ( see also @xcite , where it is shown that in langevin dynamics @xmath82 has a gaussian distribution for any kind of deterministic driving force if the properties of the bath are not modified by the driving and the potential is harmonic ) . the observation in point a ) is extremely important because it is always assumed to be true in all the theoretical analysis . in ref.@xcite this point has been precisely checked . using these experimental observations one can compute the pdf of @xmath118 and the finite time corrections @xmath178 to ssft ( see ref.@xcite ) . the continuous lines in fig . [ fig : compare_sigma ] , fig . [ fig : s_w_s_q ] and fig . [ fig : pdf_w_u ] are not fit but analytical predictions , with no adjustable parameters , derived from the langevin dynamics of eq.[eq : langevin_oscillator ] ( see ref.@xcite for more details ) . in previous sections we have studied the energy @xmath82 injected into the system in the time @xmath102 and the energy dissipated towards the heat bath @xmath105 . these two quantities and the internal energy are related by the first law of thermodynamics ( eq.[eq : qdef ] ) . around the mean trajectory @xmath179 . the continuous line is the equilibrium distribution ] following notations of ref @xcite , we define the entropy variation in the system during a time @xmath102 as : @xmath180 for thermostated systems , entropy change in medium behaves like the dissipated heat . the non - equilibrium gibbs entropy is : @xmath181 where @xmath182 denotes the set of control parameters at time @xmath183 and @xmath184 is the probability density function to find the particle at a position @xmath185 at time @xmath183 , for the state corresponding to @xmath182 . this expression allows the definition of a `` trajectory - dependent '' entropy : @xmath186 the variation @xmath187 of the total entropy @xmath188 during a time @xmath102 is the sum of the entropy change in the system during @xmath102 and the variation of the `` trajectory - dependent '' entropy in a time @xmath102 , @xmath189 : @xmath190 in this section , we study fluctuations of @xmath191 computed using ( [ eq : def_medium_entropy ] ) and ( [ eq : def_gibbs_entropy ] ) . we will show that @xmath191 satisfies a ssft for all @xmath102 . in ref . @xcite , the relevance of boundary terms like @xmath192 has been pointed out for markovian processes . we investigate the data of the harmonic oscillator described in sect.[sec : sinusexp].the probability to compute is the joint probability @xmath193 , where @xmath194 is the starting phase @xmath195 . the system is linear , so @xmath196 are independent ; thus the joint probability can be factorized into a product . the expression of the trajectory dependent entropy is : @xmath197 for computing correctly the trajectory dependent entropy , we have to calculate the @xmath198 and @xmath199 for each initial phase @xmath194 ( see fig.[fig : trajectories]a ) . these distributions turn out to be independent of @xmath194 and they correspond to the equilibrium fluctuations of @xmath35 and @xmath99 around the mean trajectory defined by @xmath200 and @xmath201 . the distribution of @xmath202 is plotted in fig.[fig : trajectories]b ) , where the continuous line corresponds to the equilibrium distribution . once the @xmath203 and @xmath199 are determined we compute the `` trajectory - dependent '' entropy . as fluctuations of @xmath35 and @xmath99 are independent of @xmath194 we can average @xmath204 over @xmath194 which improves a lot the statistical accuracy . we stress that it is not equivalent to calculate first the pdfs over all values of @xmath194 which would correspond here to the convolution of the pdf of the fluctuations with the pdf of a periodic signal and then compute the trajectory dependent entropy . the results are shown in fig . [ pendule_sinus : pdf ] . in fig . [ pendule_sinus : pdf]a ) , we recall the main results for the dissipated heat @xmath205 . its average value @xmath206 is linear in @xmath83 and equal to the injected work . the pdfs of @xmath207 are not gaussian and extreme events have an exponential distribution . the pdf of the `` trajectory - dependent '' entropy is plotted in fig . [ pendule_sinus : pdf]b ) ; it is exponential and independent of @xmath81 . we superpose to it the pdf of the variation of internal energy divided by @xmath1 at equilibrium : the two curves match perfectly within experimental errors , so the `` trajectory - dependent '' entropy can be considered as the entropy exchanged with the thermostat if the system is at equilibrium . the average value of @xmath204 is zero , so the average value of the total entropy is equal to the average of injected power divided by @xmath1 . in fig . [ pendule_sinus : pdf]c ) , we plot the pdfs of the normalized total entropy for four typical values of integration time . we find that the pdfs are gaussian for any time . the symmetry functions ( eq.[eq : ft ] ) of the dissipated heat @xmath208 and the total entropy @xmath209 are plotted in fig . [ pendule_sinus : pdf]d ) . as we have already seen in fig.[fig : s_w_s_q ] , @xmath210 is a non linear function of @xmath211 . the linear behavior , with a slope that tends to @xmath27 for large time , is observed only for for @xmath212 . for the normalized total entropy , the symmetry functions are linear with @xmath213 for all values of @xmath214 and the slope is equal to @xmath27 for all values of @xmath83 . note that it is not exactly the case for the first values of @xmath83 because these are the times over which the statistical errors are the largest and the error in the slope is large . for the harmonic oscillator we have obtained that the `` trajectory - dependent '' entropy can be considered as the entropy variation in the system in a time @xmath102 that one would have if the system was at equilibrium . therefore the total entropy is the additional entropy due to the presence of the external forcing : _ this is the part of entropy which is created by the non - equilibrium stationary process_. the total entropy ( or excess entropy ) satisfies the fluctuation theorem for all times and for all kinds of stationary external torque@xcite . more details on this problem can be found in ref.@xcite . ] the harmonic oscillator can not be driven to a non linear regime without forcing it to such a high level where thermal fluctuations become negligible . thus in order to study the non linear effects we change experiment and we measure the fluctuations of a brownian particle trapped in a non - linear potential produced by two laser beams , as shown in fig.[fig : two_laser_beams ] . it is very well known that a particle of small radius @xmath215 @xmath216 m is trapped in the focus of a strongly focused laser beam , which produces a harmonic potential for the particle , whose brownian motion is confined inside this potential well . when two laser beams are focused at a distance @xmath217 , as shown in fig.[fig : two_laser_beams]a ) the particle has two equilibrium positions , i.e. the foci of the two beams . thermal fluctuations may force the particle to move from one to the other . the particle feels an equilibrium potential @xmath218 , shown in fig.[fig : two_laser_beams]b ) , where @xmath219 , @xmath220 and @xmath221 are determined by the laser intensity and by the distance of the two focal points . this potential has been computed from the measured equilibrium distribution of the particle @xmath222 . the right left asymmetry of the potential ( fig.[fig : two_laser_beams]b ) is induced by small unavoidable asymmetries , induced by the optics focusing the two laser beams . in our experiment the distance between the two spots is @xmath223 m , which produces a trap whose minima are at @xmath224 m . the total intensity of the laser is @xmath225mw on the focal plane which corresponds to an inter - well barrier energy @xmath226 , @xmath227 , @xmath228 and @xmath229 ( see ref.@xcite for more experimental details ) . the rate at which the particle jumps from one potentials well to the other is determined by the kramer s rate @xmath230 where @xmath231 is a characteristic time . in our experiment @xmath232 hz at @xmath233 k. and @xmath234 . b ) example of trajectory of the glass bead and the corresponding perturbation at @xmath235 hz . c ) mean injected energy in the system over a single period as a function of the driving frequency.@xmath236 @xmath90 and @xmath237 @xmath88 coincide as their mean values are equal within experimental errors . the error bars are computed from the standard deviation of the mean over different runs . inset : standard deviations of work distributions over a single period normalized by the average work as a function of the frequency ( same symbols).,title="fig:",width=302 ] and @xmath234 . b ) example of trajectory of the glass bead and the corresponding perturbation at @xmath235 hz . c ) mean injected energy in the system over a single period as a function of the driving frequency.@xmath236 @xmath90 and @xmath237 @xmath88 coincide as their mean values are equal within experimental errors . the error bars are computed from the standard deviation of the mean over different runs . inset : standard deviations of work distributions over a single period normalized by the average work as a function of the frequency ( same symbols).,title="fig:",width=302 ] to drive the system out of equilibrium we periodically modulate the intensity of the two beams at low frequency . thus the potential felt by the bead is the following profile : @xmath238 with @xmath239 . the amplitude of the time dependent perturbation is synchronously acquired with the bead trajectory . , @xmath220 and @xmath240 , obtained from fitted steady distributions at given phases , vary with the phase ( @xmath241 , @xmath242 ) . ] an example of the measured potential for @xmath243 and @xmath244 is shown on the fig . [ fig : artforcingex]a ) . this figure is obtained by measuring the probability distribution function @xmath245 of @xmath246 for fixed values of @xmath247 , it follows that @xmath248 . the @xmath246 position of the particle can be described by a langevin equation : @xmath249 with @xmath250 n s m@xmath251 the friction coefficient and @xmath49 the thermal noise delta correlated in in time . when @xmath252 the particle can experience a stochastic resonance @xcite , when the forcing frequency is close to the kramer s rate . an example of the sinusoidal force with the corresponding position are shown on the figure [ fig : artforcingex]b ) . since the synchronization is not perfect , sometimes the particle receives energy from the perturbation , sometimes the bead moves against the perturbation leading to a negative work on the system . two kinds of work can be defined in this experiments @xcite @xmath253 where in this case @xmath254 is a multiple of the forcing period . the work @xmath255 is the stochastic work ( used in jarzynsky and crooks relations @xcite ) and @xmath256 is the classical work that will be discussed in this article . the results on @xmath255 are quite similar but there are subtle differences discussed in ref.@xcite . for different numbers of period @xmath257 , @xmath25 , @xmath258 , @xmath26 and @xmath259 ( @xmath260 hz ) . inset : same data in lin - log . b ) normalized symmetry function as function of the normalized work for @xmath257 ( @xmath261 ) , @xmath25 ( @xmath88 ) , @xmath258 ( @xmath92 ) , @xmath26 ( @xmath262 ) , @xmath259 ( @xmath90).,title="fig:",width=317 ] for different numbers of period @xmath257 , @xmath25 , @xmath258 , @xmath26 and @xmath259 ( @xmath260 hz ) . inset : same data in lin - log . b ) normalized symmetry function as function of the normalized work for @xmath257 ( @xmath261 ) , @xmath25 ( @xmath88 ) , @xmath258 ( @xmath92 ) , @xmath26 ( @xmath262 ) , @xmath259 ( @xmath90).,title="fig:",width=264 ] we first measure the average work received over one period for different frequencies ( @xmath263 in eq . [ eq : integr ] ) . each trajectory is here recorded during 3200 s in different consecutive runs , which corresponds to 160 up to 6400 forcing periods , for the range of frequencies explored . in order to increase the statistics we consider @xmath264 different @xmath265 . the figure [ fig : artforcingex]c ) shows the evolution of the mean work per period for both definitions of the work . first , the input average work decreases to zero when the frequency tends to zero . indeed , the bead hops randomly several times between the two wells during the period . second , in the limit of high frequencies , the particle has not the time to jump on the other side of the trap but rather stays in the same well during the period , thus the input energy is again decreasing when increasing frequency . in the intermediate regime , the particle can almost synchronize with the periodical force and follows the evolution of the potential . the maximum of injected work is found around the frequency @xmath266 hz , which is comparable with half of the kramers rate of the fixed potential @xmath267 hz . this maximum of transferred energy shows that the stochastic resonance for a brownian particle is a bona fide resonance , as it was previously shown in experiments using resident time distributions @xcite or directly in simulations @xcite . in the inset of fig . [ fig : artforcingex ] , we plot the normalized standard deviation of work distributions ( @xmath268 ) as a function of the forcing frequency . the curves present a minimum at the same frequency of 0.1 hz , in agreement again with the resonance phenomena . in order to study ft for stochastic resonance we choose for the external driving a frequency @xmath260 hz , which ensures a good statistic , by allowing the observation of the system over a sufficient number of periods . we compute the works and the dissipation using @xmath269 different @xmath183 on time series which spans about 7500 period of the driving . we consider the pdf @xmath270 which is plotted in ( fig . [ fig : wclpdfn]a ) . notice that for small @xmath81 the distributions are double peaked and very complex . they tend to a gaussian for large @xmath81 ( inset of fig . [ fig : wclpdfn]a ) . on fig . [ fig : wclpdfn]b ) , we plot the normalized symmetry function of @xmath256 . we can see that the curves are close to the line of slope one . for high values of work , the dispersion of the data increases due to the lack of events . the slope tends toward 1 as expected by the ssft . it is remarkable that straight lines are obtained even for @xmath81 close to 1 , where the distribution presents a very complex and unusual shape ( fig . [ fig : wclpdfn]a ) . we do not discuss here the case of @xmath255 as the behavior is quite similar to that of @xmath256 @xcite . the very fast convergency to the asymptotic value of the the ssft is quite striking in this example . the measurement are in fully agreement with a realistic model based on the fokker planck equations where the measured values of @xmath271 has been inserted @xcite . this example shows the application of ft in a non - linear case where the distributions are strongly non - gaussian . the fluctuations theorems have several important consequences such as the jarzinsky and crooks equalities@xcite , which are useful to compute the free energy difference between two equilibrium states using any kind of transformation@xcite . the hatano - sasa@xcite relations and the recently derived fluctaution dissipation theorems@xcite are related to fts and are useful to compute the response of a ness using the steady state fluctuations of the ness . as we have seen the ft allows the calculation of tiny amount heat , which can be useful in many applications in aging systems @xcite and biological systems . the fts for langevin systems can be used to measure an unknown averaged power . this idea has been discussed first in the context of electrical circuits @xcite and in ref.@xcite it has been applied for the first time to the measure of the the torque of a molecular motor . we discuss the method in some details in the next subsection . -atpase molecular motor composed by a rotor @xmath272 ( radius @xmath273 nm ) which rotates inside stator of radius @xmath274 nm formed by three @xmath275 subcomplexes . sequential chemical reactions between the stator and the rotor produce the motion . the @xmath276 subcomplexes are attached to a suitably coated glass plate . streptavidin is used to attach to @xmath272 either actin filaments @xcite or streptavidin - coated beads @xcite . b ) in order to follow the rotation of the rotor with a standard microscope a streptavidin - coated bead of radius @xmath277 m is glued to @xmath272 subunit ( drawing not to scale ) . the figures c),d ) and e ) ( taken from ref.@xcite ) ) illustrate the results of a measure . c ) pdfs for several @xmath102 of @xmath278 . d ) symmetry function extracted from the pdfs of c ) . e ) slopes @xmath279 of the symmetry function as a function of @xmath102 . the different colors pertain to different experimental conditions . notice the convergency to an unique value of @xmath280 for large @xmath102 . strictly speaking in this figure the function @xmath178 keeps into account also the fact that eq.[eq : langevin_motor ] is not necessarily valid for short times ( see text ) ] a molecular rotary motor , as any kind of motor , is constituted by a stator and a rotor . the movement of the rotor is provoked by chemical reactions occurring sequentially between the rotor and stator . a typical example of bio - motor is the bacterial flagellum . however it has been shown @xcite that a single molecule of @xmath281-atpase may act as a motor composed by a @xmath272 subunit ( radius @xmath282 nm ) which rotates inside a barrel of radius @xmath283 nm formed by three @xmath275 subcomplexes ( see fig.[fig : motor]a ) for a schematic diagram and ref.@xcite for more details ) . in experiments , the @xmath284 subunit is stuck on a suitably activated glass plate as shown in fig.[fig : motor]a ) . the measure of the torque of this motor is important in order to know its efficiency as a function of the concentration of the chemicals contained in the liquid surrounding it . the typical size of this molecular motor is several nanometers and the moving unit is too small to be observed with an optical microscope . therefore to measure the torque of @xmath281-atpase motor a streptavidin - coated bead of radius @xmath285 m is glued on the subunit @xmath272 , and the motion of this bead is followed by a standard microscope as sketched in fig . [ fig : motor]b ) . the motion of the bead occurs on a torus and @xmath35 is the coordinate of the motion along the torus . the time evolution of @xmath35 can be described by a langevin equation : @xmath286 where @xmath280 is the torque of the motor , @xmath287 the viscous dissipation of the fluid surrounding the bead and @xmath288 the thermal noise . in order to estimate @xmath280 , it is in principle enough to measure the mean angular velocity @xmath289 of the bead and obviously from eq.[eq : langevin_motor ] one gets : @xmath290 however the procedure is not so simple because in order to estimate @xmath280 one has to know the exact value of @xmath287 , which is a function of the viscosity of the fluid , the radius @xmath291 of the bead path , the radius @xmath3 of the bead , the shape of the bead and the distance @xmath292 of the bead from the surface of the glass plate where the experiment is performed ( see fig . [ fig : motor]b ) . the variable @xmath292 is certainly the most difficult to be determined experimentally . therefore using eq.[eq : torque_motor ] the error on @xmath280 can be really very large . there is instead another method proposed in ref.@xcite for electric circuits and first used for measurements in bio - motors in ref.@xcite . this method is based on work fluctuations and is much more precise . to apply it , we suppose that @xmath280 is constant . this is a quite reasonable hypothesis for certain regimes of @xmath281-atpase motors . we compute the work @xmath293 performed by the motor in a time @xmath102 : @xmath294 where @xmath295 and we have used the fact that @xmath280 is constant . in previous section we have seen that langevin systems satisfy the ssft , which we now apply to @xmath293 . using eq.[eq : work_motor ] and the constancy of @xmath280 ssft for the molecular motor reads : @xmath296 this equation is quite interesting because the value of @xmath280 can be determined only by the measure of the fluctuations of @xmath297 . indeed plotting @xmath298 as a function of @xmath299 we notice that the slope of the straight lines is @xmath300 . therefore studying the asymptotic value of this slope for large @xmath102 one can determine @xmath280 . it is interesting to note that in this case the knowledge of @xmath287 is not needed . this technique has been recently applied to molecular motor in ref.@xcite and their main results are plotted in figs . [ fig : motor ] c),d),e ) . the relevant parameter @xmath300 , extracted from the pdf of @xmath278 ( fig . [ fig : motor ] c ) and the symmetry function ( fig . [ fig : motor ] d ) , is plotted in fig . [ fig : motor]e ) . keeps into account the fact that for short time eq.[eq : langevin_motor ] is not a good approximation for the dynamics of the motor and eqs.[eq : work_motor],[eq : ssft_motor ] apply only for long time ] we see a clear convergence to a unique value and one gets a very precise estimation of the torque of the molecular motor independently of the size and shape of the bead glued to @xmath272 unit of the @xmath281-atpase motor . this is a very specific and interesting example of the possible applications of ft . in previous sections we have studied the probability of the instantaneous negative entropy production rates within the context of the fts for stochastic systems , where the fluctuations are produced by the coupling with a thermal bath . in sec . [ sec : gaussian_ft ] we have seen that when the energy injected into the system is larger than @xmath301 the probability of these negative events is very small and the time needed to observe them becomes extremely long . in other words the role of thermal fluctuations becomes negligible . however in the introduction we have shown that instantaneous negative entropy production rates can be observed in chaotic systems such as , for example , turbulence and granular media , where the fluctuations are produced by the non linear interactions of many degrees of freedom . we have also pointed out that for chaotic systems the amount of injected energy is order of magnitudes larger than @xmath302 and of course thermal fluctuations do not play any role in the fluctuating dynamics . the question that we want to analyze in this section is whether we can apply in these systems the fts defined in sec.[section_ft ] for stochastic systems , eqs.[eq : ft]-[eq : fts ] . for a dissipative chaotic system one could imagine to replace @xmath302 , in eqs.[eq : ft]-[eq : fts ] , with a characteristic energy @xmath303 which keeps into account the relevant energy scales of the system fluctuations . however the definition of this relevant energy scale can be in general difficult and even impossible , because it may depend on the observable and on the kind of forcing . thus the approach of introducing an @xmath303 is not very useful to compare the experimental results with the proof given for dynamical systems @xcite . indeed in this case the theorem considers a quantity : @xmath304 where @xmath305 is the instantaneous phase space contraction rate , @xmath306 the integral of @xmath307 on a time @xmath102 and @xmath308 the mean of @xmath306 . three hypothesis has been done on the dynamical system which must be : a ) dissipative , b ) time reversible c ) anosov @xmath309 where @xmath310 is the mean phase space contraction rate , which has the dimension of @xmath311 . in this equation the relevant variable is the phase space contraction rate which has been identified as the entropy production rate@xcite . the phase space contraction rate is a global variable of the system but an extension of the theorem for local variables has been done in refs.@xcite . one reason for developing local ft is that global fluctuations are usually not observable in macroscopic systems , as a consequence it is important to understand whether a local measurement is representative of the dynamics . eq.[ft_dynamical ] has been tested in several numerical simulations ( see for example ref.@xcite for a review ) , here we want to focus on experiments . the test of eq.[ft_dynamical ] in experiments is extremely useful to analyze several important questions . the first one is whether eq.[ft_dynamical ] may have a more general validity independently of the restrictive hypothesis done to prove it . indeed the hypothesis b ) is never satisfied in real systems and the hypothesis c ) does not necessarily apply to all of them . thus in general we do not even know whether eq.[ft_dynamical ] can be applied in the experimental system under study . the second question concerns the choice of the observable . indeed the direct measure of the phase space contraction rate is not possible and one has to rely upon the measure of another observable usually the energy @xmath82 injected into the system by the external forces in a time @xmath102 . in other words one is making the important hypothesis that @xmath312 , defined in eq.[eq : y_tau ] in terms of @xmath305 is equivalent to @xmath313 . this hypothesis , that is not necessarily valid , is the second question that one would like to address in experiments . the third question is related to the estimation of the prefactor @xmath314 in the right hand side of eq.[ft_dynamical ] . this prefactor , which is a function of the lyapunov exponents , is very difficult to estimate in an experimental system . finally the last question concerns the relevance of a local observable to characterize the dynamics of the system . there are not many experiments where these questions have been analyzed in some details . in several experiments @xcite only the linearity in @xmath115 of the symmetry function @xmath315 has been checked , which , for the reasons discussed in the previous paragraph , is only a partial test . for example three experiments have tried to give an answer to the question of the prefactor . two of these experiments are performed in granular media @xcite and the third on mechanical wave turbulence in a metallic plate@xcite . we will not describe in details the experiments here but we will comment the main results . the two experiments of refs.@xcite consist of diluted granular media strongly shaken by a vibrator , but the measured quantities are not the same . in ref.@xcite the authors measure the fluctuations of the energy flux in a subvolume of the system . instead ref.@xcite the work done by an external force on a ratchet inside the granular media is measured . they both find that although the system is not thermal the stochastic version of ssft ( eq.[eq : ft ] ) holds , provided that @xmath302 is replaced by a characteristic energy @xmath303 : @xmath316 with @xmath317 , @xmath116 is the integral of energy flux in ref.@xcite and the work @xmath82 performed by an external force on a ratchet in ref.@xcite . it must be pointed out that in both experiments of refs.@xcite the energy @xmath303 has been measured independently . in ref.@xcite is found that @xmath303 is about 5 times larger than the kinetic energy @xmath318 of the shaken granular medium for all the values of the control parameters used in the experiment . the fact that @xmath318 and @xmath303 have the same dependence on the control parameter have been interpreted considering that the vibrator injects into the system the amount of energy lost in the collisions but once excited in a ness the granular medium behaves like a thermal bath for the measured observable . however this interpretation is not necessarily correct . indeed for the experiment of ref.@xcite it as been shown in a numerical simulation @xcite that for the quantity measured in this experiment ft does not apply for the large deviations @xmath115 because the symmetry function @xmath154 becomes non - linear for large @xmath115 . this discrepancy between theory and experiment is obviously coming from the fact that experimentally the very large deviation are difficult to be measured , thus the non - linear part of @xmath154 can not be observed . however the experiment of ref.@xcite is certainly interesting because is the first where the question of the prefactor has been analyzed experimentally . in contrast for the experiment of ref.@xcite is observed that @xmath319 where @xmath320 is the restitution coefficient of the grains . it is interesting to notice that @xmath321 is the temperature of an intruder inside a diluted granular gas as it has been found in theoretical models @xcite . these two examples of comparison between numerical and experimental results show the difficulty of interpreting the experimental results on ft and the importance of verifying them in a precise theoretical framework . where does the difference between the experiments of ref.@xcite and ref.@xcite come from ? the answer can be found on the fact that in the two experiments two different quantities are measured . indeed in sect.[section_ft ] we have seen that , even for stochastic systems , the fluctuations of @xmath82 , @xmath105 and @xmath213 behave differently within the context of ft . this can be more complex for granular media and it will be useful to give more insight on this point . furthermore even in cases where a description in terms of @xmath322 applies , comparing eq.[ft_granular ] with eq.[ft_dynamical ] a question that arises naturally is whether @xmath323 is a good estimation of @xmath324 . this is an important question which will be interesting to analyze in the future . in the experiment on mechanical waves@xcite , a metallic plate is set into a chaotic state of wave turbulence by a periodic local forcing at @xmath325 hz . ( see fig . [ fig : cadot]a ) . the chaotic dynamics is produced by the non - linear interaction of the oscillatory modes of the plate . the authors measure the local force and displacement ( see fig . [ fig : cadot]a ) and compute the work @xmath82 done on a time @xmath102 by the external force which excites the vibrations of the plate . they find that the pdfs of @xmath313 are strongly non - gaussian ( see fig . [ fig : cadot]b ) . from these pdfs they compute the symmetry function @xmath326 which is plotted ( see fig . [ fig : cadot]c ) as a function of @xmath115 . we see that in spite of the fact that the pdf are not gaussian the function @xmath326 ( fig . [ fig : cadot]c ) converges to a unique straight line for large @xmath102 as predicted by ft . from eq.[ft_dynamical ] the slope of this straight line is @xmath310 , which the authors can estimate independently by measuring the relaxation time of the vibrational modes . they find that the values estimated with the two methods ( ft and the relaxation time ) are very close and within experimental errors . this result is quite interesting and it is probably the only experiment where a direct test of eq.[ft_dynamical ] has been done . certainly the errors of this comparison are very large but this kind of tests are useful to understand in some details the applications of fts to chaotic systems . for @xmath327 ) ( c ) functions @xmath315 obtained from the pdfs of ( b ) . inset : compensated value @xmath328 in a semi log plot . in ( c ) the dashed line corresponds to a linear law of slope @xmath329 . ] in this paper we have reviewed several experimental results on the fluctuations of injected and dissipated power in out of equilibrium systems . we considered the two cases when the fluctuations are produced by the coupling with the heat bath ( stochastic systems ) and when they are produced by the non linear interactions of many degrees of freedom ( chaotic systems ) . we have seen that in both cases we observe that the external forces may produce a negative work because of fluctuations . the probability of these negative events has been analyzed in the framework of fluctuation theorem . we have mainly discussed the stochastic systems described by langevin equations , both with harmonic and unharmonic potential . we have seen that injected and dissipated power present different behaviors . fts are valid for any value of @xmath82 whereas can be applied only for @xmath330 in the case of the heat . we have also seen the the finite time corrections to ssft depend on the driving and on the properties of the system . we have introduced the total entropy , which takes into account only the entropy produced by the external forces neglecting the the equilibrium fluctuations . for the total entropy fts are valid for all the times . we discussed the applications of fts to extract important physical properties of a stochastic system . thus one may conclude that for markovian systems driven by a deterministic force the applications of ft does not present any major problems and can be safely applied . the case of random driving has been recently discussed and several problems may arise when the variance of the driving become larger than the fluctuations induced by the thermal bath . we have not discussed this problem but an analysis of this specific case can be found in refs.@xcite . finally we discussed the applications of ft to chaotic systems . the experimental test is in this case very important and useful because many questions can be asked on the system under study which does not necessarily verify all the theoretical hypothesis . one has to say that in the case of non - gaussian statistics even the linearity of the symmetry function can be an interesting result . however we pointed out that , for a real comparison with theory , the difficulty is to estimate of the prefactor of eq.[ft_dynamical ] by an independent measurement . only a few experiments have addressed this point in some details , but many problems remain open and it seems to be difficult to find a general behavior for chaotic systems as for the case of stochastic ones . several results reviewed in this article has been obtained in collaboration , with l. bellon , r. gomez - solano , n. garnier , p. jop . we acknowledge useful discussions and collaboration during the past ten years with d. andrieux , e. cohen , r. chetrite , g. gallavotti , p. gaspard , k. gawedzky and a. imparato . we thank a. puglisi and e trizac for useful comments on ft in granular media . we are grateful to h. noji , k. hayashi and a boudauoud for having allowed us to reproduce the figures of their articles in the present one . widest - label c.w . chang , d. okawa , h. garcia , a. majumdar and a. zettl , phys . lett.101 , 075903 ( 2008 ) r. van zon , s. ciliberto , e. g. d. cohen , phys . 92 ( 13 ) 130601 ( 2004 ) n. garnier , s. ciliberto , phys . e 71 060101(r ) ( 2005 ) . s. ciliberto , n. garnier , s. hernandez , c. lacpatia j. -f.pinton , g. ruiz chavarria , physica a 340 , 240 ( 2004 ) . k. feitosa , n. menon , phys . 92 164301 ( 2004 ) . s. joubaud , talk at statphys24 , ( cairns 2010 ) . abstract , topic 3 : http://www.statphys.org.au/. a. kis andras and a. zettl , phil.trans . of the royal society a : mathematical , physical and engineering sciences , 366 , 1591 , ( 2008 ) . j. r. gomez - solano , l. bellon , a. petrosyan and s. ciliberto , europhys . lett . , 89 , 60003 ( 2010 ) . l.bellon , l. buisson , s. ciliberto , f. vittoz , rev . instrum . , 73 ( 9 ) , 3286,(2002 ) . l. bellon and s. ciliberto , physica d 168 , 325 ( 2002 ) s.joubaud , n. b. garnier , f. douarche , a.petrosyan , s. ciliberto , c.r . physique , special issue : work , dissipation , and fluctuations in nonequilibrium physics , editors : b. derrida , p. gaspard , c. van den broeck . ( elsevier 2007 ) ; cond - mat/0703695 d. j. evans and d. j. searles , phys . rev . e 50 , 1645 ( 1994 ) . d. j. evans , d. j. searles , advances in physics 51 ( 7 ) 1529 ( 2002 ) . d. j. evans , d. j. searles and l. rondoni , phys . e 71 ( 5 ) 056120 ( 2005 ) . g. m. wang , e. m. sevick , e. mittag , d. j. searles , d. j.evans , phys . rev . . 89 050601 ( 2002 ) ; g. m. wang , j. c. reid , d. m. carberry , d. r. m. williams , e. m. sevick , d. j. evans , phys . rev . e 71 046142 ( 2005 ) .

the existence of gravitational waves is an unambiguous prediction of the theory of general relativity @xcite . yet , despite efforts originating in the early nineteen sixties , gravitational waves have not been detected directly . nevertheless , observation of the binary pulsar psr 1913 + 16 convinces us that gravitational waves do exist , and that they are correctly described by einstein s theory @xcite . that gravitational waves have not yet been detected on earth is simply due to their incredible weakness : typical waves would produce in a bulk of matter a strain @xmath0 , where @xmath1 is the extension of the matter , of order @xmath2 @xcite . needless to say , to measure this effect is a great challenge for experimentalists . there is reasonable hope that gravitational waves will be detected within the next ten years , thanks to a new generation of detectors which use interferometry to monitor the small displacements induced by the passage of a gravitational wave . two groups are currently involved in building large - scale interferometers : the american ligo team , and the french - italian virgo team . the ligo ( laser interferometer gravitational - wave observatory ) project @xcite involves two detectors , to be built in hanford , washington , and in livingston , louisiana . construction has begun at both sites . each interferometer has an armlength of approximately 4 km . ligo should be completed by the turn of the century . the virgo ( so named after the galaxy cluster ) project @xcite involves a single interferometer , to be built near pisa , italy , with an armlength of approximately 3 km . virgo should also be completed by the turn of the century . the basic idea behind interferometric detectors is the following @xcite : the interferometer is composed of two long ( 4 km for ligo ) vacuum pipes forming the letter * l*. a laser beam is split in two at the corner of the * l * , and is sent into each arm of the interferometer . each beam then bounces off a mass which is suspended at each end of the * l * ( a mirror has been coated onto each mass ) . the light is finally recombined at the beam splitter , and its intensity is measured by a photodiode . when no gravitational wave is present at the interferometer , the length of each arm is so adjusted that when measured by the photodiode , the light s intensity is precisely zero ( the recombined beams are arranged to be precisely out of phase ) . however , when a gravitational wave passes through the interferometer , the armlengths are no longer constant , and the recombined beams no longer precisely out of phase . more precisely , during the first half of its cycle the gravitational wave increases the length of one arm , and decreases the length of the other . during the second half cycle , the first arm is now shorter , and the second arm longer . the light s intensity therefore oscillates with the gravitational - wave frequency . the intensity is a measure of @xmath3 , where @xmath1 denotes the interferometer armlength , and @xmath4 the gravitational - wave field . interferometers are subject to various sources of noise which limit the detector s sensitivity to gravitational waves . the relative importance of each source depends on the frequency at which the interferometer oscillates @xcite . at low frequencies ( @xmath5 ) the detector s sensitivity is limited by seismic noise , which is due to the earth s seismic activity . at frequencies larger than 10 hz the seismic noise can be eliminated with sophisticated isolation stacks ; these fail at low frequencies . at high frequencies ( @xmath6 ) the detector s sensitivity is limited by photon shot noise , which is due to statistical errors in the counting of photons by the photodiode . this source of noise can be reduced by increasing the laser power , or making use of `` light recycling '' @xcite . at intermediate frequencies ( @xmath7 between 10 hz and 100 hz ) the noise is dominated by thermal noise , which is due to spurious motions of thermal origin . for example , the suspended masses are thermally excited and vibrate with their normal - mode frequencies ; this evidently affects the recombined laser beam . interferometers are therefore broad - band detectors , with good sensitivity in the range @xcite @xmath8 the required sensitivity for full - scale interferometers is approximately @xmath9 at peak sensitivity a tall order . [ the subscript @xmath10 stands for `` noise level '' ; we will define @xmath11 precisely below . a plot of @xmath11 , appropriate for an interferometric detector with `` advanced '' sensitivity , is given in fig . 1 . ] for comparison , we may mention that the caltech 40 m prototype has already achieved @xmath12 at peak sensitivity @xmath13 . it is not implausible that improved technology and a factor of 100 in armlength will permit to reach the desired goal . the detector noise can be measured when no gravitational wave is present at the interferometer , the typical situation . then the detector output @xmath14 is given by noise alone : @xmath15 where @xmath16 represents the noise . the noise is a random process @xcite : the function @xmath16 takes purely random values . consequently , the noise can only be studied using statistical methods . in the presence of a gravitational wave , eq . ( [ 2 ] ) must be replaced by @xmath17 , where @xmath18 is the gravitational - wave field . the statistical properties of the noise can be determined by careful measurement . for example , the time average @xmath19 can be constructed . this mean value can then be subtracted from @xmath16 and , without loss of generality , we can put @xmath20 . also from measurements , the noise s _ autocorrelation function _ @xmath21 can be constructed : @xmath22 @xmath23 gives the mean squared deviation of the noise with respect to the mean value . in the following we will assume that the noise is _ stationary _ , in the sense that its statistical properties do not depend on time @xcite . this means , in particular , that the autocorrelation function does not depend explicitly on the origin of time @xmath24 , but only on the variable @xmath25 , as was expressed in eq . ( [ 4 ] ) . we shall also assume that the noise satisfies the _ ergodic hypothesis _ , so that time averages can be replaced with ensemble averages @xcite . here , the noise is imagined to be drawn from a representative ensemble , and the probability that it takes a particular realization @xmath16 is given by a specified probability distribution . the statistical properties of the noise then refer to this ( infinite dimensional ) distribution function . in full generality , the statistical properties of the noise can only be summarized by constructing all the higher moments @xmath26 . if , however , the noise is assumed to be gaussian , in the sense that its probability distribution function is an infinite dimensional gaussian distribution @xcite , then the autocorrelation function contains all the information . real detector noise is neither strictly stationary nor strictly gaussian . however , on a timescale of hours , which is long compared with typical gravitational - wave bursts , the noise appears stationary to a good approximation @xcite . and non - gaussian components to the noise can be removed , to a large extent , by cross - correlating detector outputs from two widely separated interferometers @xcite . it is therefore a satisfactory approximation to take the noise to be stationary and gaussian . under these assumptions the statistical properties of the detector noise are fully summarized by @xmath21 . it is convenient to work instead in the frequency domain , and to define @xcite the noise s _ spectral density _ @xmath27 as @xmath28 the spectral density is defined for @xmath29 only ; as @xmath21 is a real and even function , the negative frequencies only duplicate the information contained in the positive frequencies . as @xmath16 is dimensionless , the spectral density has dimensions of time . by multiplying @xmath27 with the frequency and taking the square root ( since the spectral density represents the mean squared noise ) , one obtains the _ noise level _ @xmath11 : @xmath30 this gives the equivalent gravitational - wave amplitude which would make the interferometer oscillate at just the noise level @xcite ; this quantity was introduced in the preceding section ( see fig . 1 ) . coalescing compact binaries , composed of neutron stars and/or black holes , are the most promising source of gravitational waves for interferometric detectors @xcite . consider the binary pulsar psr 1913 + 16 @xcite . this system consists of two neutron stars , each of @xmath31 , in orbital motion around each other . its present orbital period @xmath32 is approximately 8 hours , corresponding to orbital separations of about @xmath33 times the total mass . ( here and throughout we use units such that @xmath34 . ) its present eccentricity is approximately equal to 0.6 . the binary s orbital period is observed to decay at a rate @xmath35 corresponding precisely to a loss of energy and angular momentum to gravitational waves @xcite . in a timescale of approximately @xmath36 years the orbital period will have decreased to less than a tenth of a second , corresponding to orbital separations smaller than one hundred times the total mass . in this time , the eccentricity will have been reduced to extremely small values ( by the radiation reaction ) , so that the orbits are practically circular . the gravitational waves produced then have a frequency larger than 10 hz , and the frequency keeps increasing as the system evolves . _ during this late stage of orbital evolution , the gravitational waves sweep through the frequency bandwith of interferometric detectors , and thus become visible . _ by the time the orbital separation becomes as small as a few times the total mass , the neutron stars begin to merge . the gravitational waves produced during the final merger can not be detected by interferometric detectors , at least in the broad - band configuration described above @xcite : the gravitational - wave frequency is then larger than 1 000 hz , for which the detector noise is large . of course , it would be foolish to wait @xmath36 years in order for psr 1913 + 16 to produce gravitational waves with appropriate frequency . fortunately , interferometric detectors will be sensitive enough to monitor binary coalescences occurring in quite a large volume of the universe , approximately @xmath37 ( corresponding to a radius of 200 mpc @xcite ) . it has been estimated @xcite that as many as 100 coalescences could occur every year in such a volume ( this includes coalescences of black - hole systems as well ) . this potentially large event rate is one of the factors that make gravitational waves from coalescing compact binaries especially promising . the other factor comes from the fact that compact binaries are extremely clean astrophysical systems . it can be estimated @xcite that tidal interactions between the two stars are completely negligible , up to the point where the objects are about to merge . in particular , no mass transfer occurs . the system can therefore be modeled , to extremely good accuracy , as that of two point masses with a limited number of internal properties ( such as mass , spin , and quadrupole moment ) . the challenge in modeling coalescing compact binaries resides in formulating and solving the equations of motion and wave generation for a general relativistic two - body problem @xcite . at the crudest level , the gravitational - wave signal corresponding to an inspiraling binary system can be calculated by ( i ) assuming that the orbital motion is newtonian ( with the effects of radiation reaction incorporated ) and ( ii ) using the standard quadrupole formula for wave generation @xcite . as motivated above , we may also assume that the orbits are circular . we define @xmath18 to be the gravitational - wave signal . this is given by a linear combination , appropriate for interferometric detectors , of the two fundamental polarizations , @xmath38 and @xmath39 , of the gravitational - wave field . at this level of approximation , the signal is given by @xcite @xmath40 as expected , the signal decays as the inverse power of @xmath41 , the distance to the source . in eq . ( [ 7 ] ) , @xmath42 is a function of all the angles relevant to the problem : position of the source in the sky , orientation of the orbital plane , position and orientation of the detector on earth . the parameter @xmath43 is called the _ chirp mass _ and represents a particular combination of the masses , given by @xmath44 the waveform depends on the chirp mass only , and not on any other combination of the masses . the symbol @xmath7 represents the gravitational - wave frequency , which is equal to _ twice _ the orbital frequency . because the system loses energy and angular momentum to gravitational waves , the frequency is not constant , but increases in time according to @xcite @xmath45 as a consequence , eq . ( [ 7 ] ) shows that the amplitude of the signal , which is proportional to @xmath46 , also increases with time . a signal which increases both in frequency and in amplitude is known as a _ chirp _ , and this is the origin of the term `` chirp mass '' . finally , the phase function @xmath47 is given by @xmath48 because the frequency is not a constant , the phase accumulates nonlinearly with time . for a system of two neutron stars , the gravitational - wave signal undergoes approximately 16 000 oscillations as it sweeps through the frequency bandwith of an interferometric detector . the timescale for the frequency sweep is approximately 15 minutes . the orbital separation ranges from approximately 180 to 10 times the total mass @xmath49 , and the orbital velocity @xmath50 ranges from approximately 0.1 to 0.4 . this indicates that relativistic corrections must be inserted in eqs . ( [ 7 ] ) and ( [ 9 ] ) in order to obtain a satisfactory degree of accuracy . how does one go about finding a gravitational - wave signal in a noisy data stream , when typically the signal is not very strong ? and once the signal is found , how does one go about extracting the information it contains ? for signals of precisely known form , one goes about this using the technique of _ matched filtering _ signals from inspiraling compact binaries , since they can be calculated with high precision , belong to this class . the basic idea behind matched filtering is to use our knowledge about the signal in order to go find it in the data stream , after the noisy frequencies have been filtered out . suppose a signal of known form @xmath51 is present in the data stream . here , the vector @xmath52 collectively denotes all the parameters characterizing the signal . in the case of inspiraling binaries , these would be the time of arrival , the initial phase , the distance to the source , the position angles , the chirp mass , and other parameters to be introduced in sec . the detector output is given by @xmath53 where @xmath16 is the noise , whose statistical properties are fully summarized by the spectral density @xmath27 , as discussed in sec . the first step in matched filtering @xcite is to pass @xmath54 through a linear filter which removes the noisy frequencies . the idea here is to use our knowledge about the detector noise contained in @xmath27 in order to discard that part of the detector output for which the detector noise is large . the output is imagined to be decomposed into fourier modes according to @xmath55 ; the filter suppresses the modes @xmath56 for which @xmath27 is large . in mathematical terminology , a linear filter is a linear operation on a function @xmath54 . this operation can always be written as @xcite @xmath57 where @xmath58 is the filter function . the filter which removes the noisy frequencies is the one such that @xmath59 the factor of 2 is conventional . the next step in matched filtering consists of computing the overlap integral between the filtered output ( [ 13 ] ) and the known signal @xmath60 . the true value @xmath52 of the source parameters is not known prior to the measurement . this operation must therefore be repeated for a large number of trial values @xmath61 ; the corresponding signals @xmath60 are known as _ templates_. the overlap integral defines the function @xmath62 given by @xmath63 to obtain the second line we have inserted the fourier decompositions of @xmath4 , @xmath64 , and @xmath65 , and carried out the integrations over time . the inner product @xmath66 is defined as @xmath67 where @xmath68 and @xmath69 are arbitrary functions of time , with fourier transforms @xmath70 and @xmath71 . a quantity analogous to @xmath62 can be defined for the noise alone , by replacing @xmath65 to the right of eq . ( [ 15 ] ) by @xmath10 . operationally , this amounts to filtering the detector output when a gravitational - wave signal is known _ not _ to be present . because the noise is a random function , the integrals @xmath72 are random also . and because the noise is assumed to have zero mean , to take an average over all possible realizations of the noise ( by repeating the measurements many times ) would yield a zero value . none of these quantities would be especially useful . we are therefore led to consider the root - mean - square average of these integrals , @xmath73 to go from the first to the second line requires some machinery which will not be presented in this review . we refer the reader to ref . @xcite for the missing steps . the _ signal - to - noise ratio _ is defined to be the ratio of @xmath74 over @xmath75 : @xmath76 it is clear that @xmath77 is a random function , since @xmath54 is itself a random function , being the superposition of signal plus noise . its expectation value , or average over all possible realizations of the noise , is zero in the absence of signal ( since @xmath20 ) , and @xmath78 in the presence of the signal @xmath51 . it is important to notice that in eq . ( [ 19 ] ) , the numerator is the overlap integral between the true signal @xmath51 and the templates @xmath60 . in the absence or presence of a signal , the variance in the signal - to - noise ratio is precisely equal to unity @xcite , independently of the values of @xmath52 and @xmath61 . this fully summarizes the statistical properties of the signal - to - noise ratio , since @xmath77 is a gaussian random function . [ this can be seen from the fact that @xmath54 itself is gaussian , being the superposition of signal plus gaussian noise . ] it is intuitively clear that choosing a template with @xmath79 will produce the largest possible expectation value of the signal - to - noise ratio . this statement , known as _ wiener s theorem _ @xcite , can easily be shown to be true by applying schwarz s inequality to the right - hand side of eq . ( [ 19 ] ) . we have @xmath80 the fact that the signal - to - noise ratio has a variance of unity indicates that a signal can be concluded to be present only if @xmath81 is significantly larger than 1 . we will come back to this point in the next section . the true value of the source parameters can therefore be determined by maximizing @xmath82 over all possible values of the trial parameters . the fact that the signal - to - noise ratio has a variance of unity implies that this determination can only have a limited degree of accuracy . the statistical errors decrease with increasing @xmath81 ; since this is proportional to the signal s amplitude , a stronger signal gives better accuracy . maximizing the signal - to - noise ratio is essentially equivalent to maximizing the overlap integral @xmath83 it is easy to see how the choice of the parameters @xmath61 can affect the overlap integral . consider a toy waveform with three parameters : arrival time , initial phase , and chirp mass . a mismatch in the arrival times clearly reduces the overlap integral : the signal and the template , taken to be functions of time , might have support in entirely different regions of the time axis , leading to a vanishing overlap . supposing that the arrival times are matched , a mismatch in the initial phases can also reduce the overlap integral : the signal and the template might be out of phase with each other , leading to an oscillating integrand and a vanishingly small overlap . finally , supposing that both the arrival times and initial phases are matched , a mismatch in the chirp masses would also reduce the overlap integral . this is because the chirp mass governs the rate at which the signal s frequency changes with time ; cf . ( [ 9 ] ) . signal and template , starting at the same time with the same phase , might thereafter go out of phase , thereby reducing the overlap . the first order of business when analysing the output of a gravitational - wave detector is to decide whether or not a signal is present . here we assume that the signal must be of a specific form , corresponding to a coalescing binary system . in this section we discuss signal detection how the technique of matched filtering can be employed to find the signal in the noisy data stream . in the next section we will discuss signal measurement how matched filtering is used to estimate the value of the source parameters once the signal has been found . as mentioned in the previous section , a signal can be concluded to be present if the maximum value of the signal - to - noise ratio , @xmath84 , is significantly larger than unity . in fact , there exists a threshold value @xmath85 such that a signal is concluded to present , with a certain confidence level , if @xmath86 . to figure out how large this threshold must be is a standard application of the statistical theory of signal detection @xcite , which was developed largely for the purpose of detecting radar signals . the theory can easily be taken over to the case gravitational - wave signals @xcite . to go into the detail of this theory would be outside the scope of this review . we shall simply state that @xmath85 is fixed by selecting a small , acceptable value for the probability that a signal would falsely be concluded to be present the false alarm probability . ( this is the neyman - pearson criterion @xcite , which is more precisely formulated in terms of the likelihood ratio , the ratio of the probability that a signal is present to the probability that it is absent . ) once @xmath85 is fixed , the level of confidence that a signal is indeed present increases with @xmath86 . a typical ballpark value for the threshold is @xmath87 . in the preceding paragraph it was assumed that the signal - to - noise ratio @xmath88 is computed using template waveforms which are functions of the parameters @xmath52 introduced in sec . these parameters have direct physical meaning ; they include the chirp mass and other meaningful parameters to be introduced in sec . 12 . however , since these parameters are _ not _ estimated during the detection stage of the data analysis ( they are estimated only _ after _ a signal has been found ) , there is no particular need to parametrize the templates with @xmath52 . in fact , it may be desirable , in order to minimize the computational effort , to parametrize the signal in a completely different way . the new parameters , @xmath89 , would then have no particular physical significance . what is required is that the new templates , @xmath90 , reproduce the behaviour of the expected gravitational - wave signal . in other words , the templates @xmath51 and @xmath90 should span the same `` signal space '' , but @xmath90 should do so most efficiently . unlike @xmath51 , @xmath90 need not be derived from the field equations of general relativity . with these new _ detection templates _ , the signal - to - noise ratio is defined as @xmath91 where @xmath92 is defined as in eq . ( [ 18 ] ) . as before , a signal is concluded to be present if @xmath93 . once the detection templates @xmath90 have been used to conclude that a signal is present in the data stream , they are replaced with the _ measurement templates _ @xmath60 in order to estimate the value of the physical parameters @xmath52 . the procedure to estimate the source parameters was explained in sec . as was mentioned , the idea is to maximize over all possible values of the trial parameters the overlap integral @xmath94 we already have discussed the effect of a mismatch in the value of the parameters . what remains to be discussed is the effect of a mismatch in the functional form of the template with respect to that of the true waveform . the true waveform is governed by the exact laws of general relativity . the template , on the other hand , is necessarily constructed using an approximation to the exact laws . ( that approximations must be made in analytic calculations is obvious ; in a numerical treatment the approximation resides in the finite differencing of the field equations . ) this , clearly , must have an effect on our strategy for extracting the information contained in the gravitational - wave signal . this can be seen simply from the fact that wiener s theorem , as stated in sec . 8 , strictly requires signal and template to have the same functional form ; they are allowed to differ only in the value of their parameters . suppose that gravitational waves coming from a given source are received without noise , so that the true signal @xmath51 is measured accurately . suppose also that the true value @xmath52 of the source parameters is known ( god has spoken ) . then a computation of the overlap integral with @xmath79 but with an approximate template @xmath60 will not yield the maximum possible value for @xmath77 . this is because the approximation differs from the true signal , both in amplitude and in phase . since both signal and template undergo a large number of oscillations ( recall that for a system of two neutron stars , this number is approximately 16 000 ) , the overlap integral is most sensitive to phase differences : a slight phase lag causes the integrand to oscillate , thereby severely reducing the signal - to - noise ratio with respect to its maximum possible value . a gravitational - wave astronomer does nt know before the measurement the true value of the source parameters , and must work with an approximation to the true general - relativistic waveform . we have seen that the phase lag occurring between the true signal and the approximate template when the parameters are matched reduces the signal - to - noise ratio from its maximum possible value . it follows that maximizing @xmath77 with approximate templates introduce _ systematic errors _ into the estimation of the source parameters : evaluating the signal - to - noise ratio with @xmath95 will return , for some @xmath96 , a number larger than @xmath97 . the systematic errors are precisely the value of @xmath96 for which the signal - to - noise ratio is largest . if the templates are a poor approximation to the true signal , then the systematic errors will be larger than the statistical errors arising because @xmath77 is a random function ( see sec . we therefore appreciate the need for constructing measurement templates which are as accurate as possible , especially in phase @xcite . the requirement is that the systematic errors in the estimated parameters must be smaller than the statistical errors . an estimate for the required degree of accuracy comes from the observation that the overlap integral will be significantly reduced if the template loses phase by as much as one wave cycle with respect to the true signal . since the total number of wave cycles is approximately 16 000 , we have @xmath98 since the orbital velocity @xmath99 is of order @xmath100 when the gravitational - wave frequency is in the relevant bandwidth , relativistic corrections _ at least _ of order @xmath101 are required to improve the quadrupole - formula expression given in sec . as we shall see , this is an underestimate . we have seen that accurate measurement templates are required to make the most of the gravitational - wave signals originating from coalescing compact binaries . we also have seen that the quadrupole - formula waveform , eq . ( [ 7 ] ) , is not sufficiently accurate ; relativistic corrections at least of order @xmath101 are required . how does one go about calculating these ? a possible line of approach is to use a slow - motion approximation to the equations of general relativity . this is based on the requirement that if @xmath99 is a typical velocity inside the matter source , then @xmath102 we shall call this approximation `` post - newtonian theory '' @xcite . we point out that for binary systems , post - newtonian theory makes no assumption regarding the relative size of the two masses . this is to be contrasted with the perturbation approach , discussed in sec . 13 , in which the mass ratio is assumed to be small , but no restriction is put on @xmath99 . in post - newtonian theory @xcite , one starts by defining fields @xmath103 as @xmath104 where @xmath105 is the inverse of the true metric @xmath106 with determinant @xmath107 , and @xmath108 is the metric of minkowski spacetime . when the harmonic gauge conditions @xmath109 are imposed , the _ exact _ einstein field equations reduce to @xmath110 here , @xmath111 is the flat - spacetime wave operator , @xmath112 is the stress - energy tensor of the source , and @xmath113 is nonlinear in @xmath103 and represents an effective stress - energy tensor for the gravitational field . the detailed way in which one solves these equations is quite complicated , and will not be described here . the essential ideas are these @xcite : one first integrates the equations in the near zone ( @xmath114 , where @xmath41 is the flat - space radius and @xmath115 the gravitational wavelength ) assuming slow motion , or @xmath116 . one does this by iterations : the nonlinear terms in eq . ( [ 26 ] ) are first neglected , and the resulting linear equations integrated . these solutions are then used as input for the next iteration . this process is continued until the desired degree of accuracy is obtained . this is the standard post - newtonian approach @xcite . one next integrates the equations everywhere in the vacuum region outside the source . this is done once again by iterations , assuming @xmath117 , but assuming nothing about the relative size of @xmath118 with respect to the spatial derivatives . this is because the vacuum region contains the wave zone , in which the field propagates with the speed of light ; a slow - motion assumption would therefore not do for the field itself . this is the post - minkowskian approach @xcite . using the post - minkowskian approach one constructs , by successive approximations , the most general solution to the einstein equations outside the source this is characterized by two infinite sets of arbitrary multipole moments @xcite , the mass moments @xmath119 and the current moments @xmath120 , were @xmath121 and @xmath122 are the standard spherical - harmonic indices . ( in practice , the fields @xmath103 are equivalently expressed in terms of symmetric - trace - free moments , not spherical - harmonic moments . ) the general solution is then matched to the near - zone solution in the region of common validity , and the multipole moments are thus determined . finally , one expresses the radiation field the time - varying , @xmath123 part of the gravitational field in terms of the derivatives of the mass and current multipole moments @xcite . this gives the gravitational waveform . the gravitational - wave luminosity @xmath124 can also be obtained from the radiation field . to date , the post - newtonian calculation of the waveform has been carried out accurately through second post - newtonian order @xmath125 beyond the leading - order , quadrupole - formula expressions given in sec . the complete waveform will not be displayed here . instead , we will focus solely on the waveform s _ phasing_. the phasing of the waves can be determined from @xmath126 , the rate of change of the gravitational - wave frequency . this can be expressed as @xmath127 where @xmath124 is ( minus ) the gravitational - wave luminosity , and @xmath128 relates orbital energy to orbital frequency ( @xmath7 is twice the orbital frequency ; the orbits are assumed to be circular ) . both these quantities can be expanded in powers of @xmath129 with leading - order terms @xcite @xmath130 where the subscript @xmath75 stands for `` newtonian '' . we have introduced the reduced mass @xmath131 , the total mass @xmath132 , and the mass ratio @xmath133 as @xmath134 notice that @xmath133 is restricted to the interval @xmath135 , with @xmath136 for @xmath137 . it is easy to check that eqs . ( [ 27 ] ) and ( [ 29 ] ) reproduce eq . ( [ 9 ] ) above ; the chirp mass can be expressed as @xmath138 . for simplicity we will , in the following , focus on the quantity @xmath124 . to second post - newtonian order , the luminosity takes the form @xmath139 . \label{31}\ ] ] here , the terms so and ss are due to spin - orbit and spin - spin interactions , respectively @xcite ; these occur if the masses @xmath140 and @xmath141 are rotating . let @xmath142 and @xmath143 be the spin angular momentum of each mass , and define the dimensionless quantities @xmath144 , for @xmath145 . let also @xmath146 be the direction of orbital angular momentum . then @xcite @xmath147 \hat{l } \cdot \vec{\chi}_a \label{32}\ ] ] is the spin - orbit term , and @xmath148 is the spin - spin term . in eq . ( [ 31 ] ) , the leading - order term was first calculated in 1963 by peters and mathews @xcite using the standard quadrupole formula . the first post - newtonian correction , at order @xmath149 , was calculated in 1976 by wagoner and will @xcite . the @xmath150 term is due to wave propagation effects : as the waves propagate out of the near zone they are scattered by the curvature of spacetime , and this modifies both the amplitude and the phase of the waveform ; this `` tail term '' was first calculated in 1993 by this author @xcite , and then independently by wiseman @xcite and blanchet and schfer @xcite . the second post - newtonian correction , at order @xmath101 , was calculated in 1994 by blanchet , damour , iyer , will , and wiseman @xcite . finally , the spin - orbit and spin - spin corrections were calculated in 1993 by kidder , will , and wiseman @xcite . we see from eq . ( [ 31 ] ) that the post - newtonian corrections bring a number of additional source parameters into the picture . the waveform no longer depends uniquely upon the chirp mass @xmath43 ; it now depends upon the masses @xmath140 and @xmath141 separately , and upon the spin - orbit and spin - spin parameters ( which stay approximately constant as the system evolves toward coalescence ) . these new parameters must be included into @xmath52 when the signal is analyzed using matched filtering . what we have at this point is an expression for the waveform which is accurate to second post - newtonian order . the question facing us is whether this waveform is sufficiently accurate to be used as measurement templates . in other words , are the systematic errors generated by these templates guaranteed to be smaller than the statistical errors ? evidently , to answer this question is difficult , since we do not have access to the exact waveform in order to make comparisons . in the next section we will consider a model problem for which the waveform _ can _ be calculated exactly , thereby enabling us to judge the accuracy of the post - newtonian expansion . we will find , in sec . 15 , that the answer to this question is , most likely , no : the second post - newtonian waveform is not sufficiently accurate . a different line of attack for solving einstein s equations for a compact binary system is to assume that one of the bodies is very much less massive than the other @xcite . we therefore demand @xmath151 where @xmath131 is the reduced mass and @xmath132 the total mass . in this limit @xmath131 is practically equal to the smaller mass @xmath140 , and @xmath132 is practically equal to the larger mass @xmath141 . in contrast with the post - newtonian approach , we assume _ nothing _ about the size of the velocity @xmath99 . this approach takes advantage of the fact that when eq . ( [ 34 ] ) is valid , the smaller mass creates only a small perturbation in the gravitational field of the larger mass , which can be taken to be the schwarzschild field . the total gravitational field can therefore be written as @xmath152 where @xmath153 represents the background schwarzschild metric , and @xmath154 the perturbation . writing einstein s equations for @xmath106 and linearizing with respect to @xmath154 , one finds that the perturbation must satisfy an inhomogeneous wave equation in the schwarzschild spacetime . schematically , @xmath155 where @xmath156 is an appropriate curved - spacetime wave operator , and @xmath112 the stress - energy tensor associated with the orbiting mass . we will specifically assume that the central body is a schwarzschild black hole of mass @xmath132 . this assumption is made for simplicity , and removes the need to model the star s interior . as a matter of fact , the internal structure of the bodies is irrelevant , except during the last few orbital cycles before coalescence ; this was discussed in sec . taking advantage of this , we model the orbiting body as a point particle of mass @xmath131 , so that its stress - energy tensor is a dirac distribution with support on the particle s world line . for simplicity , and also because it is physically well motivated ( as explained in sec . 5 ) , we take the world line to be a circular geodesic of the schwarzschild spacetime . with @xmath157 as the usual schwarzschild coordinates , @xmath158 denotes the orbital radius , and @xmath159 is the angular velocity . we have @xmath160 and the gravitational - wave frequency @xmath7 is given by @xmath161 . we stress once more that in the perturbation approach , @xmath99 is not required to be small . the only limitation on @xmath99 comes from the fact that for @xmath162 , circular orbits are no longer stable ; this implies @xmath163 . black - hole perturbations are conveniently treated with the teukolsky formalism @xcite , in which gravitational perturbations are represented by the complex - valued function @xmath164 , a particular component of the perturbed weyl tensor ; the tensor @xmath154 can be reconstructed from @xmath164 . the equation satisfied by @xmath164 admits a separation of the variables . when @xmath164 is expanded in spherical harmonics and decomposed into fourier modes @xmath165 , one obtains an ordinary second - order differential equation the teukolsky equation for the radial function @xmath166 . here , @xmath121 and @xmath122 are the usual spherical - harmonic indices . schematically , and omitting the subscript @xmath167 , this equation takes the form @xmath168 where @xmath169 is a second - order differential operator , and @xmath170 the source , constructed from the particle s stress - energy tensor @xcite . equation ( [ 38 ] ) can be integrated in the standard way by constructing a green s function @xmath171 out of two linearly independent solutions to the homogeneous equation . these solutions , @xmath172 and @xmath173 , respectively satisfy appropriate boundary conditions at the inner ( @xmath174 ) and outer ( @xmath175 ) boundaries . schematically , @xmath176 , where @xmath177 ( @xmath178 ) is the lesser ( greater ) of @xmath41 and @xmath179 . the solution to eq . ( [ 38 ] ) can then be expressed as @xmath180 , and @xmath164 can be reconstructed by summing over all the modes . finally , the gravitational waveform @xmath4 and the luminosity @xmath124 can be obtained from the asymptotic behaviour of @xmath164 when @xmath181 . to integrate eq . ( [ 38 ] ) therefore reduces to solving the homogeneous teukolsky equation , @xmath182 , for the functions @xmath172 and @xmath173 . this , it turns out , is equivalent to integrating the regge - wheeler equation @xcite @xmath183 \biggr\ } x_{\omega\ell}(r ) = 0 , \label{39}\ ] ] for the functions @xmath184 and @xmath185 ; here , @xmath186 . this comes about because a solution to the homogeneous teukolsky equation can easily be related to a solution to the regge - wheeler equation . the relation is known as the chandrasekhar transformation @xcite . the problem of calculating the gravitational waveform for a compact binary system with small mass ratio can therefore be reduced to the simple one of integrating eq . ( [ 39 ] ) . the relevant dimensionless parameter entering this equation is @xmath187 , where , for circular orbits , @xmath188 @xcite . we therefore have , using eq . ( [ 37 ] ) , @xmath189 . for arbitrary values of @xmath187 the regge - wheeler equation must be integrated numerically @xcite . this must be done separately for each selected value of @xmath99 in the interval @xmath190 ( approximately corresponding , for systems of a few solar masses , to the frequency interval @xmath191 ) . @xcite once again we will focus on @xmath124 , the gravitational - wave luminosity . in fig . 2 we present a plot of @xmath192 as a function of @xmath99 . this is the ratio of the luminosity as calculated exactly ( numerically ) using the perturbation approach , to the quadrupole - formula expression @xmath193 the numerical results are depicted as a solid curve . apart from negligible numerical errors , these results are _ exact _ to all orders in @xmath99 ; the only assumption used in the calculation concerns the smallness of @xmath194 . the various broken curves will be described below . the figure shows that @xmath192 tends toward unity as @xmath195 , and stays within 15% of unity everywhere in the interval @xmath196 . it is now easy to judge the accuracy of the post - newtonian expansion for @xmath124 in the limit @xmath197 . one simply evaluates eq . ( [ 31 ] ) in this limit , with @xmath198 , and compare with the numerical results . this post - newtonian curve is labeled `` 4 '' in the figure , and we see that the numerical results are only imperfectly reproduced . the consequences of this will be discussed in the next section . one can in fact do better than this . the perturbation approach is not only suitable for exact , numerical computations . by combining it with a slow - motion approximation putting @xmath199 on top of @xmath200 it also becomes suitable for approximate , analytical computations . methods for integrating the regge - wheeler equations analytically in the limit @xmath201 were devised by various authors @xcite . the basic idea is to proceed by iterations : eq . ( [ 39 ] ) with @xmath202 is solved in terms of spherical bessel functions , and this zeroth order solution is used as input for the first iteration . using these methods , tagoshi and sasaki @xcite were able to calculate @xmath124 analytically , accurately through fourth post - newtonian order , two full orders beyond eq . ( [ 31 ] ) . they derive the rather impressive expression @xmath203 . \label{41}\end{aligned}\ ] ] notice the presence of @xmath204 terms in this expansion , as well as that of the euler number @xmath205 . notice also that the first four terms reproduce eq . ( [ 31 ] ) in the limit @xmath197 , when @xmath206 . this , of course , is as it should be . the broken curves in fig . 2 represent plots of eq . ( [ 41 ] ) truncated to various orders in @xmath99 . for example , the curve labeled `` 6 '' is a plot of eq . ( [ 41 ] ) with all terms of order @xmath207 and @xmath208 discarded . we see from the figure that the post - newtonian expansion converges poorly . ( the suspicion , in fact , is that the series is only asymptotic . ) witness in particular the poor quality of the curve `` 5 '' with respect to `` 4 '' , and compare also `` 8 '' to `` 7 '' . we have seen that in the @xmath197 limit , the post - newtonian expansion for @xmath124 converges poorly . a similar statement can be made about @xmath128 : the post - newtonian expansion @xmath209 where @xmath210 converges slowly to the exact result @xcite @xmath211 contrary to eq . ( [ 41 ] ) , the expansion ( [ 42 ] ) is actually _ known _ to converge ( for all values of @xmath99 in the interval @xmath196 ) . in this section we address the issue as to how much of an obstacle the poor convergence of the post - newtonian expansion poses to the construction of accurate measurement templates @xcite . we will answer this question for binary systems with small mass ratios , using the results described in the preceding section . since there is no reason to believe that the convergence of the post - newtonian expansion would be much improved for systems of comparable masses , our results should also apply , at least qualitatively , for such systems . one way to address this question is to ask , given a waveform constructed from the exact numerical results , how much signal - to - noise ratio is lost by matched filtering the exact signal with an approximate post - newtonian template ? from sec . 8 we know that filtering with the exact signal @xmath18 would give @xmath212 , the largest possible value of the signal - to - noise ratio . on the other hand , filtering with a post - newtonian template @xmath213 , where @xmath10 denotes the order in @xmath99 to which the expansion is taken ( for example , @xmath214 represents a waveform accurate to second post - newtonian order ) , gives the smaller value @xmath215 . from the results of sec . 7 and 8 we obtain @xmath216 the detail of how to compute @xmath217 is presented in ref . @xcite . the calculation described here can only be carried out for binary systems with small mass ratios , because only for these do we have access to the exact waveform @xmath18 . nevertheless , in the following we shall let @xmath194 become large , without altering our expressions for the exact and post - newtonian waveforms . this is done without justification , but reflects the adopted point of view that the quality of the post - newtonian approximation should not be appreciably affected by the finite - mass corrections . in doing this extrapolation , some thought must be given as to the interpretation of the ratio @xmath194 . in the limit @xmath197 this can be taken to be both the ratio of the individual masses or the ratio of reduced mass to total mass . in the case of comparable masses , some choice must be made . we note that as @xmath194 is allowed to grow large , expressions ( [ 40 ] ) and ( [ 43 ] ) for the newtonian quantities @xmath218 and @xmath219 must be replaced by expressions ( [ 29 ] ) , _ which are formally identical_. this shows that when extrapolating to the case of comparable masses , @xmath131 is to be interpreted as the _ reduced mass _ , and @xmath132 as the _ total mass_. .reduction in signal - to - noise ratio incurred when matched filtering with approximate , post - newtonian templates . the first column lists the order @xmath10 of the approximation , the second column lists @xmath217 as calculated using the post - newtonian approximation for @xmath128 , and the third column lists @xmath217 as calculated using the exact expression for @xmath128 . [ cols="^,^,^",options="header " , ] we quote the results for @xmath217 corresponding to a system of two neutron stars , each of 1.4 @xmath220 ; these are displayed in the second column of table 1 . we see that even at quite a high order in the post - newtonian expansion , only three quarters of the signal - to - noise ratio is reproduced by the post - newtonian template . this shows that the apparently small discrepancies between the exact and post - newtonian results for @xmath124 and @xmath128 provide a serious obstacle to the construction of accurate measurement templates . it is interesting to ask how much of the signal - to - noise ratio would be recovered if @xmath128 were kept exact instead of being expressed as a post - newtonian expansion . the third column of the table displays the values of @xmath217 calculated in this way . we see that most of the signal - to - noise is recovered : @xmath217 can now be as large as 0.9864 instead of 0.7651 . why does the exact expression for @xmath128 give such better results ? it can be established @xcite that this has to do with the following fact : while the exact expression for @xmath128 correctly goes to zero at @xmath221 ( at the innermost circular orbit ) , its post - newtonian analogue fails to do so . [ for example , the right - hand side of eq . ( [ 42 ] ) , truncated to order @xmath208 , goes to zero at @xmath222 , corresponding to a radius @xmath223 . ] when corrected for this , the post - newtonian template gives much better results . we therefore see that the poor convergence of the post - newtonian expansion is a serious obstacle to the construction of accurate measurement templates . devising ways to extract the information contained in gravitational waves produced during the late inspiral of a compact binary system poses a great challenge to theoretical physicists . it is not clear that `` simply '' pushing to higher order in post - newtonian theory will be enough to produce sufficiently accurate measurement templates ( such that the systematic errors will be smaller than the statistical errors ) . there may be a need for theorists to develop alternative ways of dealing with this problem . it should be stressed that the convergence problem does not arise when constructing _ detection templates_. indeed , as was discussed in sec . 9 , there is no particular need for these templates to be derived from the equations of general relativity . and since the detection templates need not involve any parameters of direct physical significance , the notion of systematic errors does not apply to them . the only requirement for constructing detection templates is that they should span the appropriate `` signal space '' , and that they should do so the most efficiently . the poor convergence of the post - newtonian expansion is therefore not an obstacle for _ detecting _ gravitational waves from coalescing compact binaries . it is only an obstacle for extracting the information that the waves contain . to overcome this obstacle will undoubtedly be a theorist s challenge for years to come . this state of affairs is highly interesting from a historical point of view : never before in the history of gravitational physics has experiment demanded such a high degree of sophistication on theoretical calculations . this work was supported by the natural sciences foundation under grant no . phy 92 - 22902 and the national aeronautics and space administration under grant no . nagw 3874 . this article was completed while visiting the theoretical physics institute of the university of alberta ; the author is most grateful to werner israel for his warm hospitality . 99 for broader reviews on gravitational waves , see k.s . thorne , in _ 300 years of gravitation _ , edited by s.w . hawking and w. israel ( cambridge university press , cambridge , 1987 ) ; in _ proceedings of the snowmass 95 summer study on particle and nuclear astrophysics and cosmology _ , edited by e.w . kolb and r. peccei ( world scientific , singapore , 1995 ) . r.a . hulse and j.h . taylor , astrophys . j. * 324 * , 355 ( 1975 ) ; j.h . taylor , rev . phys . * 66 * , 711 ( 1994 ) . a. abramovici _ et al . _ , science * 256 * , 325 ( 1992 ) . c. bradaschia _ el al . instrum . & methods * a289 * , 518 ( 1990 ) . see , for example , f. reif , _ fundamentals of statistical and thermal physics _ ( mcgraw - hill , new - york , 1965 ) . helstrom , _ statistical theory of signal detection _ ( pergamon , oxford , 1968 ) . schutz , nature ( london ) * 323 * , 310 ( 1986 ) ; class . quantum grav . * 6 * , 1761 ( 1989 ) . other configurations , with narrow - band sensitivity at a chosen frequency , are possible . e.s . phinney , astrophys . j. * 380 * , l17 ( 1991 ) ; r. narayan , t. piran , and a. shemi , _ ibid . _ * 379 * , l17 ( 1991 ) ; a.v . tutukov and l.r . yungelson , mon . not . * 260 * , 675 ( 1993 ) . c. kochanek , astrophys . j. * 398 * , 234 ( 1992 ) ; l. bildsten and c. cutler , _ ibid . _ * 400 * , 175 ( 1992 ) . for an overview , see t. damour , in _ 300 years of gravitation _ , edited by s.w . hawking and w. israel ( cambridge university press , cambridge , 1987 ) . see , for example , c.w . misner , k.s . thorne , and j.a . wheeler , _ gravitation _ ( freeman , san francisco , 1973 ) , chap . wainstein and v.d . zubakov , _ extraction of signals from noise _ ( prentice - hall , englewood cliffs , 1962 ) . the following discussion is based on c. cutler and e.e . flanagan , phys . d * 49 * , 2658 ( 1994 ) . finn , phys . d * 46 * , 5236 ( 1992 ) . finn and d.f . chernoff , phys . d * 47 * , 2198 ( 1993 ) . c. cutler _ et al . _ , 70 * , 1984 ( 1993 ) . for an overview , see c.m . will , in _ relativistic cosmology _ , proceedings of the eighth nishinomiya - yukawa memorial symposium , edited by m. sasaki ( universal academy press , kyoto , 1994 ) . the following discussion is based on the work of blanchet , damour , and iyer , expounded in the following papers : l. blanchet and t. damour , philos . london a * 320 * , 379 ( 1986 ) ; phys . d * 37 * , 1410 ( 1988 ) ; ann . inst . h. poincar ( phys . thorique ) * 50 * , 377 ( 1989 ) ; l. blanchet , proc . london a * 409 * , 383 ( 1987 ) ; t. damour and b.r . iyer , ann . h. poincar ( phys . thorique ) * 54 * , 115 ( 1991 ) . k.s . thorne , rev . phys . * 52 * , 299 ( 1980 ) . kidder , c.m . will , and a.g . wiseman , phys . d * 47 * , r4183 ( 1993 ) . peters and j. mathews , phys . rev . * 131 * , 435 ( 1963 ) . r.v . wagoner and c.m . will , astrophys . j. * 210 * , 764 ( 1976 ) ; * 215 * , 984 ( 1977 ) . e. poisson , phys . d * 47 * , 1497 ( 1993 ) . a.g . wiseman , phys . d * 48 * , 4757 ( 1993 ) . l. blanchet and g. shfer , class . quantum grav . * 10 * , 2699 ( 1993 ) . l. blanchet , t. damour , b.r . iyer , c.m . will , and a.g . wiseman , phys . * 74 * , 3515 ( 1995 ) . the following discussion is based on the work of this author and his collaborators . see e. poisson and m. sasaki , phys . d * 51 * , 5753 ( 1995 ) and references therein . teukolsky , astrophys . j. * 185 * , 635 ( 1973 ) . t. regge and j.a . wheeler , phys . rev . * 108 * , 1063 ( 1957 ) . s. chandrasekhar , proc . london a * 343 * , 289 ( 1975 ) . c. cutler , l.s . finn , e. poisson , and g.j . sussman , phys . d * 47 * , 1511 ( 1993 ) . m. sasaki , prog . . phys . * 92 * , 17 ( 1994 ) . h. tagoshi and m. sasaki , prog . * 92 * , 745 ( 1994 ) . this can be derived simply from the equations for circular geodesic motion in schwarzschild . e. poisson , _ gravitational radiation from a particle in circular orbit around a black hole . accuracy of the post - newtonian expansion _ d , in press .

recent works have explored the possibility to show the effects of the gravity onto quantum systems @xcite . an interesting study concerning the behaviour of a scalar field in the neighbourhood of a black hole @xcite has shown that the entanglement is degraded by the effect of the black hole . in this work , i. fuentes - schuller and r.b . mann have studied a model in which all the ingredients of quantum field theory are present , but the gravitation is only represented as a rindler spacetime corresponding to the uniform surface gravity in the neighbourhood of the event horizon . recently palmer _ etal _ @xcite have proposed a theory of localized qubits in curved spacetimes . in this paper , we want to reexamine in this framework the problem of qubits around a black hole , especially concerning the entanglement and the quantum teleportation protocol . moreover we want to analyse the physical meaning of the non - self - adjointness of the localized qubit hamiltonian . in contrast with the model of fuentes - schuller and mann , quantum field theory is not completely treated , semi - classical approximations in the localized qubit theory induce the lost of the possibility to create and annihilate particles , the lost of the unruh effect , and the lost of the delocalization of the wave packets ( but the non - locality remains in the theory with the entanglement ) . but we want to treat the complete geometry of the black hole spacetime . the interest of our approach is the possibility to treat the local effects on a fixed quantum system ( a single qubit boarded in a `` spacecraft '' with definite position and velocity and following a geodesic ) whereas no local information associated with the spacetime geometry is taken into account in the fuentes - schuller mann model since the gravitational field is considered in it as uniform and the qubits are completely delocalized in it . we want to analyse the entanglement degrading effect with respect to the position and the velocity of the qubits with respect to the black hole . + we consider a qubit realised as the spin of a fermion submitting only to the gravitational field ( no external magnetic field ) and boarded in a `` spacecraft '' following a geodesic around a black hole . by `` gravitational field '' we mean the lorentz connection associated with the spacetime geometry . in order to treat the dynamics of the qubit , we use the quantum adiabatic approximation because the qubit transport can be considered as slow with respect to its proper quantum time ( the period of its rabi oscillations induced by the gravitational field ) . section 2 summarizes the localized qubit approach by rewritting it in the language of the fibre bundle theory . the goal of this reformulation is to provide a description including spacetime geometries and qubit quantum states in a single common geometric structure . this is achieved by the introduction of the fiber bundles of the quantum adiabatic approximation . we show that the problem takes place in complex line bundles over a space constituting by the product of the space of lorentz connections by the tangent bundle of the spacetime manifold . we show that the qubit is submitted to a kind of decoherence process induced by the gravitational field and responsible to the degradation of the entanglement . section 3 presents in our framework the quantum teleportation protocol with an epr ( einstein podolsky rosen ) qubit pair when one qubit falls to the black hole whereas the other one is comoving with it . we compute a formula providing the fidelity of the teleportation . section 4 applies the formalism to two spacetime geometries , firstly to the rindler spacetime used by fuentes - schuller and mann , secondly to a schwarzschild spacetime , where we analyse the fidelity of the quantum teleportation protocol with respect to the geodesic followed by the qubit falling to the black hole . + _ throughout this paper , we consider the unit system such that @xmath0 . _ + _ note about the notations : a fibre bundle of total space @xmath1 and base space @xmath2 is denoted by its projection @xmath3 . the space of the local sections of a fibre bundle is denoted by @xmath4 . for a map @xmath5 , @xmath6 denotes the map induced by the fibration . let @xmath7 be a manifold , its tangent bundle is denoted by @xmath8 ( @xmath9 is the tangent space of @xmath7 at the point @xmath10 ) , and its differential @xmath11-form set is denoted by @xmath12 . for a map @xmath13 , @xmath14 and @xmath15 denote the associated tangent and the cotangent maps ( the pusch - forward and the pull - back ) . @xmath16 and @xmath17 denote the spaces of continuous and differentiable functions of @xmath7 . @xmath18 between two manifolds denotes a diffeomorphism . @xmath19 denotes the projection map defined by @xmath20 . we use the einstein convention concerning the up and down indices repetition . the greek indices runs in @xmath21 as curved spacetime indices , the capital latin indices runs in @xmath21 as flat minkowski auxiliary spacetime indices , the small latin indices runs in @xmath22 . @xmath23 denotes the derivation with respect to a proper time @xmath24 . _ in this section we summarize the results ( without details ) of palmer _ etal _ @xcite , we fix the notations , and we embbed the localized qubit theory into the fibre bundle theory ( for an exposition of the fibre bundle theory see for example @xcite ) . let @xmath7 be an open set of the spacetime manifold endowed with a local coordinates system @xmath25 and a metric tensor @xmath26 ( to simply we will refere to @xmath7 as the spacetime ) . let @xmath27 be a tetrad field associated with the metric : @xmath28 , where @xmath29 is the minkowski metric . let @xmath30 be the lorentz connection ( @xmath31 being the christoffel symbols ) . a dirac field @xmath32 obeys to the dirac - einstein equation : @xmath33 where @xmath34 are the dirac matrices ( in weyl representation ) and @xmath35 is the spinorial covariant derivative defined by @xmath36 where @xmath37 is the @xmath38 representation of @xmath39 ( the coverging group of the lorentz group @xmath40 ) on @xmath41 ( we denote by the same symbol the induced representation of its lie algebra ) , i.e. @xmath42 $ ] . + let @xmath43 be the tangent bundle of @xmath7 and @xmath44 be the frame principal @xmath40-bundle of @xmath7 . let @xmath45 and @xmath46 be the local trivializations of these bundles , which are defined by @xmath47 and @xmath48 ( @xmath49 is the matrix of elements @xmath50 ) . the tetrad field can be identified as the trivializing local section of @xmath51 : @xmath52 . let @xmath53 be the principal @xmath39-bundle associated with the local @xmath39 transformations of the spinors ( @xmath54 is an extension of @xmath51 such that @xmath55 ) . let @xmath56 and @xmath57 be the associated vector bundles for the representation @xmath58 and @xmath59 , i.e. @xmath56 is defined by its local trivialization @xmath60 , @xmath61_{g \in sl(2,\mathbb c)}$ ] where @xmath62 is the local trivialization of @xmath63 and @xmath64 is the projection @xmath43 , and @xmath65_{g \in sl(2,\mathbb c)}$ ] ( we have denoted simply the @xmath58-action of @xmath66 on @xmath67 by @xmath68 and the @xmath59-action of @xmath69 on @xmath70 by @xmath71 ) . + @xmath72 is a hilbert @xmath73-module endowed with the inner product : @xmath74 where @xmath75 ( @xmath76 , @xmath77 ) . let @xmath78 be a spacelike hypersurface of @xmath7 and @xmath79 be the set of futur oriented timelike normal vectors to @xmath80 . the dirac field @xmath32 is a vector of the hilbert space @xmath81 ( the space of the dirac spinor fields living at some time on the instantaneous space @xmath80 ) . + let @xmath82 be a geodesic worldline in @xmath7 and @xmath83 be a fiolation of @xmath7 along @xmath82 by spacelike hypersurfaces ( @xmath24 being the proper time along @xmath82 ) . by the wkb ( wentzel kramers brillouin ) approximation associated with the assumption that the compton wavelength is small with respect to the curvature scale @xcite , we have @xmath84 . this semi - classical approximation suppresses the space delocalisation of the fermion supporting the qubit and the absence of second quantization supresses the particle number ambiguity . since only the @xmath58-representation is needed to describe a single spin ( qubit ) , we can project onto the space @xmath85 . we work then with the composite bundle @xcite @xmath86 . it can be more easy to work with a bundle @xmath87 with structure fibre @xmath88 ( tangent vector model space and spin quantum state space ) . @xmath89 is defined by the local trivialization @xmath90 , @xmath91 . we have an action of @xmath39 onto @xmath89 defined by @xmath92 , @xmath93 , where @xmath94 is the group homomorphism associated with the quotient @xmath95 . + we can identify the space of local sections of @xmath89 , @xmath96 , to the space of @xmath40-invariant local sections of @xmath97 : @xmath98 . the restriction of @xmath99 to the invariant sections is important to ensure the physical character of the theory , more precisely to have the following property : @xmath100 i.e. the quantum propabilities are invariant under lorentz transformations . endowed with the inner product @xmath101 ( with @xmath102 , @xmath103 is the dirac distribution ) , @xmath96 is a hilbert @xmath16-module . + at this stage , the localized qubit is described by the spin state @xmath104 which obeys to the schrdinger like equation ( see @xcite ) : @xmath105 where @xmath24 is the proper time along the geodesic followed by the qubit , and @xmath106 ( with @xmath107 and @xmath108 , @xmath109 being the usual pauli matrices ) . @xmath110 where @xmath111 is the projection of @xmath97 onto @xmath8 and @xmath112 . the instantaneous qubit hilbert space @xmath113 depends on the four - velocity since the correct inner product induced by the dirac field theory is @xmath114 where the conjugate state is @xmath115 . in order to interpret the dependence of the instananeous spin hilbert space on the four - velocity , it needs to recall some elementary axioms of the quantum mechanics . an hilbert space @xmath116 constitutes the space of the states of a quantum system . but its algebraic dual @xmath117 , i.e. the space of continuous linear functionals of @xmath116 , is the space of the probability amplitudes of the elementary events : @xmath118 is a map from @xmath116 to @xmath119 , such that @xmath120 is the probability of the realization of some measurement event associated with @xmath121 when the quantum system is in the state @xmath70 . by the riesz theorem , we know that @xmath122 , @xmath123 ( up to a renormalization and phase factor ) such that @xmath124 , @xmath125 . we find @xmath126 as an eigenvector of an observable @xmath127 associated with the measure ; @xmath128 corresponds to the event `` the measure of @xmath127 has provided the result @xmath129 '' ( @xmath129 being the eigenvalue associated with the eigenvector @xmath128 ) . it is then important to note that a `` ket '' @xmath130 characterizes the quantum system as being its state , whereas a `` bra '' @xmath131 characterizes an event for an observer making measures on the quantum system . + returning to the localized qubit problem , we want to interpret the difference between the two linear functionals @xmath132 and @xmath133 ( the `` bra '' always denoting in this paper the partial inner product of @xmath134 : @xmath135 , and never @xmath136 ) . let @xmath137 be a normalized state : @xmath138 where we have introduced the magnetic four - momentum operator @xmath139 , with @xmath140 ; @xmath141 , @xmath142 and @xmath143 . the formula [ spinchangeref ] is the classical relation between a magnetic four - momentum @xmath144 measured into an inertial frame @xmath145 and the four - momentum @xmath146 measured into its rest frame @xmath147 of four - velocity @xmath148 with respect to @xmath145 ( see for example @xcite ) . for our problem , @xmath145 is a frame comoving with the black hole and @xmath147 is the frame comoving with the qubit . + with this analysis of the normalization with respect to @xmath149 we can say that : * @xmath132 is the linear functional associated with the ( non - normalized ) probability amplitude to find the spin in the state @xmath70 measured by an observer comoving with the black hole ( @xmath150 ) . * @xmath151 is the linear functional associated with the ( normalized ) probability amplitude to find the spin in the state @xmath70 measured by an observer comoving with the qubit ( @xmath152 ) . ( with @xmath153 ) . we consider the schrdinger like equation for the localized qubit : @xmath154 with the hamiltonian @xmath155 ( non - self - adjoint with respect to @xmath156 , @xmath157 ) : @xmath158 @xmath159 ( @xmath160 ) . we have @xmath161 ( dissipation operator ) and @xmath162 ( hamiltonian of the qubit rotation ) . finally we can write @xmath163 where @xmath164 ( @xmath165 is the complex self - dual lorentz connection ) . from the viewpoint of the qubit , the interaction with the gravitational field is similar to a spin submitted to a complexified magnetic field . let @xmath166 . + to integrate the dynamics involved by the schrdinger like equation , we propose to use the adiabatic approximation for the non - self - adjoint hamiltonians @xcite : @xmath167 where @xmath168 , @xmath169 and @xmath170 are respectively the instantaneous right eigenvectors , left eigenvectors and eigenvalues of @xmath171 : @xmath172 ( the overline denoting the complex conjugation ) , @xmath173 , and @xmath174 are the generators of the non - unitary geometric phases : @xmath175 @xmath176 is the path in @xmath177 defined by @xmath178 for the geodesic @xmath179 followed by the qubit . + a simple calculation shows that @xmath180 @xmath181 @xmath182 the adiabatic approximation is valid if the non - adiabatic coupling is negligible , i.e. @xmath183 let @xmath184 be the space of the @xmath39-connections of the principal bundle @xmath54 . the eigenvectors can be considered as maps @xmath185 such that @xmath186 , where @xmath187 is the inner product of @xmath7 and @xmath188 is defined by @xmath189 ( @xmath190 constitutes a set of generators of @xmath191 the lie algebra of @xmath39 ) . the eigenvectors being defined up to an arbitrary normalization and phase factor , they define @xmath119-line bundles @xmath192 with local trivializations @xmath193 with @xmath194 . @xmath70 obtained by the adiabatic transport formula [ adiabtransp ] is then a local section of @xmath195 over @xmath196 . note that the left eigenvectors do not define line bundles since their normalization factors are fixed by those of the right eigenvectors . + finally the geometric structure in which the qubit transport takes place can be summarized by the following commutative diagram : @xmath197 where @xmath198 . we can note that @xmath199 is a connection of the principal bundle @xmath54 and @xmath200 are connections of the @xmath201-principal bundles associated with @xmath202 . we have then three kinds of gauge changes associated with each floor of the composite bundle : * ground floor : @xmath203 ( diffeomorphism of the spacetime manifold ) , @xmath204 and @xmath205 . * first floor : @xmath206 ( local lorentz transformation ) , @xmath207 and @xmath208 . * second floor : @xmath209 ( normalization and phase local change ) , @xmath210 . the different steps of the construction of the localized qubit adiabatic state can be summarized as follows : @xmath211 it can be interesting to note that the holonomy of @xmath199 along @xmath82 ( between @xmath212 and @xmath24 ) is @xmath213 where @xmath214 and @xmath215 denote path and time ordered exponentials ( dyson series ) . @xmath216 which characterize the adiabatic state of the qubit can be viewed as cylindrical functions of the space of lorentz connections , @xmath217 , and @xmath70 as a linear combination of these two cylindrical functions . it can be interesting to note that ( the topological completion of ) @xmath218 constitutes the kinematical hilbert space of the loop quantum gravity @xcite , this could be indicate a possible connection of the adiabatic localized qubit formalism with a semi - classical limit of the loop quantum gravity . + note that the evolution governed by @xmath219 is unitary with respect to @xmath149 ( for a proper observer comoving with the qubit ) ( see @xcite ) : @xmath220 . but it is not unitary with respect to @xmath221 ( for an observer comoving with the black hole ) . we begin to examine this point in the next section . + _ to simplify the notation , from this point we denote @xmath222 only by @xmath223 . _ the adiabatic transport of a non - self - adjoint two - level quantum system has been extensively studied in the litterature ( see for example @xcite ) . the interesting effects in the adiabatic transport ( eq . [ adiabtransp ] ) are related to the submanifold @xmath224 of @xmath177 defined by the crossings @xmath225 . firstly because the validity of the adiabatic approximation ( eq . [ valad ] ) needs to do not approach @xmath224 ( except if @xmath226 ) . secondly because @xmath224 is a kind of hybercone separating the region of @xmath177 for which @xmath227 are real ( and generate only pure phases ) from the region for which @xmath227 are complex ( and generate non - unitary ( for @xmath156 ) evolution modifying the relative weights of the superposition of @xmath228 ) . since in the self - adjoint case , the geometric phase generator is similar to a magnetic field induced by a magnetic monopole at the eigenvalue crossing point ( see @xcite ) , for the non - unitary case , @xmath224 has been called complex magnetic monopole @xcite ( but @xmath224 is not an isolated point and is associated with exceptional crossings , i.e. @xmath219 is not diagonalizable on @xmath224 ) . let @xmath229 and @xmath230 . @xmath231 . since @xmath232 , the complex magnetic monopole is defined by @xmath233 @xmath234 . if the condition @xmath235 is satisfied , outside @xmath224 ( @xmath236 ) , @xmath237 and @xmath238 are just pure phases ; but inside @xmath224 ( @xmath239 ) , @xmath240 and @xmath241 are non - unitary dynamical phases . in this last case , the evolution modifies the weights of the superposition of @xmath228 ( with respect to @xmath221 , i.e. for an observer comoving with the black hole ) . we will call this effect a * dynamical decoherence * , because the following coherence @xmath242 falls to zero for large @xmath24 ( with @xmath243 , we have neglected the geometric phases and supposed that @xmath244 ) . we have not considered the effects of the non - unitary geometric phases @xmath245 which can induce a * geometric decoherence * if @xmath246 ( we call it geometric decoherence since the geometric phase depends only on the shape of the followed path @xmath176 and not from the proper time ) . + let @xmath247 be the preimage of @xmath224 into @xmath8 . it is important to note that the complex magnetic monopole for a fixed spacetime geometry @xmath248 is not a submanifold of the spacetime @xmath7 but a submanifold of the tangent bundle @xmath8 . the complex magnetic monopole around the black hole `` viewed '' by the qubit depends on its four - velocity . gobally the set of all complex magnetic monopoles is @xmath249 . in some cases , a class @xmath250 of geodesics can be defined with some first integrals @xmath251 and @xmath252 ( @xmath253 , @xmath254 ) depends only on @xmath251 and @xmath10 . in that case @xmath255 is a submanifold of @xmath7 which is an image ( for the qubits following geodesics of @xmath250 ) of the complex magnetic monopole in the spacetime . to understand the physical origin of the non - unitarity with respect to @xmath222 ( observer comoving with the black hole ) , consider first a more simple model constituted by a three - level system ( with levels denoted by @xmath256 ) , governed by an hamiltonian @xmath171 and with spontaneous emission from @xmath257 to the `` dark state '' @xmath258 with a rate @xmath259 . we want to consider the system restricted to @xmath260 ) as a qubit and to forget the dark state @xmath258 . the system obeys to a master equation @xcite : @xmath261 -\frac{\gamma_-}{2 } \{\sigma_{d0}^+ \sigma_{d0}^- , \rho\ } + \gamma_- \sigma^-_{d0 } \rho \sigma^+_{d0}\ ] ] where @xmath262 is the density matrix of the system , @xmath263 denotes the anti - commutator , @xmath264 and @xmath265 . the equation can be rewritten as @xmath266 where @xmath267 . the anti - self - adjoint part of the effective hamiltonian @xmath268 models the lost of population from @xmath257 to the dark state by spontaneous emission , whereas @xmath269 models the gain of population of this dark state . so , if we forget the dark state in the modelization , the qubit obeys to a schrdinger equation governed by a non - self - adjoint effective hamiltonian @xmath270 . + in a curved spacetime there is an ambiguity concerning the particle number . due to the unruh effect , the vaccum in the rest frame becomes a thermal state in an uniformly accelerated frame @xcite . at the level of the quantum field theory in curved spacetime , the evolution in the black hole frame of the dirac field spontaneously couples the one particle state to the zero particle state ( in the fermionic fock space of the system ) . but the semi - classical and wkb approximations used in our model forgets this last one ( we want to have one and only one qubit ) . we have then only two qubit states @xmath271 and @xmath272 ( forming the canonical basis of @xmath134 used in the construction of the different bundles ) , and a dark state : the vacuum @xmath273 . in the same way that for the small example of a three - level system , the qubit is then governed by a non - self - adjoint effective hamiltonian . we will study this point with more details in a concrete example in section 4 . let alice and bob be initially at the point @xmath274 of @xmath7 , supposed sufficiently far from the black hole to consider that @xmath7 is flat in the neighbourhood of @xmath274 . we set @xmath275 and @xmath276 . bob is supposed comoving with the black hole ( he stays at @xmath277 ) , but alice follows a geodesic going near the event horizon at a point @xmath278 . alice wants to teleport information when she will be at this point . at the moment @xmath279 when alice leaves bob , they have an entangled qubit pair in a bell state : @xmath280 @xmath281 and @xmath282 are the initial four - velocities of alice and bob . since alice and bob belongs to two different frames , each one has its proper definition of the qubit states . for bob , the linear functionals of finding its qubit in a particular state are @xmath283 and @xmath284 , involving that @xmath285 and @xmath286 . but for alice , her linear functionals are @xmath287 and @xmath288 since she is not comoving with the black hole . the qubit states for alice are then defined by @xmath289 ( @xmath290 ) . it follows that @xmath291 and @xmath292 ( @xmath293 ) . @xmath294 in the flat region ( @xmath295 ) , we have @xmath296 and @xmath297 . @xmath298 let @xmath299 the proper time when alice arrives at @xmath278 . we suppose that the evolution along the geodesic @xmath82 linking @xmath277 to @xmath278 is adiabatic for the alice s qubit ( as eq . [ adiabtransp ] ) . we have then for @xmath300 and @xmath301 @xmath302 ( @xmath303 ) . note that @xmath304 is defined for two proper times , one for alice and one for bob , since their clocks are asynchronous . the evolution for the bob s qubit is trivial since it is inertial in a flat part of @xmath7 . @xmath305 where @xmath306 is the alice s four - velocity at @xmath299 . @xmath307 with @xmath308 alice encodes a quantum information in a qubit @xmath309 ( @xmath310 ) . the state of the three qubits is then @xmath311 . alice performs then the operations of the usual teleportation protocol : @xmath312 where @xmath313 and @xmath314 are the cnot and hadamard gates in the alice s frame . after some algebra , we find @xmath315 alice performs a measurement of her qubits . to fix the discussion , we suppose that she finds @xmath316 ( the result can be easily adapted for another result ) . alice sends to bob by a classical communication chanel what is the operation to perform on his qubit ( in our example , the operation is the identity ) . bob receives the message at @xmath317 . the state is then for @xmath318 and @xmath319 : @xmath320 where @xmath321 is the evolution operator for an alice s qubit after @xmath299 . the fidelity of the quantum teleportation is then @xmath322 the fidelity of the teleportation is clearly degraded by the decoherence induced by the black hole which is encoded in @xmath323 . + remark : for a flat spacetime with alice having a constant four - velocity we have @xmath324 , and then @xmath325 ( we refind the efficiency of the usual teleporation protocol ) . in order to compare with the fuentes - schuller mann model @xcite , we first consider the case of the rindler spacetime defined by the metric @xmath326 which corresponds to a flat spacetime viewed in a noninterial frame uniformly accelerated ( with acceleration parameter @xmath327 ) , or to the surface gravity approximation of a schwarzschid black hole ( @xmath328 and @xmath329 , @xmath330 being the schwarzschild radius ) ( see @xcite ) . the tetrad fields are @xmath331 and @xmath332 , and the only one non - zero lorentz connection component is @xmath333 . it follows that @xmath334 and @xmath335 , and then @xmath336 with @xmath337 and @xmath338 @xmath339 . it is interesting to note that since the qubit moves in the @xmath10-direction , @xmath340 corresponds to a spin parallel to the linear momentum ( so to a positive helicity state ) and @xmath341 to a spin antiparallel to the linear momentum ( a negative helicity state ) ( @xmath342 ) . the geodesic equations are @xmath343 the first geodesic equation defines the first integral : @xmath344 the second geodesic equation becomes the autonomous equation : @xmath345 the adiabatic transport of a qubit state @xmath346 ( @xmath347 ) is then @xmath348 the dynamical decoherence kills the positive helicity state in favor of the negative helicity state . we can heuristically understand this fact as follows . the vacuum of the minkowsky spacetime becomes in the noninertial frame @xmath349 ( see @xcite ) where @xmath350 ( @xmath351 denotes the momentum , @xmath352 the helicity , @xmath353 and @xmath354 denote the two regions separated by the horizon , @xmath355 ) . it is then associated with a density matrix @xmath356 which is a thermal distribution with temperature @xmath357 ( @xmath358 is the boltzmann constant ) , corresponding to the unruh radiation . it follows that the fermion is coupled with this thermal bath . for the positive helicity mode ( which is the part of the weyl spinor with positive energy ) , we can write that the density matrix @xmath359 ( for @xmath360 and @xmath273 ) obeys to the master equation ( see for example @xcite ) : @xmath361 where @xmath362 are the fermionic creation / annihilation operators on the positive helicity mode ( @xmath363 and @xmath364 ) , @xmath365 and @xmath366 characterizes the spectral density of the bath . with some assumptions @xmath367 where @xmath368 is a constant ( see @xcite ) . since @xmath369 is very small ( the unruh temperature @xmath370 is very small ) , the master equation is dominated by the dissipation of @xmath360 in accordance with eq . [ transprindler ] . by following the approximation explained section [ origin ] ( by projecting onto @xmath371 , by neglecting the quantum jumps and the @xmath369 terms ) we have @xmath372 we refind @xmath373 by setting @xmath374 . for the negative helicity mode the problem is quite different since it is associated with the part of the weyl spinor with negative energy . it follows that the roles of @xmath375 and @xmath376 are inverted in the master equation . this one is then dominated by the increase of the population of @xmath377 . we can then postulate an effective hamiltonian creating negative helicity population , as @xmath378 . + the geodesic equation ( eq . [ geodesicrindler ] ) has for solution @xmath379 with @xmath380 . we have then an analytical expression of the dynamical phases : @xmath381 let @xmath382 be the proper time for which the qubit reaches the horizon . we have @xmath383 . + the fidelity of the teleportation protocol for the rindler spacetime is represented fig . [ fidelity_rindler ] . average fidelity @xmath384 ( eq . [ fidelity ] ) of the teleportation protocol , for alice following a geodesic of the rindler spacetime and bob being static , with respect to @xmath385 ( first integral of the geodesic ) and @xmath386 the alice s proper time when she realizes its part of the protocol . for each value of @xmath385 , the fidelity is drawn until the proper time when alice reaches the rindler horizon.,width=302 ] the fidelity of the quantum teleportation falls if alice approaches too close to the rindler horizon . we consider the metric associated with a static black hole with spherical symmetry : @xmath387 where @xmath370 and @xmath388 are the factors of time dilation and length contraction , the schwarzschild metric being obtained for @xmath389 with @xmath390 the schwarzschild radius . the tetrad fields are @xmath391 , @xmath392 , @xmath393 and @xmath394 ; and the non - zero components of the lorentz connection are @xmath395 , @xmath396 , @xmath397 and @xmath398 . because of the spherical symmetry , we can restric our attention to the equatorial plane @xmath399 . the geodesic equations are @xmath400 the first and the last geodesic equations define the first integrals : @xmath401 @xmath97 and @xmath402 being the energy and the angular momentum by mass unit . we have @xmath403 , @xmath404 , and @xmath405 . it follows that @xmath406 with @xmath407 . @xmath408 if @xmath409 ( @xmath410 \text { with } \frac{t'(r_{le } ) r^2_{le}}{t(r_{le})^2 } = \frac{l}{e}\right\}$ ] ) . for the schwarzschild case , the dynamical decoherence disapears for @xmath411 , i.e. if @xmath412 and @xmath413 . it follows that the qubit is submitted to dynamical decoherence except if it follows a strongly rotating geodesic ( @xmath402 large ) and far from the complex magnetic monopole ( which is a sphere of radius @xmath414 ) . the generators of the geometric phases are @xmath415 ( with @xmath416 , @xmath417 and @xmath418 ) . for the schwarzschild case , we have @xmath419 . geometric decoherence is always present except for the radial geodesics ( @xmath420 ) and the circular orbits ( @xmath421 constant ) . moreover the non - adiabatic coupling is @xmath422 , assuring without any assumption concerning the velocity , the validity of the adiabatic approximation for the radial geodesics ( @xmath420 ) and the circular orbits ( @xmath423 ) . we consider first the radial geodesics @xmath420 ( @xmath424 ) . the second geodesic equation ( eq . [ geodesicschwarz ] ) is then reduced to @xmath425 , which has for solution : @xmath426 . the event horizon is reached at @xmath427 . the fidelity of the teleportation protocol for this situation is drawn figure [ fidelity_schwarz_radial ] . average fidelity @xmath428 ( eq . [ fidelity ] ) of the teleportation protocol , for alice following a radial geodesic of the schwarzschild spacetime and bob comoving with the black hole , with respect to @xmath97 ( first integral of the geodesic ) and @xmath386 the alice s propre time when she realizes its part of the protocol . the fidelity is drawn until the proper time when alice reaches the event horizon.,width=302 ] as for the rindler spacetime , the fidelity of the quantum teleportation falls if alice approaches too close to the event horizon , because of the decoherence induced by the gravitational field ( all the radial geodesics are inside @xmath429\}$ ] ) . we consider the circular orbits defined by @xmath430 ( constant ) and @xmath431 . the second geodesic equation ( eq . [ geodesicschwarz ] ) involves that @xmath432 and the metric ( eq . [ metricschwarz ] ) involves that @xmath433 . the fidelity of the teleportation protocol for this situation is drawn figure [ fidelity_schwarz_circular ] . average fidelity @xmath384 ( eq . [ fidelity ] ) of the teleportation protocol , for alice following a circular orbit around the black hole and bob comoving with the black hole , with respect to @xmath434 ( radius of the orbit ) and @xmath386 the alice s propre time when she realizes its part of the protocol . the red lines indicates the proper times corresponding to the orbital periods . we start at @xmath435 ( the photon sphere ) since no closed orbit exists under this value.,width=302 ] for the circular orbits , the effect is essentially caused by the difference of four - velocity between alice and bob ( explaining why the fidelity is almost uniform with respect to @xmath434 and @xmath299 ) . no decoherence occurs since for all @xmath434 , @xmath237 ( @xmath436\}$ ] , all circular orbits are outside the complex magnetic monopole which is identified with the photon sphere ) . the adiabatic transport generates a phase difference between @xmath340 and @xmath341 which induces some interferences in the quantum teleportation explaining the small oscillations in the fidelity . we consider geodesics starting far from the event horizon and almost reaching it by an adiabatic process for the qubit evolution . since @xmath437 we can suppose that @xmath438 ( @xmath439 being the proper time needed to reach the event horizon and @xmath440 ) . moreover @xmath441 , @xmath442 , and since @xmath443 , we can compute an evaluation of the fidelity of the teleportation protocol for geodesics almost reaching the event horizon , see figure [ fidelity_schwarz_he ] . fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264]fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264 ] + fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264]fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264 ] + fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264]fidelity @xmath444 ( eq . [ fidelity ] ) of the epr teleportation protocol , for alice following a geodesic almost reaching the event horizon and bob comoving with the black hole , with respect to @xmath97 and @xmath402 ( first integrals of the geodesic ) and @xmath445 the angular position of alice when she almost reaches the event horizon ( alice being started from @xmath446 ) , its final radial position @xmath447 being such that @xmath448 . alice realizes its part of the protocol when she reaches its final point near the event horizon.,title="fig:",width=264 ] + the fidelity oscillates with the relative angular position of alice when she reaches the event horizon ( for the teleportation of a `` schrdinger cat '' , no oscillation occurs for the teleportation of @xmath257 or @xmath449 ) . the dependence from @xmath97 and @xmath402 is small except for the small values of these first integrals . a localized qubit in general relativity is described ( at the adiabatic limit ) by a geometric structure including the description of the quantum states and of the spacetime geometry , i.e. the composite bundle @xmath450 where @xmath184 is the space of lorentz connections . in this bundle , inside a particular submanifold ( the complex magnetic monopole @xmath451 ) decoherence processes appear on the qubit . this complex magnetic monopole is a sphere surrounding the event horizon for a schwarzschild black hole , with radius decreasing with the increase of the angular momentum of the qubit ( it is infinite for @xmath452 , is equal to @xmath453 ( the photon sphere ) for the circular orbits , and tends to @xmath454 ( the event horizon ) with @xmath455 ) . we have two different decoherence processes , a dynamical decoherence associated with the non - unitary dynamical phases and a geometric decoherence associated with the non - unitary geometric phases ( and depending only on the shape of the followed path and not from the proper time ) . the physical origin of these processes is related in a rindler spacetime to the unruh radiation and we can then postulate that in the general case it is related to the hawking radiation ( since the unruh effect can be considered as the near - horizon form of the hawking radiation ) . we have shown how these decoherence processes degrade the fidelity of the quantum teleportation protocol if alice falls to the event horizon , the adiabatic framework permitting to obtain a simple formula to compute this fidelity with respect to the spacetime position and to the four - velocity of alice when she realizes her part of the protocol . 0 ahmadi f and mehrafarin m 2012 _ j. phys . a _ * 45 * 485302 . brushi d e , sabin c , white a , baccetti v , oi d k l and fuentes i 2014 _ new j. phys . _ * 16 * 053041 . fuentes - schuller i and mann r b 2005 _ phys . lett . _ * 95 * 120404 . palmer m c , takahashi m and westman h f 2012 _ ann . phys . _ * 327 * 1078 . nakahara m 1990 _ geometry , topology and physics _ ( bristol : iop publishing ) . viennot d 2010 _ j. math . phys . _ * 51 * 103501 . jackson j d 1962 _ classical electrodynamics _ ( new york : john wiley & sons ltd ) . nenciu g and rasche g 1992 _ j. phys . a _ * 25 * 5741 . rovelli c and vidotto f 2015 _ covariant loop quantum gravity _ ( cambridge : cambridge university press ) . nesterov a i and aceves de la cruz f 2008 _ j. phys . a _ * 41 * 485304 . mehri - dehnavi h and mostafazadeh a 2008 _ j. math . phys . _ * 49 * 082105 . viennot d , jolicard g and killingbeck j p 2008 _ j. phys . a _ * 41 * 145303 . dridi g , gurin s , jauslin h r , viennot d and jolicard g 2010 _ phys . a _ * 82 * 022109 . breuer h p and petruccione f 2002 _ the theory of open quantum systems _ ( new york : oxford university press ) . alsing p m , fuentes - schuller i , mann r b and tessier t e 2006 _ phys . rev . a _ * 74 * 032326 . ghosh a , sinha s s and ray d s 2012 _ phys . e _ * 86 * 011138 . karmakar a and gangopadhyay g 2010 _ phys . _ * 85 * 045008 .

the spin degree of freedom of electrons in semiconductor nanostructures has attracted much interest recently . besides its applications in spintronics@xcite , which makes use of the spin in the same way as the charge in conventional devices , there has been considerable interest in the quantum coherence of spin in different materials and geometries at the nanoscale @xcite . the long coherence times measured some time ago in semiconductors @xcite supported the idea of quantum computing with spins localized in quantum dots @xcite a rapidly growing field now , where recent experimental progress has been remarkable @xcite . electron spins in semiconductor nanostructures , however , are not decoupled from the charge degree of freedom , one of the primary reasons for this coupling being the spin - orbit interaction . this coupling leads to many interesting phenomena in the physics of semiconductors @xcite , an important one being the ability to control the spin of the electron with electric fields acting on its charge degrees of freedom @xcite . in quantum dots the spin - orbit interaction manifests itself as a weak perturbation when compared with the confinement energy . the measure of smallness is given by the ratio between the dot radius @xmath2 and the spin - orbit length , @xmath3the distance over which an electron travels and thereby precesses by an angle @xmath4 about the intrinsic magnetic field induced by the spin - orbit interaction . despite its smallness the spin - orbit interaction is very important for the coherence of the spin dynamics . for example , spin - orbit interaction allows for coupling of the electron spin to ( bosonic ) environments , such as phonons @xcite or particle - hole excitations in quantum point contacts @xcite , which in turn causes relaxation and decoherence of the spins . moreover , electrons being charged particles interact via the long range coulomb forces with each other , even if they are confined to well - separated quantum dots with no overlap of their wave functions . through this electrostatic coupling and in combination with the spin orbit interaction , the spins of two electrons located in different dots become coupled even in the absence of tunneling between the dots . in this paper we provide a detailed analysis of such an effective spin - spin interaction for lateral quantum dots in a configuration as shown in fig . we will see that the origin of this interaction is the `` tidal '' effect each of the electrons produces on the charge distribution of the other electron via electrostatic forces . because of the spin - orbit interaction , the electric dipole moment ( as well as higher moments ) in each dot couples to its electron spin . as a result , the two spins experience an interaction resembling a magnetic dipole - dipole interaction@xcite with effective magnetic moments which can be strongly enhanced by up to a factor of @xmath5 compared to the bohr magneton . the magnitude of the spin - spin coupling obtained via this spin - electric effect can be efficiently controlled and even completely suppressed by adjusting external parameters such as the magnetic field direction , strength , and inter - dot distance . the spin - spin interaction can , in principle , be used to perform two - qubit operations as required in the spin - based quantum computing scheme@xcite , because it entangles spins and can easily be switched on and off . we note that a similar mechanism for spin interaction based on electrostatic coupling was studied very recently in ref . for vertically coupled quantum dots , and in ref . for the special case of one - dimensional quantum dots formed in semiconducting nanowires . besides the differences in geometry and dimensionality , both of these works treat only the case of weak coulomb interaction ( compared to the level spacing ) , while we treat here also the opposite limit of strong coulomb interaction where new and interesting features emerge . in the limiting case of strongly elliptical dots we recover the one - dimensional results obtained in ref . . we emphasize again that in the present study we exclude tunneling and thus the type of spin interaction studied in the following is fundamentally different from the heisenberg exchange interaction for which the presence of electron tunneling between the dots is crucial@xcite . similarly , the combined effect of heisenberg exchange interaction and spin - orbit coupling@xcite is also based on tunneling and should be carefully distinguished from the spin - orbit effect studied here . we also note that the heisenberg exchange coupling allows typically for much stronger spin - spin coupling than the electrostatically induced one . for instance , in gaas dots the heisenberg exchange can reach values on the order of @xmath6 mev @xmath7 mev , which , as we shall see , exceeds the electrostatically induced spin coupling by three to four orders of magnitude . nevertheless , the electrostatic spin coupling can prove useful for cases where it is difficult to get sufficient wave function overlap ( needed for large heisenberg exchange ) , and , moreover , it is also important to understand the electrostatic spin - effects in detail in order to get control over possible interference effects between different types of spin coupling . this will be for example of importance for spin qubit applications in order to minimize spin decoherence and gate errors . finally , in view of experimental tests we propose a scheme to measure the spin - spin interaction in a double dot setup with a nearby charge detector . we propose to combine the spin - measurement technique of ref . with the entangling property of the spin - spin interaction and present a gate pulsing sequence that enables one to access the coupling constant in the time domain by measuring the occupation probability of a zeeman sublevel . the paper is organized as follows . in sec . [ themodellabel ] we introduce our model for the quantum - dot system and its basic hamiltonian . in sec . [ sscouplinglabel ] we derive the effective spin - spin coupling hamiltonian within our model . in this section we analyze in detail the two limiting cases of weak and strong coulomb interaction , and obtain various forms of the spin coupling as function of magnetic field strength and orientation , as well as of the interdot distance . the most general form of the spin coupling tensor is given in the appendix . in sec . [ measurementschemelabel ] , we propose a scheme to measure the spin - spin coupling . there , we also discuss the effect of the hyperfine interaction with the nuclear spins on the measurement scheme . finally , in sec . [ discconclabel ] we discuss our results and end with some conclusions . our system consists of two electrons each of which is localized in a quantum dot , and the two dots are separated from each other , without tunneling between them . the system is composed of two gate - defined quantum dots in a two - dimensional semiconductor layer ( @xmath8 gaas or inas ) . a schematics of the system we consider is shown in fig 1 . we model the system by a harmonic confinement potential , which , for simplicity is assumed to be the same for both dots . each dot is assumed to contain one electron with charge @xmath9 and spin @xmath10 , with @xmath11 being the pauli matrices . the model hamiltonian consisting of several terms reads @xmath12 where @xmath13 is the energy of the two electrons in the confinement potentials @xmath14 here , @xmath15 is the 2d kinetic momentum of the @xmath16-th electron at position @xmath17 , @xmath18 the effective mass , @xmath19 the speed of light , @xmath20 the confinement potential for the @xmath16-th electron which is assumed to be harmonic and isotropic , and @xmath21 is the electromagnetic vector potential . the strength of the confinement energy is given by the frequency @xmath22 . the second term on the right - hand side of eq . ( [ 1 ] ) is the zeeman energy of the two electrons , @xmath23 the third term in eq . ( @xmath24 ) is the unscreened coulomb interaction between the two electrons , @xmath25 where @xmath26 is the dielectric constant of the material and @xmath1 is the geometric distance between the two dots , namely between the potential minima ( center ) of the dots . with this choice , we measure the distance for each electron from its own dot center . the last term in eq . ( @xmath24 ) is the spin - orbit coupling which for strong @xmath27-confinement is given by @xmath28 , \label{5}\ ] ] being the sum of the dresselhaus term @xcite ( @xmath29 ) coming from bulk inversion asymmetry and the rashba term @xcite ( @xmath30 ) coming from structure inversion asymmetry . we assume the same coefficients @xmath29 and @xmath30 for both dots . it is convenient to work with center - of - mass and relative coordinates @xcite , as the coulomb interaction couples only to the relative ones and the solution of the center - of - mass part is straightforward @xcite . this then involves the standard substitutions @xmath31 , @xmath32 , @xmath33 , @xmath34 , and @xmath35 and @xmath36 . we now turn our attention to the spin - orbit interaction . as was shown in ref . , the spin - orbit coupling gives non - zero first order effects only if a magnetic field is present , as a consequence of the kramers degeneracy . in order to describe the effective first order spin - orbit term in the presence of a magnetic field we make use of the schrieffer - wolff ( unitary ) transformation@xcite @xmath37 where @xmath38 is chosen such that @xmath39 , with the projector operator @xmath40 satisfying @xmath41 @xmath42 , and @xmath43 . the hamiltonian @xmath44 ( or @xmath45 in center - of - mass and relative coordinates ) . the hamiltonian @xmath46 is diagonal in the basis of @xmath47 and has the same energy spectrum as the hamiltonian @xmath48 . in first order of the spin - orbit interaction @xmath49 the transformation generator becomes @xmath50 , where @xmath51 is the dot liouvillean , @xmath52 $ ] , @xmath42 . evaluating this expression explicitly we obtain @xmath53 with @xmath54 . in this lowest order the transformed hamiltonian @xmath55 becomes @xmath56 with @xmath57 where @xmath58 = e_z(1-p)[\bm{l}\times(\bm{\xi}_r\pm\bm{\xi}_r/2)]\cdot\bm{\sigma}_{1,2}. \label{8}\ ] ] in eq . ( @xmath59 ) @xmath60 is the magnetic field direction vector , @xmath61 is the zeeman energy and the vectors @xmath62 and @xmath63 are given by @xmath64 and @xmath65 , respectively . the new coordinates correspond to a rotation by an angle @xmath66 with respect to the coordinate frame in which the direction of the @xmath67-vector is associated with the @xmath68-axis in the @xmath69 plane so that the final expressions have the simplest form @xcite . here , @xmath70 is the angle between the @xmath71 frame in fig . 1 and the normal axes of the crystal . the spin - orbit lengths @xmath72 are given in the form @xmath73 . we are now in a position to derive the coupling between the two spins . this is achieved by performing a second schrieffer - wolff ( sw2 ) transformation which excludes the first order contribution in spin - orbit interaction with no diagonal matrix elements . the new transformed hamiltonian has the form @xmath74 , with @xmath75 given by @xmath76 we assume now that the zeeman energy is smaller than the orbital confining energy , @xmath77 , which is usually the case for electrons in quantum dots , such that we can neglect in eq . ( @xmath78 ) the zeeman liouvillean @xmath79 ( for spin - orbit effects due to level crossing see ref . ) . in second - order in spin - orbit coupling the effective hamiltonian @xmath80 becomes @xmath81 . \label{10}\ ] ] the last term in eq . ( @xmath82 ) contains the desired spin - spin coupling between the two spins . however , besides this interaction it also contains some self - interaction terms which renormalizes only the zeeman splitting . we will not study those terms since they are of no practical interest in the case of identical dots . we consider a general magnetic field @xmath83 , where @xmath84 is the angle between the magnetic field and the @xmath27-axis perpendicular to the 2deg plane and @xmath85 the angle between the in - plane component of the magnetic field and the @xmath68-direction ( fig . 1 ) . the interaction between the two spins has the most general form @xmath86 , \quad i , j=1,2.\label{11}\ ] ] the spin hamiltonian is obtained by averaging over the orbital ground state , @xmath87 . we then obtain @xmath88 where @xmath89|0\rangle,\;\ ; a , b = x , y , z.\ ] ] we note that there is no component of the spin along the magnetic field direction as a consequence of the vector product in the tensor @xmath90 . by diagonalizing the above tensor , we obtain for the hamiltonian @xmath91 the reduced expression @xmath92 where the couplings @xmath93 depend on the magnetic field orientation and on the functions @xmath94|0\rangle$ ] , with @xmath95 ( for explicit expressions see the appendix ) . thus , the effective spin - spin interaction is highly anisotropic , and , in general , of the @xmath0-type . we note in particular that for an in - plane magnetic field ( @xmath96 ) , the spin hamiltonian reduces to the ising hamiltonian , @xmath97 ( in a transverse magnetic field ) . next , we rewrite @xmath91 in terms of raising / lowering spin operators @xmath98 @xmath99 with @xmath100 and @xmath101 . we recall now that the _ full _ spin hamiltonian includes the zeeman energy , given in eq . ( [ 3 ] ) , which leads to a large energy gap with @xmath102 . we will find below that typically @xmath103 under our assumption that @xmath77 and @xmath104 . as a consequence , we can neglect in eq . ( @xmath105 ) the terms proportional to @xmath106 since they cause transitions between different zeeman levels of the total spin . the relevant spin - spin interaction , @xmath107 , which acts only within the @xmath108 subspace , becomes then @xmath109 thus , we are left with the task of calculating the coupling strengths @xmath93 and @xmath110 . because of the coulomb term , eq . ( @xmath111 ) , this can not be done exactly and some approximations are required . they will depend on the ratio @xmath112 between the coulomb interaction strength , @xmath113 , and the orbital level spacing , @xmath114 , giving @xmath115 , with @xmath116 being the dot radius and @xmath117 - the bohr radius in the material . in other words , the parameter @xmath112 will dictate the physics of the system , and from now on we will speak of the ratio @xmath118 as being the coulomb interaction strength ( representing in fact the true coulomb strength for touching dots ) . for making the following analysis more transparent we focus on the case with only rashba spin - orbit coupling ( @xmath119 ) . the generalization to the case with both rashba and dresselhaus terms present is straightforward , but at the cost of more complicated expressions ( see appendix ) . one interesting case is met when @xmath120 , such that the coulomb interaction can be treated as a perturbation compared to the orbital level spacing . in this case , one can retain only the first order contribution from the coulomb interaction , which translates into the approximation @xmath121 . making use of this and after some algebra we obtain for the spin - spin coupling the following expression @xmath122 where the operators @xmath123 , @xmath16=1,2 , are the charge density distribution modifications in each dot as a consequence of the spin - orbit interaction . they are defined as @xmath124 with @xmath125 being the charge density operator in the absence of spin - orbit interaction and @xmath126 the one in the presence of spin - orbit interaction , with @xmath127 for the present approximation . from eq . ( @xmath128 ) we see that the spin interaction results from a coulomb - type of coupling between two charge density distributions which themselves depend on spin . let us now analyze in more detail eq . ( @xmath128 ) . the first task is to find @xmath129 , for @xmath130 , namely the spin - orbit induced charge distribution or the spin - dependent charge distributions for each dot . in order to do this , we give first some important relations valid in the case of harmonic confining potential , relations which are used in the following for the derivation of the main results @xmath131 @xmath132 @xmath133 making use of the relations eqs . ( @xmath134-@xmath135 ) and within the first order of spin - orbit coupling , @xmath136 @xmath137 $ ] , we obtain @xmath138,\;\;\;\;\;\;\;\;\;\;\;\;\label{53}\end{aligned}\ ] ] with @xmath139 being the bare charge density in the dot corresponding to the ground state and which assumes the well - known form for harmonic potentials @xmath140 we note that when there exist a perpendicular component of the magnetic field , the dot radius is renormalized due to the orbital effect of the magnetic field @xmath141 , with @xmath142 ( @xmath143 , @xmath144 ) . however , we will still refer to @xmath2 as being the dot radius , with the appropriate expression depending on the magnetic field orientation . we could now insert the expression eq . ( @xmath145 ) for @xmath129 in eq . ( @xmath128 ) and compute directly the spin hamiltonian . however , working with the coulomb potential , it is more convenient to work with the center - of - mass and relative coordinates and for simplicity the @xmath68-axis along the inter - dot direction @xmath67 . assuming for simplicity a perpendicular magnetic field , the spin hamiltonian @xmath91 takes the form @xmath146 with the electronic densities @xmath147 and @xmath148 . in eq . ( @xmath149 ) there are no mixed terms like @xmath150 since those terms vanish because of the odd symmetry of the integrands in the case of harmonic confinement , which reflects inversion symmetry . the limit of in - plane magnetic field is obtained very easy from eq . ( @xmath149 ) by substituting the denominator with @xmath151\sigma_{\tilde{y}}^1\sigma_{\tilde{y}}^2 $ ] . [ for general field orientation the expression for @xmath91 is more complicated ( see appendix ) . ] in order to make the following analysis more transparent , we introduce the dimensionless coordinates @xmath152 and @xmath153 . the integration over the center - of - mass coordinates is now straightforward and the reduced expression for the spin hamiltonian becomes @xmath154 for a perpendicular magnetic field and @xmath155 for an in - plane magnetic field oriented at an angle @xmath85 with respect to the inter - dot distance vector @xmath67 . the energy differences @xmath156 are given by @xmath157 @xmath158 the ground state and the first excited states of the dots in relative coordinates give rise to different charge distributions ( @xmath159 , @xmath160 , and @xmath161 , respectively ) , and thus to different potential energies seen by a test charge at a distance @xmath1 ( along @xmath68 ) away from the center of the charge distribution ( in relative coordinates ) . @xmath162 are the differences between these potential energies . before studying the distance dependence of the spin hamiltonian @xmath91 ( determined by @xmath156 ) in the entire range of distances , it is instructive to see how the expression eq . ( @xmath128 ) behaves in the large distance limit , @xmath163 , and to make some comparison with the magnetic dipolar interaction in vacuum @xcite . we perform a multipole expansion of the hamiltonian in eq . ( @xmath128 ) . the first non - zero contribution takes the form of a dipole - dipole interaction between two spin - dependent electric dipoles , or phrased differently , the interaction between two charge - induced magnetic dipoles @xmath164 with the dipole moments @xmath165 given by @xmath166=\bar{\bar{\mu}}\,\bm{\sigma}_i\;,i=1,2.\ ] ] here , the trace is taken over the orbital degrees of freedom with @xmath167 and @xmath168 being the tensor corresponding to an effective spin - orbit induced magneton @xmath169 we see from eq . ( @xmath170 ) that the tensor @xmath168 depends on the magnetic field orientation with respect to the 2deg and that it is also anisotropic , in contrast to the usual isotropic bohr magneton @xmath171 ( @xmath172 is the mass of the free electron and @xmath19 the speed of light ) . we note that the @xmath27-component of the induced magnetic moment ( with @xmath173 ) vanishes , @xmath136 @xmath174 . let us quantify the strength of @xmath168 by the norm @xmath175 , @xmath136 @xmath176 we compare now @xmath177 with @xmath178 . first of all , we note that @xmath177 vanishes when there is no zeeman splitting . however , for finite magnetic fields , @xmath177 can exceed @xmath178 by many orders of magnitude in the case of quantum dots . to give an estimate , we assume @xmath179 , @xmath180 ( @xmath181 ) and @xmath182 , @xmath183 for gaas quantum dots . with these values , and taking @xmath184 ( perpendicular magnetic field ) we obtain @xmath185 we describe now in more detail the limit of large distance between the dots . from eqs . ( @xmath186 ) and ( @xmath149 ) we find for @xmath163 @xmath187 for a perpendicular magnetic field , and @xmath188 for an in - plane magnetic field , with the coupling strength @xmath189 having the form @xmath190 from eq . ( @xmath191 ) we see a large distance decay @xmath192 , @xmath136 a long range type behavior . we note also that the large distance result in eq . ( @xmath191 ) does not depend anymore on the orbital effect of the magnetic field . working instead with the effective hamiltonian defined in eq . ( @xmath193 ) , the effective coupling strength @xmath110 for arbitrary magnetic field is given by @xmath194.\label{ferro}\ ] ] we note that the spin coupling can range from maximally ferromagnetic interaction with @xmath195 at @xmath96 and @xmath196 to maximally antiferromagnetic interaction with @xmath197 at @xmath96 and @xmath198 , and with @xmath110 passing through zero for certain angles . next , we consider the case of arbitrary distance but still with small coulomb interaction strength as defined at the beginning of this section . considering @xmath107 defined in eq . ( @xmath193 ) we obtain @xmath199 where @xmath200.\;\;\;\;\;\;\;\;\label{fa}\end{aligned}\ ] ] the function @xmath201 is plotted in fig . 2 for different angles @xmath84 and @xmath85 . as for the large distance limit in eq . ( @xmath202 ) , a similar but more complicated ferromagnetic - antiferromagnetic crossover behavior occurs as a function of the field orientation . however , in this case this behavior can also be induced by changing the distance between the dots @xmath1 ( see fig . [ fig2 ] ) . ( @xmath203 ) suggests that the condition @xmath204 is too restrictive . instead , the weaker condition @xmath205 is sufficient for the approximation to be valid . 2 shows a breakdown of the dipolar approximation ( i.e. of the @xmath206 decay ) , occurring at a dot separation @xmath207 for perpendicular magnetic fields ( @xmath184 ) , and also a cancellation of this interaction for some given distance , which is around @xmath208 . this shows that the sum of the two electrostatic energy differences @xmath209 has a non - monotonic behavior as a function of the distance @xmath1 . actually , only @xmath210 is non - monotonic , whereas @xmath211 has a positive value which decreases with @xmath1 , as can be seen from fig . if an in - plane magnetic field is applied along @xmath212 ( @xmath196 ) or @xmath68 ( @xmath198 ) direction , we obtain a dependence only either on @xmath211 or on @xmath210 . accordingly , @xmath213 will be larger in some parameter range as compared with the case of perpendicular fields , see fig . 2 . at this point it is instructive to consider numerical estimates for the coupling strength @xmath110 . for this we consider gaas quantum dots for which we assume @xmath214 , @xmath215 , @xmath216 , @xmath217 , and also consider @xmath218 , @xmath219 ( @xmath220 ) and @xmath221 . these estimations lead to a coupling strength @xmath222 , which lead to a time dynamics of the order of @xmath223 . if this time scale is longer than the decoherence times in gaas quantum dots , the system will be insensitive to the coherent dynamics induced by the coupling @xmath110 . shorter time scales are obtained for materials with larger spin - orbit coupling such as inas . the spin - orbit length @xmath3 in this material is comparable with a typical dot size of about @xmath224 . even though our perturbative approach starts to get unreliable in this case , it still can provide a rough estimate for the coupling strength . for inas we have @xmath225 , @xmath226 , @xmath217 , @xmath227 , and we choose @xmath228 , @xmath229 and @xmath230 . with those values we obtain for the coupling @xmath231 , which corresponds to a switching time of about @xmath232 for a swap of the spin states of electron one and electron two . this time scale for the spin dynamics is shorter than the expected spin decoherence time in such quantum dots . thus , this interaction mechanism provides a useful way for the dynamical control of the spin - spin coupling . as discussed before , for an in - plane magnetic field the coupling constant could even be higher , depending on the angle of the magnetic field with respect to the inter - dot axis . we briefly generalize the previous results to elliptical dot shapes . this will also allow us to study the one - dimensional limit and recover previous results obtained for one - dimensional nanowires@xcite . we consider elliptical dots which are characterized by the frequencies @xmath233 and @xmath234 corresponding to the @xmath68 and @xmath212 directions , respectively . in this case , eq . ( @xmath191 ) is replaced by @xmath235 where the electrostatic energies @xmath236 become now @xmath237 @xmath238 with the charge density distribution function , expressed in relative coordinates , @xmath239 for perpendicular magnetic fields and elliptical dots the lengths @xmath240 are given by @xcite @xmath241 where @xmath242 , @xmath243 , and @xmath244 with the explicit expressions for @xmath245 , @xmath246 , and @xmath247 @xmath248 @xmath249 @xmath250 taking now also the limit of strongly elliptical dots , @xmath136 @xmath251 , we see that this is equivalent to keeping only one component of the spin interaction , namely the @xmath252 part , and that the orbital effect of the magnetic field drops out . the resulting hamiltonian then becomes @xmath253 considering now the large distance limit , @xmath163 , analogously to eq . ( @xmath254 ) , our result reduces formally to the one in ref . , @xmath136 @xmath255 the above expression , eq . ( @xmath256 ) , could be obtained directly from eq . ( @xmath257 ) since within the considered limit there is no orbital effect of the perpendicular magnetic field on the spin - spin interaction . we note that in this limit the resulting spin - spin coupling takes the form of an ising interaction which , together with single qubit rotations , can be used@xcite to efficiently perform cnot gate operations between two qubits . we finally note that in one dimensions the rashba interaction can be treated exactly , leading to a renormalization of the @xmath258-factor@xcite , @xmath259 . this exact treatment is no longer possible in the 2d case considered here , except for the special case@xcite when @xmath260 and the problem becomes effectively 1d@xcite . we turn now to the more involved case of strong coulomb interaction strength , @xmath261 , which can not be treated perturbatively . however , some approximations are still possible and we will explore two of them in the following section . the first approximation consists in reducing the two electron system to two classical point - charge particles . the classical equilibrium condition will be obtained by minimizing the total potential energy of the two particles . by doing this , the motion of the electrons will take place around the new equilibrium positions obtained from the equation @xmath262 where @xmath1 is the initial geometric distance and @xmath263 the effective distance between the electrons in classical equilibrium . however , we are interested in the motion around the equilibrium position , which means that for small deviations , we may substitute the full coulomb interaction with an effective one , remembering that @xmath34 , @xmath264 we note that the coordinates are measured now with respect to the new equilibrium position . within this approximation , the relative hamiltonian @xmath265 is replaced with the new , renormalized one , @xmath266 @xmath267 with the definition @xmath268 and @xmath269 given by the expressions @xmath270 we see that this approximation leads only to a renormalization of the orbital frequencies of the relative hamiltonian . we proceed now to develop a second alternative for treating the coulomb interaction , namely a variational method based on the same picture of classical equilibrium . this consists in substituting the full coulomb interaction term with the same type of expression like in eq . ( @xmath271 ) , but with the effective distance obtained from the variational ansatz . for this we minimize the expectation value of the orbital hamiltonian , @xmath272 , in the ground state of the effective relative hamiltonian @xmath266[eq . ( @xmath273 ) ] with respect to the effective distance @xmath263 . this leads to the following equation @xmath274 where @xmath275 is the ground state belonging to @xmath266 , @xmath136 @xmath276 , with @xmath277 the ground state energy . since we are dealing with harmonic oscillators , those wave functions are known . however , eq . ( @xmath278 ) for the effective distance @xmath263 can be solved only numerically . we plot in fig . 3 the results obtained for the effective distance @xmath263 as a function of different parameters in both cases , namely the variational result from eq . ( @xmath278 ) and also the result obtained from the classical equilibrium condition in eq . ( @xmath271 ) . we see in fig . 3 that there is very good agreement between the two approaches in a wide parameter range and moreover , that the effective distance within the variational approach is larger then the one obtained from classical equilibrium , that means a lower ground state energy . we note that a perpendicular magnetic field practically does not change the curves in fig . 3 ( not shown ) on a large range of magnetic field strengths ( @xmath279 ) , which means that the effective distance is to a very good approximation independent of the applied magnetic field . in order to verify the accuracy of our variational method , we checked also the numerical fidelity , defined as the overlap of the wave functions in the variational case with the exact ( almost , in the sense of perturbation theory ) wave function . although the problem contains no small parameter , we can still define small matrix elements compared with level spacing in a numerical sense . for this we write the full relative hamiltonian in the following way @xmath280 with the effective coulomb interaction @xmath281 ( see eq . ( [ 20 ] ) ) expressed now also in terms of the new equilibrium coordinates ( introduced after eq . ( [ 20 ] ) ) , @xmath282 we show now that this term leads to small matrix elements so that indeed @xmath283 , where the energies @xmath284 and @xmath277 are the n - eigenvalue and ground - state energy of the hamiltonian @xmath266 , respectively . to see this numerically we introduce the fidelity @xmath285 where @xmath286 and @xmath275 are the ground state wave functions of the full hamiltonian @xmath265 and @xmath266 , respectively . we now estimate the fidelity @xmath287 by using perturbation theory to find the true ground - state wave function @xmath286 from the effective one @xmath275 @xmath288 where we retain only terms to first order in @xmath281 . taking into account eq . ( @xmath289 ) we obtain the infidelity , @xmath290 , namely the deviation of the true ground state wave function from the effective one @xmath291 we plot in fig . 4 the infidelity @xmath290 as a function of the effective distance @xmath263 for fixed coulomb strength , @xmath118 . we see that the infidelity takes very small values ( @xmath292 ) on the considered range , for two different coulomb strengths , which shows that our variational approach is very accurate . we can now evaluate the spin - spin interaction within this approximation . since we are now dealing with harmonic potentials only , the problem of finding @xmath93 from eq . ( @xmath293 ) becomes straightforward . however , in the derivation of the spin hamiltonian we need again some relations , similar to eq . ( @xmath134 ) and eq . ( @xmath294 ) , but for the present case with the harmonic oscillator renormalized . these relations read @xmath295 @xmath296 @xmath297 @xmath298 making use of the relations eq . ( @xmath299 ) and eq . ( @xmath300 ) , and also with the effective spin - orbit interaction expressed in the center - of - mass and relative coordinates , see eq . ( @xmath59 ) , the spin - hamiltonian @xmath91 takes the form @xmath301,\label{62}\ ] ] for the case of a perpendicular magnetic field and @xmath302 for the case of an in - plane magnetic field which makes an angle @xmath85 with the inter - dot distance direction . the @xmath303 is along the in - plane direction perpendicular to the in - plane magnetic field . we see that the spin hamiltonian depends on the coulomb interaction part via the difference between the inverse of the renormalized frequencies @xmath304 and the bare one @xmath22 . as expected , when there is no renormalizations of the bare frequencies ( no coulomb interaction ) the interaction vanishes . referring again to the effective spin hamiltonian @xmath107 from eq . ( @xmath193 ) , we obtain for the coupling @xmath110 for arbitrary magnetic field orientations @xmath305 where @xmath306 one can see from fig . 5b that for in - plane magnetic fields one obtains quite large values for @xmath307 in the two limiting cases @xmath196 and @xmath198 . changing the magnetic field orientation in - plane one can tune the coupling strength @xmath110 from negative to positive values , @xmath136 from ferromagnetic to antiferromagnetic regime , and make it vanish for the angle ( for in - plane magnetic field ) @xmath308 in the case of a perpendicular magnetic field , cf . 5a , we see that the coupling shows a non - monotonic behavior as a function of distance @xmath1 , and , moreover , @xmath110 vanishes for some given distance , which for @xmath309 is about @xmath310 . this could be used to tune @xmath110 on and off by changing the distance between the dots . next , we consider the case of very elliptic dots , with the bare oscillator frequencies @xmath311 corresponding to the @xmath68 and @xmath212 directions , respectively , such that @xmath312 . the spin - spin coupling becomes in this limiting case @xmath313 where both perpendicular ( @xmath196 ) and in - plane magnetic fields are contained . we see that the problem becomes effectively 1d with an ising - type spin - spin coupling , similar to the case of small coulomb coupling studied in the previous section . we mention that for finite ratio of the two bare frequencies , @xmath314 , the interaction can be varied by changing this ratio , the angle of cancellation defined in eq . ( @xmath315 ) varying as well . the behavior displayed in fig . 5 can be understood as follows . the spin - spin coupling is directly related to the deformation of the charge distributions in the two dots as a consequence of the strong coulomb interaction ( @xmath316 ) . thus , the stronger the deformation is , the stronger the spin - spin coupling becomes . or , in our case , the stronger the deviation of the renormalized orbital frequencies @xmath304 from the bare one @xmath22 is , the stronger the coupling becomes , see eqs . ( @xmath317 ) and ( @xmath318 ) . while the @xmath68 component of the spin - spin coupling is bounded because the inverse of @xmath319 tends to zero as the coulomb interaction strength @xmath112 increases , the @xmath212 component of this coupling is unbounded since the inverse of @xmath320 can grow indefinitely . consequently , the @xmath212 component will dominate the @xmath68 component for large coulomb strength and small inter - dot distance @xmath1 . however , the situation is reversed in the large distance limit , since @xmath319 increases faster than @xmath320 decreases as seen from eq . ( @xmath321 ) . these opposite limits lead to the non - monotonic behavior depicted in fig . 5a ( perpendicular magnetic field ) . we mention that the large distance limit of eq . ( @xmath322 ) converges to the large distance result obtained in the previous section , eq . ( @xmath191 ) . however , it does not converge to the results of the previous section in the case of small distance [ eq . ( @xmath203 ) ] , when going from @xmath316 to @xmath120 , a crossover description ( @xmath323 ) being needed in this situation . phrased differently , tuning the spin - spin coupling @xmath110 from strong ( @xmath316 ) to small ( @xmath120 ) coulomb interaction regime by varying the inter - dot distance reproduces the corresponding @xmath120 result in eq . ( @xmath191 ) , while by varying the ratio @xmath118 does not reproduce the corresponding @xmath324 limit , @xmath136 eq . ( @xmath203 ) . let us give now some estimates for the coupling @xmath110 when an in - plane magnetic field is applied along , say , the @xmath68-direction . assuming now gaas quantum dots , and @xmath325 ( @xmath326 ) , @xmath327 ( @xmath328 ) , @xmath329 . using these numbers and taking for the geometric inter - dot distance @xmath207 , we obtain @xmath330 . it is worth mentioning that the hyperfine interaction between the electron and the collection of nuclei in a quantum dot ( @xmath331 ) leads to similar energy scales@xcite . this shows that the spin - spin coupling derived here can be very relevant for the spin dynamics in the case of electrostatically coupled quantum dots and that it can also compete with other types of interactions . considering now the case of inas quantum dots in a magnetic field along the @xmath68 direction , with @xmath332 and @xmath333 and taking also @xmath207 , a value of @xmath334 it is obtained . however , this is just a rough estimate since the spin - orbit coupling can not be treated as a perturbation anymore and our approximation , being pushed to the limit of its range of validity , starts to break down . in this section we propose a measurement scheme for the spin - spin interaction @xmath335 . similar to the spin relaxation experiments in ref . , the left dot is monitored by a sensitive charge detector , such as a quantum point contact ( qpc ) or a single electron transistor ( set ) . we show the main steps of the scheme in fig . [ schemefigure ] . the first step is the initialization step shown in fig . [ schemefigure](a ) . at low temperatures , @xmath336 , a single - electron dot will relax to the ground state after a time larger than the spin relaxation time @xmath337 . a faster spin relaxation can be induced by cotunneling with the lead , for which the dot can be placed closer to the fermi surface for some time . in fig . [ schemefigure](a ) , the left dot is initialized in the lower zeeman sublevel @xmath338 , whereas the right dot is empty . next , the right - dot energy is lowered below the fermi energy and the dot is quickly filled with an electron in either upper or lower zeeman sublevel . this is a sequential tunneling process and we denote its rate by @xmath339 . in fig . [ schemefigure](b ) , both dots are deep below the fermi surface and @xmath335 is the energy scale that governs a coherent evolution in the subspace @xmath340 . the two spin density matrix reads @xmath341 where @xmath342 is the wave function that describes the occurrence of the state @xmath343 in the initialization step . in the ideal case , @xmath342 evolves coherently due to the spin - spin interaction , eq . ( [ 13 ] ) @xmath344 here , we neglect the cotunneling and other spin relaxation processes . in particular , the cotunneling rate @xmath345 should be much smaller than the spin - spin coupling @xmath335 . here , @xmath346 is the addition / extraction energy of the single - electron quantum dot in the coulomb blockade valley . on the other hand the sequential tunneling rate @xmath339 should be large enough , so that the spins have no time to evolve during the initialization and measurement steps . we summarize the required regime by the inequality @xmath347 after a waiting time @xmath348 , the probability of the left - dot electron to be in the upper zeeman sublevel reads , @xmath349.\ ] ] form the period of this function ( @xmath350 ) one can extract the value of the coupling constant @xmath335 . the measurement of the probability @xmath351 can be performed in the same fashion as in ref . . after the waiting time @xmath348 , the left dot is brought up to the fermi level and placed such that the electron can tunnel into the lead only from the upper zeeman sublevel . this configuration is shown in fig . [ schemefigure](c ) . tunneling of the electron out and refilling the quantum dot with an electron of the opposite spin is monitored by the charge detector close to the left dot ( not shown ) . for each value of the waiting time @xmath348 , the cycle of initialization , coherent evolution , and measurement has to be repeated many times in order to reach a good accuracy . next , we remark that the hyperfine interaction with the lattice nuclei should not impede the measurement of @xmath335 as long as @xmath352 , where @xmath245 is the atomic hyperfine coupling constant and @xmath353 is the number of nuclei in both quantum dots . note that the ratio of @xmath335 to @xmath354 for a constant @xmath355 scales with the dot lateral size as @xmath356 , for strong coulomb interaction , and as @xmath357 , for weak coulomb interaction . therefore , the regime @xmath358 can be easily achieved by taking a larger quantum dot . furthermore , the hyperfine interaction with the nuclei has only the effect of reducing the visibility of oscillations of @xmath359 , and even for @xmath360 a small part of @xmath351 shows oscillatory behavior with unchanged period , @xmath350 . in fig . [ schemefigurenucl ] , we plot the probability @xmath351 averaged over the realizations of the hyperfine field . we choose @xmath361 to show that the measurement scheme is robust against the hyperfine field . the oscillations are well visible even when @xmath354 is several times larger than @xmath362 . the averaged probability @xmath363 is obtained in the following way . for the subspace @xmath340 and in the limit @xmath364 , the coupling of spins to the hyperfine field is given by @xmath365 , where the hyperfine field @xmath366 has a gaussian distribution with zero average and a variance @xmath367 , which we take to be a measurable parameter that defines @xmath353.@xcite for more detail on the derivation of @xmath368 we refer the reader to ref . . for our description to be accurate , the time between subsequent cycles of initialization , coherent evolution , and measurement should be larger than the nuclear spin relaxation time . considering the sum of @xmath368 and @xmath369 in eq . ( [ 13 ] ) , we find that the probability @xmath351 for a fixed value of @xmath366 is given by@xcite @xmath370 the averaged probability @xmath363 is then computed by integrating eq . ( [ pldwndhz ] ) over @xmath366 with the gaussian weight factor @xmath371 , where @xmath367 . considering @xmath372 and @xmath373 , we find that the visibility of oscillations scales with @xmath374 and @xmath348 as follows @xmath375 note that the scaling law @xmath376 is weaker than what one might expect navely from eq . ( [ pldwndhz ] ) , after substituting there @xmath366 by its typical value @xmath354 , which gives @xmath377 . the reason for the weaker scaling law is the fact that @xmath366 is centered around zero and the denominator in eq . ( [ pldwndhz ] ) has nearly no effect . we find that the numerical results in fig . [ schemefigurenucl ] can be reproduced fairly accurately , if we approximate the argument of the cosine in eq . ( [ pldwndhz ] ) as follows @xmath378 this approximation is justified in the regime @xmath379 by the the minimal phase requirement , despite the fact that @xmath380 may be large . with eq . ( [ sqrtexpantion ] ) , it is easy to average @xmath351 and obtain an approximate expression , which is fairly accurate for @xmath381 and asymptotically exact in the limit @xmath379 . we thus obtain @xmath382,\label{plppm1}\\ & & \bar{p}=\sqrt{\frac{\pi}{2}}\zeta \exp\left(\frac{\zeta^2}{2}\right){\rm erfc } \left(\frac{\zeta}{\sqrt{2}}\right),\label{plppm2}\\ & & \delta p(\tau)=\frac{\zeta\cos\left[2\pi\tau/\tau_0 + \varphi_0(\tau)\right]}{\left[\zeta^4+(2\pi\tau/\tau_0)^2\right]^{1/4 } } , \label{plppm3 } \;\;\;\;\\end{aligned}\ ] ] where @xmath383 , @xmath384 is the complementary error function , and the running phase shift @xmath385 is given by @xmath386 we note that eq . ( [ plppm1 ] ) is exact in two limiting cases : @xmath387 and @xmath379 . in particular , in the limit @xmath388 , we recover @xmath351 in eq . ( [ pldwndhz ] ) for all values of @xmath348 . in the opposite limit , @xmath389 , eqs . ( [ plppm1])@xmath390([varpho0pl ] ) can be significantly simplified , yielding @xmath391.\ ] ] in the entire derivation we assumed no tunneling between the dots . even in the presence of tunneling , when direct coulomb repulsion , @xmath392 ( which is just the classical interaction between two charge distributions ) is larger than the exchange interaction , @xmath393 ( @xmath394 ) , the theory presented here is expected to remain still valid . the reason is that this type of spin - spin coupling is a direct consequence of the deformation of the electronic charge distribution due to coulomb repulsion between the two electrons . since this is given by the sum of the direct coulomb part @xmath392 , and the exchange part @xmath395 , the spin coupling @xmath110 will be insensitive to exchange in the limit @xmath396 . the point now is that with finite tunneling , even with the assumption that the coupling strength @xmath110 is not modified by the exchange coulomb interaction , the resulting heisenberg exchange coupling @xmath395 ( @xmath397 ) will start to compete with the electrostatically induced spin interaction . as a consequence , the spin coupling @xmath110 will be washed out in the limit @xmath398 . however , since the heisenberg exchange coupling decays with the inter - dot distance like@xcite @xmath399 while @xmath400 , the electrostatic spin coupling will start to dominate at not very large distances . we recall that when having exchange , another type of spin coupling , @xmath401 , induced by spin - orbit interaction comes into play , and which is proportional to the heisenberg coupling @xmath395 , i.e. @xmath402 for typical gaas dots @xcite . thus , @xmath401 can be much larger than @xmath335 for large enough heisenberg coupling , with a crossover from this exchange - type to the direct coulomb - type coupling taking place at some inter - dot distance . this crossover , however , occurs before we get @xmath403 , since @xmath401 is typically 100 times smaller than @xmath395 . to give an estimate , we assume @xmath404 for @xmath405 , which gives @xmath406 for @xmath407 , implying that @xmath408 . another important issue is the effect of screening induced by the surrounding electrons in the 2deg and the metallic gates . as is well - known , the screening effect between two charges becomes important for distances exceeding the screening length @xmath409 ( fermi liquids ) . however , the screening of bare coulomb interaction depends strongly on the dimensionality . in 3d the effect of screening is to induce an exponential decay of the bare coulomb interaction @xcite , with the decay parameter @xmath410 , while in 2d the decay follows a power law ( @xmath411 in the large distance limit ) @xcite , with @xmath410 being the relevant length scale . for gaas , the screening length is around @xmath412 . moreover , additional screening is introduced by the electrodes to gate the dots , due to their metallic character . the finite screening implies then that our theory in fact overestimates the strength of the electrostatically induced spin coupling @xmath110 for distances exceeding this screening length and the results obtained here become just an upper bound on @xmath110 for this limit . being highly controllable , the coupling @xmath110 could be used to perform two qubit gates for the realization of quantum computers with electron spins , like proposed in ref . . the switching times range between rather slow ( @xmath413 in gaas ) and reasonably fast ( @xmath414 in inas ) . when making use of the standard exchange coupling@xcite for switching ( with typical switching times of @xmath415 in gaas ) the electrostatically induced spin - coupling found here can lead to gate errors . however , this effect can be controlled by choosing the magnetic field direction or strength and/or the interdot distance such that @xmath110 becomes negligibly small ( see eq . ( @xmath416 ) ) . finally , an important question is how orbital fluctuations ( for example of the confining energy @xmath114 ) mediated via spin - orbit coupling lead to fluctuations in the coupling @xmath110 and thus eventually to spin decoherence . the relation between the orbital dephasing time ( which is assumed to be known ) and the decoherence induced by the spin coupling @xmath110 reads @xcite @xmath417 where @xmath418 is the orbital dephasing time and @xmath419 the corresponding spin decoherence time . we also assumed only fluctuations in the dot size , which means fluctuations of the confinement frequency @xmath22 . assuming an orbital dephasing time @xmath420 and also the limiting case of touching dots with the same gaas parameters as before we obtain a spin decoherence time ( lower bound ) @xmath421 . we can conclude then that the incoherent part due to this type of coupling is negligible compared with other types of decoherence mechanisms , @xmath8 induced by the hyperfine interaction@xcite . to conclude , we have derived an effective spin - spin interaction between two electrons localized in two quantum dots , spatially separated , induced by the direct coulomb interaction and mediated by the spin - orbit coupling . this interaction was found to have the form of an anisotropic @xmath0 interaction and to be proportional to the zeeman energy . the spin - spin coupling was studied both in the weak and strong coulomb interaction limits and for different magnetic field orientations and strengths . the important features are the non - monotonic behavior of this spin interaction for some magnetic field orientations , together with a vanishing of this interaction for particular inter - dot distances . this effect can be used to manipulate the spin - spin interaction in electrostatically coupled quantum dots by tuning the inter - dot distance . we proposed a measurement setup which allows one to access this spin - spin coupling experimentally . we thank d. stepanenko , g. burkard , w. a. coish , and d. klauser for useful discussions . we acknowledge financial support from the swiss nsf , the nccr nanoscience , darpa , onr , and jst icorp . in this appendix we give explicit formulas for the couplings @xmath93 for an arbitrary magnetic field orientation @xmath423 and for both rashba and dresselhaus spin - orbit couplings present . these are obtained by diagonalizing the tensor @xmath90 , which leads to with @xmath425 ( the angle @xmath70 is defined after eq . ( @xmath59 ) ) . the functions @xmath426 ( @xmath427 ) can be expressed in terms of @xmath94|0\rangle$ ] , @xmath95 , @xmath136 @xmath428 @xmath429 @xmath430 these functions can be identified very easily from our formulas derived in the paper . for example , for the case considered in eq . ( @xmath186 ) ( weak coulomb coupling regime ) we get @xmath431 , d.d . awschalom , d. loss , and n. samarth ( eds . ) , ( springer , berlin , 2002 ) . v. cerletti , w. a. coish , o. gywat , and d. loss , nanotechnology * 16 * , r27 ( 2005 ) . j. m. kikkawa and d. d. awschalom , phys . rev . lett . * 80 * , 4313 ( 1998 ) . d. loss and d. p. divincenzo , phys . a * 57 * , 120 ( 1998 ) . a. imamoglu , d.d . awschalom , g. burkard , d. p. divincenzo , d. loss , m. sherwin , and a. small , phys . lett . * 83 * , 4204 ( 1999 ) . m. ciorga , a. s. sachrajda , p. hawrylak , c. gould , p. zawadzki , s. jullian , y. feng , and z. wasilewski , phys . b * 61 * , r16315 ( 2000 ) . k. ono , d. g. austing , y. tokura , and s. tarucha , _ science _ * 297 * , 1313 ( 2002 ) . j. m. elzerman , r. hanson , l. h. willems van beveren , b. witkamp , l. m. k. vandersypen , and l. p. kouwenhoven , _ nature _ * 430 * , 431 ( 2004 ) . m. kroutvar , y. ducommun , d. heiss , m. bichler , d. schuh , g. abstreiter , and j. j. finley , _ nature _ * 432 * , 81 ( 2004 ) . j. r. petta , a. c. johnson , j. m. taylor , e. a. laird , a. yacoby , m. d. lukin , c. m. marcus , m. p. hanson , a. c. gossard , _ science _ * 309 * , 2180 ( 2005 ) . c. w. lai , p. maletinsky , a. badolato , and a. imamoglu , phys . lett . * 96 * , 167403 ( 2006 ) . f. h. l. koppens , c. buizert , k. j. tielrooij , i. t. vink , k. c. nowack , t. meunier , l. p. kouwenhoven , and l. m. k. vandersypen , _ nature _ * 442 * , 766 ( 2006 ) . s. amasha , k. maclean , i. radu , d. m. zumbuhl , m. a. kastner , m. p. hanson , and a. c. gossard , cond - mat/0607110 . melnikov and e.i . rashba , sov . jetp * 34 * , 1353 ( 1972 ) . m. dobrowolska , a. witowski , j. k. furdyna , t. ichiguchi , h. d. drew , and p. a. wolff , phys . b * 29 * , 6652 ( 1984 ) . g. salis , y. kato , k. ensslin , d. c. driscoll , a. c. gossard , and d. d. awschalom , _ nature _ * 414 * , 619 ( 2001 ) . rashba and al.l . efros , appl . lett . , * 83 * , 5295 ( 2003 ) . m. duckheim and d. loss , _ nature physics _ , * 3 * , 195 ( 2006 ) . v. n. golovach , m. borhani , and d. loss , cond - mat/0601674 . a. v. khaetskii and yu . v. nazarov , phys . rev . b * 64 * , 125316 ( 2001 ) . v. n. golovach , a. khaetskii , and d. loss , phys . rev . * 93 * , 016601 ( 2004 ) . m. borhani , v. n. golovach , and d. loss , phys . b 73 , 155311 ( 2006 ) . a. abragam , _ principles of nuclear magnetism _ ( oxford university press , 1961 ) . n. zhao , l. zhong , j .- l . zhu , and c. p. sun , cond - mat/0603363 . c. flindt , a. s. sorensen , and k. flensberg , cond - mat/0603559 . to distinguish between the two mechanisms we refer to the coupling studied in this work as spin - spin coupling as opposed to the heisenberg exchange coupling . g. burkard , d. loss , and d. p. divincenzo , phys . b * 59 * , 2070 ( 1999 ) . x. hu and s. das sarma , phys . a * 61 * , 062301 ( 2000 ) . j. schliemann , d. loss , and a. h. macdonald , phys . b * 63 * , 085311 ( 2001 ) . k. k. kavokin , phys . b * 64 * , 075305 ( 2001 ) . n. e. bonesteel , d. stepanenko , and d. p. divincenzo , phys . * 87 * , 207901 ( 2001 ) . g. burkard and d. loss , phys . lett . * 88 * , 047903 ( 2002 ) . d. stepanenko , n. e. bonesteel , d. p. divincenzo , g. burkard , and d. loss , phys . b * 68 * , 115306 ( 2003 ) . d. stepanenko and n. e. bonesteel , phys . lett . * 93 * , 140501 ( 2004 ) . g. dresselhaus , phys . rev . * 100 * , 580 ( 1955 ) . y. bychkov and e. i. rashba , j. phys . c * 17 * , 6039 ( 1984 ) . n. f. johnson and m. c. payne , phys . 67 * , 1157 ( 1991 ) . v. fock , z. phys . * 47 * , 446 ( 1928 ) . darwin , math . cambridge philos . * 27 * , 86 ( 1930 ) . d. bulaev and d. loss , phys . b * 71 * , 205324 ( 2005 ) . o. dippel , p. schmelcher , and l. s. cederbaum , phys . a * 49 * , 4415 ( 1994 ) . j. schliemann , j. c. egues , and d. loss , phys . lett . * 90 * , 146801 ( 2003 ) . a. khaetskii , d. loss , and l. glazman , phys . lett . * 88 * , 186802 ( 2002 ) . w. a. coish and d. loss , phys . b * 70 * , 195340 ( 2004 ) . more rigorously , @xmath432 , where @xmath433 is the variance of the @xmath27-component of a single nuclear spin ( assuming @xmath434 ) and @xmath435 , with @xmath436 being the electron wave function and @xmath437 the unit cell volume per nuclear spin . d. pines and p. nozieres _ the theory of quantum liquids , vol . 1 _ ( benjamin , new york , 1966 ) . f. stern , phys . rev . lett . * 18 * , 546 ( 1967 ) . t. ando , a. b. fowler , and f. stern , rev . phys . * 54 * , 437 ( 1982 ) . c. w. j. beenakker and h. van houten , in solid state physics : advances in research and applications , edited by h. ehrenreich and d. turnbull ( academic , new york , 1991 ) , vol . 44 , pp . 1 - 228 . d. saraga and d. loss , phys . rev b * 72 * , 195319 ( 2005 ) . w.a . coish and d. loss , phys . b * 72 * , 125337 ( 2005 ) .

the standard model ( sm ) has passed the tests provided by run 1 of the lhc with flying colours . many cross sections for particle and jet production have been measured at the lhc @xcite , and are in agreement with the sm predictions , as seen in fig . [ fig : sections ] . jet production cross sections agree over large ranges in energy and many orders of magnitude with qcd calculations within the sm , as do measurements of single and multiple @xmath0 and @xmath1 production and measurements of single and pair production of the top quark . the biggest headline of lhc run 1 was of course the discovery by cms and atlas of a ( the ? ) higgs boson @xcite , which has now been detected in three production channels , as also seen in fig . [ fig : sections ] , also in agreement with the sm predictions . much of this talk will concern what we already know about this newly - discovered particle , and the hints it may provide for other new physics , as well as other topics within and beyond the standard model . qcd is the basis for lhc physics : it provides many tests of the standard model as well as dominating particle production and deluging us with with backgrounds and pile - up events . the agreement between qcd predictions and measurements of large-@xmath2 jet production at the lhc over many orders of magnitude yields measurements of the strong coupling that are consistent with the world average value @xmath3 , and demonstrate that @xmath4 continues to run downward beyond the tev scale @xcite , perhaps towards grand unification , as seen in fig . [ fig : run ] . continues to run downward at energies beyond @xmath5 tev @xcite.__,height=192 ] not only are perturbative qcd calculations doing a fantastic job overall of predicting the production cross sections for jets and massive vector bosons , but also for the higgs boson . accurate higher - order qcd calculations are at a premium for the dominant gluon - fusion contribution to the higgs production cross section . several different nnlo calculations are available , and are included in various publicly - available tools @xcite . unfortunately , the agreement between them is not yet perfect . fortunately , progress is being made on nnnlo calculations @xcite . these will improve the theoretical accuracy , but progress in convergence between the parton distribution functions will also be needed in order to reduce the theoretical uncertainties below the experimental measurement uncertainties . another pillar of the sm is the cabibbo - kobayashi - maskawa ( ckm ) model of flavour mixing and cp violation . it is in general very successful , as seen in fig . [ fig : hs ] @xcite . for example , the second - greatest discovery during run 1 of the lhc was perhaps the measurement by the cms and lhcb collaborations of the rare decay @xmath6 , with a branching ratio in good agreement with the sm prediction @xcite : @xmath7 as seen in fig . [ fig : bmumu ] . however , the joint cms and lhcb analysis @xcite also has an suggestion of a @xmath8 signal that is larger than the sm prediction : @xmath9 as also seen in fig . [ fig : bmumu ] . if confirmed , this measurement would conflict not just with the sm , but also models with minimal flavour violation ( mfv ) , including many supersymmetric scenarios . something to watch during run 2 ! decays , including * b * a clear signal for @xmath6 decay that agrees with the sm , and * c * a hint of @xmath8 decay , possibly at a rate larger than expected in the sm @xcite.__,height=124 ] there is scope elsewhere for deviations from ckm predictions : for example , the data allow an important contribution to @xmath10 meson mixing from physics beyond the sm ( bsm ) @xcite . also , there are issues with @xmath11 universality in semileptonic @xmath12 decays @xcite and a persistent anomaly in the @xmath13 angular distribution for @xmath14 @xcite . could this be related to the intriguing excess in @xmath15 decay reported by the cms collaboration @xcite , which is discussed later ? other points to watch include discrepancies in the determinations of the @xmath16 ckm matrix element and the tevatron diimuon asymmetry anomaly @xcite . however , some anomalies do seem to be going away , such as the branching ratio for @xmath17 decay , which is now in good agreement with the sm @xcite and the forward - backward asymmetry in @xmath18 production @xcite , which is consistent with the latest higher - order qcd calculations @xcite , as is the @xmath18 rapidity asymmetry measured at the lhc . however , there are still plenty of flavour physics issues to be addressed during lhc run 2 . the higgs boson may be regarded as , on the one hand , the capstone of the glorious arch of the sm or , on the other hand , as the portal giving access to new physics . in this section we discuss first the extent to which the new particle discovered on july 4th , 2012 fulfils its sm rle , and then what hints it may be able to provide about possible bsm physics . the mass of the higgs boson is measured most accurately in the @xmath19 and @xmath20 final states , and atlas and cms have both reported accurate measurements in each of these final states as shown in fig . [ fig : mh ] . atlas measures @xcite @xmath21 and cms measures @xcite @xmath22 combining all these measurements , the atlas and cms collaborations find @xcite @xmath23 in addition to being a fundamental measurement in its own right , and casting light on the possible validity of various bsm models ( for example , this value is perfectly consistent with supersymmetric predictions @xcite ) , the precise value of @xmath24 is also important for the stability of the electroweak vacuum in the standard model @xcite , as discussed later . by atlas and cms in the @xmath19 and @xmath25 final states , as complied in @xcite.__,height=192 ] the fact that the higgs boson decays into @xmath19 excludes spin 1 , and spin 0 is expected , but spins 2 and higher are also possible in principle . the higgs spin has been probed in many ways @xcite , via its production and decay rates @xcite , the kinematics of higgs production in association with the @xmath0 and @xmath1 @xcite , and decay angular distributions for @xmath26 , @xmath27 and @xmath19 final states @xcite . the results of tests using the kinematics of associated @xmath28 production at the tevatron are shown in fig . [ fig : higgsspin ] @xcite . by now there is overwhelming evidence against the higgs boson having spin 2 . moreover , as also seen in fig . [ fig : higgsspin ] @xcite , it has been established that its couplings to @xmath26 and @xmath27 are predominantly cp - even , i.e. , it couples mainly as a scalar , not as a pseudoscalar . production at the tevatron @xcite.__,height=192 ] as seen in fig . [ fig : strengths ] , the strengths of the higgs signals measured by atlas in individual channels @xcite are generally compatible with the sm predictions ( as are cms measurements @xcite ) within the statistical fluctuations , which are inevitably large at this stage . atlas and cms report the following overall signal strengths after combining their measurements in the @xmath19 , @xmath29 , @xmath30 , @xmath31 and @xmath32 channels : @xmath33 both averages are quite compatible with the sm and with each other , as is also true of the tevatron measurements @xcite . , normalised to unity for the sm , as measured by atlas @xcite.__,height=220 ] because of its connection to mass generation , the couplings of the higgs boson to other particles in the sm should be related to their masses : linearly for fermions , quadratically for bosons , and be scaled by the higgs vev @xmath34 gev . these predictions are implicit in the measurements in fig . [ fig : strengths ] , and are tested directly in fig . [ fig : mass_dependence ] . the latter displays a global fit in which the higgs coupling data are parametrised as @xcite @xmath35 as seen in the left panel of fig . [ fig : mass_dependence ] , the data yield @xmath36 which is in excellent agreement with the sm predictions @xmath37 , @xmath38 gev . similar results have also been found recently in an analysis by the cms collaboration @xcite . it seems that higgs couplings indeed have the expected characteristic dependence on particle masses . couplings of the form ( [ epsilon ] ) ( central values as dashed and @xmath391@xmath40 values as dotted lines ) , which is very consistent with the linear mass dependence for fermions and quadratic mass dependence for bosons ( solid red line ) expected in the sm @xcite.__,height=206 ] according to the sm , flavour should be conserved to a very good approximation in higgs couplings to fermions . this is consistent with the available upper limits on low - energy effective flavour - changing interactions , which would , however , also allow lepton - flavour - violating higgs couplings that are much larger than the sm predictions @xcite . looking for such interactions is therefore a possible window on bsm physics . on the basis of low - energy data , we estimated that the branching ratios for @xmath15 and @xmath41 decays could each be as large as @xmath42% , i.e. , as large as br@xmath43 , whereas the branching ratio for @xmath44 could only be much smaller , @xmath45 @xcite . the cms collaboration has recently reported a measurement @xcite @xmath46 which is @xmath47 different from zero . sm flavour physics predictions are therefore being probed more stringently by the lhc than by low - energy experiments , and we are keen to see corresponding results from atlas and from run 2 of the lhc ! decay @xcite.__,height=288 ] although all the indications are that the dominant higgs couplings are cp - even , as seen , e.g. , in fig . [ fig : higgsspin ] above , there may also be an admixture of cp - odd couplings , whose fraction may depend on the particle whose coupling to the higgs boson are being probed . since the leading cp - odd @xmath48 coupling to fermions would have the _ same _ ( zero ) dimension as the leading cp - even coupling , whereas the leading cp - odd @xmath48 coupling to massive vector bosons would have _ higher _ dimension than the leading cp - even coupling , the latter may be more suppressed . ideas for probing cp violation in @xmath49 decay have been suggested @xcite , and cp violation may also be probed in the @xmath50 couplings @xcite . as seen in fig . [ fig : thcp ] , this could affect the total cross sections for associated @xmath50 , @xmath51 and @xmath52 production , shown as functions of @xmath53 arctan(cp - odd coupling / cp - even coupling ) . if @xmath54 , a cp - violating transverse polarization asymmetry is in principle observable in @xmath51 and @xmath52 production , as discussed in @xcite . , @xmath51 and @xmath52 production cross sections , taking into account the current constraints from the @xmath55 and @xmath56 couplings @xcite.__,height=240 ] one of the key questions about the higgs boson is whether it is elementary or composite . one might have thought that a composite higgs boson would naturally have a mass comparable to the scale of compositeness , but the mass can be suppressed if it is a pseudo - nambu - goldstone boson with a mass that is protected by some approximate symmetry , perhaps becoming consistent with the measured higgs mass @xmath57 gev . this possibility may be probed using a phenomenological lagrangian @xmath58 with free @xmath48 couplings , that may be constrained using @xmath48 decay and production data . since the standard model relation @xmath59 agrees well with the data , one usually assumes that the phenomenological lagrangian has a custodial symmetry : su(2)@xmath60su(2 ) @xmath61 su(2 ) . then one may parametrise the leading - order terms in @xmath58 as follows : @xmath62 where @xmath63 the free coefficients @xmath64 and @xmath65 are all normalised so that they are unity in the sm , and one searches for observable deviations from these values that could be signatures of composite models . [ fig : ey ] shows one such analysis @xcite , that looked for possible rescalings of the @xmath48 couplings to bosons by a factor @xmath66 and to fermions by a factor @xmath67 and @xmath68 @xcite , which are used by atlas and cms . ] . fig . [ fig : ey ] shows no sign of any deviation from the sm predictions @xmath69 . the yellow lines in fig . [ fig : ey ] show the predictions of specific composite models that are excluded unless ( in some cases ) their predictions are adjusted to resemble those of the sm . couplings rescaled by factors @xmath66 and @xmath67 , respectively . the sm prediction @xmath69 is shown as the green star @xcite , and the yellow lines show the possible predictions of some composite models.__,height=220 ] since the properties of the higgs boson as well as other particles continue to agree with the sm , it is increasingly popular approach to to regard the sm as an effective field theory ( eft ) valid at low energies @xmath70 tev . the effects of bsm physics at higher scales may then be parametrised via higher - dimensional eft operators constructed out of sm fields , as a first approximation , with coefficients that can be constrained by precision electroweak data , higgs data and triple - gauge couplings ( tgcs ) . @xcite discusses the operators entering electroweak precision tests ( ewpts ) at lep , together with @xmath71 cl bounds on their individual coefficients when they are switched on one at a time , and also when marginalised in a simultaneous global fit . results for the eft coefficients @xmath72 and @xmath73 , which affect the leptonic observables @xmath74 , and the eft coefficients @xmath75 and @xmath76 , which contribute to the hadronic observables @xmath77 , are shown in fig . [ fig : ewptsummary ] . the upper ( green ) bars show the ranges for each of eft coefficient when it is varied individually , assuming that the other eft coefficients vanish , and the lower ( red ) bars show the ranges for a global fit in which all the eft coefficients are allowed to vary simultaneously , neglecting any possible correlations . the ranges of the coefficients are translated in the legend at the top of the left panel of fig . [ fig : ewptsummary ] into ranges of a large mass scale @xmath78 . all the sensitivities are in the multi - tev range . for @xmath79 . @xcite _ ] other operators contribute to higgs physics and tgcs , and important information on possible values of their coefficients is provided by kinematic distributions @xcite , as well as by total rates , as illustrated in fig . [ fig : distributions ] . distribution in @xmath80 events at the lhc showing the sm expectation ( purple shading with solid outline ) , and the distributions with @xmath81 = 0.1 and 0.05 , respectively as red - dotted and blue - dashed lines @xcite . lower panel : the same - flavour @xmath2 distribution of the leading lepton in a tgc analysis . the standard model distribution is shaded blue with solid lines , and the distribution for @xmath82 is shaded green with dashed lines . in both cases the last ( overflow ) bin provides important extra information in addition to the overall normalisation @xcite.__,title="fig : " ] + distribution in @xmath80 events at the lhc showing the sm expectation ( purple shading with solid outline ) , and the distributions with @xmath81 = 0.1 and 0.05 , respectively as red - dotted and blue - dashed lines @xcite . lower panel : the same - flavour @xmath2 distribution of the leading lepton in a tgc analysis . the standard model distribution is shaded blue with solid lines , and the distribution for @xmath82 is shaded green with dashed lines . in both cases the last ( overflow ) bin provides important extra information in addition to the overall normalisation @xcite.__,title="fig : " ] fig . [ fig : fitsummary ] @xcite shows a global fit to the higgs data , including associated production kinematics , and lhc tgc measurements . the individual 95% cl constraints with one non - zero eft coefficient at a time are shown as green bars . the other lines show marginalised 95% ranges fund using the higgs signal - strength data in conjunction with the kinematic distributions for associated @xmath83 production measured by atlas and d0 ( blue bars ) , with the lhc tgc data ( red lines ) , and with both ( black bars ) . the lhc tgc constraints are the most important for @xmath84 and @xmath85 , whereas the higgs constraints are more important for @xmath86 , @xmath87 and @xmath88 . in my view , the eft approach is the most promising framework for analysing run 2 results , in particular because it can be used to tie together many different classes of measurements . production measured by atlas and d0 ( blue bars ) , with the lhc tgc data ( red lines ) , and the global combination with both the associated production and tgc data ( black bars ) . note that @xmath89 are shown @xmath90 , for which the upper axis should be read @xmath91 @xcite . _ _ ] _ the more important fundamental laws and facts of physical science have all been discovered " _ said albert michelson in 1894 , just before radioactivity and the electron were discovered . _ there is nothing new to be discovered in physics now , all that remains is more and more precise measurement " _ said lord kelvin in 1900 , just before einstein s _ annus mirabilis _ in 1905 . similarly , today there are many reasons to expect physics beyond the sm even ( particularly after the discovery of a ( the ? ) higgs boson , as i now discuss . as james bond might have said @xcite , there are 007 important reasons . 1 ) the measured values of @xmath92 and @xmath24 indicate that the electroweak vacuum is _ probably _ unstable , in the absence of some bsm physics . 2 ) the dark matter required by astrophysics and cosmology has no possible origin within the sm . 3 ) the origin of the matter in the universe requires additional cp violation beyond ckm . 4 ) the small neutrino masses seem to require bsm physics . 5 ) the hierarchy of mass scales could appear more natural in the presence of some new physics at the tev scale . 6 ) cosmological inflation requires bsm physics , since the effective higgs potential in the sm would seem to become negative at high scales . 7 ) quantising gravity would certainly require physics ( far ) beyond the sm . the first two of these issues are discussed in the following . in the sm with its su(2)@xmath60u(1 ) symmetry , the origin where @xmath93 is unstable and surrounded by a valley where @xmath94 gev , the present electroweak vacuum . at larger higgs field values , the effective potential rises , at least for a while . however , calculations in the sm show that , for the measured values of @xmath92 and @xmath24 , the effective potential turns down as a result of renormalization of the higgs self - coupling by the top quark , which overwhelms that by the higgs itself . thus , the present electroweak vacuum is in principle unstable in the sm , and quantum tunnelling generates collapse into an anti - de - sitter big crunch. the sm calculations @xcite shown in the upper panel of fig . [ fig : buttazzo ] indicate that the instability sets in at a higgs scale @xmath78 : @xmath95 uisng the world average values of @xmath92 , @xmath24 and @xmath96 , this formula yields @xmath97 however , we see in the lower panel of fig . [ fig : buttazzo ] that this calculation is very sensitive to @xmath92 . note in this connection that the d0 collaboration has recently reported a new , higher , value of @xmath92 @xcite ( tending to make the vacuum more unstable ) , whereas the cms collaboration has reported lower values of @xmath92 from new analyses @xcite ( tending to make the vacuum more stable ) . a more accurate experimental measurement of @xmath92 would help us understand the fate of the universe within the sm , and the possible need for bsm physics to stabilise the electroweak vacuum , but we also need to understand better the relationship between the parameter @xmath92 in the sm lagrangian and the effective mass parameter measured by experiments @xcite . at large scales , destabilising the electroweak vacuum . right panel : regions of vacuum stability , metastability and instability in the @xmath98 plane . both panels are from @xcite.__,title="fig:",height=192 ] + at large scales , destabilising the electroweak vacuum . right panel : regions of vacuum stability , metastability and instability in the @xmath98 plane . both panels are from @xcite.__,title="fig:",height=153 ] even if the lifetime of the electroweak vacuum is much longer than the age of the universe , as suggested by these calculations , one can not simply ignore this problem . the early universe is thought to have had a very high energy density , e.g. , during an inflationary epoch @xcite , at which time quantum and thermal fluctuations at that time would have populated the anti - de - sitter ` big crunch ' region @xcite . however , it is possible that we might have been lucky enough to live in a non - anti - de - sitter region and thereby surviving @xcite . the problem could be avoided with suitable new physics beyond the sm , of which one example is supersymmetry @xcite . one may love supersymmetry ( susy ) for many reasons , such as rendering the hierarchy problem more natural , providing a candidate for the cold dark matter , aiding grand unification and its essential ( ? ) rle in string theory . in my mind , run 1 of the lhc has added three more reasons , namely the mass of the higgs boson , which was predicted successfully by supersymmetry @xcite , the fact that the higgs couplings are similar to those of the sm higgs boson , as discussed earlier and as expected in simple realisations of the mssm @xcite , and the stabilisation of the electroweak vacuum , as mentioned just above . how can we resist susy s manifold charms ? however , so far susy has kept coyly out of sight in searches at the lhc , direct searches for the scattering of dark matter particles , indirect searches in flavour physics , etc .. where could susy be hiding ? we know that susy must be a broken symmetry , but we do not know how , so we do not know what the susy spectrum may be . it is often assumed that there is a discrete r - symmetry , which would guarantee the stability of the lightest supersymmetric particle ( lsp ) , providing the above - mentioned dark matter candidate . it is often assumed that the susy - breaking sparticle masses are universal at some high renormalisation scale , usually the gut scale , but this has no strong theoretical justification . the simplest model is one in which all the susy - breaking contributions @xmath99 to the squark , slepton and higgs masses are equal at the gut scale , and the su(3 ) , su(2 ) and u(1 ) gauging masses @xmath100 are also universal , which is called the constrained mssm ( cmssm ) . it could also be that the susy - breaking contributions to the masses of the two higgs doublets of the mssm differ from those of the squarks and leptons , and may be equal to each other ( the nuhm1 ) , or different from each other ( the nuhm2 ) . alternatively , one may consider the phenomenological mssm ( pmssm ) in which no gut - scale universality is assumed . some results of global fits to the cmssm , nuhm1 , nuhm2 and a version of the pmssm with 10 free susy - breaking parameters , combining all experimental and phenomenological constraints and requiring that the relic supersymmetric particle density be within the cosmological range , are shown in fig . [ fig : mc11 ] @xcite . the upper panel shows the profile likelihood functions for the gluino mass in these models , and the lower panel shows the likelihood functions for the first - and second - generation squarks ( which are assumed to be equal in the pmssm10 ) . in the gut - universal models the 95% cl lower limits on the squark and gluino masses are @xmath101 gev , whereas they could be significantly lighter in the pmssm10 , which offers greater prospects for discovering susy in lhc run 2 @xcite . + the pmssm revives the possibility of explaining the discrepancy between the sm calculation of @xmath102 and the experimental measurement within a susy model . this is not possible in the cmssm , nuhm1 and nuhm2 , because of the lhc constraints , and these models predict values of the @xmath102 similar to the sm prediction , as shown by the blue lines in fig . [ fig : g-2 ] . however , the black line in this figure shows that the experimental measurement could be accommodated within the pmssm @xcite . there are plans for two new experiments to measure @xmath102 @xcite , and other low - energy @xmath103 experiments will help clarify the discrepancy between the sm and experiment . if this is indeed due to susy , our pmssm10 analysis suggests that its discovery may not be far away ! in particular , there are prospects in searches for jets + missing transverse energy searches at the lhc , as well as dedicated searches for sleptons and light stop squarks @xcite . as already mentioned , a supersymmetric model that conserves r - parity has a natural candidate for a cold dark matter particle , and this is often taken to be the lightest neutralino @xmath104 @xcite ( though other candidates are also possible ) . the present limits from direct searches for the scattering of massive cold dark matter particles in underground experiments are shown in the upper panel of fig . [ fig : dmcomparison ] , together with predictions in the pmssm10 @xcite . the 68% cl region in this model ( outlined by the red contour ) lies just below the current experimental limit and within range of the planned lz experiment ( magenta line ) @xcite . other tev - scale extensions of the sm , such as some extra - dimensional models with k - parity and little higgs models with t - parity , also have possible candidates . it is therefore useful to make a model - independent comparison of the capabilities of the lhc and direct dark matter searches , and one such is shown in the lower panel of fig . [ fig : dmcomparison ] . this compares direct astrophysical searches for the scattering of generic tev - scale dark matter particles with the current reaches of the lhc via monojet searches , for the cases of spin - dependent ( axial ) couplings ( left panel ) and spin - independent ( vector ) couplings ( right panel ) @xcite . in the former case the lhc monojet searches generally have greater sensitivity than the direct searches , except for dark matter particle masses @xmath105 tev where the lhc runs out of phase space . on the other hand , direct searches for spin - independent interactions are more sensitive than the lhc searches for masses @xmath106 gev . susy models generally favour a relatively large mass for the dark matter particle , the pmssm10 being one example , as seen in the upper panel of fig . [ fig : dmcomparison ] . + -plane @xcite , showing the regions excluded by the xenon100 and lux experiments ( shaded green ) , the neutrino ` floor ' ( shaded yellow ) , and the prospective sensitivity of the lz experiment @xcite . the preferred 68% cl region is outlined in red , and the region allowed at the 95% cl is outlined in blue . the solid contours incorporate the lhc constraints , and the dashed contours omit them . lower panels : a comparison of the current 90% cl direct search limits from lux and supercdms ( red and orange lines , respectively ) , the monojet limits in simple models ( blue lines ) and the limits in an effective field theory framework ( green line ) in the cross section vs @xmath107 plane used by the direct detection community . the left and right panels show , respectively , the limits on the spin - dependent and spin - independent cross sections appropriate for axial- vector and vector mediators @xcite . _ , there is longstanding interest in building a high - energy @xmath103 collider , which might be linear such as the ilc ( @xmath108 tev ) or clic ( @xmath109 tev ) . on the other hand , there is now interest in europe and china in a possible large circular tunnel to contain an @xmath103 collider with @xmath110 gev and/or a @xmath111 collider with @xmath112 tev @xcite . a circular @xmath103 collider would provide measurements of the @xmath113 and higgs bosons of unparalleled accuracy , as seen in fig . [ fig : tlep ] @xcite . these would be able to distinguish between the predictions of the sm and various fits in the cmssm , nuhm1 and nuhm2 , as shown , if one can also reduce correspondingly the present theoretical uncertainties , which are indicated in the right panel by the shaded green bars . the other coloured bars illustrate the accuracies attainable with measurements at various accelerators . properties ( left panel ) and higgs couplings ( right panel ) , together with the prospective precisions of possible measurements at future colliders and the deviations from the sm predictions found at the best - fit points in various susy models . the right panel also shows the current theoretical uncertainties . from @xcite.__,height=192 ] a future high - energy @xmath111 collider would produce many more higgs bosons than the lhc , as seen in the upper panel of fig . [ fig : fcc - hh ] @xcite , offering the possibility of measuring higgs couplings with greater statistical accuracy , and also including the elusive triple - higgs coupling . a high - energy @xmath111 collider would also offer unique possibilities to discover and/or measure the properties of susy particles . even the susy dark matter particle could weigh several tev , as seen in the lower panel of fig . [ fig : fcc - hh ] @xcite , which illustrates a strip in the cmssm parameter space where the relic neutralino density is brought into the the range allowed by cosmology through coannihilation with the lighter stop squark . in the example shown , the lightest neutralino weighs @xmath114 tev and only a @xmath111 collider with @xmath115 tev would be able to explore all the range of particle masses compatible with susy providing dark matter ( solid and upper dashed blue lines ) . for all this range calculations of the higgs mass are compatible with the experimental value ( represented by the yellow band ) , considering the theoretical uncertainties represented by the solid and dashed green lines . colliders as functions of the centre - of - mass energy @xcite . lower panel : one of the possibilities for a relatively heavy susy dark matter particle weighing @xmath116 tev . the vertical axis is the mass difference between the dark matter particle and the next - to - lightest supersymmetric particle , in this case the lighter stop squark . the solid and upper dashed blue lines correspond to the current central and + 1@xmath40 values of the dark matter density , the horizontal yellow band represents the experimental value of the higgs mass , and the green solid and dashed lines represent the central value and @xmath391@xmath40 uncertainties in theoretical calculations of the higgs mass @xcite.__,title="fig:",width=245 ] + colliders as functions of the centre - of - mass energy @xcite . lower panel : one of the possibilities for a relatively heavy susy dark matter particle weighing @xmath116 tev . the vertical axis is the mass difference between the dark matter particle and the next - to - lightest supersymmetric particle , in this case the lighter stop squark . the solid and upper dashed blue lines correspond to the current central and + 1@xmath40 values of the dark matter density , the horizontal yellow band represents the experimental value of the higgs mass , and the green solid and dashed lines represent the central value and @xmath391@xmath40 uncertainties in theoretical calculations of the higgs mass @xcite.__,title="fig:",width=253 ] the supersymmetric dark matter particle might be even heavier in more general supersymmetric models . for example , if the lightest neutralino coannihilates with an almost degenerate gluino , it may weigh @xmath117 tev , as seen in fig . [ fig : elo ] , which would be a challenge even for a 100-tev collider . , as a function of @xmath118 . the green band corresponds to the current @xmath119-@xmath40 range of the dark matter density : @xmath120 , and the brown and red contours are for @xmath121 and @xmath122 , respectively . from @xcite.__,height=220 ] the physics cases for future large circular colliders are still being explored . there will be bread - and - butter high - precision higgs and other sm measurements to probe possible bsm scenarios for physics . as for direct searches for new physics , the search for dark matter particles may provide the strongest case , and this is under continuing study . the physics landscape will look completely different when / if future runs of the lhc find evidence for new physics beyond the sm such as susy . the lhc adventure has only just begun , and we look forward to a big increase in energy with run 2 and eventually two orders of magnitude more integrated luminosity . lovers of susy should not be disappointed that she has not yet appeared . it took 48 years for the higgs boson to be discovered , but four - dimensional susy models were first written down just 41 years ago @xcite . we can be patient for a while longer . the author is supported in part by the london centre for terauniverse studies ( lcts ) , using funding from the european research council via the advanced investigator grant 267352 , and in part by stfc ( uk ) via the research grants st / j002798/1 and st / l000326/1 . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * ( 2012 ) 1 [ arxiv:1207.7214 [ hep - ex ] ] ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * ( 2012 ) 30 [ arxiv:1207.7235 [ hep - ex ] ] . v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1410.6765 [ hep - ex ] . a. david _ et al . _ [ lhc higgs cross section working group collaboration ] , arxiv:1209.0040 [ hep - ph ] . c. anastasiou , c. duhr , f. dulat , f. herzog and b. mistlberger , arxiv:1503.06056 [ hep - ph ] . j. charles , o. deschamps , s. descotes - genon , h. lacker , a. menzel , s. monteil , v. niess and j. ocariz _ et al . _ , arxiv:1501.05013 [ hep - ph ] . v. khachatryan _ et al . _ [ cms and lhcb collaborations ] , arxiv:1411.4413 [ hep - ex ] and references therein . r. aaij _ et al . _ [ lhcb collaboration ] , phys . * 113 * ( 2014 ) 151601 [ arxiv:1406.6482 [ hep - ex ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , phys . * 111 * ( 2013 ) 19 , 191801 [ arxiv:1308.1707 [ hep - ex ] ] . v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1502.07400 [ hep - ex ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 105 * ( 2010 ) 081801 [ arxiv:1007.0395 [ hep - ex ] ] . b. kronenbitter _ et al . _ [ belle collaboration ] , arxiv:1503.05613 [ hep - ex ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 87 * ( 2013 ) 9 , 092002 [ arxiv:1211.1003 [ hep - ex ] ] . s. j. brodsky and x. g. wu , phys . d * 85 * ( 2012 ) 114040 [ arxiv:1205.1232 [ hep - ph ] ] ; s. q. wang , x. g. wu , z. g. si and s. j. brodsky , phys . d * 90 * ( 2014 ) 11 , 114034 [ arxiv:1410.1607 [ hep - ph ] ] . g. aad _ et al . _ [ atlas collaboration ] , phys . d * 90 * ( 2014 ) 052004 [ arxiv:1406.3827 [ hep - ex ] ] . v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1412.8662 [ hep - ex ] . g. aad _ et al . _ [ atlas and cms collaborations ] , arxiv:1503.07589 [ hep - ex ] . j. r. ellis , g. ridolfi and f. zwirner , phys . b * 257 * ( 1991 ) 83 ; h. e. haber and r. hempfling , phys . * 66 * ( 1991 ) 1815 ; y. okada , m. yamaguchi and t. yanagida , prog . * 85 * ( 1991 ) 1 . d. buttazzo , g. degrassi , p. p. giardino , g. f. giudice , f. sala , a. salvio and a. strumia , jhep * 1312 * ( 2013 ) 089 [ arxiv:1307.3536 ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 726 * ( 2013 ) 120 [ arxiv:1307.1432 [ hep - ex ] ] ; v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1411.3441 [ hep - ex ] . t. aaltonen _ et al . _ [ cdf and d0 collaborations ] , arxiv:1502.00967 [ hep - ex ] . j. ellis , v. sanz and t. you , phys . b * 726 * ( 2013 ) 244 [ arxiv:1211.3068 [ hep - ph ] ] . j. ellis , d. s. hwang , v. sanz and t. you , jhep * 1211 * ( 2012 ) 134 [ arxiv:1208.6002 [ hep - ph ] ] . j. ellis , r. fok , d. s. hwang , v. sanz and t. you , eur . j. c * 73 * ( 2013 ) 2488 [ arxiv:1210.5229 [ hep - ph ] ] , and references therein . atlas collaboration , https://twiki.cern.ch/twiki/bin/ + view / atlaspublic / higgspublicresults . j. ellis and t. you , jhep * 1306 * ( 2013 ) 103 [ arxiv:1303.3879 [ hep - ph ] ] . references to the original literature can be found here . g. blankenburg , j. ellis and g. isidori , phys . b * 712 * ( 2012 ) 386 [ arxiv:1202.5704 [ hep - ph ] ] . a. askew , p. jaiswal , t. okui , h. b. prosper and n. sato , arxiv:1501.03156 [ hep - ph ] . j. ellis , d. s. hwang , k. sakurai and m. takeuchi , jhep * 1404 * ( 2014 ) 004 [ arxiv:1312.5736 [ hep - ph ] ] . j. ellis , v. sanz and t. you , arxiv:1410.7703 [ hep - ph ] . references to the original literature can be found here . j. ellis , v. sanz and t. you , jhep * 1407 * ( 2014 ) 036 [ arxiv:1404.3667 [ hep - ph ] ] . j. bond _ et al . _ , http://www.imdb.com / title / tt0143145/. s. moch , s. weinzierl , s. alekhin , j. blumlein , l. de la cruz , s. dittmaier , m. dowling and j. erler _ et al . _ , arxiv:1405.4781 [ hep - ph ] . a. r. ade _ et al . _ [ bicep2 and planck collaborations ] , phys . rev . * 114 * ( 2015 ) 10 , 101301 [ arxiv:1502.00612 [ astro-ph.co ] ] ; p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1502.02114 [ astro-ph.co ] . m. fairbairn and r. hogan , phys . * 112 * ( 2014 ) 201801 [ arxiv:1403.6786 [ hep - ph ] ] . a. hook , j. kearney , b. shakya and k. m. zurek , arxiv:1404.5953 [ hep - ph ] ; j. kearney , h. yoo and k. m. zurek , arxiv:1503.05193 [ hep - th ] . j. r. ellis and d. ross , phys . lett . b * 506 * ( 2001 ) 331 [ hep - ph/0012067 ] . t. hahn , s. heinemeyer , w. hollik , h. rzehak and g. weiglein , phys . * 112 * ( 2014 ) 14 , 141801 [ arxiv:1312.4937 [ hep - ph ] ] . j. r. ellis , s. heinemeyer , k. a. olive and g. weiglein , phys . b * 515 * ( 2001 ) 348 [ hep - ph/0105061 ] . o. buchmueller , r. cavanaugh , a. de roeck , m. j. dolan , j. r. ellis , h. flacher , s. heinemeyer and g. isidori _ et al . _ , eur . j. c * 74 * ( 2014 ) 6 , 2922 [ arxiv:1312.5250 [ hep - ph ] ] . o. buchmueller , r. cavanaugh , m. citron , a. de roeck , m. j. dolan , j. r. ellis , h. flaecher and s. heinemeyer _ et al . _ , arxiv:1408.4060 [ hep - ph ] . k. j. de vries , e. a. bagnaschi , o. buchmueller , r. cavanaugh , m. citron , a. de roeck , m. j. dolan and j. r. ellis _ et al . _ , arxiv:1504.03260 [ hep - ph ] . fnal g-2 collaboration , http://muon-g-2.fnal.gov ; h. iinuma ( for the j - parc new g-2/edm experiment collaboration ) , + http://iopscience.iop.org/1742-6596/295/1/012032/ pdf/1742 - 6596_295_1_012032.pdf . j. r. ellis , j. s. hagelin , d. v. nanopoulos , k. a. olive and m. srednicki , nucl . b * 238 * ( 1984 ) 453 ; h. goldberg , phys . * 50 * ( 1983 ) 1419 [ erratum - ibid . * 103 * ( 2009 ) 099905 ] . p. cushman , c. galbiati , d. n. mckinsey , h. robertson , t. m. p. tait , d. bauer , a. borgland and b. cabrera _ et al . _ , arxiv:1310.8327 [ hep - ex ] . o. buchmueller , m. j. dolan , s. a. malik and c. mccabe , jhep * 1501 * ( 2015 ) 037 [ arxiv:1407.8257 [ hep - ph ] ] ; s. malik , c. mccabe , h. araujo , a. belyaev , c. boehm , j. brooke , o. buchmueller and g. davies _ et al . _ , arxiv:1409.4075 [ hep - ex ] . . m. bicer _ et al . _ [ tlep design study working group collaboration ] , jhep * 1401 * ( 2014 ) 164 [ arxiv:1308.6176 [ hep - ex ] ] . lhc higgs cross - section working group , as reported in m. klute , + http://indico.cern.ch/event/300048/session/13/ contribution/60/material / slides/0.pdf . j. ellis , k. a. olive and j. zheng , eur . j. c * 74 * ( 2014 ) 2947 [ arxiv:1404.5571 [ hep - ph ] ] . j. ellis , f. luo and k. a. olive , arxiv:1503.07142 [ hep - ph ] . j. wess and b. zumino , phys . b * 49 * ( 1974 ) 52 .

in an interesting saga of papers , dvali , gmez and others@xcite have put forward an intriguing suggestion : black holes ( bh ) could be understood as bose - einstein condensates of gravitons . if this view is correct , this would reveal a deep quantum nature of such fascinating objects and could lead to an alternative understanding of some of the most striking features of bh ; for instance , hawking radiation @xcite could be understood as being due to evaporation ( leakage as the authors put it ) of the condensate . besides , this picture brings new ideas about the bekenstein entropy @xcite , the absence of hair @xcite , as well as the quantum nature of information storage and the possible information loss in bhs @xcite . the main point of these works is that the physics of bh can be understood in this picture in terms of a single number @xmath0 , the number of gravitons contained in the bose - einstein condensate ( bec ) . these condensed gravitons have a wave length @xmath1 , @xmath2 being the planck length ; they have a characteristic interaction strength @xmath3 ( reminiscent of meson interactions in large @xmath4 gauge theories ) and the leakage leads to a hawking temperature of order @xmath5 , equal to the inverse of @xmath6 . the mass of the bh is @xmath7 and its schwarzschild radius therefore is given by @xmath8 , thus agreeing with the compton wavelength of the quantum gravitons @xmath6 , in accordance with the uncertainty principle that dictates @xmath9 in the ground state of the quantum system . therefore , up to various factors of the planck mass everything is governed by @xmath0 , the number of intervening gravitons . modulo some assumptions , all these results stem from the basic relation ( unless otherwise stated we work in units where @xmath10 ) @xmath11 that relates the number of gravitons @xmath0 , the mass of the gravitating object @xmath12 and its gravitational radius @xmath13 . for a schwarzschild bh @xmath14 . we found these results intriguing and we set up to try to understand them in a different language and , if possible , attempt to be more quantitative . on the process we have separated slightly from the original line of thought of the authors . the original approach of @xcite is not geometric at all . there is no mention of horizon or metric . here we will adopt a more conservative approach . we will assume the pre - existence of a classical gravitational field created by an unspecified source that generates the schwarzschild metric . as it is known , nothing can classically escape from a bh so if we wish to interpret this in potential terms ( which of course is not correct but serves us for the purpose of creating a picture of the phenomenon ) it would correspond to a confining potential . on and above this classical potential one can envisage a number of quantum fields being trapped . for sure there is a gravity quantum field present ; hence gravitons ( longitudinal gravitons that is ) . continuing with our semi - classical analogy , these gravitons have had a long time to thermalize and it is therefore expected that they can form a bose - einstein condensate . of course , one may argue at once that gravitons do not have repulsive interactions , at least naively , and that therefore a bec is impossible to sustain . to this objection one could reply in a twofold way . first , it is up to the equations to determine whether such a condensate is possible or not ( we will see below that indeed it is , at least in the case of vanishing angular momentum ) . secondly one might also answer that in theories of emergent gravity , the ultimate nature of gravitons may be some type of fermionic degrees of freedom ( such as e.g. in the model suggested in @xcite ) . then , repulsion is assured at some scale . therefore we will try to identify a consistent set of equations describing a bec constructed on top of this classical field . we will see that remarkably enough the characteristics of the bec are uniquely described in terms of the schwarzschild radius of the bh and the value of a dimensionless parameter , interpreted as a chemical potential . a condensate appears not only to be possible but actually intimately related to the classical field that sustains it that determines its characteristics . it is therefore tempting to go one step beyond and reverse the order of the logical implication and eventually attempt to derive the classical field as a sort of mean field potential la hartree - fock . however we will stop short of doing so here . in the subsequent we present the results of our investigations . in what follows we shall adhere to the following notational conventions : @xmath15 will be the einstein tensor : @xmath16 , where the ricci tensor and scalar curvature are constructed with the metric @xmath17 in the usual way . we will denote by @xmath18 the background metric that in our case it will invariably be the schwarzschild metric . perturbations above this background metric will be denoted by @xmath19 , so @xmath20 . we will use the minkowskian metric convention @xmath21 . we will leave for later discussion whether the indices of @xmath19 have to be raised or lowered with the background metric @xmath18 or the full one @xmath17 . likewise for the corresponding volume element ( @xmath22 versus @xmath23 ) . as we will see the results depend to some extent on the prescription that is chosen . to initiate our program we should , first of all , identify an equation ( or set of equations ) that could provide a suitable description of a bec in the present context . in other words , we have to find the appropriate generalization of the gross - pitaevskii equation to this case . the graviton condensate has necessarily to be described by a tensor field that in our philosophy has to be considered as a perturbation of the classical metric . only spherically symmetric perturbations will be considered to keep things as simple as possible . as we will see , every dimensionful quantity can be expressed as a function of the schwarzschild radius @xmath24 . let us consider a spherically symmetric perturbation on the schwarzschild metric ; that is we want to consider a quantum state a condensate having @xmath25 . then @xmath26 we will now expand the perturbed einstein tensor up to second order in @xmath27 . this will allow us to retain the leading non - linearities . four of the components of the einstein tensor are non - null but only three are linearly independent ; both angular components coincide when using the form @xmath28 for the einstein tensor . taking into account that @xmath29 , we have @xmath30 and analogous equations for the radial and angular components ( they are given in appendix [ einec ] ) . it is well known that birkhoff s theorem @xcite guarantees that any spherically symmetric solution of the vacuum field equations must be static and asymptotically flat . so this means that the solution of the equations @xmath31 with @xmath18 being the schwarzschild s metric , must necessarily be of the schwarzschild type again . in conclusion , a perturbation of a schwarzschild metric in vacuum must end in another schwarzschild metric but taking into account a change in the parameters responsible of inducing the perturbation . stating this in other words , a spherically symmetric gravitational field should be produced by some massive object at the origin characterized by the schwarzschild radius , hence a perturbation in itself it can only be produced by an alteration of this radius and therefore a change in its mass . indeed , imposing the following ansatz for the components of the perturbation @xmath32 and substituting @xmath33 in ( [ gtt schw ] ) is possible to obtain the coefficients for each term of the radial expansion . we obtain @xmath34 & + \frac{14a^5 + 25r_sa^4 + 20r_s^2a^3 + 10r_s^3a^2 + 5r_s^4a}{r^5}+ ... \,,\end{split}\ ] ] where @xmath35 . it is obvious that @xmath36 is a solution as implies no perturbation of the metric at all . but if @xmath37 the rest of the coefficients of the expansion are fixed once @xmath38 is given . the same happens for @xmath39 using the radial equation ( [ grr schw ] ) given in appendix [ einec ] . @xmath40 taking into account the accuracy of the expansion in @xmath27 the components of the metric become @xmath41 @xmath42 being the new mass associated to the bh and @xmath43 the perturbed schwarzschild radius . now we will use similar techniques in the ` quantum ' case and as we will see the results are dramatically different . however it is interesting to note that any perturbation @xmath27 ( i.e. each ` graviton ' ) has associated a certain amount of energy that is reflected in a change of the bh mass . perturbing the metric as @xmath44 , a completely covariant perturbative expansion for @xmath27 is obtained . by construction these equations are non - linear ( we keep only the leading non - linearities , to keep the formalism simple and analytically tractable as we will see later ) . we note that the usual gross - pitaevskii equation @xcite employed to describe bose - einstein condensates is a non - linear schrdinger equation ; i.e. an equation of motion that contains self - interactions ( hence the non - linearity ) , a confining potential for the atoms or particles constituting the condensate , and a chemical potential that is conjugate to the number of particles or atoms contained in the condensate . among all these ingredients , the perturbed einstein equations already contain most of them . they are already non - linear and while there is no confining potential explicitly included ( as befits a relativistic theory ) they do confine particles , at least classically , because if the selected background corresponds to a schwarzschild bh , the strong gravitational field classically traps particles inside the horizon . however , there is one ingredient still missing , namely the equivalent of the chemical potential . therefore we have to extend the formulation of perturbations around a classical bh solution to the grand - canonical ensemble by adding to the appropriate action a chemical potential term such as @xmath45 that is conjugate the quantity @xmath46 , which should be related to the graviton density of the condensate inside a differential volume element @xmath47 . taking into account that @xmath48 is not directly interacting with the background , but a field that propagates over the background , it may seem reasonable to take the volume element as the one corresponding to the background . still , the new term is completely covariant as can be seen in appendix [ scovariant ] . irrespective of this chemical potential term an appropriate action has to be chosen for the field @xmath19 . it is only natural to select the einstein - hilbert action expanded around the classical schwarzschild background @xmath18 to describe fluctuations of the metric in a reparametrization invariant way . one can then expand the einstein - hilbert lagrangian @xmath49 , with @xmath50 in powers of @xmath27 up to the desired order ( where @xmath51 is the planck mass ) . measure associated to the full metric has to be taken ( i.e. @xmath52 ) however this is not the only possibility to obtain diff - invariant equations ; for instance @xmath53 also could yield a covariant set of equations . ] an oddity of eq . ( [ fullaction ] ) is that , as defined , @xmath54 has dimensions of [ energy]@xmath55 . this is because the field @xmath27 does not have the canonical dimension ; the field that has canonical dimensions is @xmath56 . after scaling the two powers of @xmath57 , the chemical potential would have dimensions of [ energy]@xmath58 . still , these are not quite the correct dimensions for a chemical potential . as we shall see below the ` natural ' quantity to be identified with a chemical potential will in fact be the dimensionless combination @xmath59 that will be conjugate to @xmath0 , the number of gravitons in the bec condensate with @xmath60 . note that we assume that the chemical potential is @xmath61-dependent . the actual equations of motion derived from the previous action may take slightly different forms depending on the choice of volume elements . for the time being , let us take the simplest possibility @xmath62 where indices are assumed to be raised with the background metric only . actually eq . ( [ eeq ] ) can be derived from the action we have chosen only if we neglect terms of the form @xmath63 . we will return to this point later . it is important to keep in mind the following : in the chemical potential part @xmath54 is considered in the grand canonical ensemble ; this implies that this magnitude is an external field and does not vary in the action ; in particular , for these equations of motion , it is independent of @xmath27 , so @xmath64 . under these considerations this term resembles the chemical potential term of the gross - pitaevskii equation . this should be the way of introducing the chemical potential . otherwise , were @xmath65 not an external field , it would be necessary to take it into account when performing variations to derive the equations of motion . an equation of motion would be obtained for @xmath54 and this equation would nullify automatically the perturbation as well as the chemical potential itself , i.e. @xmath66 and @xmath67 . therefore , the external field @xmath54 is introduced in the theory as some kind of lagrange multiplier . as lagrange multipliers have constraint equations , the chemical potential of the theory may not be arbitrary at all . it must satisfy binding conditions with @xmath27 . the main difference is that this constrains are implicit in a gr theory . the restriction of @xmath54 is through the general covariance conditions for the action . in other words , the diffeomorphism covariance implies the covariant conservation of the einstein tensor , and this in itself entails the same for the chemical potential term , @xmath68 this differential equation of motion for the latter potential is valid up to every order in perturbation theory . hereinafter , we will proceed to find acceptable solutions of these equations and interpret them . we will separate the possible solutions into two regions : outside and inside the bh horizon . we certainly expect that the graviton condensate will disappear quickly in the outside region as @xmath69 and , imposing this as a boundary condition , we will see that in fact the condensate is identically zero on this side of the horizon . on the contrary , a unique non - trivial solution will be found in the interior of the bh . in this section it will be shown that even after the inclusion of @xmath70 there is not other normalizable solution outside the black hole horizon than the trivial solution for the perturbation previously discussed . in the limit @xmath71 , one needs only to keep the terms that are dominant in the einstein equations ( [ gtt schw ] ) , ( [ grr schw ] ) and ( [ gthth schw ] ) . then the equations ( [ eeq ] ) associated to @xmath72 , @xmath73 and @xmath74 in this limit are , respectively will be neglected as we are interested in the simplest approximation to a gross - pitaevskii equation . the form of these terms depends on which convention is chosen for the volume element and to raise / lower indices . ] \frac{r_s}{r^3}h_{tt } -\frac{1}{r}{h_{tt } } ' -\frac{1}{r^2}h_{rr } + \frac{r_s}{r^3}{h_{tt}}^2 -\frac{1}{r}h_{tt}{h_{tt } } ' -\frac{r_s}{r^3}h_{rr}h_{tt } + \frac{1}{r}h_{rr}{h_{tt } } ' + \frac{1}{r^2}{h_{rr}}^2 = \mu\,h_{rr } \\[4ex ] \frac{2r_s}{r^3}h_{tt } + \frac{2}{r}{h_{tt } } ' + 2\,{h_{tt } } '' + \frac{2r_s}{r^3}h_{rr } + \frac{2}{r}{h_{rr } } ' + \frac{2r_s}{r^3}{h_{tt}}^2 -\frac{4r_s}{r^3}{h_{rr}}^2 + \frac{2}{r}h_{tt}{h_{tt } } ' + 2\,h_{tt}{h_{tt } } '' \\ + { h_{tt}}'^2 -\frac{2r_s}{r^3}h_{rr}h_{tt } -\frac{2}{r}h_{rr}{h_{tt } } ' -2h_{rr}{h_{tt } } '' + \frac{r_s}{r^2}{h_{rr}}'h_{tt } -{h_{rr}}'{h_{tt } } ' -\frac{4}{r}h_{rr}{h_{rr } } ' = 0 \,.\end{split}\ ] ] as we want a vanishing perturbation at infinity , the following ansatz is imposed in the faraway region for its components and for the chemical potential : @xmath76 before proceeding , an additional consideration is needed ; namely , the perturbation must be a square - integrable function , i.e. @xmath77 as in the schwarzschild metric @xmath78 , the integral should fulfill @xmath79 for this to happen , if a power law ansatz at infinity is imposed , the exponents must obey @xmath80 . then the three equations in ( [ g schw infty ] ) are @xmath81 \left(\frac{r_s}{r } + \alpha\right)\frac{a}{r^{\alpha+2}}-\frac{b}{r^{\beta+2 } } + \left(\frac{r_s}{r}+\alpha\right)\frac{a^2}{r^{2\alpha+2 } } -\left(\frac{r_s}{r } + \alpha\right)\frac{a\,b}{r^{\alpha+\beta+2 } } + { \frac{b^2}{r^{2\beta+2 } } } = \frac{b\,c}{r^{\beta+\rho } } \\[4ex ] \left(\frac{r_s}{r}+\alpha^2\right)\frac{2\,a}{r^{\alpha+2 } } + \left(\frac{r_s}{r}-\beta\right)\frac{2\,b}{r^{\beta+2 } } + \left(\frac{2r_s}{r}+3\,\alpha^2\right)\frac{a^2}{r^{2\alpha+2 } } -\left({\frac{r_s}{r}}-\beta\right)\frac{4\,b^2}{r^{2\beta+2 } } \\ -\left[\left(2+\beta\right)\frac{r_s}{r}+2\,\alpha^2+\alpha\,\beta\right]\frac{a\,b}{r^{\alpha+\beta+2 } } = 0 \,.\end{split}\ ] ] the requirement of the solution being square - integrable exclude that @xmath82 and @xmath83 could be zero . this implies automatically the neglection of the terms proportional to the schwarzschild radius . even more , as this numbers @xmath82 and @xmath83 are positive and larger than @xmath84 , the following asymptotic identities result @xmath85 @xmath86 from the angular equation ( [ gthth schw powerlaw ansatz o(1 ) ] ) , is mandatory for both terms to contribute at infinity . if this is not the case , this would nullify @xmath38 ( or @xmath87 ) , and then , via both equations in ( [ gtt schw powerlaw ansatz o(1 ) ] ) , fix @xmath88 ( or @xmath36 ) and @xmath89 , i.e @xmath90 . the competition of both terms of the angular equation is possible if and only if @xmath91 . this condition modifies the angular equation ( [ gthth schw powerlaw ansatz o(1 ) ] ) as @xmath92 then from the second equation in ( [ gtt schw powerlaw ansatz o(1 ) ] ) , the following relationship can be read @xmath93 this automatically leads to a null chemical potential as the only possible solution for this region with the proposed ansatz . this is due to the fact that if @xmath88 , as @xmath87 is proportional to @xmath38 , then @xmath38 is also zero . in appendix [ ansatzinfty ] we explore the possibility of exponentially vanishing chemical potential when @xmath94 with analogous conclusions . likewise it can be seen that a much faster gaussian decay is also excluded ( not reported here ) . finally , let us turn to the numerical integration of the equations in the outer part of the bh . on the one hand , the following change of variables @xmath95 in eqs . ( [ eeq ] ) , maps the exterior part into a compact interval ; so as @xmath96 is transformed into the finite boundary @xmath97 ( and @xmath98 into @xmath99 ) . the numerical integration will run between @xmath100 $ ] . on the other hand , this redefinition allows rewriting the relevant equations in terms of dimensionless quantities , so they are universal as @xmath101 and @xmath70 are the only physical parameters . in spite of these simplifications , the equations are difficult to deal with because they are singular , i.e. it is not possible to isolate the higher order derivative at @xmath97 or @xmath99 . let us see how we can proceed . it is well known that schwarzschild space - time is asymptotically flat , therefore , at infinity ( @xmath97 ) it is expected for the perturbation to vanish ( @xmath66 ) . as mentioned before , the equations are singular at @xmath97 and we have to settle for a point @xmath102 but @xmath103 to set up boundary conditions for the numerical integration procedure to start . likewise setting @xmath104 at an arbitrary value for @xmath105 , even if @xmath106 immediately triggers divergences at the first step of the routine . therefore we take a small " initial value for the perturbation in a point near infinite ( @xmath97 ) and then , decrease this initial value until the numerical routine becomes unstable . with the set of solutions for the components of @xmath27 obtained for each of the initial condition imposed , it is possible to compute the quantity @xmath107 , defined in eq . ( [ h2integral ] ) , and the integral extends over the exterior of the bh ( i.e. @xmath105 runs between 0 and 1 ) . in figure [ nout ] the behaviour of the integral of @xmath108 outside the bh is presented when the initial condition for the integration approaches zero at a fixed point near infinity , @xmath109 , i.e. @xmath110 . it is observed that this integral is invariably small and actually vanishes as the initial condition for @xmath33 nullifies . outside the event horizon . the closer the limit to an asymptotically flat space - time is ( i.e. decreasing the initial condition near infinity ) , the smaller this integral is . ] this seems to confirm also numerically that the only possible solution in the outer region is the one with @xmath66 and therefore @xmath111 throughout the exterior of the bh . in order to study the behaviour of our equations when the horizon is crossed , it is convenient to keep working in the @xmath112 variable and redefine the perturbations as @xmath113 in the new variables , the horizon corresponds to @xmath99 while @xmath114 is mapped onto @xmath115 . in the new perturbation the equations become more compact . at the same order as before , the temporal , radial and angular equations read @xmath116 ( z-1)\gamma_{rr } + z{\gamma_{tt } } ' + z\gamma_{tt}\,{\gamma_{tt } } ' -z(z-1)^2\gamma_{rr}\,{\gamma_{tt } } ' -(z-1)^3{\gamma_{rr}}^2 = -x(z-1)\gamma_{rr } \\[3ex ] 2z(z-2)(z-1)\gamma_{rr } + z(z-2)(z-1)^2{\gamma_{rr } } ' -z(5z-2){\gamma_{tt } } ' -2z^2(z-1){\gamma_{tt } } '' \\ -2z(z-2)(z-1)^4\gamma_{rr}{\gamma_{rr } } ' + z(7z-2)(z-1)^2\gamma_{rr}{\gamma_{tt } } ' -z(5z-2)\gamma_{tt}\,{\gamma_{tt } } ' \\ + z^2(z-1)^3{\gamma_{rr}}'\,{\gamma_{tt } } ' -z^2(z-1){{\gamma_{tt}}'}^2 + 2z^2(z-1)^3\gamma_{rr}\,{\gamma_{tt } } '' \\ -2z^2(z-1)\gamma_{tt}\,{\gamma_{tt } } '' = 0 \,.\end{split}\ ] ] in order to get a feeling for possible solutions to these equations we consider their linearized approximation . only terms linear in @xmath117 , @xmath118 and their derivatives are kept in the region @xmath119 . that is @xmath120 ( z-1)\gamma_{rr}+{\gamma_{tt}}'= -x(z-1)\gamma_{rr } \\[3ex ] 2(z-1)\gamma_{rr}+(z-1)^2{\gamma_{rr}}'+3{\gamma_{tt}}'+2(z-1){\gamma_{tt}}''=0 \ , . \end{split}\ ] ] let us now make the following ansatz @xmath121 the three equations take the following form @xmath122 b(z-1)^{\beta+1 } + a\alpha(z-1)^{\alpha-1}=-x b(z-1)^{\beta+1 } \\[2ex ] b(2+\beta)(z-1)^{\beta+1 } + a\alpha(1 + 2\alpha)(z-1)^{\alpha-1}=0 \,,\end{split}\ ] ] respectively . it is worth noting that if @xmath123 , i.e. @xmath124 , all the terms have a contribution as we get closer to the event horizon and @xmath125 behaves as a constant . in this situation , we obtain three equations for the coefficients @xmath126=0 \,.\ ] ] the system of equations is algebraic for the variable @xmath125 ( that as we have seen should behave as a constant as @xmath119 ) ; therefore , it is possible to eliminate @xmath125 by combining the temporal and radial equation . in the present ansatz this leads to @xmath127 together with the angular equation , this determines the solutions up to a constant . there are two possible solutions for this system * * in any case , at least one of the perturbations is divergent over the event horizon . nonetheless , the first solution appears to be integrable , while this is not the case for the second one . in case i ) @xmath128 while if ii ) is taken as solution @xmath111 ( i.e. no chemical potential at all ) . however these solutions do not vanish when @xmath119 ( as one could perhaps have hoped ) , thus casting doubts on the relevance of the linearized equations . let us examine this point and let us only take into consideration the solution i ) , which is integrable . to make sure that this solution is little modified for @xmath119 when non - linearities are switched on , we substitute the solution back in the non - linear equation and verify that the most singular terms are still satisfied . calculations are presented in appendix [ solinfulleq ] . in addition , this exercise gives one interesting result : @xmath129 , where @xmath38 is so far arbitrary , also in sign . note that at the linear level we got @xmath128 and the fact that the quadratic equation gives an @xmath130 correction to this result is consistent as we are implicitly assuming that @xmath131 , i.e. @xmath132 . next we consider the region @xmath133 . we will use another dimensionless variable for this purpose : @xmath134 . in these variables the by now familiar three equations read in their linearized version @xmath135 @xmath136 let us now employ the ansatz @xmath137 while we leave @xmath138 as a free function . this function can be eliminated by combining @xmath72 and @xmath73 , leading to one equation that together with @xmath139 leads to several constraints involving the parameters of the ansatz . of all possibilities , the one that gives the same weight to all terms in the equations is the one corresponding to @xmath140 , @xmath141 . then , close to the origin it would appear as if @xmath142 it remains to be seen if this solution matches with the one obtained in the inner vicinity of the horizon . so far we have seen that the solution i ) given in eq . ( [ behaviourw=1 ] ) satisfies not only the linearized approximation but also the full quadratic equations . could it happen that the previous analysis misses some solution ? to try to answer this question we have performed a numerical integration of the basic equations ( [ eeq ] ) . the integration will be done in the variable @xmath143 . it is impossible to start integrating from either @xmath144 ( the bh center ) or @xmath145 ( the bh horizon ) exactly because as can be seen from eqs . ( [ gtt schw ] ) , ( [ grr schw ] ) or ( [ gthth schw ] ) the system of equations is singular at @xmath114 and @xmath98 ; i.e. one can not determine in either point the value of the higher order derivatives because precisely at the origin and at the horizon they are multiplied by factors that vanish or diverge . in order to proceed we start the integration procedure at a small distance @xmath146 from the origin and following ( [ behaviourw=0 ] ) set @xmath147 we could have equally started the numerical integration routine from a point close to @xmath145 , i.e. in the vicinity ( the inner vicinity that is ) of the horizon . the boundary conditions according to ( [ behaviourw=1 ] ) would then be @xmath148 the respective remaining degrees of freedom are the constants @xmath38 , @xmath149 and @xmath150 , and obviously they must be chosen in such a way that the solution near @xmath145 ( @xmath99 ) and the one near @xmath144 ( @xmath115 ) match . it turns out that there are unique values of @xmath38 and @xmath151 that make both solutions match . namely , @xmath152 . even more , when the previous boundary conditions are assumed for the numerical integration , @xmath117 as well as the function related to the chemical potential @xmath125 , turns out to be constant in the interior of the bh . the constant @xmath38 itself is arbitrary and is not determined by the structure of the equations . changes in @xmath38 appear as an overall factor in the solution . for a certain value of @xmath38 the curves presented in figure [ gammasandxin ] are obtained . the numerical solution reproduces the general features of the analytical study . is represented by a black line , @xmath118 by a dashed line and the dimensionless chemical potential @xmath125 by a dotted line . the results depend on an arbitrary constant as discussed in the text . for the plot we have taken @xmath153 that implies @xmath154 and @xmath155 inside the bh . note that the numerical integration does not extend all the way to @xmath144 and @xmath145 because these are singular points in the differential equation . ] it is most interesting to plot the results of the numerical integration for @xmath156 and @xmath157 ( i.e. considering @xmath158 and then raising one index , taking into account that in the current approach @xmath159 ) . in figure [ hcte ] , a plot of the numerical solution reveals that to a very good approximation throughout the interior of the bh horizon @xmath160 ( represented by a gray solid line ) and @xmath161 ( by a black dashed line ) , as obtained by numerical integration of the @xmath162 equations , are constant and equal . ] actually raising an index with @xmath163 or @xmath164 makes no qualitative difference at all because the solution for the perturbation @xmath19 turns out to be proportional to the initial background metric element itself , as can be seen from the results obtained in the previous section for @xmath117 and @xmath118 ( even though these results in principle take into account the leading behaviour when @xmath119 ) . using the full metric to raise the index changes the numerical value ; yet , the qualitative conclusions do not change . inspired by the previous numerical solution we reformulate our main equations derived at @xmath162 and write them in terms of the metric fluctuation with mixed indices ( one covariant , one contravariant ) . let us write @xmath165 the full metric would become @xmath166 \,.\ ] ] if indices are lowered and raised with the full metric itself . otherwise , if the background metric is used to raise and lower indices , one obtains @xmath167 \,.\ ] ] and the corresponding volume elements are @xmath168 or @xmath169 , respectively . if one assumes constant values for @xmath170 and @xmath171 , as the solution in the previous subsection suggests , the corresponding @xmath162 equations simplify enormously and reduce to the three following ones @xmath172 and @xmath173 therefore we are retrieving the solution found in the previous subsection : @xmath174 plays the role of the integration constant @xmath38 . now let us see that the problem actually admits an exact solution if we assume that the solution is of the form @xmath175 with @xmath174 being a constant . in fact , there are several possible solutions depending on how one chooses to treat the separation between background and fluctuation metrics . if one chooses to consistently raise and lower indices with the background metric only , one has to plug eq . ( [ metricbackground ] ) into the einstein tensor and gets the following closed algebraic equation : @xmath176 that agrees with the @xmath162 perturbative solution found above . note that @xmath111 implies @xmath177 but not the other way round . however , in deriving the equations of motion we have actually neglected a possible contribution of the corresponding volume measures , assuming that eq . ( [ eeq ] ) holds exactly . if we decide to raise and lower indices with the full metric , it is natural to keep the volume element as the one given by @xmath22 , @xmath17 being the full metric . the lhs of the equation of motion now reads @xmath178 ( this is obtained after substituting eq . ( [ metricfull ] ) into the einstein tensor ) . the rhs of the equation also gets modified . on the one hand @xmath23 is replaced by @xmath22 , as mentioned , that depends on @xmath19 . on the other hand , because the indices of @xmath19 are raised with the full metric ( that depends on @xmath19 itself ) , deriving @xmath108 w.r.t . @xmath19 is not totally immediate . all things considered , the resulting equation of motion for a constant @xmath174 is now @xmath179 that can be interpreted as a mean field - like gross - pitaevskii equation for the condensate wave function @xmath174 . using the fact that @xmath180 , we ( re)obtain that @xmath125 is a constant and @xmath181 the previous results are valid to all orders in the expansion . note that @xmath182 . before moving on , it is mandatory to make a comment on the covariant conservation of our equations . at the end of section 2.2 we have pointed out that the diffeomorphism invariance entails a differential equation for the chemical potential , namely ( [ diffeqchempot ] ) . in appendix [ diffecforchempot ] a short discussion of this equation is presented . we see that the actual relation between @xmath174 and @xmath125 depends from the precise way the fluctuations around the background metric are assumed to interact with the metric itself . in any case the qualitative result that there is a normalizable solution that can be interpreted as the collective wave function of a graviton condensate remains . a unique relation is obtained between this ( constant ) wave function and a dimensionless chemical potential . in order to present a physical interpretation for the obtained results , undoing the redefinition of the components of the wave function ( i.e. @xmath183 ) is mandatory . the left graph in figure [ hxh2forh ] presents the curves for the two covariant components of the perturbation . the dimensionful chemical potential is plotted also for comparison . as @xmath184 is a constant function , @xmath185 and it is _ not null _ over the event horizon . of course it is simpler to represent just @xmath174 or @xmath108 , which are just constants . ( black solid ) , @xmath33 ( dashed ) line and the dimensionful chemical potential @xmath70 ( dotted line ) . on the right plot , @xmath186 is shown . as already emphasized , a whole family of solutions are obtained , parametrized by a real arbitrary constant . one can simply trade this constant for the value of @xmath174 . ] it is immediate to see that the solution found is of finite norm . this can be computed numerically using the numerical data into the equation ( [ h2integral ] ) . taking into account that the perturbation @xmath27 is null from the event s horizon onwards , the endpoint on the integration limit can be chosen as the schwarzschild radius . this way , the integral is @xmath187\simeq h^2 \ , r_s^3 \,.\ ] ] then the integral over the interior of the bh of @xmath108 is given by @xmath188 . this quantity should , in a way that we will discuss below , be related to the total number of gravitons of the condensate . @xmath108 would be related to its density in the bh interior that turns out to be constant . note that @xmath19 is dimensionless and that in order to have a properly normalized kinetic term we have to divide the einstein - hilbert action by @xmath189 . at this point we should attempt to make contact with the results of @xcite . possibly our more striking results are that the dimensionless chemical potential @xmath190 stays constant and non - zero throughout the interior of the bh and that so does the quantity @xmath108 previously defined . it is totally natural to interpret @xmath125 as the variable conjugate to @xmath0 , the number of gravitons . how could we determine the value of @xmath0 from our solution ? even after using a properly normalized @xmath191 , the dimensions of ( [ integralh2 ] ) are not appropriate to deliver to us the value of @xmath0 . as it is well known since the early days of quantum field theory @xcite there are no conserved currents or continuity equations for fundamental fields that are chargeless ( such as a real klein - gordon field ) . this also applies to @xmath192 . however , the quantity @xmath193 with @xmath6 being the graviton wave length , has the right ingredients to play the role of probability density in the present context . note that in the bec @xmath194 . if we assume this , then @xmath195 that agrees nicely with @xcite under the maximum packaging condition @xmath194 . recall that @xmath196 is a constant that is entirely determined by the value of the dimensionless chemical potential @xmath125 . the rest of relations can be basically derived from this . indeed within the present philosophy we assign to each graviton in the bec an energy @xmath197 . while there is no formally conserved current , the total energy of the system will be given , by analogy with the case of a massless neutral scalar field , @xmath198 by @xmath199 if the energy @xmath200 is a constant and given by @xmath201 , then eq . ( [ h2integral ] ) has to be interpreted as the number of gravitons in the bec , i.e. the integral of the graviton density in the interior of the bh . as seen above the dimensionless chemical potential @xmath125 has a rather peculiar behaviour . it has a constant value inside the bh and it appears to be exactly zero outside . this behaviour is summarized in figure [ muxvsw ] . let us now for a moment forget about the geometrical interpretation of bh physics and let us treat the problem as a collective many body phenomenon . from this figure it is clear why gravitons are trapped behind the horizon : the jump of the chemical potential at @xmath98 would prevent the ` particles ' inside to reaching infinity . however this is not completely true as the picture itself suggests that _ one _ of the modes can scape at a time without paying any energy penalty if the maximum packaging condition is verified . let us do a semiclassical calculation inspired by this picture ; using @xmath202 : @xmath203 to estimate @xmath204 ( which is negative ) we can use geometrical arguments to determine the flux . if we assume that for a given value of @xmath101 only one mode can get out ( as hypothesized above ) and that propagation takes place at the speed of light , elementary considerations we have to multiply the surface ( @xmath205 ) times the flux ; i.e. the density of the mode times the velocity , assumed to be @xmath206 in our units . since the density of the mode is constant in the interior , it is just @xmath207 . ] lead to @xmath208 this agrees with the results of @xcite for instance eq . ( 35) and yields @xmath209 . the approximations made in the discussion could only modify the coefficient by a numerical factor of @xmath210 . within this picture , several questions concerning the long - standing issue of loss of information may arise as the outcoming state looks thermal @xcite but apparently is not ; or at least not totally so . however we shall refrain of dwelling on this any further at this point . in solid line and the dimensionless @xmath125 in dashed line , across the horizon of the bh ( dotted line ) . in this case the jump at the horizon amounts to @xmath211 . ] the main result from the previous rather detailed analysis is that the bh is able to sustain a graviton bec ( and surely similar bec s made of other quanta ) . but is that condensate really present ? our results do not answer that question , but if we reflect on the case where the limit @xmath212 is taken , without disturbing the bh geometry ( i.e. keeping @xmath101 constant ) this requires taking @xmath213 in a way that the ratio @xmath214 is fixed . then one gets @xmath128 . this value appears to be universal and independent of any hypothesis . the metric is 100% schwarzschild everywhere . we conclude that a bh produces necessarily a ( trapping ) non - zero chemical potential when the physical system is expressed in terms of the grand - canonical ensemble . outside the bh , @xmath111 . another way of reaching this conclusion is by taking a closer look to our exact equations in the previous section . if one sets @xmath111 then necessarily @xmath177 and one gets the classical schwarzschild solution everywhere , but the reverse is not true . one can have @xmath177 but this does not imply @xmath111 . let us emphasize that these results go beyond the second order perturbative expansion used in parts of this article . as gravitons cross the horizon and are trapped by the bh classical gravitational field they eventually thermalize and form a bec . the eventual energy surplus generated in this process is used to increase the mass and therefore the schwarzschild radius of the bh . @xmath174 is now non - zero ; it is directly proportional to @xmath215 , the number of gravitons , and the chemical potential departs from the value @xmath128 , presumably increasing it . as soon as @xmath216 the metric inside the bh is not anymore schwarzschild ( but continues to be schwarzschild outside ) . note that the metric is destabilized and becomes singular for @xmath217 , so surely this is an upper limit . yet another possible interpretation , that we disfavour , could be the following . each bh has associated a given constant value of @xmath125 , hence of @xmath174 . then eqs . ( [ mistery ] ) would imply that after the emission of each graviton , the value of @xmath101 is readjusted . the problem with this interpretation is that it would require a new dimensionless magnitude ( @xmath125 ) to characterize a schwarzschild bh ; something that most bh practitioners would probably find hard to accept . in either case the metric in the bh interior is different from the schwarzschild one . we shall now conclude . the purpose of this work was to have some insights in the reformulation of a quantum theory of black holes in the language of condensed matter physics . the key point of the theory is to identify the black hole with a bose - einstein condensate of gravitons . we have conjectured the set of equations that play the role of the gross - pitaevskii non - linear equation ; they are derived from the einstein - hilbert lagrangian after adding a chemical potential - like term . we have used a number of different techniques to analyze these equations when the perturbation ( i.e. the tentative condensate ) has spherical symmetry . the equations appear at first sight rather intractable , but by doing a perturbative analysis around the bh schwarzschild metric at quadratic order ( i.e. including the leading non - linearities ) we found that the chemical potential necessarily vanishes in the exterior of the bh . on the contrary , in the interior we have found two set of solutions , one of them has to be discarded as producing a non - normalizable result . the other one leads to a non - zero chemical potential in the interior of the bh that behaves as @xmath218 . there is a finite jump of the chemical potential at the bh horizon . surprisingly or maybe not so this solution modifies the coefficient of the @xmath219 and @xmath220 terms in the schwarzschild solution , but not its functional form . of course if there is no chemical potential at all , the modification vanishes , in accordance with well known theorems . however , if the former is non zero , the modification is also necessarily non - zero . from the existence and knowledge of this solution , most relations obtained in @xcite can be rederived . we have seen that there is some ambiguity in the way the effect of the background metric is taken into account and probably more work is needed to produce a unique fully convincing answer . these ambiguities get reflected in the relation between the number of gravitons and the schwarzschild radius , where the value of the ( constant ) condensate enters . the relation between the number of gravitons and the geometric properties of the bh involve an a priori independent and tuneable parameter , the dimensionless chemical potential @xmath125 ( related to the mean - field wave function of the condensate @xmath174 ) . we find this somewhat strange as this would be a new black hole parameter . therefore we favour the universal value @xmath128 as discussed in the text . as should be obvious to the reader who has followed our discussion , our approach is somewhat different from the one developed in the initial papers by dvali , gmez and coworkers . we assume from the start the existence of a classical geometry background that acts as confining potential for the condensate . the fact that the functional form of the metric perturbation induced by the condensate is exactly the same as the original background , of course gives a lot of credence to the possibility of deriving the latter from the former in a sort of self - consistent derivation . we have not explored this possibility in detail yet . it is quite plausible that one could entertain the presence of condensates of other quantum fields inside the bh horizon ( why only gravitons ? ) . while we do not expect much of a conceptual difference , it would be very interesting to see the similarities and differences with the case of quantum gravitons . we acknowledge the financial support of projects fpa2013 - 46570-c2 - 1-p ( mineco ) and 2014sgr104 ( generalitat de catalunya ) . this research was supported in part though the e.u . eplanet exchange program fp7-people-2009-irses , project number 246806 . the work of j.a . is partially supported by grants fondecyt 1150390 and conicyt - pia - act14177 ( government of chile ) . 99 g. dvali and c. gomez , black hole s 1/n hair . lett . _ * 719 * , 419 ( 2013 ) ; g. dvali and c. gomez , landau - ginzburg limit of black hole s quantum portrait : self similarity and critical exponent . b * 716 * , 240 ( 2012 ) ; g. dvali , d. flassig , c. gomez , a. pritzel , n. wintergerst , scrambling in the black hole portrait . d * 88 * , 124041 ( 2013 ) ; g. dvali and c. gomez , black holes as critical point of quantum phase transition . phys . j. c * 74 * , 2752 ( 2014 ) . g. dvali and c. gomez , black hole s quantum n - portrait . fortsch . phys . * 61 * , 742 ( 2013 ) . j. alfaro , d. espriu , d. puigdomnech , spontaneous generation of geometry in four dimensions . d _ * 86 * 025015 ( 2012 ) ; j. alfaro , d. espriu , d. puigdomnech , the emergence of geometry : a two - dimensional toy model . _ * 82 * , 045018 ( 2010 ) . s.w . hawking , particle creation by black holes . 43 * , 199 ( 1975 ) . bekenstein , black holes and entropy . d * 7 * , 2333 ( 1973 ) ; j.d . bekenstein , generalized second law of thermodynamics in black - hole physics . d * 9 * , 3292 ( 1974 ) ; s.w . hawking , black holes and thermodynamics . d * 13 * , 191 ( 1976 ) . bekenstein , phys . * 28 * , 452 ( 1972 ) ; j.d . bekenstein , phys . d * 5 * , 1239 ( 1972 ) ; j.d . bekenstein , phys . d * 5 * , 2403 ( 1972 ) ; c. teitelboim , phys . d * 5 * ( 1972 ) 294 ; j. hartle , phys . d * 3 * , 2938 ( 1971 ) . hawking , the unpredictability of quantum gravity . phys . * 87 * , 395 ( 1982 ) . for an excellent review on information loss paradox , see , j. preskill , do black holes destroy information ? international symposium on black holes , membranes , wormholes , and superstrings , houston advanced research center ( 1992 ) . arxiv : hep - th/9209058 . birkhoff and r. langer , _ relativity and modern physics _ ( harvard univ . press , 1923 ) . f. dalfovo , s. giorgini , l.p . pitaevskii and s. stringari , theory of bose - einstein condensation in trapped gases . phys . _ * 71 * , 463 ( 1999 ) . w. pauli and v. weisskopf , helv . acta * 7 * , 709 ( 1934 ) . s.w . hawking , breakdown of predictability in gravitational collapse . d * 14 * 2460 ( 1976 ) . the two remaining components of the einstein tensor are @xmath221 = 0 \end{split}\ ] ] and @xmath222=0 \,.\end{split}\ ] ] here @xmath174 stands for any of the angular coordinates : @xmath223 or @xmath224 . it is worth to note that if a rescaling on the schwarzschild radius @xmath225 is done , fixing it to zero , then we end up with minkowski space - time in the three components of @xmath226 : ( [ gtt schw ] ) , ( [ grr schw ] ) and ( [ gthth schw ] ) . in the following part , the invariance under the gauge group of diffeomorphisms of the chemical potential term in the action ( [ fullaction ] ) will be shown . let s start by pointing out that both metrics , @xmath227 and @xmath228 , as well as the perturbation , transform covariantly . in fact , the full metric @xmath44 , under an infinitesimal displacement d in the coordinates of the form @xmath229 x^\mu=\xi^\mu$ ] , changes as @xmath230 the same gauge transformation rules the background metric under the general coordinate transformation mentioned above . this is , @xmath231 as the perturbation is defined by @xmath232 , under the same perturbation of the coordinate system , from the subtraction of ( [ fullmetrictransf ] ) and ( [ backgroundmetrictransf ] ) , the following equality is obtained @xmath233 the latter identity implies @xmath27 has also a lie derivative transformation . the fact that the perturbation transforms in the same way than the metric tensor , and the chemical potential @xmath54 behaves as a scalar under a general coordinate transformation , ensure automatically the general covariance of the theory . let s change the ansatz and impose an exponential decreasing solution for the perturbations at infinity : @xmath234 as before for this type of ansatz , the linear order dominates in front of the higher ones . inserting this ansatz into a linearization of the einstein equations partially evaluated at infinity ( [ g schw infty ] ) @xmath235 @xmath236 @xmath237 as decreasing solutions are expected , the parameters @xmath82 , @xmath83 and @xmath238 must be not null . this allows one to neglect all the terms besides the principal ones inside each parenthesis . in the angular equation there is no way for both terms to compete between each other @xmath239 in conclusion , depending on the fact whether if @xmath82 is bigger or not than @xmath83 , @xmath36 or @xmath88 . for the first case , if @xmath36 , the temporal equation @xmath240 fixes @xmath88 . being both , @xmath38 and @xmath87 null , the chemical potential disappears form the theory in this region . for the second case , if @xmath88 , the same equation @xmath72 nullifies @xmath38 ( as no null solutions for @xmath241 are expected ) however , no perturbation makes the chemical potential senseless . to sum up , this ansatz implies that the only solution at infinity is a null perturbation , @xmath66 , and the disappearance of the chemical potential @xmath70 from the theory . the solution of the linearized differential equations leads to the expressions @xmath242 ; @xmath243 . let us now plug this solution in the non - linear @xmath244 system and see if the solution survives or how could it get modified . near @xmath245 the full temporal , radial and angular equations are @xmath246 @xmath247 @xmath248 respectively . replacing the possible solution , @xmath242 this eliminates any derivative of @xmath117 and @xmath243 into these equations , we get in turn @xmath249 \longrightarrow\ , 3\,a - 2\,a + 5\,a^2 - 4\,a^2 = a+a^2 = ( 1+a)a = -x\,a \end{split}\ ] ] @xmath250 \longrightarrow\ , -\frac{a+a^2}{(z-1 ) } = -\frac{(1+a)a}{(z-1 ) } = \frac{x\,a}{(z-1 ) } \end{split}\ ] ] @xmath251 \longrightarrow\ , -\frac{2\,a}{(z-1 ) } + \frac{2\,a}{(z-1 ) } -\frac{4\,a^2}{(z-1 ) } + \frac{4\,a^2}{(z-1 ) } = 0 \,.\end{split}\ ] ] therefore , quite surprisingly , the linear solution is still an exact solution of the non - linear quadratic differential equations and we conclude that @xmath252 are solutions . the only difference is that the dimensionless chemical potential is at this order @xmath253 rather than @xmath128 as suggested by the linearized approximation . in this appendix , a derivation of the differential equation for the chemical potential @xmath70 is presented . we are going to derive this equation for a theory where only the background metric is considered for the integration measure and to raise and lower indices . covariant conservation of the einstein tensor implies automatically the same for the rhs of our equations @xmath254 \end{split}\ ] ] in the last equality , the covariant derivative is expressed in terms of the christoffel symbols of the theory . performing the sum in @xmath255 because @xmath256 appears always in the perturbation and only exists in the temporal and radial components , and doing the same procedure with @xmath257 the following relation is obtained @xmath258 \,\end{split}\ ] ] where we have taken into account that there is only radial derivatives . the latter equation is a set of 4 equations , @xmath259 . the only non trivial equation is the radial one , i.e. @xmath260 . dropping all the null christoffel symbols and the off - diagonal elements of the @xmath27 matrix , makes the equation reads as @xmath261 \end{split}\ ] ]

many communication protocols implicitly require that the communicating parties share a reference frame @xcite . for instance , if one party , alice , transmits qubits to another party , bob , using spin-1/2 particles , the quantum state of the collection can only be recovered by bob if he and alice share a reference frame for orientation . lacking such a shared reference frame ( for instance , by lacking knowledge of the relation between their local reference frames ) is equivalent to having a noisy channel ; the density operator relative to bob s local frame is the average over rotations of the density operator relative to alice s local frame . for a single qubit , such an average over rotations yields complete decoherence no information about the quantum state survives nonetheless , alice and bob can still achieve perfect classical and quantum communication by encoding the information into the rotationally - invariant degrees of freedom of many qubits @xcite . indeed , in the limit of large numbers , the cost of not sharing a reference frame is only logarithmic in the number of systems . similarly , if alice and bob lack a phase reference , they can still encode classical and quantum information in phase - invariant states of composite systems @xcite . however , such communication schemes are technically challenging to implement because they make use of highly entangled states of many qubits @xcite . such schemes will be referred to here as `` calibration - free '' . a more straightforward strategy for coping with the lack of a shared reference frame is for alice and bob to begin their communication protocol by setting up a shared reference frame , that is , they begin by calibrating or aligning their local reference frames . thereafter alice transmits her quantum systems normally . this strategy is illustrated in figure [ fig : figure1 ] . the problem of aligning reference frames using finite communication resources has been well studied @xcite . if only finite communication resources are devoted to the task , then the token of alice s reference frame that is transmitted to bob will be of bounded size . assuming the reference frame in question is associated with a continuous degree of freedom , this bound leads to a nonzero probability of error in the decoding of messages . nonetheless , this scheme has two advantages over its calibration - free counterpart : ( i ) alice does not need to implement any entangling operations to encode classical bits or logical qubits into the physical qubits , nor does bob require such operations to decode ( they may require entangling operations to prepare and measure the reference frame token , but the preparation and measurement they require is always the same and their effort does not scale with the size of the message ) ; ( ii ) if alice wishes to communicate a classical bit string or a string of logical qubit states , she can encode one logical bit or qubit per physical qubit , and bob can decode one logical bit or qubit per physical qubit ( in other words , the blocks of physical qubits into which they encode and decode their logical bits and qubits can be as small as one , unlike the calibration - free scheme ) . by virtue of ( ii ) , alice does not need to know the entire message string at the outset to achieve her optimal communication rate , nor does bob need to store all of the systems coherently until he has received the entire sequence of physical qubits , whereas such capabilities are required to achieve the optimal rate of communication in the calibration - free scheme . we are therefore motivated to explore how well alice can communicate quantum information to bob after supplying him with a bounded - size token of her reference frame . by the state @xmath0 and the system @xmath1 by the state @xmath2 . bob , who is not correlated with alice s reference frame , describes the joint state of @xmath3 and @xmath1 as the twirling of alice s description , namely @xmath4 . b ) we can consider alice s act of adjoining the reference frame token to the system together with the twirling as an encoding operation @xmath5 . composing this with bob s decoding operation @xmath3 , the resulting channel can be described as an effective @xmath6.,width=302 ] the general case we consider is that of a reference frame associated with a compact lie group @xmath7 . if the system is prepared in the state @xmath2 and the reference frame token in the state @xmath8 relative to alice s local reference frame , then relative to bob s local reference frame , the pair is in the state @xmath9 = \mathcal{g } ( |e\rangle \langle e| \otimes \rho ) \,,\ ] ] where @xmath10 averages over the collective action of the group and is termed the @xmath7-twirling operation ( see figure [ fig : figure1 ] ) . this state encodes @xmath2 . we consider the case where the reference frame token is prepared in a particular state @xmath8 , suggested by previous investigations @xcite , which is near - optimal for transmitting information about the group element and which makes the mathematics particularly simple . we then determine how well bob can reconstruct the state @xmath2 . it turns out that the recovery ( or decoding ) operation @xmath11 that is equal to the hilbert - schmidt adjoint of @xmath5 , normalized to be trace - preserving , is provably near - optimal ( in a sense we will define later ) . the composition of encoding and decoding yields an effective decoherence of the form @xmath12 where @xmath13 is a probability distribution over the group with @xmath14 the group - invariant measure , @xmath15 and @xmath16 are unitary representations of @xmath7 on the system @xmath1 and the reference frame token @xmath3 respectively . with our particular choice of reference state @xmath8 , we find that @xmath17 for all @xmath18 , ensuring that @xmath19 is a @xmath7-invariant map ( it commutes with the action of every @xmath20 ) . the figure of merit relative to which the decoding operation is judged to be near - optimal is the entanglement fidelity . we also demonstrate two distinct ways of implementing this decoding operation . the first is an obvious scheme : bob estimates the orientation of the token relative to his local frame and then re - orients the system appropriately . we call this the `` measure and re - orient '' scheme . the second is less intuitive , but reveals more about the structure of the problem : bob projects into the virtual subsystems that support the representation of the group induced by _ relative _ transformations of the system and token and implements an isometry that maps these onto a single hilbert space . we call this the `` extract from the relational subsystems '' scheme . these relational subsystems are the places in the hilbert space of the combined token and system where the quantum information associated with @xmath2 is to be found . their characterization is the key technical result of the article . the second scheme is also particularly interesting because , with a slight modification , it can yield a decoding that is probabilistically perfect , that is , one which sometimes fails but which yields a perfect decoding when it succeeds . we work out the explicit form of the recovery operation @xmath11 and the effective decoherence @xmath19 for several interesting examples : ( i ) a phase reference ; ( ii ) a cartesian frame ( representing an orthogonal triad of spatial directions ) ; ( iii ) a reference direction ( representing a single spatial direction ) . in each case , we consider a system consisting of a single qubit . the explicit form of the decoherence map is actually quite simple in this case . because @xmath19 is a @xmath7-invariant map , it follows from the results of ref . @xcite that it is a sum of irreducible @xmath7-invariant maps called moments . but in the case of a qubit , there is only a single nontrivial moment the @xmath7-twirling map @xmath10 . thus , we have @xmath21 where @xmath22 is the identity map . we show that for our examples , @xmath23 is inversely proportional to the size of the reference frame token , @xmath24 where the size of the reference frame is given by the quantum number of the highest irreducible representation appearing in the state of the token . it is also shown that in probabilistically perfect decoding schemes , the probability of failure is inversely proportional to the size of the reference frame token . these results are specific to the special form of reference frame state that we consider here . the idea that bounded - size quantum reference frames induce an effective decoherence is not new . the effect of a bounded - size clock , which is a kind of phase reference , has been considered previously by many authors @xcite and that of a bounded - size directional frame has been considered by poulin @xcite . our results go beyond this work in several ways . for one , the case of a cartesian frame , which is particularly significant given that it is representative of the general non - abelian case , has not been examined before . more significantly , we consider many different sorts of reference frames within a unified framework and we provide insight into the structure of the problem . it should be noted that although our method can be applied to a system of arbitrary dimension , we here obtain explicit expressions for the effective decoherence only in the case of the simplest possible system : a qubit . we hope to provide a discussion of the foundational implications of modeling bounded - size reference frames by effective decoherence in a subsequent paper . finally , it is worth pointing out that the problem of communication in the presence of a bounded - size reference frame has interesting connections with a disparate set of topics in quantum information theory and the theory of quantum reference frames : _ partially correlated reference frames . _ when some information is known about the relative orientation of alice and bob s local reference frames , they are said to share partially correlated reference frames . this is a resource that interpolates between having and lacking a shared reference frame . its quality can be characterized by the probability distribution over the relative orientation the more peaked the distribution , the better the correlation . what can be achieved with this resource is an interesting question that has only begun to be addressed . we gain some insight into the question in this article because the `` measure and re - orient '' implementation of the recovery operation begins with a reference frame alignment protocol @xcite that leaves alice and bob holding partially correlated reference frames . _ programmable operations . _ there have been many investigations into the possibility of encoding an operation into a quantum state such that the state can subsequently be used to implement the operation on another system . if the system into which the operation is encoded is bounded in size , then it is known that one can only achieve an approximate version of the operation or a perfect version with non - unit probability @xcite . the token of the quantum reference frame in our communication protocol is an instance of a program that encodes the unitary that relates alice s local reference frame to bob s . bob subsequently uses it to implement the inverse of this unitary on the system . our results therefore provide interesting examples of both approximate and unambiguous programmable operations . _ measures of the quality of a quantum reference frame . _ our results also contribute to the project of quantifying the extent to which a quantum reference frame of bounded size can stand in for one of unbounded size @xcite , an important element of a resource theory of quantum reference frames . for instance , a measure of the strength of the effective decoherence associated with the bounded - size reference frame may serve as an operational measure of its quality . _ private channels . _ if no party besides bob has a sample of alice s rf , then alice and bob are said to possess a _ private _ shared reference frame . these have been shown to constitute a novel kind of key that is useful for private communication schemes @xcite . our results establish a lower bound on the fidelity between input and output in private communication schemes that rely on bob possessing a bounded - size token of alice s reference frame . they also establish a bound on the probability of achieving perfect fidelity by post - selection . this cryptographic application provides a particularly useful perspective on bob s recovery operation : the message is encoded in the relative orientation between the system and the token of alice s reference frame in the same way that the plain - text in classical cryptography is encoded in the bit - wise parity of the cypher - text message and the key . bob decodes by using the token of alice s frame as a key . _ dense coding . _ as noted above in point ( ii ) , the calibration - free communication scheme requires alice to know the entire message to be sent prior to transmitting any systems to bob if she is to achieve the maximum rate . in the communication scheme that first sets up a bounded - size reference frame , alice can transmit to bob the quantum token of her local reference frame prior to knowing anything about the message . consequently , alice can use the quantum channel at an early time , when it is perhaps cheaper , to enhance the communication capacity at a future time , when it is known what message is to be sent . sending the frame token is therefore akin to establishing entanglement in a dense coding scheme @xcite . one difference , however , is that the fidelity of communication is never perfect for a bounded - size token whereas a single maximally - entangled state allows for perfect communication of one qubit . in this section , we present some formal mathematical tools that are useful for describing classical and quantum reference frames . we follow the notation of @xcite , to which we refer the reader for further details . suppose alice and bob are considering a single quantum system described by a hilbert space @xmath25 . let this system transform via a representation of a group @xmath7 relative to some reference frame . we will restrict our attention to lie groups that are compact , so that they possess a group - invariant ( haar ) measure @xmath14 , and act on @xmath25 via a _ unitary _ representation @xmath26 , ensuring that they are completely reducible @xcite . let @xmath20 be the group element that describes the passive transformation from alice s to bob s reference frame . furthermore , suppose that @xmath27 is completely unknown , i.e. , that alice s reference frame and bob s are uncorrelated . it follows that if alice prepares a system in the state @xmath2 on @xmath25 relative to her frame , then it is represented relative to bob s frame by the state @xmath28 = \int_g \text{d}g\ , u(g ) \rho u^\dag(g ) \ , . \label{eq : averagedstate}\ ] ] the action of the representation @xmath26 of the ( compact lie ) group @xmath7 on @xmath25 yields a very useful structure . it allows for a decomposition of the hilbert space into a direct sum of _ charge sectors _ , labeled by an index @xmath29 , where each charge sector carries an inequivalent representation @xmath30 of @xmath7 . each sector can be further decomposed into a tensor product of a subsystem @xmath31 carrying an irreducible representation ( irrep ) of @xmath7 and a subsystem @xmath32 carrying a trivial representation of @xmath7 . that is , @xmath33 note that this tensor product does not correspond to the standard tensor product obtained by combining multiple qubits : it is _ virtual _ @xcite . the spaces @xmath31 and @xmath32 are therefore _ virtual subsystems_. expressed in terms of this decomposition of the hilbert space , the map @xmath10 takes a particularly simple form , given by @xmath34 = \sum_q ( \mathcal{d}_{\mathcal{m}^{(q ) } } \otimes \mathcal{i}_{\mathcal{n}^{(q)}})[\pi^{(q ) } \rho \pi^{(q ) } ] \,,\ ] ] where @xmath35 is the projection into the charge sector @xmath29 , @xmath36 denotes the trace - preserving operation that takes every operator on the hilbert space @xmath37 to a constant times the identity operator on that space , and @xmath38 denotes the identity map over operators in the space @xmath39 . a proof of this result is provided in ref . @xcite . note that the operation @xmath10 has the general form of _ whereas decoherence typically describes correlation with an environment to which one does not have access , in this case the decoherence describes correlation with a reference frame to which one does not have access @xcite . consider a communication scheme wherein alice prepares a system @xmath3 in a pure quantum state @xmath8 and sends it to bob as a quantum sample of her reference frame , together with a system @xmath1 ( a collection of qubits for example ) that is described by a quantum state @xmath2 relative to her reference frame . let @xmath3 transform via the unitary representation @xmath16 of @xmath7 , and @xmath1 via the unitary representation @xmath15 . the lack of a shared reference frame between alice and bob implies that the transmitted composite @xmath40 is described relative to bob s reference frame by the @xmath7-invariant state @xmath41 \ , , \label{eq : ginvstate}\ ] ] where @xmath42 is the @xmath7-twirling operation of ( [ eq : averagedstate ] ) for the representation @xmath43 of @xmath7 . this map @xmath5 will be referred to as the _ encoding map_. note that its input space is @xmath44 ( the bounded operators on @xmath45 ) , while its output space is @xmath46 . it maps @xmath2 to a @xmath7-invariant state of the composite @xmath40 . it is useful to define the set of states @xmath47 which form the orbit under the representation @xmath16 of @xmath7 of the fiducial state @xmath8 ( associated with the identity element of the group ) . by expressing the @xmath7-twirling operation explicitly and making use of these , we can express the encoding operation as @xmath48 the encoding map clearly depends on the choice of the state @xmath8 for the reference frame . we turn to this choice now . we begin by considering what properties of a quantum state make it a good representative of a reference frame for the group @xmath7 ( the case wherein the reference frame is associated with a coset space will be considered in section [ sec : direction ] ) ; for a more complete discussion , see @xcite . states @xmath49 corresponding to different orientations of the reference frame must be distinct , so at the very least one requires that the the fiducial state @xmath8 is not invariant with respect to @xmath7 or any subgroup thereof . to emulate a perfect reference frame for @xmath7 , these states must in fact be perfectly distinguishable , @xmath50 where @xmath51 is the delta - function on @xmath7 defined by @xmath52 for any continuous function @xmath53 of @xmath7 , where @xmath54 is the identity element in @xmath7 . if the states latexmath:[$\ { requirements , then @xmath16 is the _ left regular representation _ of @xmath7 . in the case of a lie group , the dimensionality of any system @xmath56 that carries the regular representation must necessarily be infinite . we refer to such an infinite - dimensional quantum rf as _ unbounded _ ; such systems and states were considered in ref . @xcite . if the hilbert space dimensionality of the system @xmath3 serving as a quantum rf is finite , then we say that the quantum rf is _ of bounded size_. if the rf is associated with a lie group , having a continuum of elements , then a bound on the size of the rf implies that the condition ( [ eq : deltafunction ] ) can not be satisfied precisely . in this case , a key question is : what state on @xmath3 is the best approximation to a perfect reference frame ? the answer will depend on the figure of merit for the task at hand , but we will make use of a generic construction @xcite that illustrates the key features . suppose the representation @xmath16 reduces to a set of irreps @xmath57 , @xmath58 where the tensor product is the one appearing in the decomposition ( [ eq : hdecompfull ] ) of @xmath59 and where @xmath60 is the identity on @xmath61 . we are interested in a special subset of these irreps , namely , the @xmath62 that appear in the decomposition of @xmath16 a number of times greater than or equal to their dimension @xmath63 , i.e. those for which @xmath64 we denote the set of @xmath29 that label such irreps by @xmath65 and , in what follows , we will be restricting our attention to only these irreps . also , for irreps @xmath66 , choose an arbitrary subspace @xmath67 with dimension @xmath63 , i.e. , with dimension equal to that of @xmath68 . we now define a new hilbert space @xmath69 as @xmath70 which is of dimension @xmath71 the state of @xmath3 that we will use for our quantum rf is @xmath72 where @xmath73 is an arbitrary basis for @xmath31 , and @xmath74 an arbitrary basis for @xmath75 . note that the orbit of @xmath8 under @xmath7 has support in @xmath69 . the embedding @xmath76 provides a way of embedding @xmath8 in the original hilbert space @xmath59 , and in addition @xmath16 acts on the hilbert space @xmath69 in the obvious way . if @xmath65 contained all irreps of @xmath7 , then @xmath16 would be the ( left ) regular representation , and for a lie group the hilbert space @xmath69 would be infinite dimensional . if the quantum rf is of bounded size , then a limited set of irreps appear in @xmath65 . we note that , for the problem of optimally encoding a reference frame relative to a maximum likelihood figure of merit , given a general hilbert space @xmath59 the optimal states will not have this precise form @xcite . however , such optimal states do take the form of eq . when restricted to @xmath69 . these are the states of interest here . the restriction to irreps having the special property of eq . ( [ eq : specialproperty ] ) is in fact critical for our analysis , because only in this case can we define a useful _ right _ action of @xmath7 on the hilbert space @xcite . consider the representation @xmath77 of @xmath7 defined by its action on the covariant set as @xmath78 to obtain an explicit form for this right action in terms of the decomposition of eq . , we make use of the fact that the state @xmath8 is maximally entangled across the virtual tensor products @xmath79 . thus , we have for any transformation @xmath80 on a subsystem @xmath68 the identity @xmath81 where @xmath82 denotes the transpose , @xmath83 the complex conjugate , and we have made use of the fact that @xmath62 is unitary . given that the complex conjugate of a representation @xmath62 of @xmath7 is also a representation of @xmath7 ( called the conjugate representation and denoted by @xmath84 ) , we can define a representation @xmath77 by @xmath85 in contrast to @xmath16 , the representation @xmath77 acts on the subsystems @xmath86 irreducibly according to the conjugate representation @xmath87 , and leaves the subsystems @xmath68 invariant . clearly , the two actions @xmath16 and @xmath77 commute . furthermore , it is easy to verify that @xmath77 satisfies eq . ( [ eq : rightaction ] ) . as the states of the reference frame are restricted to the hilbert space @xmath69 , it is useful to consider our encoding operation @xmath5 of eq . ( [ eq : encoding ] ) with fiducial state @xmath8 of eq . ( [ eq : chiribella ] ) as a map from @xmath44 to @xmath88 . for the remainder of this paper , we consider the encoding map to be defined thus . finally , we note that the map @xmath5 with the fiducial state @xmath8 chosen to be of the form ( [ eq : chiribella ] ) is unital . ( because the input and output spaces of @xmath5 are of differing dimension , we define _ unital _ for such a trace - preserving map as one which maps the ( normalized ) completely mixed state to the completely mixed state . ) this result is seen as @xmath89 where @xmath90 is the dimension of @xmath45 . here , we have used the fact that the maximally - entangled states @xmath91 in eq . ( [ eq : chiribella ] ) have reduced density matrices on @xmath92 that are proportional to the identity . it is illustrative to investigate the action of the encoding map ( the fiducial state @xmath8 will be assumed to be given by eq . ( [ eq : chiribella ] ) except in the final section of the article ) and to explicitly identify the subsystems of @xmath93 into which the system s state is encoded . the details of this section require extensive use of the virtual tensor product structure of @xmath94 induced by the unitary representation @xmath95 of @xmath7 , given in eq . , as well as an application of the stinespring theorem for covariant maps @xcite ; this section may be skipped on first reading . to facilitate this , we first state the main result of this section prior to our detailed investigation of the encoding map . main result : : : according to eq . , the joint hilbert space @xmath94 can be decomposed under the representation @xmath95 of @xmath7 as @xmath96 where @xmath97 are the set of irreps @xmath29 of @xmath7 that appear in the decomposition of @xmath95 . the encoding map yields @xmath7-invariant density operators which , in terms of the decomposition , are block - diagonal in the irreps @xmath98 and , within each block , have the form of a tensor product of the completely mixed state on the subsystem @xmath99 and some non - trivial state on the subsystem @xmath100 . thus , the action of the encoding can be expressed as @xmath101 where @xmath102 is the identity operator on @xmath99 , and @xmath103 is a trace - decreasing map from states on @xmath45 to states on @xmath100 . we show below that , under the assumption that @xmath45 is an irrep of @xmath7 , each of these encodings @xmath103 takes the form @xmath104 where @xmath105 is the identity operator on a hilbert space @xmath106 carrying an irrep @xmath87 of @xmath7 , and @xmath107 is a linear map satisfying @xmath108 . in addition , each map @xmath109 takes a very simple form , which depends on the irrep @xmath29 . specifically , there is a subset of irreps @xmath110 such that , for @xmath111 , the map @xmath109 is a bijective isometry , that is , a unitary ; in these instances , the map @xmath103 can be inverted and @xmath2 can be recovered perfectly . for @xmath29 not in @xmath112 , the map @xmath109 is an isometry that is not surjective , i.e. , it maps onto a _ proper subspace _ of @xmath113 . the map @xmath103 is not invertible in these cases . we can identify the relational degrees of freedom in which the message is encoded by investigating how relational transformations act on the hilbert space @xmath94 . the subsystems @xmath99 carry an irreducible representation of @xmath7 corresponding to the collective action @xmath95 and describe collective degrees of freedom . in contrast , the subsystems @xmath100 are relational . however , not all degrees of freedom in @xmath100 describe relations of the system @xmath1 to the reference frame @xmath3 ; some of these describe relations among the parts of @xmath3 ( or among the parts of @xmath1 if the latter are composite systems ) . we seek to identify , for each irrep @xmath29 , the precise subsystem of @xmath100 into which the message state @xmath2 is encoded . the system hilbert space @xmath45 carries a representation @xmath15 of @xmath7 . if we act with @xmath7 on the system but not on the rf , this will induce a relative transformation of the two ; however , this action alone is not @xmath7-invariant . while it is possible to construct a @xmath7-invariant action of @xmath15 by using the techniques of refs . @xcite , it is much more straightforward to make use of the right action @xmath77 of @xmath7 defined in eq . ( [ eq : rightaction ] ) . this action commutes with the left action @xmath16 , and thus also commutes with the collective action @xmath95 of @xmath7 . by acting with @xmath114 for @xmath18 on a state @xmath115 of the form , we have @xmath116 where we have used the invariance of the haar measure . the action of @xmath114 on @xmath117 yields another invariant state , but one which is now an encoding of the transformed state @xmath118 . thus , @xmath114 acts as a transformation of the relation between @xmath1 and @xmath3 . a map @xmath5 satisfying eq . is called _ @xmath7-covariant_. as @xmath77 is a relational action , it acts on the subsystems @xmath100 in eq . ; we now decompose these subsystems according to the irreps of @xmath7 under the action of @xmath77 , and in doing so identify the subsystems in which we find the image of @xmath2 under the encoding map . at this stage , we restrict our attention to the case where @xmath15 is an irrep of @xmath7 , labelled @xmath119 . it appears straightforward ( although with substantially more burdensome notation ) to extend our results to the general case wherein this restriction is relaxed . indeed , the u(1 ) example considered in sec . [ sec : phase ] provides evidence of the generality of our theorem . however , we do not consider the general case here . recall that the reference frame @xmath3 has a hilbert space given by eq . , and the system s hilbert space is @xmath120 . thus , @xmath121 where @xmath122 denotes that the irreps @xmath123 and @xmath119 couple to the irrep @xmath29 , @xmath124 is the multiplicity space for the irrep @xmath29 in tensor representation @xmath125 , and where @xmath97 is the set of all irreps that are obtained by coupling some irrep @xmath126 to @xmath119 . comparing the expression above with eq . , the subsystems @xmath100 are given by @xmath127 we use the fact that if @xmath122 , then @xmath128 @xcite , and that @xmath129 @xcite . thus , @xmath130 where each subsystem @xmath131 on the right - hand side carries an irrep @xmath132 of @xmath7 under the action @xmath77 . ( we note that in the examples presented in the latter sections , with groups @xmath133 and @xmath134 , the subsystems @xmath135 are trivial and can be ignored . ) at this stage , we will make use of the @xmath7-covariance of the encoding map @xmath5 , given by eq . , to determine how the message is encoded into the relational subsystems @xmath100 . as the state @xmath136 is @xmath7-invariant under the action of @xmath95 for any @xmath2 , it can be expressed according to the hilbert space decomposition as @xmath137 where @xmath102 is the identity operator on @xmath99 and @xmath138 is an ( unnormalized ) density operator on @xmath100 . this expression defines a set of trace - decreasing superoperators @xmath139 ; note that the latter can be naturally embedded in the full multiplicity spaces @xmath140 of the combined system . from eq . ( [ eq : unital ] ) , we see that the maps @xmath103 are also unital in that @xmath141 is proportional ( because @xmath103 is trace - decreasing ) to the identity on @xmath100 . also , as @xmath77 commutes with @xmath95 , we have that each term @xmath103 is itself @xmath7-covariant , satisfying @xmath142 . we now make use of the stinespring theorem for covariant cp maps @xcite ; in particular , we use a form due to keyl and werner @xcite for unital covariant cp maps . there exists another unitary representation @xmath143 of @xmath7 on a space @xmath106 and an _ intertwiner _ ( linear map ) @xmath144 satisfying @xmath108 with @xmath145 such that @xmath146 the form of eq . allows us to identify a suitable stinespring extension . the representation @xmath147 acts on @xmath100 through what appears to be ( ignoring the limits on the sum @xmath148 ) a tensor representation of an irrep @xmath87 with an irrep @xmath119 . thus , we can choose our stinespring extension in a minimal way such that @xmath143 acts on @xmath106 irreducibly as the irrep @xmath87 of @xmath7 . the operators @xmath109 intertwine the representation @xmath147 with the collective representation @xmath149 on @xmath113 . we now consider two cases : * case a : * if @xmath98 is such that , for all @xmath123 obtained via @xmath150 then @xmath126 , ( i.e. , @xmath65 contains all of the irreps @xmath123 that one can obtain by @xmath150 ) , then the direct sum hilbert space in eq . is given by @xmath151 where @xmath152 denotes that these spaces are unitarily equivalent ; that is , the map @xmath109 is a bijective isometry and simply represents the clebsch - gordan transformation relating the tensor product of two irreps with the direct sum decomposition of @xmath100 given in . let @xmath153 denote the set of irreps satisfying this condition . * case b : * if , however , @xmath98 is such that the condition of case a fails ( i.e. @xmath65 does _ not _ contain all of the irreps @xmath123 that one can obtain by coupling @xmath87 and @xmath119 ) , then the intertwiner is no longer surjective . rather , the intertwiner maps @xmath100 onto a proper subspace of @xmath113 , specifically the subspace defined by the carrier space of the irreps @xmath132 , with @xmath126 , obtained through the coupling of the irrep @xmath87 on @xmath106 with the irrep @xmath119 on @xmath45 . ( this space is necessarily a proper subspace of @xmath154 because , by the conditions of case b , @xmath65 does not contain all irreps obtained in this coupling . ) a set of basis states for this subspace can be calculated explicitly in any particular instance using the clebsch - gordan coefficients for the group @xmath7 . we now turn to the probabilities of each of these cases . we note that @xmath155 = { \rm tr}[\mathcal{e}^{(q)}(\rho)]$ ] is the probability that the system is encoded into the irrep @xmath29 . we now prove that , for the case where @xmath45 carries an irrep @xmath15 of @xmath7 , this probability is independent of @xmath2 . using eq . ( [ eq : eq ] ) , we have @xmath156 \nonumber \\ & = \int { \rm d}g \frac{d_q}{d_r } { \rm tr}_{\bar{\mathcal{n}}^{(q)}_{rs}}\bigl[v^{(q^*)}(g ) a^{(q)\dag } \bigl(i_{\mathcal{k}^{(q ) } } \otimes \rho \bigr ) \nonumber \\ & \qquad \times a^{(q ) } v^{(q^*)}(g)^{-1 } \bigr ] \nonumber \\ & = \frac{d_q}{d_r } { \rm tr}_{\bar{\mathcal{n}}^{(q)}_{rs}}\bigl [ a^{(q)\dag } \bigl(i_{\mathcal{k}^{(q ) } } \otimes \int { \rm d}g\ , u_s(g)\rho u_s(g)^{-1 } \bigr)a^{(q ) } \bigr ] , \nonumber \\ % & = \frac{d_q}{d_s d_r } { \rm tr}_{\bar{\mathcal{n}}^{(q)}_{rs}}\bigl [ a^{(q)\dag}a^{(q)}\bigr]\end{aligned}\ ] ] where in the second line we have used the g - invariance of @xmath103 and in the third line we have used eq . ( [ eq : intertwiner ] ) . because @xmath45 is an irrep , it follows that @xmath157 where @xmath158 , and therefore @xmath159 is independent of @xmath2 for all @xmath29 . the su(2 ) case , presented in sec . [ sec : cartesian ] , provides an explicit example of this . we note that for the general case , where the system does not carry an irrep of @xmath7 , then this probability can be state - dependent . for example , if @xmath45 is a direct sum of irreps @xmath119 , then @xmath160 where @xmath161 is the projector onto the @xmath119 irrep of @xmath45 . in this case , for @xmath162 ( where @xmath163 is unitary ) , we have @xmath164 , independent of @xmath2 . however , for @xmath165 , the weight @xmath159 can depend on @xmath2 . this occurs for the u(1 ) case , as seen explicitly in sec . [ sec : phase ] . with each map @xmath103 now defined through eq . ( [ eq : eq ] ) , we can explicitly express @xmath5 in kraus operator form as @xmath166 , where @xmath167 and where @xmath168 is a basis for @xmath99 and @xmath169 is a basis for @xmath106 . finally , we point out a useful expression for @xmath103 ( which applies regardless of whether @xmath45 carries an irrep of @xmath7 ) . from eq . ( [ eq : eofrho ] ) , it is clear that @xmath170 combining this with the expression for @xmath5 in eq . ( [ eq : ginvstate ] ) and making use of eq . ( [ eq : generaldecoherencedfulldfree ] ) , we obtain @xmath171\,.\ ] ] this form will be used frequently when working out explicit examples . in the communication protocol we are considering , bob s task is to recover the quantum message @xmath2 by implementing a decoding map @xmath172 . a useful recovery map to consider is the following @xmath173 where the adjoint for superoperators is defined relative to the hilbert - schmidt inner product , @xmath174)= \mathrm{tr}(\mathcal{e}[a ] b)$ ] . the map @xmath11 is completely positive and linear , because the superoperator adjoint preserves these features . it is also trace - preserving . to see this fact , observe that @xmath175 , where @xmath176 is the identity operator on @xmath69 ; consequently , for @xmath177 , we have @xmath178=\mathrm{tr}[\rho_{rs}\mathcal{r}^{\dag}(i_{s } ) ] = \mathrm{tr}[\rho_{rs}]$ ] . we have therefore verified that @xmath11 is a valid quantum operation that can be implemented deterministically . assuming that there is no prior information about the input state to @xmath5 , the map @xmath179 is precisely the `` approximate reversal '' operation for @xmath5 proposed by barnum and knill @xcite which yields an error no more than twice that of the optimal reversal operation . the error here is defined in terms of the deviation from unity of the average entanglement fidelity for an arbitrary input ensemble . given a superoperator @xmath180 , the adjoint @xmath181 is easily determined through a kraus decomposition of @xmath180 . specifically , if @xmath182=\sum_{i } k_{i}\rho k_{i}^{\dag}$ ] then @xmath183=\sum_{i}k_{i}^{\dag } \rho k_{i}$ ] . the expression ( [ eq : encoding ] ) for the encoding map @xmath5 provides one kraus decomposition : the covariant set of operators @xmath184 , where @xmath185 . it follows that a covariant set of kraus operators for @xmath186 is @xmath187 where @xmath188 and consequently @xmath189 = d_r \int\mathrm{d}g\bigl ( \langle g| \otimes u_{s}^{\dag}(g)\bigr ) \rho_{rs}\bigl ( |g\rangle \otimes u_{s}(g)\bigr ) \,.\ ] ] from this expression , we see that one way in which @xmath11 can be implemented is as follows : measure the covariant povm @xmath190 on @xmath1 , and finally discard @xmath3 and the measurement result @xmath27 . we refer to this as the `` measure and re - orient '' implementation of the decoding map . ( it is the adjoint of the `` prepare and @xmath7-twirl '' implementation of the encoding map . ) so we see that the decoding map we are considering is in fact the most obvious recovery scheme one can imagine ! bob simply estimates the relative orientation between alice s reference frame and his own by measuring how the sample @xmath3 of alice s rf is oriented , then re - orients the system appropriately ( i.e. in such a way that it is finally oriented relative to his rf in precisely the way that it was initially oriented relative to alice s rf ) . one can view the system @xmath3 as a cryptographic key or calibrating system that contains the information for how to recover the quantum state of @xmath1 . it is noteworthy that this implementation of bob s decoding map does not require any entangling operations between @xmath3 and @xmath1 . bob can achieve it with local operations and classical communication ( locc ) between @xmath3 and @xmath1 . because bob does not need to possess @xmath1 to implement the appropriate measurement on @xmath3 , it follows that alice can subsequently transmit an arbitrary number of systems and bob can decode these with the same fidelity as the first . consider the action of the decoding map on states @xmath115 , i.e. , on states of the form of eq . . after a measurement on @xmath3 having outcome @xmath191 , followed by a transformation @xmath192 to system @xmath1 , the reduced density operator on @xmath1 is @xmath193 & = d_r \int \text{d}g\ , |\langle g|g'\rangle |^{2}\ , u_s((g')^{-1}g)\rho u_s^{\dag}((g')^{-1}g ) \nonumber \\ & = d_r \int \text{d}g\ , | \langle e|g\rangle |^{2}\ , u_s(g)\rho u_s^{\dag } ( g)\ , , \label{eq : method2}\end{aligned}\ ] ] where the simplification occurs because d@xmath27 is invariant . note that the result is independent of the outcome @xmath191 . it is straightforward to check that this state is normalized , as @xmath194 . this is precisely how a state @xmath2 relative to alice s frame would be redescribed relative to bob s frame if their relative orientation @xmath27 was known to be distributed according to the probability distribution @xmath195 . if @xmath196 as a function of @xmath27 is highly peaked around the identity group element @xmath197 then the only unitary that will contribute significantly in the integral will be the identity operation , and we will have @xmath198\simeq\rho$ ] . it is the narrowness of the distribution @xmath196 , a measure of the quality of the quantum reference frame , that determines the degree to which one can recover the quantum information . we see that for bounded - size samples of alice s reference frame , the decoding map we have described achieves _ approximate error correction_. further on , we will show that the degree to which it deviates from perfect error correction is inversely proportional to the size of the quantum reference frame . recall that the `` measure and re - orient '' implementation of the recovery operation @xmath11 was inferred from the adjoint of the kraus decomposition @xmath199 of @xmath5 . we exhibited a different kraus decomposition of the encoding operation in eq . ( [ eq : krause ] ) . the adjoint of the latter provides a novel kraus decomposition of @xmath11 and therefore also a new way of implementing the recovery operation . we will refer to it as the `` extract from the relational subsystems '' implementation . we find that @xmath200 can be written as @xmath201\bigr ) \,,\ ] ] where we define @xmath202 \,,\ ] ] as a map from @xmath203 to @xmath44 . recalling the form of the encoding map @xmath103 of eq . ( [ eq : edecomp2 ] ) , we see that @xmath204 . this implementation of the decoding map @xmath11 differs from the `` measure and re - orient '' scheme in that it requires joint ( i.e. nonseparable ) operations on @xmath3 and @xmath1 . specifically , it is implemented via a joint unitary on @xmath40 followed by a trace on @xmath3 . finally , we highlight another decomposition of @xmath205 that will be useful to us further on . it is the one obtained by taking the adjoint of eq . ( [ eq : equsefulform ] ) , @xmath206 given eqs . ( [ eq : edecomp1 ] ) and ( [ eq : decodingdecomposition ] ) , the composition @xmath19 can be written as @xmath207 = \sum_{q\in q_{rs } } \mathcal{r}^{(q)}\circ\mathcal{e}^{(q)}[\rho]\ , . \label{eq : composition}\ ] ] substituting the expressions for @xmath103 and @xmath205 in eqs . ( [ eq : eq ] ) and ( [ eq : rq ] ) , we obtain @xmath207 = \sum_{q\in q_{rs } } \frac{d_q}{d_r } \mathrm{tr}_{\mathcal{k}^{(q)}}\bigl [ a^{(q)}a^{(q)\dag}\bigl ( i_{\mathcal{k}^{(q)}}\otimes\rho\bigr ) a^{(q)}a^{(q)\dag}\bigr]\ , . \label{eq : composition2}\ ] ] we now consider the two subsets of @xmath97 from sec . [ subsec : relsubsystems ] . in case a , where @xmath208 , the intertwiner @xmath109 is a bijective isometry , and consequently @xmath209 is the identity and @xmath210 = ( d_q^2/d_r ) \rho$ ] . therefore , in this case the quantum information is perfectly recovered by the decoding map . in case b , however , @xmath211 is a non - trivial projection on @xmath113 and the recovery is not perfect . we can express eq . ( [ eq : composition2 ] ) as @xmath212 = \bigl ( \sum_{q\in q_{rs}^{\mathrm{ok } } } \frac{d_q^2}{d_r } \bigr ) \rho \\ + \sum_{q\notin q_{rs}^{\mathrm{ok } } } \frac{d_q}{d_r } \mathrm{tr}_{\mathcal{k}^{(q)}}\bigl[p^{(q)}\bigl ( i_{\mathcal{k}^{(q)}}\otimes\rho\bigr ) p^{(q)}\bigr]\ , . \label{eq : composition3}\end{gathered}\ ] ] this is just an alternative kraus decomposition of the effective decoherence map of eq . ( [ eq : method2 ] ) . we have shown two very distinct ways of implementing one and the same decoding operation . if we describe the reference frame token @xmath3 as an ancilla , then what we have is an example wherein a single map can be implemented either by a joint unitary followed by a trace on the ancilla , or by a measurement of the ancilla followed by a unitary rotation on the system that depends on the outcome of the measurement . the existence of many different implementations of an operation is familiar in quantum information theory . for instance , griffiths and niu have made use of a similar multiplicity of possibilities for the optimal eavesdropping strategies in quantum cryptography @xcite . the multiplicity of ways of implementing a single operation is analogous to the multiplicity of mixtures that lead to the same density operator . two kraus decompositions of our @xmath7-invariant recovery operation differ in their transformation properties under the group : one is a @xmath7-covariant set of operators ( a continuous set in the case of a lie group ) and the other is a discrete set of @xmath7-invariant operators . similarly , a @xmath7-invariant density operator @xmath2 on a finite - dimensional hilbert space admits two sorts of convex decompositions : a spectral decomposition with a discrete number of @xmath7-invariant elements , and the decomposition induced by @xmath2-distortion of a g - covariant povm ( continuous if @xmath7 is a lie group ) @xcite . recognizing this multiplicity of convex decompositions and the fact that no particular decomposition is preferred has been important for resolving many conceptual confusions @xcite . furthermore , each decomposition may yield important insights . in quantum optics , for example , a poissonian mixture of number eigenstates is equivalent to a uniform mixture over coherent states with the same mean number but differing in phase . the decomposition into states with well - defined phase is particularly useful for making predictions about wave - like phenomena , such as interference experiments , whereas the number state decomposition is best for particle - like phenomena , such as determining number statistics @xcite . similarly , each of the two decompositions we have provided of our decoding operation provides some insight into our problem . the `` measure and re - orient '' scheme is clearly the most intuitive and demonstrates that joint operations on reference token and system are not necessary to implement our recovery map . the `` extract from relational subsystems '' scheme demonstrates that if bob begins by measuring the irrep of the composite of @xmath3 and @xmath1 , he learns whether the state was in a `` good '' irrep or not and consequently whether or not he has achieved a perfect decoding . this sort of post - selectively perfect decoding operation is discussed in the following section . thus far we have only judged decoding schemes by their average performance . it is also possible to say something about the best and worst case performance . the `` measure and re - orient '' scheme is not particularly interesting in this regard : one achieves precisely the same fidelity of recovery regardless of the outcome of the covariant measurement on the rf token , so that the best and worst case recoveries are equivalent to the average . on the other hand , in the `` extract from the relational subsystems '' scheme , we found that the fidelity of the recovery depends on the irrep of the composite of rf token and system into which the input state was encoded . furthermore , given that the decoding operation was incoherent over these irreps , it is always possible to make a projective measurement that distinguishes these . depending on the measurement outcome , one can achieve decodings with fidelities that are sometimes better and sometimes worse than the average . indeed , by enhancing the `` extract from the relational subsystems '' scheme with such a measurement , bob can achieve _ perfect _ decoding with some probability . specifically , if he finds one of the `` good '' irreps , @xmath162 , then @xmath103 is invertible and the decoding operation @xmath205 of eq . ( [ eq : rq ] ) recovers the quantum message perfectly . ( of course , if he achieves one of the `` bad '' irreps , @xmath213 , then he achieves a decoding that is worse than the average . ) recalling eq . ( [ eq : composition3 ] ) and making use of eq . ( [ eq : dr ] ) , the probability of perfect recovery is @xmath214 such a decoding scheme achieves post - selectively perfect error correction @xcite . it is akin to achieving unambiguous discrimination of a set of nonorthogonal quantum states @xcite . for this implementation of the decoding , note that bob must be able to store the quantum token of alice s reference frame coherently until the time when he receives the message systems . another point worth noting : if alice is sending a large number of systems and bob wishes to implement probabilistically perfect error correction on some subset of them , he must wait until he has collected all of the systems in that subset . the reason is that he must perform a joint measurement on the composite of these and the rf token . furthermore , after his measurement is complete , he has disrupted the state of the rf token and he can no longer achieve perfect error correction for any other systems . the tradeoffs involved in such postselectively perfect decoding schemes are an interesting topic for future research . finally , we can consider modifying the encoding operation rather than the decoding in a similar way . note that if , immediately after implementing her encoding operation , alice implements a projective measurement of the irrep of the composite of rf token and system , she can come to know whether a subsequent decoding operation will achieve a perfect recovery or not . in addition , if it happens that alice has a classical description of the quantum message rather than merely having a sample , then she can prepare the quantum state many times and only initiate transmission when her measurement finds one of the `` good '' irreps . in this case , the quantum message is perfectly encoded into a pure @xmath7-invariant state of the composite of system and rf token . such a scheme therefore achieves a relational encoding akin to the one presented in bartlett _ the precise connection of our results to the latter encoding is an interesting topic for future research ( as is the application of the mathematical tools developed here to the general problem of calibration - free communication schemes discussed in the introduction ) . the quantum state of a harmonic oscillator is always referred to some phase reference @xcite . in this example , we consider using one quantum harmonic oscillator ( a single mode ) as a phase reference for another , and investigate the effect of bounding the maximum number @xmath215 of excitations in the phase reference . specifically , consider the single - mode rf to be prepared in the bounded - size phase eigenstate @xmath216 where @xmath217 is the fock state with @xmath218 excitations . this state is of the form of our general state ( [ eq : chiribella ] ) for the case of @xmath219 u(1 ) . for the system , we consider a qubit encoded in the two - dimensional subspace spanned by @xmath220 and @xmath221 . note that the system we consider does not carry an irrep of u(1 ) , and in fact u(1 ) has only one - dimensional irreps . because our main result concerning the representation of the encoding map , presented in sec . [ subsec : relsubsystems ] , was only proven under the assumption that @xmath45 is an irrep , the u(1 ) example can not be presented as a special case of this result . nonetheless , we find the u(1 ) example to be in accord with the general result , suggesting that our theorem applies more generally . for simplicity , we consider a system prepared in an arbitrary pure state @xmath222 our results will directly extend to the mixed - state case via the linearity of convex combination . the overlap of the rf state @xmath223 with its rotated version is @xmath224 rotations in u(1 ) act on the qubit system state as @xmath225 evaluating eq . ( [ eq : method2 ] ) then gives @xmath226}{1-\cos \theta } \left(\begin{smallmatrix } |\alpha |^{2 } & \alpha \beta ^{\ast } e^{i\theta } \\ \alpha ^{\ast } \beta e^{-i\theta } & |\beta |^{2 } \end{smallmatrix}\right)\,.\ ] ] noting that @xmath227}{1-\cos \theta } & = n_{r}+1\ , , \\ \int \tfrac{d\theta } { 2\pi } \tfrac{1-\cos [ ( n_{r}+1 ) \theta ] } { 1-\cos \theta } e^{i\theta } & = n_{r}\,,\end{aligned}\ ] ] which also gives the normalization , we have @xmath228 \ , , \end{aligned}\ ] ] where @xmath10 here denotes the u(1)-twirling operation ( the dephasing map ) . it follows that in the `` measure and re - orient '' scheme for decoding , regardless of the outcome of the measurement , the reduced density operator is with probability @xmath229 the state @xmath230 , while with probability @xmath231 it is completely dephased in the @xmath232 basis . the overall effect of encoding and decoding is to implement a partial dephasing . because the irreps of u(1 ) are all one - dimensional , we have @xmath233 and consequently , by eq . ( [ eq : edecomp1 ] ) , the encoding operation @xmath5 may be expressed simply as @xmath234 . by virtue of eq . ( [ eq : equsefulform ] ) , each operation @xmath235 may in turn be expressed as @xmath236\pi^{(n ) } \,,\ ] ] which evaluates for different values of @xmath237 as : @xmath238 the decoding operation has the form @xmath239 where @xmath240 . one easily verifies that @xmath241 maps @xmath242 to @xmath220 and @xmath243 to @xmath221 , so that @xmath244 \propto \begin{cases } \left\vert 0\right\rangle , & n=0 \\ \alpha \left\vert 0\right\rangle + \beta \left\vert 1\right\rangle , & 0<n < n_{r}+1 \\ \left\vert 1\right\rangle , & n = n_{r}+1\ , . \end{cases}\ ] ] the probability of the outcome @xmath245 is @xmath246 , of @xmath247 is @xmath248 , and of each of the other outcomes is @xmath231 . weighting the decoded states @xmath249 by these probabilities , we can verify that eq . ( [ eq : partialdephasing ] ) is recovered . in this particular example , taking the adjoint of the encoding operation as one s recovery operation is actually optimal . the proof is as follows . for @xmath250 , the recovery operation is perfect and consequently optimal . otherwise , the action of the encoding map is to measure the system in the @xmath251 basis and update it to one of two orthogonal states [ as can be inferred from eq . ( [ eq : u(1)encodingcases ] ) ] . it is a well - known result that the update map that maximizes the entanglement fidelity is simply the lders rule ( or projection postulate ) @xcite , and this is precisely what the composition of the encoding with the recovery operation achieves . the fact that the optimal recovery operation can be achieved using a `` measure and reorient '' scheme shows that having the classical resource of partially correlated reference frames that is obtained by this scheme is just as good as having the quantum reference frame token , at least for the purpose of optimizing average - case performance in decoding . this is a surprising result because one might have expected the quantum resource to always do better . finally , by implementing a projective measurement of the total number and post - selecting on finding @xmath252 , it is clear that bob can achieve perfect decoding . this occurs with probability @xmath253 for a cartesian frame , the relevant group is the rotation group . rather than @xmath254 to allow for spinor representations of the rotation group . ] the charge sectors ( irreps ) are labeled by a nonnegative integer or half - integer @xmath255 , and the irreps are @xmath256-dimensional with the standard basis @xmath257 . we bound the size of our reference frame token by bounding @xmath255 . recall that the fiducial state for the frame , eq . ( [ eq : chiribella ] ) , requires us to work in a subspace @xmath258 satisfying eq . ( [ eq : specialproperty ] ) . in the cartesian case , we are confined to @xmath255 values such that @xmath259 . we denote the largest such value by @xmath260 . ( as an example , for an even number @xmath237 of spin-1/2 particles , only the highest irrep , @xmath261 , fails to satisfy eq . ( [ eq : specialproperty ] ) , and consequently @xmath262 . see @xcite . ) for simplicity , we restrict our attention to integer values of @xmath260 ( similar results can be obtained if one also allows non - integer values ) . the fiducial state of the rf token is @xmath263 where @xmath264 the system is taken to be a spin-@xmath265 particle . because this is an irrep of su(2 ) , the general results of sec . [ subsec : relsubsystems ] apply . we choose the following parametrization of su(2 ) , @xmath266 describing a rotation by angle @xmath267 about the unit vector @xmath268 . let @xmath269 where @xmath270 is the fiducial rf state of eq . ( [ eq : spinchiribella ] ) . then @xmath271 where @xmath272 is the spin-@xmath255 irrep of su(2 ) , and @xmath273 $ ] are the characters of su(2 ) . these characters are independent of @xmath274 and @xmath275 . they are given by @xmath276}{\sin[\omega /2]}\,,\ ] ] using the following identity @xmath277}{2\sin[\omega/2 ] } = 1/2 + \sum_{k=1}^n \cos(k\omega)\,,\ ] ] we find that @xmath278(1+\cos\omega)}{\sin\omega(1-\cos\omega ) } \\ - 2 ( { j_r}+1)\frac{\cos[\omega({j_r}+ 1)]}{(1-\cos\omega)}\bigr)^2\,.\end{gathered}\ ] ] in terms of this parametrization , the su(2 ) invariant measure is @xmath279 where @xmath280 , @xmath281 , and @xmath282 . for the rotation @xmath283 on the qubit system in this parametrization , we have @xmath284 it follows that the composition of encoding and decoding maps , given by eq . ( [ eq : method2 ] ) , is @xmath285 \label{eq : su(2)method1}\end{aligned}\ ] ] where @xmath10 is the su(2)-twirling operation ( which is completely decohering for a single qubit ) . next , we determine the precise nature of the relational subsystems where the quantum information is encoded . in this example , the multiplicity spaces play a key role . we begin by describing the group - induced structure of the hilbert spaces , both for the rf and the total system . under the action of the representation @xmath16 of su(2 ) , the hilbert space for the reference frame is decomposed as @xmath286 the joint system @xmath40 , consisting of the rf plus a spin-1/2 qubit , carries a collective representation @xmath287 of su(2 ) which can easily be determined using standard angular momentum coupling . for coupling a spin-@xmath255 irrep to a spin-1/2 irrep , we have @xmath288 . thus , the hilbert space of the joint system @xmath40 has a similar decomposition under the action of @xmath95 , given by @xmath289 the multiplicity spaces for the joint system @xmath40 are related to those of the rf as @xmath290 for simplicity , we consider the qubit state to be pure , expressed in the standard angular momentum basis as @xmath291 the encoded state within the @xmath292th irrep is @xmath293 to evaluate this expression , we first evaluate @xmath294 where we recall that @xmath295 is defined on the subsystem @xmath296 . therefore , the state ( [ eq : jmpurestate ] ) is an unnormalized vector on @xmath297 . we transform @xmath298 to a coupled basis using clebsch - gordan coefficients @xmath299 . in terms of the bases used in ( [ eq : spinchiribella ] ) and ( [ eq : spinqubitpure ] ) , we have @xmath300 we note that the states @xmath301 for @xmath302 ( @xmath303 ) form a basis of @xmath304 ( @xmath305 ) . it follows that the full set of states @xmath306 form a basis of @xmath297 . we have @xmath307 we use the following clebsch - gordan identity , @xmath308 to obtain @xmath309 we now consider two cases for @xmath292 separately . for @xmath310 , we note that the multiplicity space @xmath297 is unitarily equivalent to the tensor product of a spin-@xmath292 and a spin-1/2 system coupled to total angular momentum @xmath311 . that is , @xmath312 where @xmath313 carries an irrep @xmath292 of su(2 ) and @xmath152 denotes unitary equivalence . we explicitly define the bijective isometry @xmath314 via its adjoint action on a basis for @xmath315 as @xmath316 in terms of this new subsystem structure for the multiplicity spaces , we can express ( [ eq : intermediatestep ] ) as @xmath317 where @xmath318 is defined by eq . ( [ eq : spinqubitpure ] ) . it follows that the encoded state for an irrep @xmath292 where @xmath319 is @xmath320 \nonumber \\ & = \sum_{m =- j}^j \langle j , m| ( |e_{j_r } \rangle \langle e_{j_r}| \otimes \rho)|j , m\rangle \nonumber \\ & = \frac{2j+1}{d_r } a^{(j)\dag}\bigl ( i_{\mathcal{k}^{(j ) } } \otimes \rho \bigr ) a^{(j ) } \ , , \quad j < j_r+\tfrac{1}{2}\ , . \label{eq : usualjqubit}\end{aligned}\ ] ] because @xmath321 is bijective , ( specifically , because the set of states @xmath322 are orthogonal ) we find the qubit faithfully encoded into the relational subsystem whenever @xmath323 . however , for the result @xmath324 , the multiplicity space @xmath297 is exceptional ; see ( [ eq : su2multspacedirectsum ] ) . we can not factorize @xmath297 as in eq . ( [ eq : magicdecomp ] ) . nonetheless , we can still introduce a hilbert space @xmath325 which carries an irrep @xmath326 of su(2 ) and in terms of it we can define an isometry @xmath327 , by modifying eq . ( [ eq : intertwineraction ] ) to include only the @xmath328 term in the sum . this isometry is simply not surjective . it follows that the set of states @xmath329 are no longer orthogonal . therefore , the map @xmath330 is no longer invertible . the action of @xmath331 can be viewed as a projection of uncoupled states on @xmath332 onto the subspace of states which couple to total angular momentum @xmath260 . the probability assigned to each irrep @xmath292 is @xmath333 \nonumber \\ & = \begin{cases } \frac{(2j+1)^2}{d_r } \ , , & j < j_r+\frac{1}{2}\ , , \\ \frac{(2j_r+1)(j_r+1)}{d_r } \ , , & j = j_r+\frac{1}{2}\ , . \end{cases}\end{aligned}\ ] ] we note that these probabilities satisfy @xmath334 . the decoding map within each irrep @xmath292 takes the form @xmath335 \nonumber \\ & = ( 2j+1 ) \langle e_{j_r}| i_{\mathcal{m}^{(j)}_{rs } } \otimes ( \cdot ) |e_{j_r}\rangle \,.\end{aligned}\ ] ] for @xmath319 , eq . ( [ eq : usualjqubit ] ) gives @xmath336 for @xmath337 , we make use of eq . ( [ eq : exceptionaljqubit ] ) and find @xmath338\,,\end{gathered}\ ] ] where @xmath10 is the su(2)-twirling operation ( complete decoherence ) . averaging over the irreps with the weights given in eq . ( [ eq : probscartesian ] ) , we find the decoded state to be @xmath339\nonumber \\ & \qquad + \tfrac{(2j_r+1)(j_r+1)}{d_r } \bigl ( \tfrac{2j_r}{6({j_r}+1)}\mathcal{i } + \tfrac{(2j_r+3)}{3({j_r}+1)}\mathcal{g } \bigr ) [ \rho ] \nonumber \\ & = \bigl ( \tfrac{j_r}{{j_r}+1}\mathcal{i } + \tfrac{1}{{j_r}+1}\mathcal{g } \bigr ) [ \rho ] \,,\end{aligned}\ ] ] thereby verifying that eq . ( [ eq : su2effectivedecoherence ] ) is recovered . we note that our recovery operation is perfect in the non - exceptional irreps and therefore optimal there . in the exceptional irrep , the error incurred in recovery is at most twice that of the optimal recovery , as described in sec . [ sec : decoding ] , by virtue of being of the form of barnum and knill s approximate reversal . we leave it as an open problem to to prove whether this recovery is in fact optimal , or if not , to identify what form an optimal recovery map would take . post - selectively perfect decoding is achieved if bob implements a projective measurement that distinguishes the irreps and obtains @xmath340 . by eq . ( [ eq : probscartesian ] ) and eq . ( [ eq : drcartesian ] ) ( or by eq . ( [ eq : pperfect ] ) directly ) , we see that the probability for this to occur is @xmath341 a directional rf identifies only a single direction in space , as opposed to a full set of axes . such an rf is not associated directly with a lie group , but rather with a coset space . specifically , although su(2 ) may provide a group of transformations between all possible directional rfs , any one directional rf is invariant under u(1 ) rotations about its axis of symmetry ; the relevant coset is then su(2)/u(1 ) . because of this distinction , we expect this example to proceed differently from the other two . the distinction is immediately apparent because there is no obvious candidate for a fiducial reference state as in eq . ( [ eq : chiribella ] ) . instead , we take the directional rf to be in an su(2)-coherent state of size @xmath260 , so that the fiducial state is @xmath342 we consider a qubit system which is described relative to alice s local cartesian frame by the state @xmath343 where @xmath344 . note first of all that even if bob shared a classical reference direction with alice , her @xmath345 axis for instance , his description of the system is still related to alice s by a dephasing operation . the reason is that he _ only _ shares alice s @xmath345-axis , and so the rotation about @xmath345 that relates his local @xmath346-axis to hers is completely unknown . averaging over rotations @xmath347 yields the dephasing operation @xmath348 = \int_0^{2\pi } \frac{d\theta}{2\pi } r_z(\theta ) \rho r^\dag_z(\theta ) \\ = |\alpha|^2 |0\rangle \langle 0| + |\beta|^2 |1\rangle \langle 1|\,.\ ] ] consequently , if bob has a bounded - size token of alice s @xmath345-axis , his decoding will yield a state that approaches @xmath349 rather than @xmath2 as one increases the size of the token . define @xmath350 where @xmath351 . the encoding of @xmath2 relative to bob s local cartesian frame is the following state of the composite of rf token and system @xmath352 to decode , bob measures the covariant povm @xmath353 on the rf token and reorients the system accordingly . the net result is @xmath354 the effect of this map will be particularly simple given that @xmath270 is invariant under u(1 ) rotations about the @xmath355-axis . for this calculation , it will be easiest to use euler angles to parametrize su(2 ) : @xmath356 where @xmath357 $ ] . ( we note this parametrization is different from that used in sec . [ sec : cartesian ] . ) with this parametrization , @xmath358^{2{j_r}}\,,\end{aligned}\ ] ] and thus @xmath359^{4{j_r}}$ ] . we can express @xmath360 as a @xmath361 matrix in the @xmath355-basis as @xmath362 and our qubit system in bloch vector notation as @xmath363 using the identities @xmath364 we find that @xmath365 \,,\end{aligned}\ ] ] where @xmath10 is the su(2)-twirling operation , and @xmath366 is the dephasing operator defined in eq . ( [ eq : dephasing ] ) . first , we note that the rf token , consisting of only a spin-@xmath260 system , does not possess a multiplicity space . when coupling this spin-@xmath260 system to the spin-1/2 qubit , the resulting collective system is described by @xmath367 and does not possess any multiplicity spaces either . given that the fiducial state , eq . ( [ eq : fiducialdirection ] ) , is not of the form of eq . ( [ eq : chiribella ] ) , the derivation of eq . ( [ eq : edecomp1 ] ) is no longer valid . nonetheless , the encoding map defined by eq . ( [ eq : encodingdirection ] ) may still be written in the form of eq . ( [ eq : edecomp1 ] ) , namely , @xmath368 where @xmath369\,,\ ] ] which is of the same form as eq . ( [ eq : equsefulform ] ) . to see that this decomposition exists , simply express eq . ( [ eq : encodingdirection ] ) as @xmath370 where @xmath10 is the su(2)-twirling operation . then , using eq . ( [ eq : generaldecoherencedfulldfree ] ) and eq . ( [ eq : hdecomp_direction ] ) , we have @xmath371 \,,\end{aligned}\ ] ] which is equivalent to eqs . ( [ eq : qqq ] ) and ( [ eq : qqq2 ] ) . note that @xmath372 is still a map from @xmath44 to @xmath373 , but in this case @xmath374 , so that it maps density operators to scalars . specifically , @xmath375 we see that the encoding operation in this case can be described as follows : after adjoining the rf token to the system , destructively measure the total angular momentum @xmath376 of the composite and upon obtaining quantum number @xmath292 , reprepare the system in the associated su(2)-invariant state @xmath377 . the decoding operation defined in eq . ( [ eq : decodingdirection ] ) is clearly proportional to the hilbert - schmidt adjoint of the encoding operation . consequently , it admits a decomposition into irreps via the adjoints of eq . ( [ eq : qqq ] ) , namely , @xmath378 \,,\ ] ] where @xmath379 is a map from @xmath380 to @xmath44 . this is of the same form as eq . ( [ eq : decodingdecomposition ] ) . to determine @xmath381 , we calculate the adjoint of eq . ( [ eq : qqq2 ] ) and determine the normalization by requiring that @xmath382=1 $ ] for all @xmath292 . we obtain @xmath383 where @xmath384 . except for the constant of proportionality , this has the form of eq . ( [ eq : rqusefulform ] ) . it evaluates to @xmath385 [ which one could also infer by taking the adjoint of eq . ( [ eq : ejexplicit ] ) ] . consequently , the decoding operation may be described as follows : destructively measure the total angular momentum ( squared ) , @xmath376 , on the composite of rf token and system and upon obtaining quantum number @xmath386 , reprepare the system in one of the two states in eq . ( [ eq : rjp ] ) . the composition of encoding and decoding yields @xmath387 \,,\end{aligned}\ ] ] in agreement with eq . ( [ eq : effectivedecoherencedirection ] ) . we note that this coincides with the result obtained by poulin @xcite . this recovery operation is of the form of the approximate reversal operation described by barnum and knill @xcite and is therefore near - optimal in the sense described in sec . [ sec : decoding ] . although it is not itself the optimal recovery map , the latter is easy to find in this example and we do so presently . given that the only pure states of the system that one can hope to reconstruct in the limit of an unbounded rf token are the @xmath388 eigenstates , denoted here by @xmath389 and @xmath390 , it is natural to take as our figure of merit the average input - output fidelity equally weighted over these two input states ( because then one can achieve fidelity 1 in the limit of an unbounded rf token ) . the only information about the state of the system that is encoded in @xmath136 is encoded in the relative weights of its two su(2)-invariant terms . consequently , the optimal decoding operation must consist of a determination of @xmath292 followed by a repreparation of the system state . we make use of previous work on the optimal estimation of the relative angle between a spin-1/2 system and a spin-@xmath260 reference frame @xcite . these results show that a measurement of @xmath376 on the composite is in fact optimal for estimating whether the system state was @xmath389 or @xmath390 given a uniform prior over the two . it is also shown there that the posterior probabilities one ought to assign to these two states upon obtaining outcomes @xmath391 and @xmath392 are @xmath393 in order to optimize the fidelity , one must reprepare the state that is most likely given the outcome , so that one should reprepare @xmath389 given the ` @xmath394 ' outcome and @xmath395 given the ` @xmath396 ' outcome . @xmath11 falls short of this optimal recovery because , by eq . ( [ eq : rjp ] ) , it does not reprepare @xmath389 given the ` @xmath394 ' outcome ; instead , it prepares a mixed state reflecting bob s knowledge of the state given the measurement outcome . we thank robin blume - kohout , j .- c . boileau , matthias christandl and renato renner for helpful discussions . sdb acknowledges the support of the australian research council . rws acknowledges support from the royal society . pst acknowledges the support of the aif , icore , mitacs , and jsps .

the study of charged relativistic fluid balls attract the interest of researchers of different areas of physics and astrophysics . there exists a general consensus that astrophysical objects with large amounts of charge can not exist in nature @xcite , @xcite . this point of view had been challenged by several researchers @xcite , @xcite , @xcite , @xcite . it can not be discarded the possibility that during the gravitational collapse or during an accretion process onto a compact object the matter acquires large amounts of electric charge . this has been considered in @xcite and @xcite . in this paper we study the stellar structure and temporal evolution of compact charged fluid spheres , independently on the mechanism by which the matter acquires an electric charge . from a pure theoretical point of view , the collapse of charged fluid balls and shells are connected with the laws of black hole thermodynamics . the third law of black hole thermodynamics states that no process can reduce the surface gravity of a black hole to zero in a finite advanced time @xcite , @xcite , @xcite . the surface gravity , @xmath0 , plays the role of a temperature while the area of the event horizon is equivalent to the entropy . israel @xcite gave a formulation and proof of the third law . it was demonstrated that the laws of black hole thermodynamics are analogous to the common laws of thermodynamics @xcite . translated to the physics of fluid collapse , the third law implies the impossibility of forming an extremal black hole , for which @xmath1 . an extremal black hole has a total charge @xmath2 : @xmath3 , where @xmath4 is the gravitational constant , and @xmath5 is the total mass of the black hole . the extremal black hole has not an horizon and constitutes a naked singularity , contradicting the cosmic censorship hypothesis . so , in this paper we explore the possibility of forming an extremal black hole from the collapse of a ( charged ) compact object . we will show that the answer is no . the relativistic equations for the collapse of a charged fluid ball were obtained by bekenstein @xcite . so far in the literature , as long as the author knows , it was studied the dynamics of charged shells or scalar fields falling onto an already formed charged black hole ( see @xcite , @xcite , @xcite , and references therein ) , and solutions of the tolman - oppenheimer - volkoff @xcite equations for charged stars ( see @xcite , @xcite , @xcite , @xcite , etc ) . however , in the stellar collapse , the collapsing matter determines the background geometry and the metric of space - time is changing with time as the star collapses . moreover , in general relativity , the pressure of the fluid contributes to the gravitational field and strengthen it . from the present study it is possible to understand if the coulomb repulsion will prevent or not the total collapse of a charged fluid ball , and which are the stability limits for a charged relativistic star . we emphasize that we will not be concerned here with the mechanism of electric charge generation , nor with effects that could neutralize or discharge the star . in the present paper , the equations for the evolution of charged fluid spheres in schwarzchild - like coordinates are presented in sec . [ relatequ ] . in the sec . [ sec2 ] , we re - derived the equations of hydrostatic equilibrium and the relativistic equations for the temporal evolution of charged spheres in a form closer to that obtained by may and white @xcite , @xcite ( for the case of zero charge ) . the equations obtained are well suited for performing a finite difference scheme , and allows a direct implementation of very well known numerical techniques @xcite , @xcite , @xcite . the formalism is also compatible with the bekenstein equations @xcite , and with the misner and sharp equations @xcite for the case of zero charge . the calculations performed in this study are rather lengthy in general and we tried to be the more self - consistent as possible . however , we apologize that in most of the paper we only give hints for the computation of the intermediate algebraic steps in order to keep the paper at a reasonable length . the matching conditions between the exterior and interior solutions of the einstein - maxwell equations are given in the sec . [ match ] . for completeness , in sec . [ seceos ] it is considered an equation of state of a zero temperature neutron gas . two codes were built and are used in the present study : one to obtain neutron stars models in hydrostatic equilibrium , presented in the sec . [ sectov ] ; and the other code is to evolve the stars forward in time , integrating the einstein - maxwell equations . this code is introduced in sec . [ sechydro ] . in all the simulations , the charge distribution is taken proportional to the rest mass distribution . we found that although the total binding energy grows with the total amount of charge , the binding energy per nucleon tends to zero as the charge tends to the extremal value . this is related to the impossibility of obtaining bounded solutions of extremely charged fluid configurations . in addition , it results that the upper limit for the total amount of charge could be slightly lower than the extremal value . this is discussed in the sec . [ secresults ] . in the sec . [ secresults ] , is discussed the stability and evolution of the stellar models . we study , as well , the evolution of stars that were suddenly discharged to take into account the case of charged spheres being a metastable state of more complex scenarios . in a recent work ghezzi and letelier @xcite presented a numerical study of the collapse of charged fluid spheres with a polytropic equation of state , and with an initial uniform distribution of energy density and charge . they found that during the evolution a shell of higher density is formed near the surface of the imploding star . from that work we can not say if a shell will always form when changing the initial conditions and the scenario for the collapse . in the present study , we checked that the effect is negligible during the collapse of a charged neutron star ( see sec . [ secresults ] ) . in the sec . [ secremarks ] , we end with some final remarks . considering a spherical symmetric fluid ball , the line element in schwarzchild - like , or standard form , is ( see @xcite , @xcite , @xcite ) : @xmath6 an observer moving radially have a 4-velocity , and the electric 4-current is . the energy - momentum tensor of the charged fluid is given by : @xmath7\end{aligned}\ ] ] where @xmath8 is the density of mass - energy , @xmath9 is the scalar pressure , and @xmath10 is the electromagnetic tensor . the electromagnetic field satisfies the maxwell equations : @xmath11 and @xmath12}=0\,.\ ] ] only the radial component @xmath13 is non - zero , and the last equation is satisfied if @xmath14 . the covariant derivative in eq . ( @xmath15 ) can be written as @xmath16_{,\nu}=4 \pi j^{\mu}\,,\ ] ] with @xmath17 . defining @xcite @xmath18 the eq . ( [ maxwell2 ] ) gives @xcite @xmath19 and @xmath20 integration of eq . ( [ mw1 ] ) gives @xcite @xmath21 where @xmath22 and from eq . ( [ mw2 ] ) @xmath23 from this equation we see that the charge is a constant @xmath24 outside the fluid ball , where @xmath25 . far away from the sphere , where @xmath26 ( see below ) , the eq . ( [ campoe ] ) gives the electric field of a classical charged particle . the einstein s equations to be solved are : @xmath27 here ( and only here ) , @xmath28 denotes the scalar curvature , and @xmath29 is the ricci tensor . the components of the einstein equations are ( see @xcite , @xcite ) : @xmath30= \nonumber\\ & & \hspace{0.8 cm } e^{-\lambda } \,\biggl(\frac{\lambda'}{r}-\frac{1}{r^2}\biggr)+\frac{1}{r^2}\,,\end{aligned}\ ] ] @xmath31= \nonumber\\ & & \hspace{0.8 cm } -\,e^{-\lambda}\ , \biggl(\frac{\nu'}{r}+\frac{1}{r^2}\biggr)+\frac{1}{r^2}\,,\end{aligned}\ ] ] @xmath32 multiplying the right hand side of eq . ( [ e1 ] ) by @xmath33 and rearranging , the equation can be cast in the form @xmath34\ , r^2\,.\ ] ] setting @xmath35 , with @xmath36 and @xmath37 : @xmath38 the eq . ( [ re1 ] ) becomes @xmath39\ , r^2\,.\ ] ] integrating this equation , we get the equation for the mass @xcite : @xmath40\ , r^2\ , dr \nonumber\\ & & \hspace{0.8cm}+\frac{1}{2 c^2 } \frac{q^2}{r}+ \frac{1}{2 c^2 } \int^r_0 \frac{q^2}{r^2 } dr+m_0\ , , \end{aligned}\ ] ] where @xmath41 is an integration constant , and can be taken as zero for the purpose of this work . outside the fluid ball , the mass do not depend on @xmath42 and then : @xmath43 for @xmath44 , where @xmath45 is the coordinate @xmath42 of the surface of the sphere . actually the mass do not depend on @xmath46 for @xmath44 ; from eqs . ( [ e3 ] ) and ( [ gamma0 ] ) : @xmath47 as @xmath2 is constant and @xmath48 outside the fluid ball , @xmath49 is independent of @xmath46 . subtracting eq . ( [ e1 ] ) from eq . ( [ e2 ] ) it is @xmath50 so , @xmath51 is independent of @xmath42 . to get an asymptotic flat solution it must be @xmath52 as @xmath53 . thus a solution of eq . ( [ metrica0b ] ) is @xcite @xmath54 using the eq . ( [ gamma0 ] ) , ( [ coefmetrica1 ] ) , and substituting into eq . ( [ metrica0 ] ) , the line element outside the ball is @xmath55 which is the reissner - nordstrm space - time . the gravitational mass is @xmath56 . we see that although the ( interior ) mass distribution and the metric depends on time , the external space - time is static . this is the birkhoff theorem , and implies that a spherical distribution of mass and charge can not emit gravitational waves . in this section we will specialize the equations for the case of a coordinate system co - moving with the fluid . we will use a notation closer to that used in the may & white papers @xcite . the 4-velocity for an observer co - moving with the fluid is @xmath57 , and the measured 4-current is @xmath58 . the 4-velocity satisfies : @xmath59 . the line element for co - moving coordinates in standard " form ( see @xcite , pag . 336 ) is : @xmath60 note that now @xmath61 , @xmath62 , and @xmath28 are functions of @xmath63 and @xmath46 . we will see in the next subsection that the coordinate @xmath63 can be chosen to be the total rest mass inside a sphere , such that each spherical layer of matter can be labeled by the rest mass it contains . the stress tensor in co - moving coordinates is obtained using the solution of the maxwell equations ( [ campoe ] ) : @xmath64 from now on we will restore in the equations the gravitational constant @xmath65 and the speed of light @xmath66 . similarly as was defined above , @xmath67 is the density of energy in @xmath68 $ ] and @xmath9 is the scalar pressure in the same units . it will be useful to perform the split @xmath69 where @xmath70 is the rest mass density and @xmath71 is the internal energy of the gas . we want to set the radial coordinate equal to the rest mass enclosed by co - moving spherical layers . we performed a calculation analogous to that of may and white @xcite , and used the same notation for clarity and comparison . in this work is assumed that there is only one species of particles , and there are no particles created nor destroyed . the number density of particles will be denoted by @xmath72 and the rest mass density by @xmath70 . if the rest mass of the particles is @xmath73 then : @xmath74 the conservation of the baryon number current is expressed : @xmath75 or equivalently : @xmath76 the line element in spherical coordinates written in standard " form is : @xmath77 observe that this line element is not identical with that given in eq . ( [ comline ] ) . the conservation of baryons in this frame is reduced to ( see eq . [ baryons ] ) : @xmath78 thus the quantity @xmath79 is a function of @xmath28 only @xmath80 choosing @xmath81 : @xmath82 on the other hand , the proper mass is @xmath83 where @xmath84 is the volume element of the space section , and @xmath85 is the volume of integration . then @xmath86 is the proper mass enclosed by a sphere of circumference @xmath87 , and radial coordinate @xmath88 . with this election of coordinates : @xmath89 so we can perform the coordinate transformation @xmath90 and in this case the metric functions change to @xmath91 replacing these functions on the line element ( [ comline2 ] ) , dropping the primes , and using the eq . ( [ diffs ] ) we obtain the line element given in the eq . ( [ comline ] ) . so , the radial lagrangian coordinate was gauged to be the rest mass `` @xmath63 '' of each spherical layer of matter . the proper time derivative of the charge is ( see eq . [ dqdt ] ) @xmath92 because it is possible to write the 4-current as a product of a scalar charge density times the 4-velocity , i.e. : @xmath93 . thus the electric charge is conserved in spherical layers co - moving with the fluid . the einstein equations ( see the appendix ) for the components @xmath94 are @xmath95 where we use the notation @xmath96 , @xmath97 , etc . the equation for the component @xmath98 is , must be a plus . this is immediately evident when obtaining eq . ( 42 ) from eq . ( 35 ) , in that work . ] : @xmath99_{,\mu}\ , \nonumber\\ & & \hspace{0.45cm}-\frac{g q q_{,\mu}}{c^2 r}=g m_{,\mu}-\frac{g q q_{,\mu}}{c^2 r}\ , , \end{aligned}\ ] ] and the equation for the component @xmath100 is ) to eliminate @xmath101 from the component @xmath100 of the einstein s equations , given in the appendix . in addition , use that @xmath102 . ] : @xmath103_t\,\nonumber\\ & & \hspace{2.02cm}=-g m_{,t}\ , . \end{aligned}\ ] ] we introduced into eqs . ( [ e00 ] ) and ( [ e11 ] ) , the definition of the total mass : @xmath104 we give further definitions that simplify the aspect of the equations : @xmath105 with this definitions an integral of the eq . ( [ e00 ] ) and ( [ e11 ] ) is : @xmath106 where @xmath49 is given by the eq . ( [ totalmass ] ) . combining eqs ( [ metricgrr ] ) and ( [ r01 ] ) we get the mass conservation equation @xmath107 it can be seen that the quantity @xmath108 ( see eq . [ udefin ] ) is the radial component of the 4-velocity of the fluid in schwarzschild coordinates . using the transformation rule for the contra - variant 4-vector @xmath109 from co - moving to schwarzschild coordinates : @xmath110 , yields @xcite : @xmath111 @xmath112 from the einstein equations and the contracted bianchi identities follows that : @xmath113 in co - moving coordinates the components are : @xmath114 expanding the eq . ( [ dmetb ] ) and using the eq . ( [ em0 ] ) we get : @xmath115 which is identical to the non - relativistic adiabatic energy conservation equation , and express the first law of thermodynamics . it could be surprising for the reader that there is not an electromagnetic term on this equation : this is due to the spherical symmetry of the problem . ( [ em1 ] ) can be written : @xmath116_{,\mu}}{a ( \delta+p / c^2)/\rho } = \frac{1}{\bigl[(\delta+p / c^2)/\rho\bigr ] c^2 } \biggl[\epsilon_{,\mu } + \nonumber\\ & & \hspace{3.3cm}p \biggl(\frac{1}{\rho } \biggr)_{,\mu } + \frac{q q_{,\mu}}{4 \pi r^4 \rho } \biggr]\ , , \end{aligned}\ ] ] or using the eq . ( [ split ] ) and the definition of the relativistic specific enthalpy ( see @xcite , @xcite ) @xmath117 : @xmath118 we can write the eq . ( [ gtt1 ] ) in a form closer to that used by may & white @xcite : @xmath119\,.\ ] ] the equation of motion can be obtained expanding the eq . ( [ e11 ] ) , using the eq . ( [ r01 ] ) to eliminate the factor @xmath120 , the eq . ( [ em1 ] ) to eliminate @xmath101 and the definitions given in the eqs . ( [ udefin ] ) , ( [ gammadef ] ) , ( [ gamma2 ] ) , ( [ enthalpy ] ) . after lengthy calculations we obtain : @xmath121\,.\end{aligned}\ ] ] this equation reduce to the equation of motion obtained by misner and sharp @xcite and may and white @xcite for the case @xmath122 , and is compatible with the equation derived by bekenstein @xcite . ( [ motion ] ) reduce to the newtonian equation of motion of a charged fluid sphere letting @xmath123 , and consequently : @xmath124 ( see eq . [ gamma2 ] ) ; @xmath125 ( see eq . [ enthalpy ] ) ; and @xmath126 ( see eq . [ em1 ] ) . the equation for charged fluid spheres in hydrostatic equilibrium is a generalization of the tolman - oppenheimer - volkov @xcite equation obtained by bekenstein @xcite . we can obtain it from the equation of motion ( [ motion ] ) , taking . in hydrostatic equilibrium , the factor @xmath127 ( see eq . [ gammadef ] ) is @xmath128 rearranging ) can be simplified using the chain rule : @xmath129 , and using the eq . ( [ gammadef ] ) . ] the eq . ( [ motion ] ) and using the eqs . ( [ enthalpy ] ) and ( [ gammahe ] ) , the tolman - oppenheimer - volkov ( tov ) equation for charged fluid spheres is : @xmath130 this equation gives the well known tolman - oppenheimer - volkov @xcite equation when @xmath122 . in the section [ relatequ ] we find an exterior solution to the problem which agrees with the reissner - nordstrm solution . in the preceding section we derived an interior solution that we want to integrate with a computer . however , we must show that the interior and the exterior solutions match smoothly . the matching conditions between the interior and the exterior solutions are obtained by the continuity of the metric and its derivatives along an arbitrary hypersurface . first of all we must establish the equality between the exterior and the interior metric at the surface of the star . this can be done transforming the interior solution given in co - moving coordinates into the schwarzschild frame . the metric tensors in the two frames are related by the tensor transformation law : @xmath131 the line element in the schwarzschild frame is given by @xmath132 with @xmath133 . using the eq . ( [ tensorlaw ] ) we obtain the non - trivial relations between the metric coefficients : [ metriceqs ] @xmath134 from eq . ( [ tensor2 ] ) and using eqs . ( [ gammadef ] ) and ( [ udefin ] ) we get : @xmath135 or equivalently , @xmath136 from eqs . ( [ tensor1 ] ) and ( [ tensor3 ] ) : @xmath137 with a little amount of algebra we can transform this equation into : @xmath138 , and using the eq . ( [ b1 ] ) : @xmath139 this last equation can be obtained independently by comparing the induced metric , from interior and exterior solutions , on the hypersurface @xmath140 : @xmath141 . the line element corresponding to the exterior solution is given by : @xmath142 from now on a plus ( minus ) subscript or superscript denotes exterior ( interior ) solutions . the line element for the interior solution is ( see eq . [ comline ] ) : @xmath143 due to the spherical symmetry : @xmath144 . the line element compatible with the induced metric on the hypersurface , from the exterior solution is : @xmath145 and the metric induced on @xmath140 by the interior solution gives : @xmath146 on @xmath140 we have @xmath147 , so we obtain : @xmath148 and we recovered the eq . ( [ a1 ] ) . so eq . ( [ a2 ] ) ( or equivalently eq . [ a1 ] ) are conditions for the continuity of the metric along @xmath140 . from now on , we will use indistinctly @xmath149 . the relation between the exterior and interior metric coefficients constitutes three equations with the four unknowns ( see eq . [ metriceqs ] ) : @xmath150 . observe that @xmath151 and @xmath152 are obtained as part of the interior solution . in addition , the solution exterior to the sphere of matter must satisfy the field equations : @xmath153 , here @xmath154 is the ricci tensor . so we obtain ( see for example @xcite pag . 180 and pag . 337 ) : @xmath155 the continuity of the metric is almost already established . only for completeness , and after a little algebra , we can find an equation for @xmath156 : @xmath157 ( although we will not use it ) . the equation of the hypersurface containing the surface of the star is @xmath158 where @xmath159 is the lagrangian mass at the surface of the star , so the metric coefficient @xmath160 is : @xmath161 with @xmath162 , @xmath163 , and @xmath164 . the coefficient @xmath165 is @xmath166 , and we recover the reissner - nordstrm solution : @xmath167 it is important to remark that we obtained the reissner - nordstrm solution by performing a tensor transformation of the interior solution . the exterior solution must satisfy the einstein field equations , and so we used them to find the eq . ( [ a3 ] ) . thus , the exterior and interior metric are consistently matched . it remains to prove the continuity of the derivatives of the metric along @xmath168 . this is a more lengthy calculation and we will give only the the main algebraic steps and the final results . the continuity of the derivatives of the metric is established from the continuity of the second fundamental form , or extrinsic curvature tensor @xmath169 of the hypersurface @xmath168 . by definition ( see @xcite , pag . 59 ) : @xmath170 here @xmath171 is a unit vector normal to @xmath168 ; @xmath172 , are basis vectors on @xmath168 ; the coordinates on @xmath168 are denoted with @xmath173 , while @xmath174 are the space - time coordinates . equivalently : @xmath175 , here @xmath176 is a lie derivative . we will choose @xmath177 as coordinates on @xmath168 . the equation of the hypersurface approaching from the exterior is : @xmath178 the basis vectors on @xmath179 are : @xmath180 in the coordinate basis @xmath181 , and the normal co - vector is : @xmath182 the equation for @xmath168 as seen from the interior is ( see eq . [ sigmas ] ) : @xmath183 the basis vectors on @xmath184 are : @xmath185 in the coordinate basis @xmath186 , and the normal co - vector is : @xmath187 if there are no surface distributions of energy - matter on @xmath168 , it must be : @xmath188 \equiv k^+_{ab}-k^-_{ab } = 0\,.\ ] ] the angular components of this equation are @xmath189 , and @xmath190 . for @xmath191 we have : @xmath192 here @xmath193 is the connection . from now on , in the present section , we are not using the sum rule over repeated indexes . in the last two steps we used the eqs . ( [ a2 ] ) and ( [ a3 ] ) . for @xmath194 , we find @xmath195 , and using the eqs . ( [ gammadef ] ) and ( [ gamma2 ] ) : @xmath196 , so eq . ( [ keq ] ) gives : @xmath197 or equivalently : @xmath198 here @xmath199 and @xmath200 , with @xmath201 a positive arbitrary small number or zero . the definition of @xmath108 assures its continuity across @xmath168 , so @xmath202 . moreover , we already found that @xmath5 and @xmath2 are constants independent of @xmath46 and @xmath63 , so the eq . ( [ gammaspm ] ) implies that @xmath203 . then , using the eq . ( [ dmdt ] ) in co - movel coordinates ( @xmath204 in both coordinate systems , see eq . [ dqdt ] and eq . [ chargeconsintime ] ) , we obtain the boundary condition at the surface : @xmath205 the equation @xmath206=0 $ ] gives an analogous result . the component @xmath207 is : @xmath208 the external component is : @xmath209 the sum rule is not used here . for clarity , we will define @xmath210 , so eq . ( [ a2 ] ) becomes @xmath211 and an over - dot means @xmath212 . applying the over - dot operator to this equation we obtain an equation that will be used later : @xmath213 so , @xmath214 expanding this equation , and using the eq . ( [ ddott ] ) to simplify the result we obtain : @xmath215 further simplifications can be made using : @xmath216 thus we obtain : @xmath217 comparing this equation with the eq . ( [ kminus ] ) we obtain an expresion for @xmath218=0 $ ] : @xmath219 now rearranging and remembering that @xmath220 and @xmath221 , using the eq . ( [ em1 ] ) to replace @xmath222 we get : @xmath223\,.\end{aligned}\ ] ] then , we see that the equation @xmath218=0 $ ] is satisfied by virtue of the equation of motion ( see eq . [ motion ] ) , and thus the exterior solution matches with the interior solution . the equations obtained above must be supplemented with an equation of state for the gas ( eos ) . in particular we choose to represent the gas of neutrons as obeying the quantum statistic of a fermion gas at zero temperature . this let us use the oppenheimer - volkov numerical results as a test - bed for the code . however , any other eos can be equally implemented in the two codes presented in this paper . the energy per unit mass is given by @xcite @xmath224 where @xmath225 is the relativity parameter , and @xmath226 is the momentum of the particles . the fermi parameter is : @xmath227 , with @xmath228 , where @xmath72 is the number density of neutrons . it can be written ( we suppress the sub - index f from now on ) : @xmath229 where @xmath230\,.\ ] ] the eq . ( [ eose ] ) can be integrated to give : @xmath231\ , , \end{aligned}\ ] ] the quantity @xmath232 is measured in [ @xmath233 , or equivalently in [ @xmath234 . this is an expression for the total energy of the gas , so it includes the rest mass energy density . although this is obvious by definition ( see eq . [ eose ] ) , it can be useful to check this expanding eq . ( [ eose2 ] ) for @xmath235 ( or @xmath236 ) . in this two cases an split similar to eq . ( [ split ] ) is obtained . the pressure is given by @xcite : @xmath237 and performing the integral : @xmath238\ , , \end{aligned}\ ] ] measured in [ @xmath234 . for charged matter , the equation of state must describe several species of particles , i.e. , neutrons , protons and electrons . however , the amount of charge in all the models studied here is very small to change the eos . for an extremal charged fluid ball there are only one unpaired ( not screened ) proton over @xmath239 particles . in this case , the change on the chemical potential of the neutrons is negligible and charge neutrality can be assumed from a microscopic point of view . ] relaxing the condition of charge neutrality , it is straightforward to show that for the case of extremely charged matter , the chemical potential of the neutrons is @xmath240 thus , the eos is slightly modified , and charge neutrality can be assumed from a microscopic point of view . we will come back to this issue in a forthcoming paper . ] the simulations were extended beyond the densities at which the eos is valid . however , we want to take into account possible charge and pressure regeneration effects , that occurs at high densities . the equations that must be integrated to obtain the stars in hydrostatic equilibrium are conveniently written in dimensionless form . from eqs . ( [ metrica0 ] ) and ( [ re1 ] ) it is possible to obtain an equation for the metric component , @xmath241 in addition we use the chain rule on the tov equation ( eq . [ toveq ] ) : @xmath242 the dimensionless equations ( [ toveq2 ] ) and ( [ eqlambda ] ) can be obtained with the replacements : @xmath243 where the bar denotes dimensionless variables , and the subscript @xmath244 indicates the dimensional constants . the full set of dimensionless equations to be integrated are : @xmath245 the dimensional constants are given by the set of equations : @xmath246 it is natural to choose : @xmath247 since @xmath248 is the factor in the expressions for the pressure ( eq . [ press ] ) and energy ( eq . [ eose ] ) . hence : @xmath249 moreover the set of equations ( [ lambdaadim])-([massadim ] ) are invariant under the transformations : @xmath250 , @xmath251 , and @xmath252 , etc ( the other constants can be obtained from these ) , with @xmath253 an arbitrary number different from zero . so , it is possible to choose any other set of constants consistent with these transformations . in particular , the dimensional constants obtained by oppenheimer and volkov @xcite , are recovered by setting : @xmath254 . in this case : @xmath255 cm ; @xmath256 g , etc , equivalent to the constants obtained in that paper @xcite . in the eq . ( [ tovadim ] ) the term @xmath257 , is obtained deriving the eq . ( [ press ] ) : @xmath258 it is used a @xmath259 order runge - kutta scheme to integrate simultaneously the set of equations ( [ lambdaadim])-([massadim ] ) . the charge distribution is chosen proportional to the rest mass distribution : @xmath260 . for simplicity , in this paper we will concentrate on the case @xmath261 . the code implemented to integrate the equations above is called he05v1 . the he05v1 is used to build neutron star models in hydrostatic equilibrium . this models constitutes an initial data set for another code presented below . in this section we collect the equations that must be integrated numerically in order to simulate the temporal evolution of the stellar models constructed with the he05v1 code . the set of equations is : [ allmotion ] @xmath262\,dt \hspace{1.56cm}[\ref{motion}]\\ % & & r_{,t}=u\,a \\ & & w=1+\frac{\epsilon}{c^2}+\frac{p}{\rho\ , c^2 } \hspace{2.96cm}[\ref{enthalpy } ] \\ & & \gamma^2=1+\frac{u^2}{c^2}-\frac{2 m g}{r c^2 } + \frac{g q^2}{c^4 r^2 } \hspace{1.335cm}[\ref{gammahe } ] \\ & & d\mu=4 \,\pi\ , \rho\ , r^2\ , dr/\gamma\hspace{2.8cm}[\ref{gammadef } ] \\ & & { a w}={a_0 w_0}\,{\rm exp}\,\biggl[\,\int_0^\mu\,\biggl ( d\epsilon + p \,d \biggl(\frac{1}{\rho } \biggr)+ \nonumber\\ & & \hspace{0.8 cm } \frac{q \,d q}{4\ , \pi\ , r^4\ , \rho } \biggr ) /w\,c^2\biggr ] \hspace{2.45cm}[\ref{gtt2}]\\ & & dm=4\ , \pi\ , \rho\ , \biggl(1+\frac{\epsilon}{c^2 } \biggr ) \,r^2\ , r_{,\mu}\ , d\mu + \nonumber \label{hamilton}\\ & & \hspace{0.8 cm } \frac{1}{c^2 } \frac{q q_{,\mu}}{r } d\mu \hspace{3.25cm}[\ref{totalmass}]\\ & & { \rho\,r^2}={\rho_0\ , r_0 ^ 2}\,\,{\rm exp}\ , \biggr[-\int_0^t a \frac{u_{,\mu}}{r_{,\mu } } \,dt\biggl ] \label{momentum } \hspace{0.57cm}[\ref{massconservation}]\\ & & d\epsilon =- p\ , d\biggl(\frac{1}{\rho}\biggr ) \hspace{3.39cm}[\ref{1stlaw}]\\ & & dq=4\ , \pi\ , \rho_{ch } \,r^2\ , dr/\gamma \label{egamma}\\ & & d{\mu}=4 \,\pi\ , \rho\,{r}^2\,dr/\gamma\,.\hspace{2.7cm}[\ref{gammadef } ] \label{emu}\end{aligned}\ ] ] where we mean @xmath263 and @xmath264 . the references to other equations are enclosed in square brackets . comparing the eqs . ( [ allmotion ] ) with the may and white equations @xcite , it can be seen that they share a very similar structure when @xmath265 . we have chosen a charge distribution proportional to the rest mass distribution so we obtain the eq . ( [ egamma ] ) . the eq . ( [ dynam ] ) is the only equation that have a second time derivative and constitutes the dynamical part of the einstein - maxwell equations . ( [ hamilton ] ) and ( [ momentum ] ) represent the hamiltonian and momentum constraint equations , respectively . the numerical code to integrate the equations above are called . the initial conditions at @xmath266 are obtained with the he05v1 for each stellar model , and consist on : the initial mass density distribution @xmath267 ; the initial electric charge density distribution @xmath268 ; the kinetic energy @xmath269 ; the cell spacing @xmath270 and the surface radius @xmath271 . here , @xmath272 is the mass coordinate of the surface . then , the output of the he05v1 is used as an initial cauchy data set in the space - time hypersurface @xmath266 , @xmath273 , @xmath274 , @xmath275 , for the . the boundary conditions are : @xmath276 the condition given in eq . ( [ bca ] ) , makes the time coordinate synchronized with an observer co - moving with the surface of the star . the conditions expressed in the eqs . ( [ bcp])-([bcr ] ) , also gives : @xmath277 the finite difference algorithm was implemented using a method similar to that developed by may and white @xcite . ( [ me1])-([emu ] ) were integrated with a leap - frog finite difference method plus a predictor - corrector step . we call this numerical code as . the integration is iterated according to a desired error control . the method is second order accurate in space and time . each experiment was repeated with different number of particles and with different values of the numerical viscosity parameter , in order to check the convergence of the results . in general , good results are obtained using @xmath278 points along the coordinate @xmath63 . the compatibility between the two codes is an indirect check of the convergence properties of the , since the he05v1 is @xmath279 order accurate . for example , at @xmath266 there exists differences in the integrated mass between the two codes of the order of @xmath280 , using @xmath281 points in the . this difference can be made @xmath282 by taking , instead , @xmath283 points , and so on . for the integration in the time dimension is needed to take initially small time - steps @xmath284 s , but the code is capable of self - adjust the time - step so as to take small changes in the physical variables . therefore , with this algorithm it is indifferent to take an initial time - step of @xmath285 s or @xmath286 s , the code always self - adjust the time - step and prove to converge always to the same solution . the shocks are treated by adding an artificial viscosity , which is non - zero only on discontinuities . its effects are to smear out the discontinuities over several cells , and to reduce the post - shock oscillations and the numerical errors related to the leap - frog method . we experienced with distinct functional forms for the viscosity term , probing the method of may and white @xcite to be the best in the present algorithm . however , there are not strong shocks ( like in ref . @xcite ) in the set of simulations studied in this paper and with the particular initial conditions chosen . in the the eqs . ( [ egamma ] ) and ( [ emu ] ) are not integrated in time , and the rest mass and charge are constrained to be constant in the layers of matter ( as it must be ) . we check that this kind of algorithm reduce the numerical errors . as was showed in the sec . ( [ relatequ ] ) the total mass - energy at the surface of the star is also a constant @xmath287 . the numerical errors have an impact on the constancy of the total mass - energy @xmath5 , and this is the physical quantity with which is checked the accuracy of the simulation . in all the runs @xmath5 is very well conserved , with the exception at the time when a trapped surface forms . in that case , very large gradients forms and serious numerical instabilities appear . although , it is possible to manage the code to keep the error in mass - energy conservation below @xmath288 at the time the apparent horizon forms . of course , once the trapped surface forms nothing can change the fate of the matter inside the apparent horizon ( see the sec . [ subsubsec : ah ] ) . the simulation must be discontinued soon after the matter cross the gravitational radius because the time - step attains very small values . in general , when @xmath289 is roughly @xmath290 s , it is needed a very large number of time - steps to obtain further progress and this consume too much computer time . the reason is that the large gradients produce huge changes in the physical variables and a very small time - step is required to keep the errors at a low level . thus , it is not possible to follow the dynamical evolution until a stationary regime is reached ( or the formation of an event horizon , if ever possible ) . when the outer trapped surface forms , some layers of matter external to it can be collapsing or expanding and is not possible in general to say which will be their fate . however , the matter that falls in the trapped surface can not escape from it and this is enough to say that the star ( or some part of it ) collapsed . there are two different sources of perturbation introduced in the models : the first source of perturbation is of numerical origin and cames from the mapping of the equilibrium models in the evolution code . this mapping is not exact , and there is a difference in the integrated rest mass and total mass between the two codes that is taken @xmath291 . this difference can be made as small as wanted by taking a higher number of points to perform the integration with the ; the second source of perturbation , is introduced by multiplying the kinetic energy by a small constant grater than one , i.e. , @xmath292 . in the present simulations we choose @xmath253 to take the values @xmath293 . the structure of the space - time of a collapsed charged star is very complex , with the possible existence of time - like singularities and tunnels connecting several disjoint asymptotically flat space - time regions . this tunnels and the true singularities , pertains to the analytically extended portions of the space - time , and the present code is not prepared to reach that regions of the total manifold . in fact , with the present code is not possible to reach the cauchy horizon where important physical effects must occur @xcite , @xcite . however , from an astrophysical point of view is interesting to know what happens outside the apparent horizon , where the astronomers live . we will not describe the numerical methods in more detail because is not the objective of this paper . moreover , the numerical methods are very well known and better explained in textbooks and papers ( see for example ref . the is an extended version of a preliminary code developed by ghezzi ( 2003 ) , earlier test - beds and results obtained with this code were published in collaboration @xcite , @xcite . .[tab : table1]mass , radius , and central density for neutron star models with zero charge . [ cols=">,>,^,^,^,^,^,^ " , ] a trapped surface is a space - like orientable two dimensional compact surface , such that inward and outward null geodesics normal to it converge . the apparent horizon is the outer boundary of all the trapped surfaces ( see @xcite , @xcite , @xcite , @xcite , @xcite , and @xcite , for definitions and related theorems ) . a black hole in asymptotically flat space - time is the region from where no causal or light signals can reach @xmath294 ( the future null infinity , see for example @xcite and @xcite for definitions ) . the event horizon @xmath295 is the boundary of the black hole region @xmath296 . the formation of an apparent horizon is a sufficient , although not necessary , condition for the formation of a black hole ( see ref . in fact , in the case an apparent horizon forms the theorems of hawking and penrose ( 1970 ) can be applied to know that the outcome will be the formation of a singularity , i.e. : the spacetime contains at least one incomplete timelike or null geodesics ( see the theorems 9.5.3 and 9.5.4 in @xcite , pags . 239 - 241 , known as `` singularity theorems '' ) . moreover , the trapped surface ( and the apparent horizon ) is contained within a black hole as a mathematical proposition asserts ( see the propositions 12.2.2 and 12.2.3 in @xcite , pags . 309 - 310 , and proposition 9.2.1 in @xcite , pag . so , we assumed that if an apparent horizon is formed in the stellar collapse , the result will be the inexorable formation of a black hole . in a numeric simulation , in general , is needed an algorithm to find out the apparent horizon s location . however , in the present study it is easy to guess that the trapped surfaces are spheres . with this in mind , we can describe the apparent horizon formation and evolution . the vector @xmath297 is tangent to the outgoing null geodesics , and @xmath298 is tangent to the ingoing null geodesics , in co - moving coordinates . they are also orthogonal to the constant @xmath299 surfaces . following mashhoon and partovi @xcite , we define the quantities @xmath300 and @xmath301 which are related by a positive multiplicative factor to the expansion factor for radially ingoing and outgoing null geodesics , respectively . thus , the coordinates of the apparent horizon are found solving the equation : @xmath302 using the eqs . ( [ gammadef ] ) and ( [ udefin ] ) , and squaring , this is equivalent to @xmath303 and using eq . ( [ gamma2 ] ) , this gives : @xmath304 the solution of this equation is : @xmath305 where @xmath306 is the inner boundary , while @xmath307 is the outer boundary of the trapped surfaces . so , the apparent horizon is the surface ( in schwarszchild coordinates ) : @xmath308 in order to know the space - time coordinates of the apparent horizon in co - moving coordinates , we must solve the eq . ( [ ahrp ] ) for @xmath46 and @xmath63 . this is done with a numeric subroutine that for each time @xmath46 and for each coordinate @xmath63 asks if @xmath309 . some of the stars when evolved in time give rise to the formation of an apparent horizon , that is the coordinate @xmath28 contracted to the value @xmath307 for some @xmath46 and @xmath63 . the table [ tab : table4 ] summarize the results of the simulations : it is shown that some of the stars collapsed , and it is given the time elapsed from the beginning of the simulation and the formation of the apparent horizon . ( [ apparenteq ] ) is also equivalent to : @xmath310 this equation can also be obtained from the equation for light rays @xmath311 . from this equation we see that outgoing light rays have zero expansion in the co - moving frame when @xmath312 . however , from the simulation was obtained that the apparent horizon forms before @xmath313 , this means that radial light rays enter the trapped region ( at its boundary @xmath314 ) with @xmath315 . in the figs . ( [ fig : v08m21]b ) and ( [ fig : v05m21]b ) is shown the evolution of the metric coefficient @xmath61 for model number @xmath316 with @xmath317 , and for model number @xmath316 with @xmath318 ( this model is one of the entries of the table [ tab : table2 ] ) . as we see , @xmath319 for this models , so they collapse . the event horizon can not be described in the simulations , since it is the result of the complete history of the collapsing matter ( while the simulation lasts a finite amount of time ) . but we know that the trapped region is contained in the black hole region . there exists only one situation in which the formation of the event horizon can occur in a finite coordinate time : when all the matter composing the star collapses in a finite time . only in the special case that the star collapse completely , i.e. : if @xmath320 for some finite time @xmath321 , the apparent horizon will coincide with the event horizon of a reissner - nordstrm space - time : @xmath322 , where @xmath323 is the reissner - nordstrm radius of a black hole of total mass - energy @xmath5 and total charge @xmath2 . in all the simulations performed the apparent horizon is formed for some coordinate @xmath63 with @xmath324 . so , in the present set of simulations , the apparent horizon never coincides with the event horizon . we picked up some stable stars and followed their temporal evolution after a sudden discharge of its electric field . the scenario is a pure hypothetical one , in which the neutron star is leaved in a charged metastable state after its formation , or during an accretion process onto it . the charged matter could suffer a discharge by recombination of charges of opposite sign , or by effect of the high electrical conductivity . we emphasize that we are not claiming that this is a possible astrophysical scenario , but we want to explore the different theoretical possibilities for the stellar dynamics . the simulation is performed with the worst conditions for the stellar stability : the electric field is simply turned off after the first time - steps of temporal evolution of an otherwise stable star . the energy is maintained constant in the process , corresponding to a conversion of the electromagnetic energy into heat or internal energy . this is easily performed with the code . the results of the simulation of the stars suddenly discharged are summarized in the table [ tab : table4 ] . the models discharged are indicated with primed numbers . we see that not all the models collapse to form black holes : the model number @xmath325 with an initial charge to mass ratio @xmath326 oscillate , while the same model with @xmath327 collapsed after a sudden discharge ; the model number @xmath328 oscillate if its initial charge to mass ratio is @xmath326 , while the same model collapses if @xmath327 . the models number @xmath316 always collapse , as they fall on the unstable branch , then it is unnecessary to consider them . in the reference @xcite it is described the formation of a shell of higher density formed near the surface of the star . although this effect must happen in newtonian physics , its evolution in the strong field regime is highly non - linear and far from obvious . the weight of the star is supported by the gas pressure and by the coulomb repulsion of the matter . if the coulomb repulsion is important respect to the gravitational attraction -although not necessarily stronger- a shell of matter can form ( see @xcite for an explanation ) . the contrast of density between the shell and its interior depends on the energy and charge distribution . so , it is important to check if the effect take place in the present simulations . we found that the shell forms only mildly , or do not form , and its late behavior can not be followed clearly . from all the numerical experiments performed , it is observed that the shell formation effect arise more clearly when the initial density profile is flat ( like the simulations of ref . @xcite ) . in consequence , the shell formation is not an important physical effect in the collapse of charged neutron stars . however , it could be important for the collapse of the core of super - massive stars @xcite . @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiimassrad0.eps } & \epsfig{width=7cm , height=6 cm , file = prdiimassrad3.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiibindingdens.eps } & \epsfig{width=7cm , height=6 cm , file = prdiimassadens3.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiivel08m21.eps } & \epsfig{width=7cm , height=6 cm , file = prdiigtt08m21.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiivel05m21b.eps } & \epsfig{width=7cm , height=6 cm , file = prdiigtt05m21.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiigamma08m21.eps } & \epsfig{width=7cm , height=6 cm , file = prdiigamma05m21b.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] @xmath329 \epsfig{width=7cm , height=6 cm , file = prdiivel05m13.eps } & \epsfig{width=7cm , height=6 cm , file = prdiigtt05m13.eps } \\ [ 0.4 cm ] \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] it is usually assumed in astrophysics that stars havent important internal electric fields . whether an star can have large internal electric fields or a net total charge is not yet clear . however , several features could be common to the evolution of rotating collapsing stars , with the angular momentum playing the role of the electric charge . so , the present study can shed some light on more realistic astrophysical scenarios . in the models studied in this paper we assumed , for simplicity , that the charge density is proportional to the rest mass density . we found that in hydrostatic equilibrium the charged stars have a larger mass and radius than the uncharged ones . this is , as expected , due to the coulomb repulsion . the mass of the models tends to infinity as the charge approaches the extremal value @xmath330 . the hydrostatic equilibrium solutions with @xmath331 gives models with an infinite mass and radius ( see also @xcite ) . this means that in this particular cases the integrated mass and radius diverges within the computer capacity . all the models with charge less than the extremal ( @xmath332 ) , have a mass limit and there are unstable and stable solutions . we checked the stability of the solutions integrating forward in time the models , and applying strong perturbations to them . some of the models collapse directly to form black holes , without ejecting matter . other models oscillate . for a given model , with fixed central density , the frequency of oscillation is lower when the charge is higher . the frequency of oscillation is weakly dependent on the numerical viscosity . the models that collapse are solutions with @xmath333 , while the oscillating models have @xmath334 . so , the stability of the models agree with the predictions of the first order perturbation theory @xcite . it seems that there is a limit for the charge that a star can have , which is lower than the extremal case . this limit arise because the binding energy per nucleon of the models with @xmath335 , is lower than the binding energy per nucleon on an atomic nucleus . then , it is possible that this stars disintegrate to reach a more bounded energy state . this point deserves further study . from a pure theoretical point of view , the issue of the collapse of charged fluid spheres , is related to the third law of black hole thermodynamics @xcite , and with the cosmic censorship hypothesis @xcite , @xcite . the third law of black hole physics states that the temperature of a black hole can not be reduced to zero by a finite number of operations . the impossibility of transforming a black hole into an extremal one , in a finite number of steps , is related to the impossibility of getting @xmath330 in some particular experiment @xcite , @xcite . in agreement with this law , we found that it is not possible to form extremal black holes from the collapse of a charged fluid ball . in fact , any charged ball with a charge to mass ratio greater or equal than one explodes , or its matter spreads . the black holes with @xmath331 represent naked singularities , and the impossibility of getting black holes with this charge to mass ratio in the present simulations are in agreement with the cosmic censorship hypothesis " . it must be remarked that once an apparent horizon forms the formation of an event horizon is inevitable , and as we proved the matching of the interior solution with an exterior reissner - nordstrm solution , we simulated here the formation of a reissner - nordstrm space - time from a gravitational stellar collapse . although we are not using a realistic equation of state we think that the results are of general validity , at least from a qualitative point of view . other fields or more exotic physics must be considered in order to form extremal black holes . for completeness we reproduce here the einstein equations in co - moving coordinates as derived by landau and lifshitz ( see @xcite , pag . 311 ) . the equations of section [ sec2 ] were derived from these by making the replacements @xmath336 , @xmath337 , and @xmath338 . the einstein equations are : @xmath339 the author thank professor dr . jacob d. for clarifications about his equations . this work was performed with the computational resources of professor dr . patricio s. , so the author thank him . the author thank an anonymous referee for useful comments that help to improve this paper . the author acknowledge the brazilian agency fapesp for support .

the isoperimetric problem is a classical subject in differential geometry with its origin in ancient greece . it consists in finding on a riemannian manifold @xmath2 the regions that minimize the perimeter among sets enclosing the same volume . the solutions are called isoperimetric regions and their boundaries isoperimetric hypersurfaces . the euclidean plane is historically the first space where the problem started to be investigated rigorously . it is now a well known fact that the round circles are the optimal curves for the problem . this geometric fact is often seen through the following classical inequality : @xmath3 where @xmath4 and @xmath5 stand for the length and the enclosed area of a simple closed curve @xmath6 respectively . the framework of geometric measure theory and its tools work successfully well in tackling the aspects of existence and regularity of this variational problem . when @xmath7 is closed or homogeneous , then isoperimetric hypersurfaces do exist and are smooth up to a closed set of hausdorff dimension @xmath8 . the regular part is a stable hypersurface of constant mean curvature . this major contribution to the isoperimetric problem was achieved thanks to the efforts of many people , including f. almgren , r. schoen , l. simon , f. morgan , and others ( see @xcite for a comprehensive list ) . despite the long history of the problem , it remains largely open with few 3-manifolds where the problem is completely understood . the simply connected space forms , @xmath9 , @xmath10 and @xmath11 , are the most appealing spaces to begin the study of the isoperimetric problem . it turns out that their symmetries are enough to characterize the geodesic spheres as the isoperimetric hypersurfaces . a complete solution on @xmath12 with the standard product metric can be found in @xcite . for other homogeneous manifolds with certain product structure , such as @xmath13 , @xmath14 , and @xmath15 , see @xcite , @xcite , and @xcite respectively . the case @xmath16 is also treated in @xcite and they show that when @xmath17 unduloids are minimizers rather than cylinders for certain volumes . one key idea exploited in the results listed above is the use of symmetry to reduce the problem to an ode analysis . the case @xmath18 where @xmath19 is a flat torus is not solved in full ; great progress can be found in @xcite and in @xcite . more generally , it is known that boundaries of small isoperimetric regions in closed manifolds are nearly round spheres , see @xcite . to finish this brief and not exhaustive account of results on the isoperimetric problem we mention that bray and morgan ( @xcite and @xcite ) classified the horizon homologous isoperimetric surfaces in the schwarzschild manifold . the work in @xcite highlighted the interesting relationship between isoperimetric surfaces and the concept of mass in general relativity . we will be interested in spherical space forms in this paper . a significant result in this direction was given in @xcite where a. ros and m. ritor solved the isoperimetric problem in the projective space @xmath20 . they show that the solutions are geodesic spheres or flat tori . later , a. ros @xcite used the above result to give a proof of the willmore conjecture in @xmath21 for the special case of surfaces that are invariant by the antipodal map . the real projective space is a special case of an important family of riemannian manifolds , namely the lens spaces @xmath22 . these are elliptic space forms obtained as a quotient of @xmath21 by a finite group of isometries that are isomorphic to @xmath23 but which also depend on @xmath24 . they are , along with @xmath25 , characterized by having heegaard genus one . we give a complete solution for the isoperimetric problem in the lens spaces with large fundamental group : [ theorem 1 ] _ there exists a positive integer @xmath26 such that for every @xmath27 and every @xmath28 the isoperimetric surfaces in @xmath22 are either geodesic spheres or tori of revolution about geodesics . _ the isoperimetric problem in dimension three was previously solved for only a finite number of non - diffeomorphic 3-manifolds . we also add to the literature the observation that the proof of the willmore conjecture by marques and neves @xcite can be applied to extend the work of ros and ritor @xcite , on the classification of stable cmc surfaces in @xmath20 , to @xmath0 and @xmath1 : and @xmath1 are diffeomorphic but not isometric . ] [ theorem 2 ] _ the immersed stable cmc surfaces in @xmath0 and @xmath1 are either geodesic spheres or flat tori . moreover , the minimal clifford torus is , up to ambient isometries , the only index one minimal surface in @xmath0 and @xmath1 . _ the idea of the proof of theorem [ theorem 1 ] is as follows . stability implies that every isoperimetric surface is connected and its genus is @xmath29 or @xmath30 . if follows from a classical result of hopf that if the genus is 0 , then it is a geodesic sphere . from @xcite we know that if the genus is 1 , then it is flat , and this forces the surface to be of revolution about a geodesic . we are left to rule out other topological types . to do so we argue by contradiction . we assume that there exists a sequence of lens spaces with increasing fundamental group containing isoperimetric surfaces of genus @xmath31 . after a suitable rescaling on the metrics we use compactness results to obtain a limit for the sequence of lens spaces which will be a flat three manifold of rank one . in the same way , the sequence of isoperimetric surfaces will converge to a flat surface in the respective ambient manifold . the topology of the surfaces will force the limit to be an union of planes . on the other hand , the minimization property rules this configuration out . the arguments in the proof of theorem [ theorem 1 ] generalize naturally for the berger spheres @xmath32 . this is a well known one parameter family of homogeneous metrics on the 3-sphere ; the case @xmath33 corresponds to the round metric . all the hopf fibers in @xmath32 have the same length @xmath34 . when @xmath35 is not too small , spheres are the only solutions of the isoperimetric problem , see @xcite . when @xmath35 is very small , it is pointed out in @xcite that some tori are better candidates to solve the isoperimetric problem rather than spheres for certain volumes . _ there exists @xmath36 such that for every @xmath37 the isoperimetric surfaces in the berger spheres @xmath32 are either rotationally invariant spheres or tori . _ it follows from the work in @xcite and @xcite that the minimal clifford torus in @xmath38 ( resp . @xmath32 ) is isoperimetric for every @xmath39 ( resp . @xmath40 ) , see section 3 . _ acknowledgements . _ i am grateful to my advisor andr neves for his constant encouragement and support during the course of this work . i also thank jason lotay for many helpful comments on an earlier version of this manuscript . let @xmath41 be an orientable riemannian manifold of dimension @xmath42 . the @xmath42-dimensional hausdorff measure of a region @xmath43 is denoted by @xmath44 . similarly , we denote the @xmath45-dimensional hausdorff measure of the hypersurface @xmath46 by @xmath47 . the class of regions considered here are those of finite perimeter , see @xcite . a region @xmath43 is called an _ isoperimetric region _ if @xmath48 in this case , the hypersurface @xmath49 is called an _ isoperimetric hypersurface_. in general , the existence of isoperimetric surfaces is handled by a compactness theorem from geometric measure theory . for non - compact manifolds one needs to be careful since a minimizing sequence of regions of fixed volume may drift off to infinity . we recommend @xcite for a recent reference on the regularity of isoperimetric hypersurfaces : _ let @xmath41 be a closed riemannian manifold . for every @xmath50 there exists an isoperimetric region @xmath51 satisfying @xmath52 . moreover , @xmath49 is smooth up to a closed set of hausdorff dimension @xmath8 . _ isoperimetric hypersurfaces are stable critical points of the area functional for variations that preserve the enclosed volume ; thus , the regular part of isoperimetric hypersurfaces has constant mean curvature . more generally , we say an isometric immersion @xmath53 has _ constant mean curvature _ ( cmc ) if it is a critical point of the area functional for volume preserving variations . critical points are called _ stable cmc _ if the second derivative of the area is non - negative for such variations . equivalently , @xmath54 is stable if for every @xmath55 with compact support such that @xmath56 , we have @xmath57 @xmath58 is the unit normal vector of @xmath59 and @xmath5 is the _ second fundamental form _ of the immersion @xmath54 mean curvature _ of @xmath59 , denoted by @xmath60 , is defined by @xmath61 . the study of immersed stable cmc hypersurfaces started in @xcite and @xcite with a new characterization of the geodesic spheres in the simply connected space forms @xmath10 , @xmath9 and @xmath11 . the classification of stable cmc surfaces is often a way to approach the isoperimetric problem in reasonable spaces . with this purpose in mind , a. ros and m. ritor in @xcite used the hersch - yau trick to study orientable stable cmc surfaces on three manifolds of positive ricci curvature . [ up bound w energy , g=2 ] _ let @xmath62 be a closed three manifold with positive ricci curvature . if @xmath63 is a stable cmc immersion , then @xmath64 . moreover , if @xmath65 or @xmath30 , then @xmath66 _ [ wilmore energy genus 2 ] the first statement follows from the previous theorem since @xmath67 . let @xmath68 be the induced map in fundamental groups . as @xmath69 has finite index there exists a finite covering @xmath70 such that @xmath71 and @xmath72 . this means there exists a lifting of this map into @xmath21 and we denote it by @xmath73 . it follows that @xmath74 . if @xmath54 is not an embedding , then @xmath75 is not embedding either . by the work of li and yau @xcite the willmore energy of the immersed surface @xmath76 , i.e. @xmath77 , is strictly greater is discussed in @xcite . ] than @xmath78 . therefore , if @xmath79 , we obtain a contradiction and @xmath75 is an embedding . moreover , for closed surfaces with genus greater than or equal to @xmath80 in @xmath21 the willmore conjecture , recently proved in @xcite , states that @xmath81 . let @xmath82 be the pre - image of @xmath59 by the universal covering map , then @xmath83 therefore , if @xmath84 , then @xmath85 and @xmath86 . in particular , if @xmath87 or @xmath30 , then there exist no stable cmc surface @xmath59 with @xmath88 . for each @xmath89 we define the clifford torus @xmath90 as : @xmath91 every flat tori with constant mean curvature in @xmath21 is congruent to a clifford torus @xmath90 . this follows from the rigidity theorem , pg . 49 in @xcite , for they have the same second fundamental form . _ the stable cmc surfaces in @xmath0 and @xmath1 are totally umbilical spheres or flat tori . in addition , the index one minimal surfaces in @xmath0 and @xmath1 are congruent to the the projection of minimal clifford torus . _ let @xmath92 , @xmath93 , be in the conditions of the corollary . by proposition [ wilmore energy genus 2 ] , @xmath94 or @xmath80 . if @xmath94 , then it follows from the hopf holomorphic quadratic differential that @xmath59 is totally umbilical . if @xmath95 , then it is proved in @xcite that @xmath59 is flat . the isoperimetric properties of @xmath2 can be encapsulated in a single function called the _ isoperimetric profile_. this is the function @xmath96 \rightarrow [ 0,+ \infty)$ ] defined by @xmath97 we finish the section with some well known facts on the analytic nature of @xmath98 . these will be used later in section 3 . let @xmath51 be an isoperimetric region in @xmath2 such that @xmath99 for some @xmath100 . the function @xmath98 has left and right derivative @xmath101 and @xmath102 . in addition , if @xmath60 is the mean curvature of @xmath49 in the direction of the inward unit vector , then @xmath103 the second derivative also exists but weakly in the sense of comparison functions . more precisely , we say @xmath104 weakly at @xmath105 if there exists a smooth function @xmath106 such that @xmath107 , @xmath108 , and @xmath109 . in this sense we have @xmath110 the equations ( [ first derivative ] ) and ( [ second derivative ] ) are first presented on @xcite . we sketch the proof of ( [ first derivative ] ) and ( [ second derivative ] ) . let @xmath111 be the variation @xmath112 of @xmath59 reparametrized in terms of the enclosed volume @xmath113 . in addition , let @xmath114 ( resp . @xmath115 ) be the area of @xmath116 ( resp . @xmath117 ) . by the first variation formula for the area and volume we have @xmath118 and @xmath119 respectively . since @xmath120 , we conclude that @xmath121 and also that @xmath122 . on the other hand , the second derivative of area for general variations implies the following : @xmath123 hence , in the sense of comparison functions , ( [ second derivative ] ) follows from : @xmath124 in order to define the lens spaces , we first recall the round three sphere as : @xmath125 fix @xmath126 integers with the following property @xmath127 and @xmath128 . let @xmath23 be the group @xmath129 acting on @xmath21 as follows : @xmath130 the group @xmath23 acts freely and properly discontinuous on @xmath21 . the orbit space @xmath131 is a closed three manifold called the _ lens space _ , it is denoted by @xmath22 . the hopf fibration , which is the riemannian submersion @xmath132 defined by @xmath133 , can be extended naturally to @xmath22 . indeed , the group @xmath23 acts on the set of _ hopf fibers _ through the following cyclic action on @xmath134 : @xmath135 the hopf fibration for @xmath22 is then defined as @xmath136 . the set @xmath137 is a two dimensional orbifold with conical singularities at the north and south pole . the preimage of horizontal slices of @xmath137 via @xmath138 correspond to the clifford torus described in ( [ clifford torus ] ) . they are natural candidates to solve the isoperimetric problem in @xmath22 . steiner and schwarz symmetrization theorems were proved in @xcite for certain fiber bundles such as the lens spaces . to explain this symmetrization procedure we restrict to the case @xmath38 where the hopf fibration @xmath139 is a smooth riemannian submersion . the symmetrization consist of associating to each set of finite perimeter @xmath140 the set @xmath141 in the product manifold @xmath142 defined by replacing the slice of @xmath143 in each fiber with a ball of the same volume about the respective fiber in the product . the coarea formula for riemannian submersions implies that @xmath141 encloses the same amount of volume as @xmath143 . it is proved in @xcite that @xmath141 has no greater perimeter than @xmath143 . one immediate consequence is that @xmath144 . applying the classification of the isoperimetric problem on @xmath142 , @xcite , we conclude that @xmath145 in a interval around @xmath146 . in particular , the minimal clifford torus is isoperimetric in @xmath38 for every @xmath39 . the isoperimetric profiles , however , do not coincide as the profile of geodesic spheres on the respective spaces are different . therefore , this technique is not enough to completely solve the isoperimetric problem . d_{\mathbb{s}^3}[(\theta^{kq } z , \theta^k w),(z , w)]&=&2\arcsin(\frac{a}{2})\geq 2\arcsin\big({\frac{2}{2}\sin{\frac{2\pi}{2p}}}\big)=\frac{2\pi}{p}.\end{aligned}\ ] ] however , is not true in general that @xmath153 as @xmath154 . let s consider @xmath155 , @xmath156 . we show that the injectivity radius at points far away from the critical fibres are @xmath157 . if the round metric is scaled by the factor @xmath158 , then we have the riemannian submersion : @xmath159 the fibers have constant length @xmath160 except the critical fibres which have length @xmath161 . the right hand side will converge , as @xmath162 , to @xmath163 . it follows from the coarea formula for riemannian submersions that the volume of the geodesic ball @xmath164 in @xmath165 is bounded from below . therefore , by cheeger s inequality , lemma 51 in @xcite , the injectivity radius of the sequence @xmath165 is bounded from below . this sequence converges to a flat @xmath166 . let @xmath173 and recall that @xmath174 . without loss of generality , let s assume that @xmath175 . hence , under the rescaled metric @xmath176 , the extrinsic diameter of @xmath170 is greater than or equal to @xmath177 . let @xmath178 be a geodesic segment realizing the intrinsic diameter of @xmath170 . thus , we can find disjoint balls @xmath179 , with @xmath180 , @xmath181 , and @xmath182 + 1 $ ] . hence , @xmath183 + 1}\mathcal{h}^2(b_r(x_{i_0})\cap t_{\frac{\pi}{4}}/\mathbb{z}_p ) \leq |t_{\frac{\pi}{4}}/\mathbb{z}_p|= \lambda_p^2\,\frac{2\pi^2}{p}.\ ] ] therefore , there exists @xmath184 + 1\}$ ] such that : @xmath185 the first inequality follows from the monotonicity formula , proposition [ monotonicity ] in the appendix , applied to @xmath186 . this finishes the proof of the lemma . let s use the notation @xmath187 to represent the group @xmath23 acting on @xmath21 and its dependence on the parameter @xmath24 . by using the toroidal coordinate system for @xmath21 , @xmath188 the action of @xmath189 on @xmath90 corresponds to the following action on @xmath190 : @xmath191 in these coordinates , the @xmath189 orbit at the point @xmath192 is given by : @xmath193 to prove the lemma it is enough to consider @xmath194 . if there is a subsequence for which the diameter of @xmath195 is going to zero as @xmath154 , then @xmath196 is clearly becoming dense on @xmath197 and item 1 is proved . let s consider now the case where the diameter of @xmath197 in @xmath22 is bounded away from zero . from the equivalence between extrinsic and intrinsic distance and by lemma [ injective radius ] we conclude that the euclidean injectivity radius satisfies @xmath198 . in particular , there exist @xmath199 such that @xmath200 therefore , there exists @xmath201 such that @xmath202 infinitely often and @xmath203 is the euclidean injectivity radius of @xmath170 at @xmath204 for this subsequence . hence , the sub - orbit generated by the translation @xmath205 is contained in the line @xmath206 . it follows that the @xmath207 is contained in a union of equidistant lines parallel to the one described above by homogeneity . modulo @xmath208 the number of such lines is finite , let s denote it by @xmath209 . modulo @xmath208 there are @xmath210 points of @xmath207 in each of these lines . hence , @xmath211 therefore , @xmath209 divides @xmath212 and is independent of @xmath213 . in other words , @xmath196 is contained in @xmath214 integral curves of @xmath215 . [ main theorem ] _ there exists a positive integer @xmath26 such that for every @xmath27 and every @xmath28 the isoperimetric surfaces in @xmath22 are either geodesic spheres or tori of revolution about geodesics . _ [ proof of theorem 1.1 ] arguing by contradiction , let s assume that there exists an infinite sequence of lens spaces @xmath22 containing an isoperimetric surface @xmath216 of genus @xmath217 or @xmath30 for each @xmath213 . 1 . there exists an exhaustion of @xmath2 by compacts @xmath220 , i.e. @xmath221 and @xmath222 . 2 . there exists a family of diffeomorphism onto their images , @xmath223 , such that @xmath224 and @xmath225 in the @xmath226 topology . we consider pointed manifolds @xmath227 with base points @xmath228 belonging to @xmath229 . by the cheeger - gromov compactness theorem we have that @xmath230 , where @xmath231 is a flat three manifold . the inclusion of @xmath216 into @xmath2 through the diffeomorphism @xmath232 is still denoted by @xmath216 . let @xmath235 be such that @xmath236 and define @xmath237 . arguing by contradiction , let s assume that @xmath238 . in local coordinates around @xmath239 we consider the surface @xmath240 on the euclidean ball @xmath241 endowed with the rescaled metric @xmath242 . therefore , @xmath243 converges to @xmath244 as @xmath154 . the surface @xmath245 now has the property that @xmath246 . by the strong compactness for a sequence of isoperimetric surfaces with bounded second fundamental form , see corollary [ compactness iso ] in the appendix , there exists a subsequence converging to a properly embedded surface @xmath247 , the convergence is in the sense of graphs and with multiplicity one . moreover , @xmath248 is also stable , i.e. : @xmath249 if @xmath248 is compact , then it has to be a round sphere by alexandrov s theorem , which is a contradiction since strong convergence preserves topology . if @xmath248 is non - compact , then it has infinite area by the monotonicity formula : indeed , by proposition [ monotonicity ] in the appendix there exists a positive constant @xmath250 such that @xmath251 in particular , @xmath252 . as @xmath248 is properly embedded , it has infinite extrinsic diameter and the claim follows . therefore , @xmath248 is totally geodesic by da silveira s theorem [ silveira ] in the appendix , which is a contradiction since @xmath253 . below we denote @xmath254 by @xmath90 . let @xmath255 be the clifford torus through @xmath228 enclosing a region @xmath256 . under the scaling by @xmath257 we have that @xmath258 . if @xmath259 , then @xmath260 . moreover , the second fundamental form @xmath261 of @xmath255 satisfies @xmath262 . the critical fiber @xmath263 is distant from @xmath228 by @xmath264 since @xmath256 is converging to a compact region in @xmath2 . instead of using base points @xmath228 we choose new base points @xmath265 ; it follows that @xmath266 and @xmath267 . we claim that rank of @xmath58 is at most one : @xmath268 let s assume now that @xmath269 , consequently @xmath270 and @xmath271 . recall the function @xmath272 , the distance from the clifford torus through @xmath149 to the critical fiber @xmath273 with respect to the round metric . the unit vector field @xmath274 is orthogonal to @xmath90 for every @xmath275 and it is well defined on @xmath276 . let @xmath277 be the geodesic whose velocity is @xmath274 and such that @xmath278 . consider @xmath279 . by the triangle inequality @xmath280 under the metric @xmath281 . applying the coarea formula for @xmath282 , @xmath283 , we obtain : @xmath284 where @xmath285 $ ] is from the mean value theorem for integrals . the last inequality is justified as follows . either the extrinsic diameter of @xmath286 is going to infinity and @xmath286 is converging with multiplicity to a flat surface or the extrinsic diameter of @xmath286 is bounded . the former implies that @xmath287 . the latter implies that @xmath288 , which is a contradiction since @xmath289 . we have conclude that @xmath290 and rank of @xmath2 is at most one . [ isoperimetric small volume ] _ for @xmath213 large enough there exist @xmath292 and @xmath293 such that @xmath291 is given by the profile of spheres on @xmath294 $ ] and by the profile of flat tori on @xmath295 . moreover , if @xmath216 is an isoperimetric surface such that @xmath296 , then @xmath297 . _ for each @xmath213 we consider the first volume , @xmath292 , for which there is transition on topology of isoperimetric surfaces from spheres to something else . if @xmath298 is the volume for which the profile of geodesic spheres intersect the profile of flat tori , then @xmath299 . the value of @xmath298 is computed by solving the following system of equations : @xmath300 the left hand sides ( right hand sides ) correspond to the enclosed volume and area of the clifford torus @xmath90 ( geodesic spheres @xmath301 of radius @xmath302 ) , respectively . it follows that @xmath303 ; another way to see this is by recalling that the injectivity radius of @xmath38 is @xmath304 at every point . therefore , @xmath305 . let @xmath216 be an isoperimetric surface with the property that @xmath296 . by lemma [ bound second fundmental form lemma ] the sequence @xmath306 has bounded second fundamental form in @xmath227 ; thus , it strongly converges to a properly embedded surface @xmath59 of finite area in some orientable flat three manifold @xmath231 of rank at most one by lemma [ rank one ] . by the monotonicity formula , proposition [ monotonicity ] in the appendix , @xmath59 is a closed surface . it follows that the pre - image @xmath307 of @xmath59 in @xmath308 is contained in a solid cylinder . hence , @xmath307 is an union of round spheres by alexandrov s theorem or is a surface of revolution about the axis of the cylinder by theorem [ kks ] in the appendix . therefore , @xmath94 or @xmath80 . from the strong compactness for isoperimetric surfaces , see corollary [ compactness iso ] , we have that @xmath297 , @xmath309 and the existence of the desired @xmath293 . by the strong compactness for isoperimetric surfaces , corollary [ compactness iso ] , there exists @xmath310 such that if @xmath59 is an isoperimetric surface enclosing volume @xmath311 $ ] , then @xmath59 is a flat torus . it follows from lemma [ isoperimetric small volume ] and for large @xmath213 that the isoperimetric profile @xmath291 is given by the area of geodesic spheres for volumes in @xmath294 $ ] and by the area of flat tori for volumes in @xmath312\cup [ \widehat{v},\frac{\pi^2}{p}]$ ] . in other words , if @xmath313 is the function defined by @xmath314 , where @xmath315 is the clifford torus enclosing a volume equal to @xmath316 , then @xmath317 on @xmath312\cup [ \widehat{v},\frac{\pi^2}{p}]$ ] . it follows that @xmath318 has a local maximum point at @xmath319 $ ] . the claim will follow by exploring the weak differential equation for @xmath291 . from ( [ second derivative area ] ) we have @xmath320 let @xmath59 be an isoperimetric surface such that @xmath321 . if @xmath117 is the unit normal variation of @xmath59 parametrized by the enclosed volume , then we define @xmath322 . since @xmath323 , we have that @xmath324 has also a local maximum point at @xmath325 . hence , @xmath326 , i.e. , @xmath327 , and @xmath328 . applying equation ( [ second derivative area ] ) for @xmath138 together with the gauss equation and the gauss - bonnet theorem we obtain : @xmath329 therefore , @xmath95 and @xmath330 in @xmath331 $ ] . we claim that the injectivity radius of @xmath22 at every point is @xmath264 . indeed , let @xmath207 be the orbit of @xmath334 with respect to @xmath189 . as before , @xmath335 points of @xmath207 lie on the curve @xmath336 . here , @xmath337 is the one parameter family of diffeomorphisms associated to @xmath338 : @xmath339 when ordered according the orientation of @xmath340 , those points determine a piecewise closed geodesic @xmath178 with @xmath341 . as @xmath154 , @xmath178 converges to @xmath340 . the claim now follows from : @xmath342 by the cheeger - gromov compactness theorem we obtain , up to subsequence , the following convergence : @xmath343 where @xmath344 is a flat three manifold of rank one by lemma [ rank one ] . by ( [ injective radius ] ) we conclude that @xmath345 . the curves @xmath346 represent integral curves of @xmath338 through @xmath148 ; they have bounded geodesic curvature and @xmath347 . hence , the integral curves of @xmath338 converge to closed geodesics in @xmath2 under ( [ convergence ] ) . as sets , they coincide with the standard fibers of @xmath348 . by the poincare - hopf index theorem there exists a zero for the vector field @xmath349 since @xmath350 . hence , we can choose the base points @xmath228 to satisfy @xmath351 , here @xmath58 is the unit normal vector of @xmath216 . by lemma [ bound second fundmental form lemma ] the sequence of isoperimetric surfaces @xmath356 has uniformly bounded second fundamental form . applying the strong compactness theorem for isoperimetric surfaces , corollary [ compactness iso ] , we conclude that @xmath216 converges smoothly and with multiplicity one to a properly embedded stable cmc surface @xmath357 . if @xmath358 , then the monotonicity formula , proposition [ monotonicity ] , implies that the extrinsic diameter of @xmath216 and @xmath354 are bounded . this is impossible since the sequence @xmath216 separates @xmath22 in two regions of the same volume that goes to infinity as @xmath359 . therefore , @xmath354 is a complete properly embedded stable cmc surface in @xmath355 with infinite area . applying da silveira s theorem [ silveira ] once more , we obtain that @xmath354 is totally geodesic . as @xmath351 , we conclude that @xmath354 is orthogonal to the standard fibers of @xmath355 . we claim that @xmath354 separates @xmath355 . if it does not separate , then there exists a loop @xmath360 intersecting @xmath354 at a single point . as @xmath361 with multiplicity one , the same conclusion holds for @xmath216 , which contradicts the fact that @xmath216 separates @xmath22 . therefore , there is @xmath362 such that @xmath363 , where @xmath364 is a flat plane for each @xmath365 . we regard @xmath355 as a slab in @xmath308 with height @xmath160 . now , we construct a deformation of @xmath354 which decreases its area as follows . first , we cut off the @xmath366 solid cylinders obtained from the intersection of @xmath367 with a vertical solid torus of radius @xmath143 . to balance the enclosed volume we add a vertical solid torus of radius @xmath275 , see figure 1 for the case @xmath368 . if @xmath369 is the distance between @xmath370 and @xmath371 , then the radius @xmath275 is given by : @xmath372 the boundary of this new region is denoted by @xmath373 and @xmath374 if @xmath143 is large enough , then @xmath373 has less area than @xmath354 . this is impossible since the strong multiplicity one convergence allow us to carry out this deformation of @xmath354 to @xmath216 which contradicts the fact that @xmath216 is an isoperimetric surface . it is proved in @xcite that if @xmath49 has positive mean curvature , then @xmath377 applying this formula to the sequence of isoperimetric surfaces , we conclude that the mean curvature of @xmath216 satisfy @xmath378 . we have used that @xmath216 separates @xmath22 into regions of the same volume . as @xmath379 has area and mean curvature bounded , we apply _ allard s compactness theorem _ , theorem 42.7 and remark 42.8 in @xcite , to obtain an integral varifold @xmath380 that , up to subsequence , is @xmath381 . recall that @xmath382 acts on @xmath21 via @xmath383 . we claim that @xmath384 is @xmath382 invariant : indeed , if @xmath385 and @xmath386 , then for each @xmath213 there is @xmath387 such that @xmath388 . on the other hand , as @xmath389 , there is @xmath390 such that @xmath391 and @xmath392 . it follows that @xmath393 and @xmath394 . in particular , @xmath395 where @xmath396 is a clifford torus . the monotonicity formula implies that the convergence @xmath397 is also in hausdorff distance ; hence , we have that @xmath398 and @xmath399 . since @xmath245 is @xmath332 invariant there exists @xmath400 for which @xmath401 . by taking the limit we obtain @xmath402 for some @xmath403 . as this is impossible , we conclude that @xmath368 and all @xmath245 are connected for @xmath213 large . now we consider @xmath404 , the space of 2-dimensional integral _ currents _ on @xmath21 . each @xmath405 and @xmath406 . as @xmath407 is a region of finite perimeter ( @xmath408 is bv function with uniform bounded variation ) , then @xmath409 and @xmath410 as currents , @xmath411 is an open set of finite perimeter , see theorem 6.3 and proof of theorem 37.2 in @xcite . since @xmath412 we conclude that @xmath413 . applying the _ constancy theorem _ , theorem 26.27 in @xcite , we conclude that @xmath411 is the handlebody bounded by the clifford torus @xmath414 , and consequently @xmath415 . we proved that @xmath416 for some positive integer @xmath417 . since @xmath216 is isoperimetric , it follows that @xmath418 . indeed , @xmath419 as @xmath197 is smooth , we have for @xmath420 sufficiently small that the density @xmath421 , where @xmath422 is from theorem [ allard regularity ] in the appendix . on the other hand , as @xmath245 is converging to @xmath197 with multiplicity one , then @xmath423 for @xmath213 large enough . now we invoke the smooth version of _ allard s regularity theorem _ , theorem [ allard regularity ] , to concluded that the convergence @xmath424 is strong , i.e. , graphical with multiplicity one . as strong convergence preserves topology , we conclude that @xmath425 . this completes the proof of theorem [ theorem 1 ] . the _ berger metrics _ are riemannian metrics @xmath430 on @xmath21 defined as : @xmath431 the riemannian manifolds @xmath432 are called the _ berger spheres _ , they are denoted by @xmath32 . geometrically , the metric @xmath430 shrinks the hopf fibers to have length @xmath34 . the berger metrics are also homogeneous and their group of isometries has dimension four . it follows from the work of abresch and rosemberg @xcite that every constant mean curvature surface in @xmath32 admits a holomorphic quadratic differential . in particular , every cmc sphere in @xmath32 is rotationally invariant . a precise study of closed orientable surfaces with constant mean curvature on the berger spheres is given in @xcite . it is proved there the existence of @xmath433 with the following property : if @xmath434 $ ] , then every stable constant mean curvature surface in @xmath32 has genus zero or one . moreover , if @xmath435 $ ] , then these stable cmc surfaces are totally umbilical spheres or the minimal clifford torus , the latter only occurring when @xmath436 . in particular , rotationally invariant spheres are the only solutions of the isoperimetric problem in @xmath32 for @xmath435 $ ] . we rescale the metric @xmath430 of @xmath32 by the factor @xmath441 . the hopf fibers have constant length equal to @xmath160 under the new metric @xmath442 . it follows that the injective radius of @xmath32 at a point @xmath213 is equal to @xmath443 for every @xmath444 . since @xmath138 is a local trivial fibration , we have that for each @xmath445 there exist a neighbourhood @xmath446 of @xmath447 and a diffeomorphism @xmath448 such that @xmath449 , where @xmath450 given by @xmath451 . moreover , @xmath452 in the @xmath226 topology . therefore , in the sense of cheeger - gromov we have : @xmath453 we pick the points @xmath454 with the property that @xmath455 , this means @xmath427 and @xmath456 are parallel at @xmath457 . these points exist by the poincar - hopf index theorem . by lemma [ bound second fundmental form lemma ] the inclusion of @xmath439 in @xmath458 has the following property : @xmath459 by the strong compactness theorem for isoperimetric surfaces , corollary [ compactness iso ] in the appendix , we can extract a subsequence , @xmath460 , which converges with multiplicity one to a properly embedded surface @xmath461 if @xmath462 , then the monotonicity formula , proposition [ monotonicity ] , implies that @xmath354 is compact . we apply theorem [ kks ] to conclude that @xmath354 is either a round sphere or torus . this is impossible since we have strong convergence and @xmath463 . therefore , @xmath354 is a complete non - compact surface with infinite area . moreover , @xmath354 is also a stable cmc surface in @xmath355 : @xmath464 it follows from theorem [ silveira ] that @xmath354 is totally geodesic . by the choice of @xmath457 we conclude that @xmath354 is orthogonal to the @xmath465 fibers of @xmath355 . since @xmath354 separates @xmath355 , we also conclude that @xmath354 is an union of at least two totally geodesic planes . as shown in the proof of theorem 1.1 , this configuration can not be a limit of isoperimetric surfaces . [ monotonicity ] _ let @xmath466 be a three manifold with bounded curvature , @xmath467 , and with positive lower bound on the injectivity radius @xmath468 . if @xmath469 is a smooth surface with mean curvature which satisfies @xmath470 , then there exists a positive constant @xmath471 such that @xmath472 for every @xmath473 and @xmath474 . _ [ allard regularity ] _ let @xmath466 be a closed manifold and @xmath476 . there exist @xmath477 and @xmath478 with the following property : if @xmath479 is a smooth embedded cmc surface satisfying @xmath480 for every @xmath481 and @xmath482 , then its second fundamental form is uniformly bounded , i.e. , @xmath483 . _ let @xmath484 be a sequence of surfaces in a manifold @xmath2 . we say that @xmath485 converge to @xmath59 in the sense of graphs if near any point @xmath486 and for large @xmath45 the surface @xmath485 is locally a graph over an open set of @xmath487 and these graphs converge smoothly to the graph of @xmath59 . in addition , we say that @xmath488 satisfy _ local area bounds _ if there exist @xmath420 and @xmath169 such that @xmath489 for every @xmath490 . [ compactness ] _ let @xmath491 be a sequence of embedded surfaces with constant mean curvature satisfying local area bounds and such that @xmath492 . let s assume that @xmath493 converges to a metric @xmath494 in the @xmath226 topology . if @xmath495 has an accumulation point , then we can extract a subsequence that converges to a properly weakly embedded cmc surface @xmath59 in @xmath231 . _ let s first recall the constant mean curvature equation for graphs . if @xmath248 is a surface with constant mean curvature @xmath60 in @xmath62 , then @xmath248 can be written locally as a graph over a neighbourhood @xmath496 : @xmath497 in coordinates @xmath498 where @xmath499 is the coordinate base associated to @xmath500 . let @xmath501 be the coordinate base for @xmath485 , i.e. , @xmath502 . the induced metric @xmath138 is expressed by @xmath503 . a simple computation gives : @xmath504 we also have @xmath505 where @xmath506 are the christoffel symbols of @xmath106 . therefore , @xmath507 finally , the mean curvature equation is written as : @xmath508 since @xmath509 and @xmath510 , then @xmath511 . the equation ( [ mean curvature equation ] ) is uniformly elliptic as long as @xmath512 . let @xmath513 be an accumulation point for the sequence @xmath488 . by the upper bound on the second fundamental form , @xmath514 , there exists @xmath515 such that for every @xmath516 we have that @xmath517 is locally a graph @xmath518 over a neighbourhood @xmath519 . moreover , there exists @xmath520 for which @xmath521 . as @xmath522 , the schauder estimates for solutions of elliptic equations , see @xcite , imply that @xmath518 have @xmath523 estimates on @xmath524 , i.e. , @xmath525 . therefore , @xmath526 are uniformly bounded and equicontinuous . as @xmath484 satisfy local area bounds , then @xmath527 . on the other hand , the monotonicity formula , proposition [ monotonicity ] , gives that @xmath528 , where @xmath529 is a connected component of @xmath530 . it follows that the number of components of @xmath531 is finite and independent of @xmath45 . by the arzel - ascoli theorem we can extract a subsequence for which @xmath532 converges to @xmath533 for every @xmath534 . moreover , @xmath533 also satisfies the constant mean curvature equation ( [ mean curvature equation ] ) . as the set of accumulation points of @xmath488 is compact in @xmath535 we can cover this set by finite balls @xmath536 with @xmath537 . repeating the arguments in each of these balls and applying a diagonal argument we obtain a properly immersed surface @xmath59 on @xmath538 with constant mean curvature @xmath60 . since the surfaces @xmath485 are embedded , it follows that @xmath59 does not cross itself though it may have tangential self - intersections . therefore , @xmath59 is properly weakly embedded in @xmath2 . [ compactness iso ] _ let @xmath539 be a sequence of isoperimetric surfaces with @xmath540 . assume that @xmath541 converges , in the sense of cheeger - gromov , to a three manifold @xmath542 . there exists a properly embedded surface @xmath543 such that @xmath544 in the sense of graphs and the convergence is with multiplicity one . _ first we remark that @xmath488 satisfy local area bounds . indeed , @xmath545 . by proposition [ compactness ] we only need to rule out possibly multiplicities for the convergence @xmath544 and points where @xmath59 fails to be embedded . if the multiplicity of the limit is bigger than two , then @xmath546 has several components getting arbitrarily close . this allow us to do a local cut and past deformation , as shown in figure 2 , that preserves the enclosed volume . if @xmath494 is the euclidean metric , then @xmath547 and @xmath548 . thus , if @xmath549 , then @xmath550 the construction to deal with points where @xmath59 has tangential self intersections is similar to the multiplicity two case . the corollary now follows since these constructions contradict the fact that @xmath485 is isoperimetric for every @xmath45 . [ silveira ] _ let @xmath554 be a complete orientable surface conformally equivalent to a compact riemann surface punctured at a finite number of points . let @xmath555 be an operator on @xmath59 satisfying @xmath556 and @xmath557 . if @xmath59 has infinite area , then there exists a piecewise smooth function @xmath558 with compact support such that @xmath559 _ u. abresch and h. rosenberg . _ generalized hopf differentials_. mat . contemp . 28 ( 2005 ) , 1 - 28 . h. bray , _ the penrose inequality in general relativity and volume comparison theorems involving scalar curvature _ ( thesis ) . arxiv:0902.3241v1 , ( 1998 ) . d. gilbarg and n.s . trudinger . _ elliptic partial differential equations of second order_. second edition . fundamental principles of mathematical sciences , 224 . springer - verlag , berlin , 1983 . xiii+513 pp . p. petersen . _ riemannian geometry_. second edition . graduate texts in mathematics , 171 . springer , new york , 2006 . xvi+401 pp . r. pedrosa and m. ritor . _ isoperimetric domains in the riemannian product of a circle with a simply connected space form and applications to free boundary problems_. indiana univ . j. 48(4 ) , ( 1999 ) , 1357 - 1394 .

the leading order bfkl kernel @xcite , derived from feynman diagrams in momentum space , has been investigated in much detail . as a function of the transverse momenta , it is a meromorphic function . in the color singlet exchange channel , it describes the pomeron contribution in pqcd , consisting of ladder diagrams with reggeized gluons . this bfkl pomeron couples to the impact factors of colourless particles which , because of gauge invariance , vanish as one of the two reggeized gluons carries a zero transverse momentum . this property allows us to modify the space of functions to which the pomeron wave function belongs . on the other hand , in the color octet exchange channel the bfkl equation has the bootstrap solution , which represents a fundamental consistency property derived from the s - channel unitarity . a manifestation of this bootstrap property also takes place for the color singlet state of two gluons and inside the coupling of three or more gluons to colorless particle impact factors . in particular , it plays an important role in the odderon solution @xcite which appears as a bound state of three reggeized gluons . an important virtue of the color singlet kernel is its invariance under the mbius transformations in the space of transverse coordinates . exploiting the gauge invariance of the colorless impact factor to which the bfkl pomeron couples , the mbius symmetry allows to search solutions of the bfkl eigenvalues equation in the space of mbius functions which vanish as the coordinates of the two reggeized gluons coincide , and to define a spectral representation of the bfkl kernel in terms of conformal eigenfunctions . in this space of functions , the bfkl kernel also enjoys the property of holomorphic separability . translating back to the momentum representation , the kernel in the corresponding spectral form acquires @xmath4 function - like pieces which , because of the special properties of the impact factors , do not contribute to physical amplitudes . as mentioned above , the bfkl kernel , when applied to the color octet wave function in momentum space , satisfies the bootstrap condition , which reflects the s - channel consistency of the bfkl calculation . when transforming to transverse coordinates one finds that , in the space of mbius functions , this bootstrap condition can not be satisfied , i.e. the colour octet wave function lies outside this space of functions . in this paper we will define a similarity transform , @xmath5 , which takes us from the mbius space of functions ( m space ) to another space of functions ( named ` analytic feynman diagram ' ( af ) space ) in which the bootstrap holds . with the same transformation we can also define ` deformed ' mbius transformations and a ` deformed ' spectral representation of the bfkl kernel . the paper is organized as follows . in the following section we review the properties of the bfkl kernel , both in momentum space and in the space of transverse coordinates . in section 3 we introduce the deformed representation , and we define the similarity transform which takes us from the mbius space ( @xmath6 space ) to the new deformed space ( @xmath7 space ) of functions . in section 4 we explicitly show that , in the @xmath7 representation , the bootstrap properties are fulfilled . section 5 is devoted to the deformed representation of the conformal algebra which follows from the particular choice of the scalar product ; in particular we compute the transformed generators of the mbius group @xmath0 . some details of our calculations are collected in two appendices . in this section we give a brief review of the bfkl approach describing the dynamics of the reggeized gluons in lla of perturbative qcd . let us start from the schrdinger - like bfkl equation @xcite describing the compound state of two reggeized gluons , @xmath8 here @xmath9 labels the colour representation of the two gluon state and in the singlet and octet channel one has respectively @xmath10 and @xmath11 . the symbol @xmath12 denotes an integration in the transverse space in the case of the integral operator @xmath13 , while @xmath14 where @xmath15 is the @xmath16-channel angular momentum . the eigenvalues of @xmath17 give the positions of singularities of the @xmath16 channel partial waves , related to the scattering amplitude by the mellin transformation in the variable @xmath18 , where @xmath19 is the squared total energy of colliding particles . in lla , in the momentum representation , the gluon trajectory ( scaled by @xmath20 ) is given by the well known expression @xmath21 and the interaction term is defined by its action on the wave functions @xmath22 \phi(\mbox{\boldmath $ k$}_1^{\prime},\mbox{\boldmath $ q$}-\mbox{\boldmath $ k$}_1^{\prime } ) \nonumber\end{aligned}\ ] ] and @xmath23 . let us note that this form is proper when the kernel acts on amputated ( without gluon propagators ) impact factors , otherwise the propagators should be removed from the kernel , as for example is the case for the expression in eq . ( [ bfkl_mom ] ) . we stress that this form of the ll bfkl kernel is obtained directly from the feynman diagrams analysis in perturbative qcd and has a simple analytic behavior in the transverse momentum space . in particular it does not contain @xmath4-functions @xmath24 and @xmath25 . in the construction of the bfkl kernels a very important property is the gluon reggeization which can be verified with the bootstrap relation for the amplitude with the octet quantum numbers @xcite . the bootstrap relation is a consequence of the @xmath19-channel unitarity . it claims , that the scattering amplitude with octet quantum numbers obtained as a solution of the corresponding bfkl equation should coincide with the born term multiplied by the regge factor @xmath26 . because the bfkl equation was obtained by summing contributions from the multi - particle production described by the multi - regge amplitudes , the bootstrap relation valid in the leading and next - to - leading orders allows to connect the gluon trajectory and interaction terms . in lla the bootstrap condition for the bfkl kernel can be written as @xmath27 where the constant @xmath28 is the wave function which can be conveniently obtained after rescaling any function depending only on @xmath29 , and the kernel @xmath30 acts on the amputated amplitudes . let us note that it is crucial in the bootstrap relation in lla that the two gluons are located at the same point in the transverse coordinate plane . it has played a crucial role in the discovery of the leading odderon solution in the lla of perturbative qcd @xcite and also some general relation @xcite between bound states of @xmath31 and @xmath32 reggeized gluons . in the following we shall use its equivalent form @xcite , which has the virtue of being infrared finite and which uses the standard bfkl singlet kernel : @xmath33 on using an infrared mass regularization ( @xmath34 ) one may write @xmath35 where @xmath36 and @xmath37 , so that a relation equivalent to the bootstrap , but explicitely infraredly finite , is given by @xmath38 the 2-gluon kernel ( [ bfkleq ] ) in the singlet channel has been investigated in great details in the coordinate representation @xcite . the amplitude for the scattering of colorless objects is factorized , in the high energy limit , in the product of the green s function ( which exponentiates the bfkl kernel ) and two impact factors . the impact factors vanish as the momentum of one of the attached reggeized gluons goes to zero . this permits to choose a special representation for the two - gluon propagator and for the full bfkl kernel acting in the space of the so called mbius functions ( for two gluon states these are functions @xmath39 such that @xmath40 ) @xcite . this is a special choice among infinite others compatible with the gauge freedom of integrating a colorless impact factor with any function belonging to the set defined by the equivalence relation @xmath41 since the corresponding shift in the momentum representation is proportional to @xmath42 . in such a representation the operators are mbius ( conformal ) invariant @xcite . using complex coordinates , the bfkl hamiltonian @xmath43 acting on functions with propagators included can be written in the operator form @xcite @xmath44 where @xmath45 , and we introduced the gluon holomorphic momenta @xmath46 . on the mbius space of functions the holomorphic separability applies : @xmath47 and it is possible to find more easily the solutions of the homogeneous bfkl pomeron equation which belong to irreducible unitary representations of the mbius group . in particular the symmetry generators in this representation are @xmath48 for two reggeized gluons one has @xmath49 and the casimir operator is defined as follows @xmath50 where @xmath51 and @xmath52 . the eigenfunctions of the bfkl kernel are also eigenstates of two casimir operators and given by @xcite @xmath53 where @xmath54 , @xmath55 are conformal weights for the principal series of the unitary representations of the mbius group , @xmath31 is the conformal spin and @xmath56 is the anomalous dimension of the operator @xmath57 describing the compound state of two reggeized gluons ( note , that here and below we use other notations for conformal weights @xmath58 in refs . the corresponding eigenvalues of the bfkl kernel are given by @xmath59 the action of the bfkl kernel can also be written , again on the space of mbius functions , after a duality transformation @xcite , in terms of integral operator appearing in the dipole picture @xcite @xmath60 we now consider the pomeron eigenstates in the momentum representation . different forms of the fourier transform of the function in eq . ( [ pomstatescoord ] ) have been given in the most general case . one finds a sum of an analytic term @xmath61 , which has been given in an explicit way or in a integral form , and a @xmath4-like one @xmath62 . more precisely we obtain @xmath63 where a bra - ket compact notation is also introduced . by explicit computation one finds @xmath64 \frac{i^n}{2\pi } 2^{1-h-\bar{h } } \frac{\gamma(1-\bar{h})}{\gamma(h ) } q^{\bar{h}-1 } q^{*\ , h-1 } \ , , \label{pom_delta}\ ] ] where the complex form @xmath65 of any bidimensional transverse vector @xmath66 ( coordinate or momentum ) is used . in the calculation of a physical cross section one performs integrations with two - gluon colourless impact factors and this @xmath4-like term gives zero contribution , so that all the physics is related to the analytic term . in fact the @xmath4-terms have been introduced when one has chosen to move to the mbius representation which is characterized by the simple and beautiful form in the coordinate representation . the analytic part can be written , for example , as given in @xcite wherein one finds @xmath67 with the coefficient @xmath68 given by @xmath69 the expression for @xmath70 in complex notation was written in terms of the hypergeometric functions @xmath71 another expression for @xmath72 which will be very useful for our purposes has been given in @xcite in an integral form : @xmath73^{\bar{h}-1 } \left [ \frac{p^ * ( q^*-p^*)}{k_1^*-p^*}\right]^{h-1 } \ , , \label{ana_lip}\ ] ] where @xmath74 below we shall consider the completeness relation and the spectral representation for the bfkl kernel ( cf . @xcite ) . let us before discuss the scalar product in the space of functions where the kernel is acting on . for the kernel in the momentum space associated to the feynman diagram derivation there is only one scalar product available . if we consider non amputated functions of two momenta @xmath75 , i.e. with 2-dimensional propagators @xmath76 included , the scalar product is defined as @xmath77 with the integration measure @xmath78 which kills the extra propagators . in the mbius space of functions ( in coordinate space ) for the principal series of the mbius group representation , whenever functions with conformal weight @xmath79 or @xmath80 are not considered , two possible choices are available @xcite . since the mbius functions contain propagators , one possibility is to remove , as before , the propagators in one function , acting with the operator @xmath81 . the other possibility , related to the form of the casimir operator of the mbius group , is to use instead the measure @xmath82 . the latter is not equivalent to the former when we consider the conformal wheights @xmath79 or @xmath80 . note , that , providing that we go from the regge kinematics to the deep - inelastic scattering in the bjorken regime , the additional series of the unitary representations of the mbius group should be also used @xcite ( cf . ref . @xcite ) . we now consider the completeness relation in coordinate space @xcite , in the space of mbius functions , where the eigenfunctions of the casimir operator of the mbius group , defined in ( [ pomstatescoord ] ) , constitute a suitable spectral basis for the operators acting on the space of the colourless impact factors , which , we stress , never `` feel '' the presence of the terms with a @xmath4-distribution behavior in momentum space , i.e. those of eq . ( [ pom_delta ] ) . for the operators amputated ( spatial propagators removed ) to the left ( with notation @xmath83 and @xmath84 ) one can write @xcite @xmath85 where we have defined the weights @xmath86 , as well as @xmath87 for later use , @xmath88[\nu^2+(n+1)^2/4]}\ ] ] and @xmath89 . if we try to extrapolate these operators to a wider domain which contains functions which no longer vanish for zero momenta , contrary to the colorless impact factor case , one needs some care . for example , for the function , which in momentum space depends only on the total momentum , just the case we meet in the bootstrap relation in lla , one obtains the the result @xmath90 in coordinate space ( for the amputated impact factor ) . but such generalized function is orthogonal to the basis constituted by the functions in eq . ( [ pomstatescoord ] ) since @xmath91 and therefore the application of the operators in eq . ( [ coord_relations ] ) on this function will give zero . in other words the bootstrap relation can not be fullfilled in such a framework with the spectral representations given in eq . ( [ coord_relations ] ) . but this should not be surprising , since the choice done thanks to the equivalence relation in eq . ( [ uv_transf ] ) , is based on a restriction of the action of the kernel to the functions similar to colorless impact factors . moving to the momentum representation , one may write the fourier transform of the relations in ( [ coord_relations ] ) and the same considerations done above may be repeated in this case , leading to the same conclusions . in particular one can not write a bootstrap relation in a spectral basis built upon the states of eq . ( [ pom_mom ] ) . on the other hand one notices that the bfkl kernel @xmath92 in momentum space , as constructed from the feynman diagram analysis , has an analytic behavior which must be reflected also in its spectral decomposition . it can also be noted that the behavior in @xmath93 of the @xmath4-like term ( [ pom_delta ] ) is @xmath94 , which is singular in the forward limit where analytic behavior is expected for fixed @xmath95 . therefore it is natural to suggest the modified relations for the eigenfunction completeness and for the spectral representation of the bfkl kernel : @xmath96 where only the analytic contribution from each state of the spectral basis is used and the measure for integration is defined in eq . ( [ scal_prod ] ) , according to our choice of considering functions with propagators included . below we shall verify this anzatz . firstly we consider a space of functions ( with removed propagators ) which includes functions corresponding to the colorless impact factors and at least one function depending only on the total momentum , which is associated to a particular colored impact factor . secondly we modify the spectral representation of the ll bfkl kernel , wherein the eigenvalues are assumed to be the same expressions of the conformal weights as for the restriction on the mbius space of function , and the eigenfunctions are deformed in order to be analytic in the momentum space . a similar idea was already considered in the literature in an attempt to couple the ll bfkl pomeron to a quark @xcite . we proceed now with the definition of a transformation between the mbius ( m ) basis and the analytic feynman ( af ) basis . let us observe that one can go from one basis to the other by a simple transformation , which reads in coordinate representation as follows @xmath97 these relations are clearly non local and have been obtained by noting that @xmath98 for the case of even conformal spins ( @xmath31 ) the second transformation can be written in a simple local form on observing that @xmath99 while for odd conformal spin such a simple transformation is not possible since @xmath100 . it is related to the fact , that in the last case both functions @xmath101 and @xmath102 satisfy the colour transparancy property @xmath103 , but only @xmath101 has the simple conformal properties . it can be easily checked that @xmath104 so that @xmath5 is a 1 - 1 mapping with inverse really given by @xmath105 . the green s function for the evolution in the rapidity @xmath106 based on the deformed analytic spectral basis was also written above ( see ( [ mom_relations ] ) ) . in the next section we shall show that with this prescription one is recovering the correct relation compatible with the bootstrap requirement for the gluon reggeization in lla . since the spectrum in both representations is the same it is natural to expect that the bfkl kernel satisfying the bootstrap relation is conformal invariant , in the new deformed analytic feynman basis obtained by a similarity transformation given by the operator @xmath5 . namely , the well known generators of the mbius group @xmath107 could be extended to this basis as follows @xmath108 let us summarize the result . if one considers the impact factors derived from feynman diagrams calculations , they are the functions belonging to the af - space and not to the m - space . we remind that we have two different completeness relations based on the two spectral basis @xmath109 and @xmath110 and in these two spaces there are different definitions for scalar products and normalizations of wave functions . nevertheless the cardinality of the two basis is the same . the two spaces are related by the 1 - 1 mapping @xmath111 af - space @xmath112 m - space as previously discussed ( see also fig . [ fig_mapping ] ) and one space is _ shifted _ with respect to the other . moreover any operator @xmath113 in the af - space which can be decomposed on the spectral basis can be defined on the m - space as @xmath114 and viceversa @xmath115 . this can be seen as an isometry between the mbius space and the analytical feynman space with the different scalar products in these spaces . -@xmath28 mapping between the af - space and the m - space.,width=264 ] the next interesting question is therefore if it is possible to construct explicitely the new representation defined in eq . ( [ derep ] ) . we shall study this problem in the section @xmath116 , starting from analyzing the fourier transform of the mbius algebra in the m - space and later on moving to the af - space . in the next section we will show that with the meromorphic prescription we indeed recover the bootstrap properties of the bfkl kernel . for the sake of using a compact notation , let us introduce for some functions of transverse gluon momenta the corresponding wave functions for a color singlet state . these are the _ power _ @xmath117 , the _ unity _ @xmath118 and the _ log _ @xmath119 states ( modulus the propagators ) , defined by their representations in the momentum space : @xmath120 since , as previoulsy discussed , one has for the mbius - invariant wave function @xmath121 the equality @xmath122 we obtain @xmath123 let us recall the completeness relation and the bfkl kernel written as @xmath124 on using the previous relations we write the bootstrap relation in the form of eq . ( [ bootstrapirfinite ] ) as @xmath125 ) was used . in order to show that our definitions ( [ gen_relations ] ) are correct , we should prove that the coefficients of the conformal basis @xmath126 on both the l.h.s . and the r.h.s . do coincide , i.e. @xmath127 it is therefore enough to calculate the integral given by @xmath128 . verifying that this expression is finite it is sufficient also to calculate the first two terms of the taylor expansion in @xmath129 to verify the eq . ( [ coeff_bootstrap ] ) , thanks to eq . ( [ power_rel ] ) . we also note that the left hand side of eq . ( [ coeff_bootstrap ] ) can be calculated easily in an independent way using the relations in eq . ( [ ana_delta ] ) and ( [ pom_delta ] ) , which give : @xmath130 where @xmath131/2 $ ] has support ( and value @xmath28 ) for any even integer number . let us now find explicitely @xmath132 . it is convenient to perform the calculations in the momentum space where we have defined the analytic element @xmath133 of the conformal basis . taking for it the expression given in eq . ( [ ana_lip ] ) we write : @xmath134 using new integration variables induced by the change @xmath135 and @xmath136 with analogous relations for the complex conjugated ones , we obtains @xmath137 where @xmath138 it is convenient to define the function @xmath139 because one can find an explicit expression for it in terms of the generalized hypergeometric functions @xmath140 . we illustrate some details of the calculation in appendix a. it is sufficient to represent the result of the integration in power series of @xmath129 for the point @xmath141 : @xmath142 recalling the first two terms in the expansion ( see eq . ( [ d01 ] ) in appendix a ) one can see , on using the definitions of eq . ( [ power_rel ] ) and ( [ hapl ] ) , which imply @xmath143 and @xmath144 , that the relation in eq . ( [ coeff_bootstrap ] ) is verified and therefore also the bootstrap relation . as a check one may evaluate explicitely the quantity @xmath145 using the eq . ( [ cl ] ) . it is easy to show that it coincides with the expression of eq . ( [ hau ] ) , computed in a completely different way.in order to give an alternative check of the representation proposed , we have also calculated explicitly the third term of the expansion of the bfkl green s function @xmath146 ( the first term is trivial and the second corresponds to the bootstrap relation ) , which contains the double iteration of the kernel acting on the unity ( momentum space ) . the details are in the appendix b. using the spectral representation one can compute an arbitrary number of iterations of the bfkl kernel or the action of the bfkl green s function on the impact factor of a quark or gluon line . in this section we will examine the representation of the @xmath0 algebra in the @xmath7 space of funtions . to this end we start from the familiar mbius algebra in the @xmath6 space and transform it into momentum space . we then switch to the @xmath7 space : in order to retain the correct structure of the algebra we have to construct a new representation of the generator @xmath147 . let us reconsider the generators of the conformal mbius group in @xmath148 dimensions in coordinate space given in eq . ( [ moebius_gen ] ) and the corresponding casimir operator ( [ moebius_cas ] ) . we will be interested in writing similar relations in momentum space . we start from the conformal eigenfunctions ( [ pomstatescoord ] ) which from now on will carry the superscript @xmath6 , and we define their fourier transforms : @xmath149(\mbox{\boldmath $ q$};\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2 } ) & = & \int d\mu ~e^{-i\mbox{\boldmath $ q$}\cdot \mbox{\boldmath $ \rho$}_{0}+i\mbox{\boldmath $ k$}_{1}\cdot \mbox{\boldmath $ \rho$}_{1}+i\mbox{\boldmath $ k$}_{2}\cdot \mbox{\boldmath $ \rho$}_{2}}e_{h{\bar{h}}}^{m}(\mbox{\boldmath $ \rho$}_{0};\mbox{\boldmath $ \rho$}_{1},\mbox{\boldmath $ \rho$}_{2})= \nonumber \\ & = & ( 2\pi ) ^{2}\delta ^{(2)}(\mbox{\boldmath $ q$}-\mbox{\boldmath $ k$}_{1}-\mbox{\boldmath $ k$}_{2}){\tilde{e}}_{h{\bar{h}}}^{m}(\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2})\ , , \\ { \tilde{e}}_{h{\bar{h}}}^{m}(\mbox{\boldmath $ \rho$}_{0};\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2})=\tilde{\mathcal{f}}[e_{h{\bar{h}}}^{m}](\mbox{\boldmath $ \rho$}_{0};\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2 } ) & = & \int d\tilde{\mu}~e^{i\mbox{\boldmath $ k$}_{1}\cdot \mbox{\boldmath $ \rho$}_{1}+i\mbox{\boldmath $ k$}_{2}\cdot \mbox{\boldmath $ \rho$}_{2}}e_{h{\bar{h}}}^{m}(\mbox{\boldmath $ \rho$}_{0};\mbox{\boldmath $ \rho$}_{1},\mbox{\boldmath $ \rho$}_{2})= \nonumber \\ & = & e^{i(\mbox{\boldmath $ k$}_{1}+\mbox{\boldmath $ k$}_{2})\cdot \mbox{\boldmath $ \rho$}_{0}}{\tilde{e}}_{h{\bar{h}}}^{m}(\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2})\ , , \\ { \tilde{e}}_{h{\bar{h}}}^{m}(\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2})=\tilde{\mathcal{f}}[e_{h{\bar{h}}}^{m}](\mbox{\boldmath $ \rho$}_{0}=0;\mbox{\boldmath $ k$}_{1},\mbox{\boldmath $ k$}_{2 } ) & = & \int d\tilde{\mu}~e^{i\mbox{\boldmath $ k$}_{1}\cdot \mbox{\boldmath $ \rho$}_{1}+i\mbox{\boldmath $ k$}_{2}\cdot \mbox{\boldmath $ \rho$}_{2}}e_{h{\bar{h}}}^{m}(0;\mbox{\boldmath $ \rho$}_{1},\mbox{\boldmath $ \rho$}_{2})\,,\end{aligned}\ ] ] where @xmath150 , @xmath151 and @xmath152 , @xmath153 are the fourier transform operators in three and two variables respectively . in coordinate space the action of the generators on the eigenfunctions @xmath154 is : @xmath155 note , that the functions @xmath156 can be considered as the clebsch - gordon coefficients in the expansion of the product of the reggeon wave functions @xmath157 in the sum of the irreducible representations @xmath158 where the conformal weights for the principal series of unitary representations are @xmath159 and the coefficients @xmath160 are fixed with the use of the normalization conditions for the wave functions @xmath157 and @xmath161 . in the irreducible representations @xmath161 the mbius group generators are given below @xmath162 the generators of the mbius group can be translated into the momentum space , using the relation @xmath163 : @xmath164 where @xmath165 one can calculate also the action of some bilinear combinations of the generators in the momentum representation @xmath166 @xmath167 where the casimir operator in the short momentum representation is @xmath168 note that in our notations @xmath169 is a generator while @xmath170 is the casimir operator . the algebra of these generators in momentum space is the same as in coordinate space : @xmath171={\tilde{m}}_{r}^{+}\,,\quad \lbrack { \tilde{m}}_{r}^{+},{\tilde{m}}_{r}^{-}]=-2{\tilde{m}}_{r}^{3}\,,\quad \lbrack { \tilde{m}}_{r}^{-},{\tilde{m}}_{r}^{3}]=-{\tilde{m}}_{r}^{-}\,.\ ] ] while the action of the generator @xmath172 does not need special comments , it is interesting to look at others . one finds @xmath173 the last relation follows from the action of the casimir operator @xmath174 and from relation ( [ momm3 m ] ) . in particular , given that @xmath175 + \left ( { \tilde{m}}^{3}\right ) ^{2}={\tilde{m}}^{+}{\tilde{m}}^{-}+{\tilde{m}}^{3}\left ( { \tilde{m}}^{3}+1\right ) \ , , \label{casimir}\ ] ] where the commutation relation @xmath176=-2{\tilde{m}}^{3}$ ] has been used , one can see that indeed the relation ( [ mommminusm ] ) is satisfied . these relations mean that in the m - space in momentum representation the mbius algebra is projected to @xmath177 { \tilde e}^m = { \tilde m}^+ { \tilde e}^m\ , , \quad { \tilde m}^-{\tilde m}^+ { \tilde e}^m = 2 { \tilde m}^3 { \tilde e}^m \ , , \quad 0\ , { \tilde e}^m=0\ , { \tilde e}^m\,.\end{aligned}\ ] ] let us now move to the @xmath7 space . we start , in coordinate space , from the definition @xmath178 where , according to eq . ( [ mapping_on_e ] ) , @xmath179 and @xmath180 is a projector ( @xmath181 ) defined by its action @xmath182 in order to compute the action of the mbius generators on the analytic part , @xmath183 , of the conformal eigenfunction we first examine the singular piece , @xmath184 . first of all let us observe that the casimir operator acting on @xmath185 gives zero , as it is trivially checked in the coordinate representation . this means that @xmath121 in such a case can be obtained with the action of a function of the casimir ( solving the eq . ( [ casimiraction ] ) for @xmath121 ) only before applying the projector @xmath180 . after the application of @xmath180 one needs to define a new operator to extract the scaling parameter @xmath121 . it is related to the fact , that @xmath186 is not an eigenfunction of the usual casimir operator . we should construct a new casimir operator @xmath187 for this representation . in coordinate representation we can deduce the following relations for generators @xmath188 which coincides with their action for @xmath189 while the action of @xmath190 is different . according to this fact the action of mbius generators on @xmath191 can be written as @xmath192 and @xmath193 2^{-h-{\bar{h}}}\,.\end{aligned}\ ] ] therefore only the last generator @xmath190 , needs to be redefined in the @xmath7 space . let us consider now again the momentum space for these functions . first we analyze the action of the @xmath194 generator on @xmath195 , which we have written in terms of hypergeometric functions . one can immediately see that @xmath196 for any @xmath197 . this means again that we are not forced to use a non local operator @xmath198 ( solving an equation with the corresponding casimir @xmath199 ) to extract the dimension @xmath121 of an eigenstate , as we had to do in the coordinate representation . the @xmath200 function is instead very peculiar since we know that the casimir gives zero . this fact is also verified directly in momentum space , due to the deltas present in @xmath201 . later we will show it explicitely . instead we note that the action of the @xmath194 operator is good . indeed , from the explicit form given in eq . ( [ pom_delta ] ) , we obtain @xmath202 this is compatible with the fact that the @xmath194 operator has the same action on the @xmath203 and @xmath195 functions . the fact that we may choose to extract the dimension @xmath121 in momentum space using the @xmath194 operator may be useful to simplify the construction of the mappings @xmath5 and @xmath105 . on defining @xmath204 one may write the mapping introduced in eq . ( [ mapping_on_e ] ) as @xmath205 where in general @xmath206 and which imply @xmath207 and we have denoted @xmath208 , @xmath209 and similarly @xmath210 . clearly in the relation for the mappings one can use different forms for the operator @xmath211 depending on which space it acts . in momentum space @xmath212 then one can write @xmath213=0 $ ] ( with some abuse of notation ) and things are further simplified . proceeding along this line one can give a deformed representation @xmath214 which lives on the af - space . in order to do this , let us first study @xmath215 and @xmath216 c_h ( k_1+k_2)^{{\bar{h}}-1 } ( k_1^*+k_2^*)^{h-1 } \nonumber \\ & = & 2i h ( h-1 ) \frac{1}{k_1^*+k_2^ * } { \tilde e}^\delta_h \,=\ , -h(h-1 ) \left ( { \tilde m}^+ \right)^{-1 } { \tilde e}^\delta_h \nonumber \\ & = & -\left ( { \tilde m}^+ \right)^{-1 } h(h-1 ) \frac{f}{1-f } p { \tilde e}_h^a \ , . \label{rel1}\end{aligned}\ ] ] note that @xmath217 has the same form on any of the spaces of functions we consider . the resulting action of @xmath218 on @xmath219 has to be studied in order to see if it is possible to define a new representation for @xmath218 on af - space which satisfies the conformal algebra . but before let us note that from the previous relation we have @xmath220 so that we reobtain , recalling eq . ( [ casimir ] ) , @xmath221 in general we have defined the mbius generators on the af space ( spanned by the basis functions @xmath219 ) using the isometry with the m - space : @xmath222 let us look at each of the three generators . from the previous discussion , also in the coordinate representation , we observe that @xmath223 which means that in momentum space the actions of these two operators is the same in both m - space and af - space . we therefore define for the af - space a new generator in momentum representation such that @xmath224 which puts into relation , as before , the action of the casimir operator such that @xmath225 together with the piece of the mbius algebra @xmath226=-2{\tilde m}_{af}^3 $ ] and the action of @xmath227 . according to the previous result ( see eqs . ( [ mminusa ] ) and ( [ rel1 ] ) ) , by construction the choice is @xmath228 where , in the last form , g is a non local operator such that @xmath229 let us finally verify that , with this new representation of @xmath214 , the algebra of the mbius generators is correct . it is easy to see that @xmath230=-{\tilde m}_{af}^- \,,\ ] ] which is a relation that is projected to zero when applied to the function @xmath219 . in order to check that the mbius algebra is closed we need to examine the relation @xmath231=+2{\tilde m}_{af}^3\ ] ] understood to act on @xmath219 . one then has to verify that @xmath232 using eq . ( [ reldelta ] ) we derive also @xmath233 and making the difference between eqs . ( [ v1 ] ) and ( [ v2 ] ) we therefore obtain @xmath234 if @xmath217 commutes with @xmath235 this is equivalent to @xmath236 which coincides with the starting definition of the generator . the commutation relation is fullfilled since the operator @xmath217 is just a multiplicative operator associated to the total momentum and therefore commutes with the action of the projector @xmath180 and moreover it commutes with the function of the casimir @xmath237 . finally let us look at the relation between the two casimir operators in m - space and af - space : using @xmath238 + \left ( { \tilde{m}}_{af}^{3}\right ) ^{2}$ ] we find from the previous relations @xmath239 this last relation , applied to @xmath240 , gives @xmath241 which indeed is compatible with @xmath242 . among the properties of the ll bfkl kernel the bootstrap relation connected to the gluon reggeization is a fundamental consistency condition . this bootstrap property leads to an important feature of the bfkl kernel also in the color singlet state : this can be demonstrated most easily in momentum space where the bfkl kernel is meromorphic . many previous investigations , exploiting the conformal symmetry of the kernel , have been carried out in the space of mbius functions ( @xmath6 ) . in this space of functions , however , the bootstrap relation for the bfkl kernel is not satisfied . in this paper we have defined a modified space of functions ( @xmath7 ) , in which the bootstrap property is valid . in particular , we have derived a spectral representation of the kernel which is also useful in evaluating the coupling of the bfkl green s function to colored impact factors , and we have verified explicitly that the bootstrap relation is fulfilled . we also have derived the corresponding represention of the mbius algebra ; in order to act in this modified space of functions , one of the mbius generators has to be deformed . our discussion has been limited to the the case of two - gluon green s functions . since in both spaces , @xmath6 and @xmath7 , the eigenvalues of the casimir operator , @xmath170 and @xmath187 , resp . , are the same , and since the hamiltonian for the pairwise interaction of two reggeized gluons can be expressed in terms of this casimir operator , one can expect that , for states consisting of more than two gluons all remarkable properties of the multi - colour bfkl dynamics derived in the mbius picture can be generalized to the af picture . this includes , in particular , the holomorphic separability and integrability @xcite . we are going to return to these interesting problems in future publications . part of this work has been done while one of us ( l.n.l ) has been visiting the ii.institut f.theoretische physik , university hamburg , and the university of bologna , bologna . two of us ( j.b . and g.p.vacca ) gratefully acknowledge the hospitality of the nuclear physics institute in gatchina , st.petersburg . the calculation of the integral in eq . ( [ integral ] ) is carried on noting that one can find a third order differential operator in @xmath243 and another in @xmath244 , related to the @xmath140 functions , which , when applied to the integrand , gives a total derivative in an integration variable so that , on applying the stokes theorem , one finds that @xmath245 satisfies the related differential equation in both the holomorphic and antiholomorphic sectors . this allows to write it as a sum of products of the linearly independent solutions of the two differential equations and on imposing the single valueness and the correct normalization ( at some convenient point ) the full expression is found @xcite . in particular this has the following , so called , conformal block structure @xmath246 where @xmath247 and @xmath248 are three independent solutions of the generalized hypergeometric differential equations in @xmath243 and @xmath249 . the coefficients @xmath250 depend through @xmath251 functions on the conformal weights . the general form of @xmath139 is pretty complicated but it simplifies considerably for @xmath141 using the relations @xcite @xmath252 on using the above relation , after algebraic simplifications , we find @xmath253 with @xmath254 the expressions of @xmath255 and @xmath256 have a simple pole at @xmath257 and opposite residues for this pole , while @xmath258 is holomorphic at @xmath257 , therefore the full function in eq . ( [ resulti ] ) is not singular at @xmath257 . using the transformation properties of the gamma function , each of three terms can be reduced to a common multiplicative factor and expressions involving only trigonometric functions . these terms are combined together to give the following compact form for @xmath259 @xmath260 now the first few derivatives with respect to @xmath129 can be easily calculated . in particular we obtain , after some manipulations , @xmath261 \ , c_0 \ , . \label{d01}\end{aligned}\ ] ] we recompute here the double iteration of the bfkl kernel in momentum space acting on a constant function @xcite and compare it with the expressions obtained with the use of the above spectral representation of the kernel ( and therefore of the green s function ) . let us write before the result of the direct integration in momentum space : @xmath262 \ , . \label{ksquare1analytic}\end{aligned}\ ] ] before giving some details of the computation let us consider what we obtain from the spectral approach . on using the spectral representation for the kernel , the relation ( [ ksquare1analytic ] ) can be written as @xmath263 a relation which we have checked numerically . this check has been also performed in the case of the single iteration ( bootstrap relation ) . the argument of the integration over @xmath264 oscillates with a very slow decay for fixed conformal spin , but summing over some tens of @xmath31 gives a good suppression of the tails and the integral can be computed . finally we sketch the derivation of eq . ( [ ksquare1analytic ] ) . there are three kinds of integral involved ; they can be calculated with the dimensional regularization and the feynman parameterization , in the case @xmath265.the needed integrals are : @xmath266 where @xmath267 , @xmath268 and @xmath269 is the gluon trajectory ( [ gluontrajectory ] ) rescaled by @xmath270.the first two integral can be easily calculated exactly , giving : @xmath271 \ , , \nonumber \\ j_1 ( \mbox{\boldmath $ k$}_1,\mbox{\boldmath $ k$}_2 ) & = & \pi^{1+\epsilon } \frac{\gamma(1-\epsilon)\gamma^2(\epsilon ) } { \gamma(2 \epsilon ) } \left [ \pi \cot \pi \epsilon + \log \frac{\mbox{\boldmath $ q$}^2}{\mbox{\boldmath $ k$}_1 ^ 2 } + \psi(2 \epsilon ) - \psi(1 ) \right ] ( \mbox{\boldmath $ k$}_1 ^ 2)^\epsilon = \nonumber \\ & = & \pi^{1+\epsilon } \gamma(1-\epsilon ) ( \mbox{\boldmath $ q$}^2)^\epsilon \left [ \frac{1}{\epsilon^2 } + \frac{1}{\epsilon } \log \frac{\mbox{\boldmath $ q$}^2}{\mbox{\boldmath $ k$}_1 ^ 2 } - \frac{\pi^2}{6 } -\frac{3}{2}\log^2 \frac{\mbox{\boldmath $ q$}^2}{\mbox{\boldmath $ k$}_1 ^ 2 } + \mathcal{o}(\epsilon ) \right ] \ , .\end{aligned}\ ] ] the third integral is calculated to the order @xmath272 : @xmath273 \ , .\end{aligned}\ ] ] putting together all the pieces to construct the complete action of the kernel , all the singulatities cancel and the result , which is finite in the limit @xmath274 , correspond to the expression given in eq . ( [ ksquare1analytic ] ) . j. bartels , l. n. lipatov and g. p. vacca , phys . b * 477 * ( 2000 ) 178 , [ arxiv : hep - ph/9912423 ] . g. p. vacca , phys . b * 489 * , 337 ( 2000 ) , [ arxiv : hep - ph/0007067 ] . l. n. lipatov , sov . jetp * 63 * ( 1986 ) 904 [ zh . * 90 * ( 1986 ) 1536 ] . j. bartels , l. n. lipatov and g. p. vacca , nucl . b * 706 * ( 2005 ) 391 , [ arxiv : hep - ph/0404110 ] . l. n. lipatov , phys . lett . * b309 * ( 1993 ) 394 ; _ high energy asymptotics of multi - colour qcd and exactly solvable lattice models _ , hep - th/9311037 , unpublished . j. bartels , m. a. braun , d. colferai and g. p. vacca , eur.phys.j.c * 20 * ( 2001 ) 323 , [ arxiv : hep - ph/0102221 ] . h. j. de vega and l. n. lipatov , phys . d * 64 * , 114019 ( 2001 ) , [ arxiv : hep - ph/0107225 ] , phys . d * 66 * ( 2002 ) 074013 , [ arxiv : hep - ph/0107225 ] . s. e. derkachov , g. p. korchemsky , j. kotanski and a. n. manashov , nucl . b * 645 * ( 2002 ) 237 [ arxiv : hep - th/0204124 ] . a. h. mueller and w. k. tang , phys . b * 284 * ( 1992 ) 123 . j. bartels , j. r. forshaw , h. lotter , l. n. lipatov , m. g. ryskin and m. wusthoff , phys . b * 348 * ( 1995 ) 589 [ arxiv : hep - ph/9501204 ] . j. s. geronimo and h. navelet , j. math . * 44 * ( 2003 ) 2293 . bateman project , a. erdelyi , higher trascendental functions , vol . 189 , mcgraw - hill ( 1953 ) .

the study of open heavy flavor production in relativistic @xmath18 collisions is sensitive to different kinds of strong interaction physics . because the leading order ( lo ) production mechanism is gluon fusion @xcite , open heavy flavor production rates are directly related to modification of the gluon parton distribution function ( pdf ) , i.e. shadowing or saturation @xcite . also , the initial and/or final state partons can scatter and lose energy in the cold nuclear medium @xcite , thereby modifying and producing a nuclear modification of open heavy flavor production . recently , the possibility of flow even in small collision systems such as @xmath18 has raised the question of modified charm momentum distributions @xcite . modification to heavy quark production rates and kinematics in collisions at the relativistic heavy ion collider ( rhic ) is well established . electron production from open heavy flavor decay is enhanced @xcite , while @xmath19 production is suppressed @xcite at midrapidity . at positive rapidity , defined with the positive @xmath20 axis as the direction of the deuteron , there is a suppression of heavy - flavor decay muons @xcite and a larger suppression of @xmath19 @xcite . while @xmath21@xmath13 correlations from open heavy flavor decays have not been published at rhic to date , correlations involving light flavor hadrons have shown modification in collisions at rhic . a suppression has been observed of positive rapidity @xmath22 mesons associated with midrapidity trigger hadrons , especially in the back - to - back peak at @xmath23 , indicating @xmath24 scatterings @xcite . this suppression increases as @xmath25 , the fraction of the nucleon momentum carried by the gluon , decreases . these results are in quantitative agreement with energy loss models @xcite and saturation models @xcite . this paper presents measurements of azimuthal correlations of electron - muon pairs produced from heavy flavor decays , primarily @xmath17 , in and collisions using the phenix detector at rhic . the heavy - flavor @xmath21@xmath13 correlations are free of backgrounds from other sources that contribute to other dilepton analyses ( @xmath26 or @xmath27 ) , such as resonance decay and drell - yan . while analysis of dilepton mass and @xmath28 provides a way to separate charm and bottom contributions , the azimuthal correlations have an important advantage for studying the charm production process . the leading order production , @xmath29 and @xmath30 , will produce back - to - back open heavy flavor pairs that can semileptonically decay and produce azimuthally correlated @xmath21@xmath13 pairs . next - to - leading order ( nlo ) processes like flavor excitation and gluon splitting produce much less correlated @xmath31 and thus much less correlated @xmath21@xmath13 pairs . therefore , modification to different portions of the azimuthal correlations can be attributed to modifications of @xmath17 pairs from different production mechanisms . in energy loss models such as ref . @xcite , a broadening of the back - to - back azimuthal correlation should accompany a suppression of the peak due to the multiple scattering that the incoming gluons and/or the outgoing @xmath17 undergo in the cold nuclear medium . this paper is organized as follows . the phenix detector is outlined in section [ sec : phenix ] . section [ sec : method ] describes the details of the method used to measure the correlations , the background subtraction method , and the tests of the method . section [ sec : results ] presents the results in and compares them to monte carlo models . the results are presented and compared to the results in section [ sec : dauresults ] . conclusions are given in section [ sec : end ] . the phenix detector at rhic is multi - purposed and optimized for precision measurements of electromagnetic probes for relativistic hadronic and heavy ion collisions . a complete overview of the detector can be found in ref . the data presented here are from 2006 and 2008 data taking at rhic . figure [ fig : phenix ] shows a schematic of the detector during those years . this analysis uses the central spectrometer arms for electron detection and the forward rapidity muon spectrometer arms , labeled north and south in fig . [ fig : phenix ] , for muon identification . for the 2008 collisions , the deuteron beam moves toward the north arm , which defines positive rapidity for both and . the forward produced muons come from a high-@xmath25 parton in the deuteron interacting with a low-@xmath25 parton in the gold . pythia@xcite indicates that the average @xmath25 of a parton producing a heavy flavor muon from 1@xmath32 gev/@xmath3 in the forward muon spectrometer is about @xmath33 . this analysis focuses only on the muons measured in the north arm utilizing the deuteron beam as a probe of low-@xmath25 partons in the gold nucleus . the central spectrometer comprises two arms subtending @xmath34 in azimuth and covering @xmath1 . charged tracks are measured using a drift chamber ( dc ) and a set of multi - wire proportional chambers with pad readout ( pc1 and pc3 ) . the dc measures the bend angle in the @xmath35 plane due to a central magnetic field directed along the beam axis . pc1 is used to measure the longitudinal coordinate of the track . these tracks are then projected into pc3 , where a hit is required to ensure high track quality . the momentum resolution of the tracks in this data is @xmath36%@xmath371.16%@xmath7 , where @xmath7 is the total momentum measured in gev/@xmath3 . electrons can be identified from associated hits in the ring imaging erenkov ( rich ) detector and the electromagnetic calorimeters ( emcal ) . electrons above 17 mev/@xmath3 passing through the co@xmath38-filled rich will emit erenkov radiation . the emcal comprises eight sectors , six of lead - scintillator and two of lead - glass , used to collect the energy from electron and photon showers . the nominal energy resolution for the lead - scintillator and lead - glass is 8.1%@xmath39}}\oplus$]2.1% and 6.0%@xmath40}}\oplus$]0.9%@xcite , respectively . the north muon spectrometer is located at @xmath41 and covers 2@xmath42 in azimuth . the spectrometer measures tracks in the muon tracker ( mutr ) and the muon identifier ( muid ) . prior to entering the muon arm , particles pass through approximately 20 cm of copper and 60 cm of iron . particles that are not absorbed pass through the mutr , which comprise three stations of cathode strip chambers with multiple ionization regions and located inside a radial magnetic field . after the mutr , particles pass through the muid , which comprises five alternating steel absorbers and muid detector planes , called gaps , with iarocci tubes . muid roads reconstructed from muid hits are projected back to mutr tracks and to the measured vertex to provide the complete information for a track through the spectrometer . trigger and global event characterization in and are provided by the beam - beam counter ( bbc ) . the bbc is a set of 64 hexagonal erenkov counters located from @xmath43 and covering full azimuth . the vertex of the collision along the beam line ( @xmath44 ) is determined by the time difference between the bbcs on either side of the collision region . the minimum bias ( mb ) trigger requires that there is at least one hit in each of the bbcs . from vernier scans and verified by monte carlo studies , the bbc mb trigger is sensitive to 55@xmath125% of the inelastic cross section and 88@xmath124% of the inelastic cross section@xcite . the trigger used for this analysis is a combination of the bbc trigger and a deep muon trigger . the deep muon trigger requires three or more muid gaps with a signal in both the @xmath25 and @xmath45 direction tubes and that the last pair of hits be in the last ( 5th gap ) or next to last gap ( 4th gap ) . after quality cuts and requiring a vertex within 25 cm of the @xmath20=0 vertex , an integrated luminosity of 2.1 pb@xmath46 in and a -equivalent of 7.7 pb@xmath46 in was sampled . the primary goal of this analysis is to identify @xmath47 where the opposite sign electron - muon pair is from the @xmath17 pair decay . only muon candidates with @xmath48 gev/@xmath3 are used in the analysis because real muons with total momentum less than about 2.7 gev/@xmath3 are stopped in the muon arm before reaching the 5th ( and last ) gap . single muon candidates are constructed from muid roads projected and matched to mutr tracks . cuts on muid roads and mutr tracks are designed to reject hadrons that mimic a muon signal and to reject tracks that did not originate from the collision vertex . for the muid roads , at least three of five gaps with @xmath49 hit information are required , including a pair of hits in the 5th gap . these muid roads must project back near the nominal vertex position . those muons that do not typically come from beam - related backgrounds . for the mutr tracks , cuts that reject hadrons are detailed in ref . the muid roads are then projected and matched to mutr tracks at the 1st muid gap . an identified muon candidate is the closest mutr track that matches a muid road within at least 10@xmath50 in slope and 10 cm in distance . muon candidates are further restricted to @xmath4 . during both the and data taking periods , there were backgrounds primarily from beam - related particles interacting with material in the accelerator upstream of phenix . collimators were used in the accelerator to reduce this background but it was not totally eliminated . restricting the @xmath51 range of the muon candidates helped minimize this background . electrons with @xmath2 gev/@xmath3 are identified by matching a track in dc , pc1 , and pc3 to a signal in the rich and a cluster in the emcal . the relevant details on measuring electrons in phenix are given in ref . @xcite . for this analysis , the projected track must match within 3@xmath52 in position to a cluster in the emcal . clusters are also required to have a matching profile , when compared to an electromagnetic shower shape profile at the measured energy . once a track matches both the rich and the emcal , an @xmath53 cut is applied , where it is required that the energy measured in the the emcal @xmath54 be approximately equal to the reconstructed track momentum @xmath7 . this is sufficient to remove most combinatorial matches and background from real electrons resulting from long - lived particle decays occurring near the dc , which have mismeasured momentum . a cut of -2@xmath52 to + 3@xmath52 from the mean @xmath53 in the data and -1.5@xmath52 to + 3@xmath52 from the mean in the data is applied . the asymmetry of the cuts is due to the dominance of backgrounds below 2 or 1.5@xmath52 of the mean . the tighter cut in the data was necessary because of the increased background from the hadron blind detector ( hbd ) support material not present during 2006 data taking . after particle identification cuts have been applied to an event , all pairs of identified electrons and muons are formed in each of the four charge - sign combinations . the fully corrected invariant - pair yield , calculated for each sign combination , is @xcite @xmath55 where @xmath56 is the number of sampled bbc triggered events , @xmath3 is the mb trigger bias accounting for events missed by the bbc trigger@xcite , @xmath57 and @xmath58 are the rapidity ranges of the electrons and muons , respectively , @xmath59 is the inclusive electron - muon pair yield , and @xmath60 is the mixed - event electron - muon pair distribution . the two - particle acceptance and efficiency is corrected by the mixed - event technique , where electrons from one event are paired with muons from a different event . pools of inclusive electrons and muons are kept in bins 2.5 cm - wide @xmath20 vertex bins , and , in the case of , 10%-wide centrality bins . when mixing events , the pair distribution is weighted by the @xmath45- and @xmath61-averaged efficiency of each particle , @xmath62 and @xmath63 . both @xmath62 and @xmath63 were determined by generating single electrons and single muons with a flat distribution in @xmath28 , @xmath61 , @xmath64 or @xmath65 and collisions @xmath20-vertex location and running them through a geant-3 simulation of the phenix detector . the output was subjected to the same analysis cuts applied to the data . the efficiency is defined as the ratio of particles reconstructed through the analysis to the number simulated . these simulations demonstrated that @xmath62 and @xmath63 are independent of the @xmath20-position of the event vertex , @xmath66 is independent of @xmath51 and @xmath67 has a slight @xmath51-dependence . pair yields are reported with the average pseudorapidity @xmath68 , which include the @xmath51-dependence of both single inclusive muons and the single particle efficiency . inclusive muon and electron candidates come from both heavy- and light - flavor decays and from misidentified hadrons . the fully - corrected inclusive electron - muon pair yield for each sign combinations can be written as @xmath69 here @xmath70 indicates the fully - corrected inclusive pair yield defined in eq . [ eq : pairyield ] ; @xmath71 is the fully - corrected pair yield produced from a heavy flavor pair decay ; @xmath72 is the fully - corrected pair yield from correlating a heavy flavor decay product with a light flavor decay product ; and @xmath73 is the fully - corrected pair yield from correlating pairs of light - flavor decay products or misidentified hadrons . pairs from the semileptonic decay of a @xmath17 pair have opposite signs . [ eq : basecorr ] can be decomposed into its like- and unlike - sign pieces : @xmath74 while semileptonic decays of @xmath75 can also produce both like- and unlike - sign @xmath21@xmath13 signals , in this analysis , pythia indicates that only about 1% of the final heavy - flavor @xmath21@xmath13 pair yield is from @xmath75 and is neglected . if we assume muon ( electron ) candidates from light flavors are not charge - correlated with electron ( muon ) candidates from light flavors , then @xmath76 if only one of the pair is from heavy flavor , then again , we assume they are not charge correlated , and @xmath77 therefore , the heavy flavor @xmath21@xmath13 signal distributions is the difference between the unlike - sign and the like - sign inclusive correlations : @xmath78 figure [ fig : sb ] shows the fully - corrected inclusive like - sign ( @xmath79 ) and unlike - sign ( @xmath80 ) @xmath21@xmath13 pair distributions in and . the inset figures show the signal - to - background distributions given the assumptions above . ) and unlike - sign ( @xmath81 ) distributions for ( a ) and ( b ) , as a function of @xmath6 . the inset shows the unlike - like difference divided by the like - sign distribution , which is the heavy flavor signal - to - background in the inclusive unlike - sign distribution . , title="fig : " ] ) and unlike - sign ( @xmath81 ) distributions for ( a ) and ( b ) , as a function of @xmath6 . the inset shows the unlike - like difference divided by the like - sign distribution , which is the heavy flavor signal - to - background in the inclusive unlike - sign distribution . , title="fig : " ] we have checked the like - sign subtraction method using pythia leading order quantum chromodynamics ( qcd ) events . with all events containing a heavy quark in the final state removed , the pair yields as a function of @xmath6 for like - sign and unlike - sign electron - muon pairs were the same within 3% over all @xmath6 . while this corroborates the basic idea of the subtraction , the assumption was further tested with data . in the following sections we detail the results of different methods to tag electrons and muons from light flavor decay to examine the validity of eq . [ eq : subtrcorr ] and to quantify the systematic uncertainty of the method . the general method is to use a sample of single electrons paired with single muons , where one or both are likely from light - hadron decays . if the method is correct , the like - sign subtraction should produce no correlation at all . if there are statistically significant correlations after like - sign subtraction , these are subtracted from the final @xmath21@xmath13 pair yield and uncertainties on the residual correlation strength are propagated as a systematic uncertainty on the final pair yield . if no statistically significant yield is found after like - sign subtraction , the statistical uncertainty on the zero yield is propagated as the systematic uncertainty . one source of background to the single muons is from hadrons that penetrate to the 5th gap , called punch - through hadrons . after single particle cuts there is some small fraction ( roughly 1 out of every 250 @xcite ) of candidate tracks with @xmath48 gev/@xmath3 that are hadrons that punch through . while this represents an irreducible background to the single muons , we can obtain a clean sample of hadrons that punch through and stop in the 4th gap of the muid . [ fig : pz ] shows the @xmath82 distribution of muon candidates that stop in the 4th gap . the peak at 2.3 gev is composed of muons that have insufficient energy to penetrate further . the broader portion of the distribution comprises light hadrons that are not stopped by the upstream absorber materials but are subsequently absorbed in the steel just after the 4th gap , thus not leaving a hit in the 5th gap . we identify punch - through hadrons as having stopped in the 4th gap with @xmath82 larger than 3 gev . for tracks that stop in the next - to - last muid gap ( 4th gap ) . the peak at lower @xmath82 is due to muons , while the broad distribution is from hadrons that punch through the absorber to the 4th gap . the solid line is a two - gaussian fit to this distribution with the solid line indicating the hadronic background in the muon peak region . ] /ndf = 22.7/24 . in it is a flat line and a gaussian centered at @xmath42 with @xmath83/ndf = 26.3/22.,title="fig : " ] /ndf = 22.7/24 . in it is a flat line and a gaussian centered at @xmath42 with @xmath83/ndf = 26.3/22.,title="fig : " ] fig . [ fig : punch - throughcorr ] shows the fully corrected like - sign subtracted pair yield of central - arm electrons and the punch - through hadrons in the muon arms for both and collisions . if both the like- and unlike - sign pair yields were dominantly from light hadron decays , the like - sign subtraction should produce zero pair yield . to determine the magnitude of the residual correlation strength after like - sign subtraction , the data were fitted with a flat line . this is shown as the solid line in fig . [ fig : punch - throughcorr]a . the fit uncertainty is shown as the shaded band around the solid line . the flat fit in had a @xmath83/ndf of 22.7/24 and gave a value that was nonzero with greater than 1@xmath52 significance . this means there is yield in the final @xmath21@xmath13 correlations from these punch - through hadrons . the fitted yield was subtracted from the final pair yield and its uncertainty was propagated as a systematic uncertainty on the final pair yield . for the case , we fitted the residual correlation to a flat line and found reasonable agreement with a @xmath83/ndf of 30.9/24 or a @xmath7-value of 14% . however , there is a possible excess of counts near @xmath14 , which when included as a gaussian component fixed at @xmath14 and the width and yield as free parameters , a slightly better @xmath83/ndf of 26.3/22 or a @xmath7-value of 26% was found . if there is any correlated yield beyond a pedestal , it would show up in the back - to - back peak . therefore , we subtract the gaussian fit , shown as the solid line in fig . [ fig : punch - throughcorr ] from the final pair yield and propagate the uncertainty , shown as the shaded region around the solid line , on the fit to the systematic uncertainty in the final pair yield . two additional corrections to this data are applied before subtraction from the final pair yield . because the punch - through hadrons are measured in the 4th gap , the yields need to be scaled to match the rate of hadrons at the last gap . the rate of hadrons at the 5th gap was determined by using pion and kaon nlo perturbative qcd spectra @xcite and passing them through a geant-3 model of the phenix muon arms . the muid absorber steel cross section was modified until there was agreement between data and the simulation for the rate of punch - through hadrons in the 3rd and 4th gap . we extrapolated to the 5th gap and find the rate of hadrons is 2.81@xmath120.30 times the rate of punch - through hadrons in the 4th gap @xcite . the 3 gev/@xmath3 @xmath82 cut removes some fraction of the punch - through hadrons . based on the two - component fit to the @xmath82 distribution shown in fig . [ fig : pz ] , the yield is scaled up to account for those hadrons rejected by the @xmath82 cut . in the end , the pair yield uncertainty is 2.17@xmath8410@xmath85 ( rad)@xmath46 in . in there is a @xmath6-independent uncertainty on the final pair yield that is 1.42@xmath8410@xmath86 ( rad)@xmath46 and the gaussian uncertainty that ranges from 0 to 6.30@xmath8410@xmath86 ( rad)@xmath46 . one source of real muons is from decays of light hadrons , predominantly charged pions and kaons , before the absorber material . the observed rate of muons into the north arm is higher , when the collision vertex is farther from the spectrometer arm . because heavy flavor decays ( including drell - yan , heavy quarkonia , etc . ) have a much shorter @xmath87 than light flavor decays , heavy flavor decay muons have a much weaker vertex dependence . therefore , we assume there are two components to the muon rate : a component that follows the primary vertex distribution , attributable to heavy flavor decays , and a component that folds the linear component due to light hadron decays with the primary vertex distribution . pair yield in ( a ) and ( b ) collisions . both are consistent with no residual correlation after like - sign subtraction . the solid lines and shaded bands indicate the flat line fits and their uncertainty with @xmath83/ndf of 27.1/24 and 18.0/24 in and , respectively . , title="fig : " ] pair yield in ( a ) and ( b ) collisions . both are consistent with no residual correlation after like - sign subtraction . the solid lines and shaded bands indicate the flat line fits and their uncertainty with @xmath83/ndf of 27.1/24 and 18.0/24 in and , respectively . , title="fig : " ] muons that are near the detector ( @xmath88 cm ) and far from the detector ( @xmath89 cm ) , where @xmath44 is the measured collisions vertex , are separately correlated with central arm electrons . because the signal heavy flavor muons follow the primary collision vertex distribution , subtracting the near - vertex pair yield from the far - vertex pair yield , should remove these and only residual correlations from decay muons should be present . the pair yields in and after subtracting near- and far - vertex muons and after like - sign subtraction are shown in fig . [ fig : decaycorr ] . the correlations are consistent with a flat line with zero yield with a @xmath83/ndf of 18.0/24 . the data seems to have a residual shape . however , this shape is asymmetric about @xmath6 of zero and is not physical . therefore , we fit with a flat line that results in zero correlation yield and a @xmath83/ndf of 27.1/24 . the fits are shown in fig . [ fig : decaycorr ] as solid lines and shaded bands indicating the statistical uncertainties . these uncertainties were propagated into the systematic uncertainties of the final pair yields . to propagate the uncertainties , additional corrections are needed . first , in the far - near subtraction , some fraction of the decay muons are removed . second , light hadron decays outside the @xmath1230 cm vertex cut are not counted in the subtraction . to account for both effects , a fit to the vertex dependence of the muon yield is extrapolated to a point one interaction length inside the absorber , a distance of about 56 cm from the nominal @xmath20 vertex and about 16 cm into the absorber . it is assumed that the decay contribution to the muons is negligible at that point , which fixes the fraction of muons that are from light decays within the measured vertex window of the analysis . under this assumption , only 22% of the decay muons are measured within the vertex window after the like - sign subtraction . the fit uncertainties are increased to account for those muons not measured . the final systematic uncertainties on the final pair yield are 1.13@xmath8410@xmath86 ( rad)@xmath46 and 5.05@xmath8410@xmath86 , independent of @xmath6 for and , respectively . electrons can result from light hadron decays through internal and external photon conversions . the dominant photonic source of electrons are from @xmath22 decays . we assume that , if we measure the @xmath22-decay electrons correlations with muons , this will represent the other photonic sources ( such as @xmath51 and @xmath90 decay ) in shape and yield . to tag decay or converted electrons , we construct the invariant mass distribution of all pairs of electrons and photons in an event . electrons paired with photons within the @xmath22 mass peak are then correlated with muon candidates . the signal - to - background of pairs in the @xmath22 mass range is about one . to remove correlations from combinatorial electron - photon pairs that fall within the @xmath22 mass window , muon candidates were also correlated with the @xmath21@xmath91 pairs in a `` sideband '' @xmath22 mass region from 0.20.4 gev/@xmath92 . after scaling by the appropriate signal - to - background under the @xmath22 mass region , the `` sideband '' correlations were subtracted from the in - mass electron - muon correlations for each of the @xmath21@xmath13 charge types . [ fig : photoniccorr ] shows the `` sideband''-subtracted and like - sign subtracted correlation between electrons tagged in the @xmath22 mass region with muons from and data . flat fits to these correlations produced a yield consistent with zero with @xmath83/ndf of 33.2/24 and 20.2/24 in and data , respectively . the statistical uncertainty from the fitted yield to the @xmath22-tagged correlations is a factor of 10 smaller than the other background correlations after accounting for reconstruction efficiency and additional sources of photonic electrons . this uncertainty is negligible compared to those from the muon backgrounds . pair yield in ( a ) and ( b ) collisions . both are consistent with no residual correlation after like - sign subtraction . the solid lines and shaded bands indicate the flat line fits and their uncertainties with @xmath83/ndf of 33.2/24 and 20.2/24 in and , respectively . , title="fig : " ] pair yield in ( a ) and ( b ) collisions . both are consistent with no residual correlation after like - sign subtraction . the solid lines and shaded bands indicate the flat line fits and their uncertainties with @xmath83/ndf of 33.2/24 and 20.2/24 in and , respectively . , title="fig : " ] in this analysis there are three general types of uncertainty that we identify as type a , point - to - point uncorrelated , type b , point - to - point but correlated , and type c , total normalization uncertainty . except for statistical uncertainties there are no type a uncertainties in this analysis . the type b uncertainties are from the subtraction of known backgrounds discussed in section [ sec : subtr ] . the fully corrected pair yield uncertainties in are @xmath93 ( rad)@xmath46 and @xmath94 ( rad)@xmath46 from punch - through hadron and decay hadron subtraction uncertainties , respectively . these values are independent of @xmath6 . in the flat - line fit contributions to the systematic uncertainty are @xmath95 ( rad)@xmath46 and @xmath96 ( rad)@xmath46 from punch - through hadron and decay hadron subtraction uncertainties , respectively . the additional uncertainty from the gaussian fit to the punch - through hadron correlations resulted in a @xmath6-dependent uncertainty ranging in absolute value of 0 at @xmath97 2 rad to @xmath98 ( rad)@xmath46 at @xmath99 . the type - b systematics are summarized in table [ tab : sys ] . cccc type & description & & + b & @xmath6 dependent & & 0%6.30@xmath8410@xmath86(rad)@xmath46 + b & punch - through & @xmath93 ( rad)@xmath46 & @xmath95(rad)@xmath46 + b & decay muons & @xmath94 ( rad)@xmath46 & @xmath96(rad)@xmath46 + + c & muon cuts & 7.8% & 8.3% + c & electron cuts & 8.3% & 9.3% + c & muon efficiency & 2.2% & 2.2% + c & electron efficiency & 1.0% & 1.0% + c & trigger efficiency & 11.1% & 4.2% + + c & total & 16.1% & 13.4% + the type c uncertainties are attributable to several sources and are given in table [ tab : sys ] . one source of systematic uncertainty is evaluated by tightening the single particle cuts for this analysis . each single particle cut was tightened independently and the analysis , including reevaluation of the single particle efficiency , was performed . the uncertainty from each of the individual single particle cuts was combined using the correlation amongst the cuts . the values of these are different in and data , because of the higher backgrounds in collisions . however , these uncertainties are highly correlated between and , because the same cuts are applied to both data sets . another source of uncertainty is in the evaluation of the single particle efficiencies . the single particles were generated flat in @xmath28 and then weighted to match the measured phenix heavy flavor lepton spectra @xcite . for the uncertainty determination , the single particle efficiency was re - evaluated without the weighting applied . this was estimated to be 1.0% for the electrons and 0.8% for the muons . for muons there is an additional 2.0% uncertainty due to the run - by - run variation in muon acceptance . the final portion of the type c systematic uncertainty is due to the trigger efficiency . to evaluate this uncertainty , the data were analyzed for several data - taking periods defined by the muon trigger performance . the difference in fully - corrected yields between data sets was taken to be the uncertainty in the muon trigger efficiency . this is combined with the uncertainties in the bias factor @xmath3 in eq . [ eq : pairyield ] . the total uncertainty for the trigger is 11.1% for and 4.2% for . as indicated in table [ tab : sys ] , combining all type c uncertainties gives 16.1% for data and 13.4% for data . the fully - corrected like - sign subtracted @xmath21@xmath13 pair yield as a function of @xmath6 for electrons with @xmath2 gev/@xmath3 and @xmath100 , with opposite - signed forward muons with @xmath5 gev/@xmath3 and @xmath4 , in is shown in fig . [ fig : ppcorr ] . the average muon @xmath51 in these correlations is 1.75 . the error bars are statistical uncertainties only , while the boxes are the type b systematic uncertainties . we note that the distribution has two components : a nonzero continuum as well as a back - to - back peak near @xmath101 . @xmath13 pair yield in . the error bars are statistical only . the boxes show the type b systematic uncertainty from the punch - through hadron and light hadron decay muon background subtraction . the 16.1% type c systematic uncertainty is not shown . ] points ) with heavy flavor production in powheg ( [ blue ] dashed line ) , pythia ( [ black ] solid line ) and mc@nlo ( [ green ] long dashed line ) . the @xmath21@xmath13 pair yield from the subset of pythia events , when the @xmath17 is not produced at the event vertex is plotted as the dotted [ black ] line . each monte carlo curve was scaled by a single parameter to match the observed yield . the resulting cross sections are consistent with the previously measured phenix results ( see table [ tab : xsect ] ) . ] to interpret these data , we compare the results to several different monte carlo generators , pythia , powheg@xcite , and mc@nlo@xcite . the pythia mb qcd events were generated to model the lo gluon fusion process and also model next - to - leading order processes , like flavor excitation and gluon splitting . events with a @xmath17 pair and an electron and a muon in the measured kinematic range as the corrected data ( @xmath102 gev/@xmath3 and @xmath103 , @xmath104 gev/@xmath3 and @xmath105 ) were correlated and a like - sign subtraction was performed . an overall scale factor was used to fit the pythia curve to the data . in the fit , the @xmath83 was calculated for different scale parameters using the statistical error on the data . we report the cross section for the scale factor that minimizes that @xmath83 and report a statistical error on the cross section as the value that changes the @xmath83 by one unit . to evaluate the systematic uncertainty on the cross section , the data were increased and decreased by their combined type b and type c systematic uncertainty and the process to determine the scale factor by finding a minimum @xmath83 using the statistical uncertainty in the data was repeated . we find the pythia correlation is consistent with the data with a @xmath17 cross section of @xmath106 = 340@xmath1229(stat)@xmath12116(syst ) @xmath13b with a @xmath83/ndf of 20.5/24 . this is shown as the solid curve in fig . [ fig : compare ] . the other model comparisons are from nlo generators , powheg and mc@nlo . events were generated to produce the hard scattering heavy flavor event vertex and then interfaced to pythia , which performed the fragmentation and underlying event generation . the qualitative features of the data are present in these correlations : the continuum and the back - to - back peak . as described for the pythia fit , a single scale parameter was used to calculate a @xmath83 between the generated @xmath21@xmath13 correlations and the data using the data s statistical uncertainty . the resulting best fits for powheg and mc@nlo are shown in fig . [ fig : compare ] as the short dashed and the long dashed lines , respectively . the extracted cross sections are @xmath106 = 511 @xmath12 44 ( stat ) @xmath12 198 ( syst ) @xmath13b with @xmath83/ndf of 23.5/24 for powheg and @xmath106 = 764 @xmath12 64 ( stat ) @xmath12 284 ( syst ) @xmath13b with @xmath83/ndf of 19.2/24 for mc@nlo . we combine the cross sections from the three models and report a measured cross section of @xmath11 538 @xmath12 46 ( stat ) @xmath12 197 ( data syst ) @xmath12 174 ( model syst ) . the central value of the cross section is the average of the three model cross sections , while the model systematic uncertainty is the standard deviation of the three model cross sections . this value can be compared with previous phenix measurements . from the heavy flavor electron spectra at midrapidity , phenix found @xmath11 567 @xmath12 57 ( stat ) @xmath12 224 ( syst ) @xcite and from the dielectron mass spectrum at midrapidity , phenix extracted @xmath11 554 @xmath12 39(stat ) @xmath12 142 ( data syst ) @xmath12 200 ( model syst ) @xcite . within the data systematics the value extracted here is consistent with previously published phenix results . cccc & description & @xmath106 ( @xmath13b ) & + & pythia @xmath21@xmath13 & 340@xmath1229(stat)@xmath12116(syst ) & + & powheg @xmath21@xmath13 & 511@xmath1244(stat)@xmath12198(syst ) & + & mc@nlo @xmath21@xmath13 & 764@xmath1264(stat)@xmath12284(syst ) & + & combined @xmath21@xmath13 & 538@xmath1246(stat)@xmath12197(data syst)@xmath12174(model syst ) & + + & phenix single @xmath107 @xcite & 567@xmath1257(stat)@xmath12224(syst ) & + + & phenix dilepton ( @xmath26 ) @xcite & 554@xmath1239(stat)@xmath12142(data syst)@xmath12200(model syst ) & + [ tab : xsect ] using the pythia event record , it is possible to separate the @xmath17 production into an lo component , where the @xmath108 and a component from the pythia model of nlo mechanisms of flavor excitation and gluon splitting , where the @xmath17 pair is produced in the initial or final - state shower . the `` pythia ( no lo ) '' dashed line in fig . [ fig : compare ] shows the correlations from the sample of produced pythia events , where the @xmath17 were not generated in the primary event vertex of pythia . the back - to - back peak at @xmath14 is dominated by the lo gluon fusion process while the continuum is due to the correlations from the higher order processes . from an accounting from pythia , we find that 32% of the @xmath21@xmath13 pair yield results from gluon fusion , consistent with the expectations from charm production @xcite . throughout the analysis it has been assumed that semileptonic @xmath17 decay is the dominant contribution to the correlations . however , @xmath75 semileptonic decays would produce a signal in both the like- and the unlike - sign pair distributions . up to four semileptonic decays can occur where @xmath109-quarks semileptonically decay to @xmath3-quarks , which subsequently semileptonically decay . we have used pythia and powheg to check these contribution from bottom . in both cases , for electrons and muons in the kinematic region that we measure , the bottom contribution is about a factor of 100 below the charm yield . this is further corroborated by the phenix heavy flavor electron measurements that show that bottom becomes significant only at @xmath28 above 3 gev/@xmath3 @xcite . in this analysis only 3% of the sampled electrons have a @xmath28 above 3 gev/@xmath3 , so we expect that the bottom contribution is negligible in this measurement especially compared to the background subtraction systematic uncertainties . @xmath13 pair yield in . the error bars are statistical only . the boxes show the type b systematic uncertainty from the punch - through hadron and light hadron decay muon background subtraction . the 13.4% type c systematic uncertainty is not shown . ] the fully - corrected like - sign subtracted pair yield as a function of @xmath6 for electrons with @xmath102 gev/@xmath3 and @xmath103 with forward muons with @xmath111 gev/@xmath3 and @xmath105 in 0%100% , corresponding to the total inelastic cross section , is shown in fig . [ fig : daucorr ] . a nonzero correlations strength is observed . however , unlike the data , there is a much less distinct back - to - back peak near @xmath6 of @xmath42 . [ fig : paircorr ] shows the overlay of the and pair correlations . the pair correlations are scaled by the @xmath112 @xcite . the peak in is suppressed compared to , indicating a medium modification to the yield per collision in . @xmath13 pair yield in ( [ red ] circles ) @xmath113-scaled ( [ blue ] boxes ) , shifted in @xmath6 for clarity . the bars are statistical uncertainty . the boxes are the type b systematic uncertainty from the decay and punch - through background subtraction . the overall normalization uncertainties of 16.1% and 13.4% in and , respectively , and 5.7% uncertainty from @xmath114 are not included . ] is plotted as a function of @xmath6 . the vertical bars are statistical uncertainties , the black boxes are the type b systematic uncertainties , and the gray band around 1.0 on the left is the type c systematic uncertainty . the type b systematics are symmetric around the central value but in some cases are outside the range of the plot . ( b ) alternative rebinning of the data in ( a ) . , title="fig : " ] is plotted as a function of @xmath6 . the vertical bars are statistical uncertainties , the black boxes are the type b systematic uncertainties , and the gray band around 1.0 on the left is the type c systematic uncertainty . the type b systematics are symmetric around the central value but in some cases are outside the range of the plot . ( b ) alternative rebinning of the data in ( a ) . , title="fig : " ] to quantify the difference between and yields , we calculate the ratio @xmath16 defined as the ratio of a pair yield in to the @xmath114-scaled pair yield in . @xmath115 any deviation from unity of this ratio would indicate modification to the yield . when taking this ratio several systematic uncertainties common to and cancel . these are dominantly from identical cuts used in the analyses with the same systematic uncertainties . the noncanceling type c systematic uncertainties in the and yields are 7.7% and 8.9% , respectively . [ fig : jdadphi]a shows a plot of @xmath16 as a function of @xmath6 for all bins in @xmath6 . the bars are statistical uncertainties and the type b systematic uncertainties are plotted as boxes . the noncanceling type - c uncertainty is 14.1% and is indicated by the shaded box around one on the left . while the points near @xmath116 are consistent with unity with large error bars , the points near @xmath14 , where @xmath16 is about 0.4 . for clarity , [ fig : jdadphi]b shows a rebinning of fig . [ fig : jdadphi]a . we find @xmath117 for the bin near @xmath14 . this value is 3.5@xmath52 different from unity after combining the statistical and systematic uncertainties . these results show that , in the measured kinematics , charm pairs are modified in the cold nuclear medium . these results are in a different kinematic region than either the single electrons , which are enhanced at midrapidity @xcite , or the single muons , which are suppressed at forward rapidity @xcite . from the pythia simulation , the @xmath21@xmath13 correlations arise from partons in the gold nucleus with @xmath118 at @xmath119 , on the edge of the shadowing region . as discussed in section [ sec : yields ] , the back - to - back peak is dominated by leading order gluon fusion , while the continuum is dominated by other processes like flavor excitation and gluon splitting . the observed back - to - back peak and pedestal in and should help lead to an understanding of the mechanism or mechanisms responsible for the modification . for example , the back - to - back peak is dominated by low-@xmath25 gluons participating in the hard scattering , whereas the continuum has a larger contribution of quarks participating in the hard scattering . quarks are probably less shadowed than gluons at the @xmath25 and @xmath120 where this analysis is measured . it is possible that there are kinematic differences between the final state charm quarks in the peak and the continuum . these differences could affect the amount of final state energy loss and multiple scattering that modify the measured pair yields . it may be possible to combine these results with other cold nuclear matter charm measurements to disentangle the effects of shadowing , saturation , and energy loss . we presented phenix results for heavy flavor production of azimuthally - correlated unlike sign @xmath21@xmath13 pairs in and collisions at @xmath121 of 200 gev . the yield shows a nonzero continuum as well as a back - to - back peak structure centered at @xmath101 . when compared with several models , we find the charm cross section @xmath11 538 @xmath12 46 ( stat ) @xmath12 197 ( data syst ) @xmath12 174 ( model syst ) @xmath13b . this is also consistent with previously measured @xmath17 cross sections at this center of mass energy . in collisions a yield reduction in the back - to - back peak is observed , where we measure @xmath122 = 0.433 @xmath12 0.087 ( stat ) @xmath12 0.135 ( syst ) . this indicates that the nuclear medium modifies the @xmath17 correlations . such a suppression could arise due to nuclear pdf shadowing , saturation of the gluon wavefunction in the au nucleus , or initial / final state energy loss and multiple scattering . we thank the staff of the collider - accelerator and physics departments at brookhaven national laboratory and the staff of the other phenix participating institutions for their vital contributions . we acknowledge support from the office of nuclear physics in the office of science of the department of energy , the national science foundation , a sponsored research grant from renaissance technologies llc , abilene christian university research council , research foundation of suny , and dean of the college of arts and sciences , vanderbilt university ( u.s.a ) , ministry of education , culture , sports , science , and technology and the japan society for the promotion of science ( japan ) , conselho nacional de desenvolvimento cientfico e tecnolgico and fundao de amparo pesquisa do estado de so paulo ( brazil ) , natural science foundation of china ( p. r. china ) , ministry of education , youth and sports ( czech republic ) , centre national de la recherche scientifique , commissariat lnergie atomique , and institut national de physique nuclaire et de physique des particules ( france ) , bundesministerium fr bildung und forschung , deutscher akademischer austausch dienst , and alexander von humboldt stiftung ( germany ) , hungarian national science fund , otka ( hungary ) , department of atomic energy and department of science and technology ( india ) , israel science foundation ( israel ) , national research foundation and wcu program of the ministry education science and technology ( korea ) , physics department , lahore university of management sciences ( pakistan ) , ministry of education and science , russian academy of sciences , federal agency of atomic energy ( russia ) , vr and wallenberg foundation ( sweden ) , the u.s . civilian research and development foundation for the independent states of the former soviet union , the us - hungarian fulbright foundation for educational exchange , and the us - israel binational science foundation .

setup for performing homodyne detection of matter - wave fields . a gas of identical , bosonic atoms is confined in two distinguishable regions of space . in the left region it forms a bose - einstein condensate , that acts as reference system for measuring the condensate fraction of the right system . a pump - probe laser setup outcouples the atoms from both wells . the photon flux of the probe beam is constituted by the photon scattered in the outcoupling process and exhibits oscillations whose amplitude depend on the condensate fraction of the right system , and which vanish when this is zero . the first - order correlation function of a quantum gas can be determined in an extension of this setup , where the shaded regions denoted by @xmath0 and @xmath1 correspond to the spatial regions of the quantum gas trapped in a single well . in this case , the amplitude of the photon flux oscillations provides a measurement of the spatial correlation function as a function of the distance @xmath2 . detail on the parameters are reported in the text . , scaledwidth=34.0% ] the measurement technique , which is mostly employed in experiments with ultracold atoms , is based on time of flight @xcite . used in different setups , it allows one to determine the quantum state of matter @xcite . the most recent demonstration of in - situ imaging of atoms in optical lattices @xcite has opened further exciting directions for probing and manipulating atoms in periodic arrays @xcite . in addition , the remarkable progress in coupling ultracold atomic gases with high - finesse optical resonators @xcite has generated a renewed interest in revealing the correlation functions of matter by photodetection of the scattered light . this progress opens , amongst several others , the possibility to monitor properties of the quantum state of matter in a non - destructive way @xcite . photons emitted by rayleigh scattering , and in general in a pump - probe type of experiment , have been used in order to characterize the quantum state of atomic gases @xcite . they deliver information on the structure form factor of the quantum gas and thus on the density and density - density correlations @xcite . proposals for optical detection of certain quantum states of matter have appeared in the literature @xcite . to our knowledge , however , a setup has not yet been identified , which would allow one measure the mean value of the atomic field operator by means of the scattered light . this measure would be analogous to homodyne detection of light fields @xcite , and specifically allow one to determine the condensate fraction of a quantum gas @xcite by photodetection . in this article we propose a setup in which detection of the photons scattered by a quantum gas allows one to determine the mean value of the atomic field . the setup for determining the condensate fraction is sketched in fig . [ fig : homodynesetup ] , and is based on an interferometer for bose - einstein condensates ( bec ) realized in refs . we establish a direct connection between our theoretical model and the experimental setup in ref . @xcite , and then focus on a scheme in which the second system is not necessarily a bec . we then show that this setup allows one to determine the condensate fraction of the second system by measuring the flux of the scattered photon . furthermore , we argue that this scheme can be extended to determine the first - order correlation function of an atomic gas , when the two illuminated regions belong to the same quantum gas . the dynamics is based on first creating entanglement between the scattered photon and the scattering atom of the quantum gas , and then on performing an operation similar to a quantum eraser @xcite , thereby allowing one to measure the correlation functions of matter in the flux of the scattered light . this article is organized as follows . in sec . [ sec:1 ] the setup and the theoretical model are introduced . in sec . [ sec:2 ] a formal connection is derived between the photon flux and the correlation functions of the scattering atoms . in sec . [ sec:3 ] the photon flux is evaluated for three exemplary cases : when the scattering systems are two bose - einstein condensates at zero temperature , when they are at different temperatures , and when one of the systems is composed by ultracold atoms in a two - dimensional optical lattice . in sec . [ sec:4 ] it is discussed how this setup may allow one to measure the first - order correlation function . the conclusions are drawn in sec . [ sec:5 ] . the appendices report details of the calculations presented in secs . [ sec:2 ] , [ sec:3 ] , and [ sec:4 ] . a realization of the setup we consider in this paper is depicted in fig . [ fig : homodynesetup ] . here , ultracold bosonic atoms of mass @xmath3 are initially prepared in the stable electronic state @xmath4 and confined by the ( state - dependent ) potential @xmath5 . the atoms , that are illuminated by the lasers , are coherently coupled to the electronic state @xmath6 , which is stable and not confined : the outcoupled atoms propagate freely in space . in this section we introduce the hamiltonian and the physical quantities which are at the basis of our analysis . we denote by @xmath7 the hamiltonian governing the dynamics of photons and atoms , which we decompose into the sum @xmath8 with @xmath9 the hamiltonian governing the atoms dynamics in absence of interaction with the lasers , @xmath10 the hamiltonian for the free , transverse electromagnetic ( e.m . ) field , while @xmath11 describes the coupling between states @xmath4 and @xmath6 induced by coherent raman scattering . hamiltonian @xmath12 includes the dynamical stark shift due to off - resonant coupling between atoms and e.m .- field modes . these hamiltonians are reported in a second - quantized description of matter- and photon - fields . in detail , @xmath13 gives the energy of the transverse electromagnetic field in free space ( and is reported without the vacuum energy ) , where @xmath14 labels a mode of the e.m .- field at wave vector @xmath15 , polarization @xmath16 , and frequency @xmath17 , with @xmath18 the velocity of light . the operators @xmath19 and @xmath20 annihilate and create , respectively , a photon in mode @xmath14 , and obey the bosonic commutation relation @xmath21=\delta_{\lambda,\lambda'}$ ] . the atomic hamiltonian @xmath9 is conveniently rewritten as @xmath22 , where @xmath23 describes the atoms dynamics when the atoms are in the electronic state @xmath24 ( at frequency @xmath25 ) . here @xmath26 and @xmath27 are the annihilation and creation operator , respectively for a bosonic atom at position @xmath28 and in the electronic state @xmath29 , obeying the commutation relation @xmath30=\delta_{j , k}\delta({{\bf r}}-{{\bf r}}')$ ] . parameter @xmath31 is the strength of @xmath32-wave scattering between the atoms in state @xmath24 , with @xmath33 the @xmath32-wave scattering length in the corresponding electronic state @xcite . hamiltonian term @xmath34 describes @xmath32-wave scattering between an atom in electronic state @xmath4 and an atom in electronic state @xmath6 : we do not report its explicit form as we will consider situations for which these types of collisions can be neglected . the interested reader can find the corresponding term , for instance , in refs . @xcite . the hamiltonian terms describing the interactions between photons and atoms are given using normal ordering @xcite . they include a term describing coherent raman coupling between the states @xmath4 and @xmath6 , in which a photon in mode @xmath35 is scattered into mode @xmath36 and vice versa . raman transitions follow from a pump - probe type of excitation , and the corresponding hamiltonian reads @xmath37 \ , , \ ] ] with the position - dependent coupling strength @xmath38 . this interaction can be tailored by means of a laser , pumping the atoms , and a cavity mode , acting as a probe , that are both far - detuned from the atomic excited state but are set close to resonance with an atomic raman transition . for later convenience we denote by the quantity @xmath39 the difference between the frequencies of the two e.m.f .- modes . moreover , we assume @xmath40 , i.e. , the two lasers drive quasi - resonantly the raman transition coupling the stable electronic states @xmath4 and @xmath6 . the interaction with the pump and probe laser also induces a dynamical stark shift , whose corresponding hamiltonian reads ( in normally - ordered form ) @xmath41 where the parameter @xmath42 has the dimension of an angular frequency . it can be written as @xmath43 , with @xmath44 the rabi frequency of the mode coupling the transition @xmath45 and @xmath46 the detuning between the mode and the atomic transition frequencies ( the lamb shifts are included in the frequencies of the atomic states ) . in the following we will assume that the pump and probe fields are traveling waves with wave vectors @xmath47 . in this case , the raman - coupling strength @xmath48 reads @xmath49 with @xmath50 and @xmath51 . we will now provide more details on the potential which confines the atoms . we will assume that the atoms in state @xmath6 propagate freely , namely , @xmath52 in eq . ( [ eqn : hamoutcouplgen:1 ] ) , which we set equal to zero . the atoms in state @xmath4 are trapped by potential @xmath53 . this potential confines the atoms in two spatially separated regions around the two potential minima , which are at distance @xmath2 and , specifically , are localized at the points @xmath54 and @xmath55 . the potential can be then decomposed in the sum @xmath56 , with @xmath57 the potential center at @xmath58 and @xmath59 . the two atomic clouds at each well are initially uncorrelated , and there is no tunneling between the two spatial region . in principle , hence , the atoms in the two regions of space are distinguishable . it is then useful to consider the partition @xmath60 , where we denote by @xmath61 and @xmath62 the field operators which do not vanish on the left and right region of space , respectively , such that @xmath30=\delta_{j , k}\delta({{\bf r}}-{{\bf r}}')$ ] for @xmath63 @xcite . using these definitions , we can write @xmath64 , with @xmath65 and where we have set @xmath66 . with this representation , the hamiltonian describing the interaction with the lasers takes the form @xmath67 \,.\ ] ] we note that our model is an extension to the one studied in refs . @xcite , where the authors studied the outcoupling of atoms from a single bec by means of _ classical _ raman lasers . by considering the quantum dynamics of the interaction between photons and atoms , we take into account the quantum fluctuations of the light field due to the scattering process , which are discarded in @xcite . we will indeed show that these fluctuations give access to some correlation functions of the scattering system . we now study the properties of the scattered photons by considering the photon field operators in heisenberg picture . the heisenberg equations of motion are determined assuming that the coupling between matter and photons is sufficiently weak to be treated in second - order perturbation theory . in this limit the operators for the photonic modes read @xmath68 \psi_2^{(0)}({\bf r},\tau ) a_2(0 ) \ , , \\ \label{eqn : heisenberg:2 } a_2(t)&=&a_2(0 ) { \rm e}^{-{\rm i}(\omega_2 t+\phi_2(t))}-{\rm i}\,{\rm e}^{-{\rm i}\omega_2 t}\int_0^t { \rm d}\tau\ , { \rm e}^{-{\rm i}\omega_{12}\tau } \int { \rm d}{\bf r}\;\gamma({\bf r } ) \psi_2^{(0)\dagger}({\bf r},\tau)\bigl[\psi_l^{(0)}({\bf r},\tau)+\psi_r^{(0)}({\bf r},\tau)\bigr ] a_1(0 ) \ , , \end{aligned}\ ] ] where @xmath69 . the physical origin of the individual terms on the right - hand side ( rhs ) can be simply identified . the first term on the rhs of both equations is the free - field component . it is characterised by a time - dependent phase @xmath70 , which is proportional to the atomic density in the electronic state @xmath24 . when the born approximation can be performed , it is a density - dependent phase shift of the field mode , with the form @xmath71 since all atoms are initially prepared in the electronic state @xmath4 , this density - dependent shift can be measured in the frequency shift the field - mode @xmath35 after it has interacted with the atomic medium . this would allow a quantum - non - demolition measurement of the density of the medium integrated over the path along which light propagates @xcite . it is interesting to note that in the considered setup this shift contains also the operator @xmath72 and its adjoint . hence , its measurement would reveal tunneling events between the wells . this is however not relevant for the case considered in this work , since we assume that there is no tunneling between the two separate wells . the second term on the rhs establishes a direct proportionality relation between the photonic and the matter - wave fields @xmath73 and @xmath74 ( @xmath75 and @xmath76 ) . it shows , in particular , that the source term of field @xmath75 is the coherent overlap of the field scattered from the right and from the left wells . in this shape the outcoupling process in this system resembles a beam - splitter operation . this property establishes an analogy with an interferometric setup , which we will exploit in order to perform homodyne detection of matter - wave field . differing from a simple interferometer , however , atoms and photons are correlated by the scattering process . this important difference is at the basis of the dynamics we observe . before we discuss the signal at the photodetector , we will introduce few approximations which will notably simplify the treatment . in first place we neglect atomic collisions between outcoupled atoms in state @xmath6 , since we assume that the gas of outcoupled atoms is at very low densities . this regime also permits us to neglecting collisions between outcoupled atoms and trapped atoms in state @xmath4 @xcite . the quantity we analyse in this paper is the flux of photons of mode 2 , namely , the rate of change of the photon number in the mode at frequency @xmath77 . formally , the photon flux is given by the equation @xmath78 where the expectation value @xmath79 is taken over the initial state of the atoms and the e.m .- field . the photon flux , integrated over the detection time , gives the integrated intensity of the field at the detector . it can be verified that @xmath80 where the proportionality factor is the mean number of photons in mode 1 . this equality shows indeed that flux of scattered photons and of the corresponding outcoupled atoms carry the same information . for the specific setup we consider the photon flux can be rewritten as the sum of two contributions , @xmath81 with @xmath82 where @xmath83 is a scaling factor , proportional to the number of photons in mode 1 . here , @xmath84}\left\langle \psi_2({\bf r'},t ' ) \psi_2^{\dagger}({\bf r},t)\right\rangle \ , , \label{fun : f } \nonumber \\\end{aligned}\ ] ] where the initial state is the vacuum state of the e.m . field except for mode 1 and 2 , which are assumed to be in coherent states with non - vanishing photon number , while all atoms are in internal state @xmath85 and at equilibrium in the grand - canonical ensemble at temperature @xmath86 . the atoms in state @xmath6 propagate freely and do not undergo collisions , since we assume that the density is very low . therefore , eq . ( [ fun : f ] ) can be cast in the form @xmath87 } \ , , \ ] ] with @xmath88 the recoil frequency and @xmath89 the two - photon detuning . in order to obtain explicit expressions for the correlation functions @xmath90 it is convenient to work in the interaction picture with respect to the grand canonical ensemble @xmath91 , where @xmath92 is the chemical potential of the atoms in either the right or left cloud and @xmath93 . the new atomic field operators are obtained from the ones in heisenberg picture with respect to @xmath9 by replacing @xmath94 . from now on we will denote by @xmath95 the atomic field operators in heisenberg picture with respect to @xmath96 . we now discuss how the considered setup can be used in order to measure the mean value of the atomic field operator by means of photons . we first note that the photon flux is the sum of two terms : the component @xmath97 , that is the sum of the flux from each well , and the component @xmath98 , that arises from the coherent superposition of a photon ( atom ) scattered by the wells . we will denote @xmath97 by `` background contribution '' and @xmath98 by `` interference contribution '' . in absence of initial correlations , this latter term is proportional to the product of the mean value of the field operators @xmath99 . let us now consider the case in which , say , the left well confines weakly - interacting atoms forming a bose - einstein condensate . under this assumption the atomic field operator can be written as @xcite @xmath100 where @xmath101 is the macroscopic wave function , which solves the gross - pitaevskii equation for the quantum gas in the left well of the potential ( with chemical potential @xmath102 ) . then , when the order parameter of the left condensate is known , the interference term will deliver the mean value of the field operator in the right well . in the following we use eq . ( [ eqn : psisplit ] ) in the equations for the background contribution to the photon flux , eq . ( [ f : b ] ) , and for the interference term , eq . ( [ f : i ] ) . in particular , we will take @xmath103 and assume that both @xmath104 and @xmath105 are real valued . hence , @xmath106 is the density of condensed atoms , while the phase @xmath105 is assumed to be constant in space ( hence discarding the possibility of superfluid currents in the bose - einstein condensate @xcite ) . the background contribution can be decomposed into the sum of the photon flux from the left and from the right well , @xmath107 . since we make a specific assumption on the quantum state of the gas trapped in the left well , we can provide an explicit form for @xmath108 . for this purpose , we consider the integrand @xmath109 in @xmath108 and observe that it can be splitted into two terms : @xmath110\,,\nonumber\\\end{aligned}\ ] ] where @xmath111 accounts for the contribution of the noncondensed atoms to the photon flux . here , @xmath112 \right\rangle \,\ ] ] is the spectral density @xcite , with @xmath113^ { -1}$ ] . similar expressions can be found for the background contribution of the quantum gas in the right well . in order to gain insight on the equations we just derived , we assume that the gas can be considered homogeneous . this assumption allows us to write the spectral density as @xmath114 . we then inspect the equations for the flux using the momentum representation for the atomic field operator . the terms giving the contribution to the photon flux from the left cloud can be cast in the form @xcite [ eqnket : fcb ] @xmath115 where @xmath116 is the spectral weight function , which is the fourier transform of @xmath117 and gives strength of the collective excitations at wavevector @xmath118 and frequency @xmath119 , weighted by the bose function @xmath120 @xcite . the other parameters are the volume @xmath121 in which the atoms of the left well are confined , the fourier transform of the macroscopic wave function @xmath122 , and the frequency @xmath123 giving the effective detuning of the laser from the collective transition . moreover , in eqs . we have introduced the quantity @xmath124 that is proportional to the diffraction function @xmath125 , enforcing energy conservation for long interaction times @xmath126 . in particular , @xmath127 for @xmath128 , with @xmath129 dirac - delta function @xcite . definition contains a time - dependent phase and highlights that the corresponding factor tends to unity in the limit in which energy conservation applies . nevertheless , since we are going to also consider intermediate times , we will systematically keep this time - dependent phase in the equations for the photon flux . the integrals in eqs . runs over all values of the atomic momentum @xmath118 . the integrands are the product of the momentum distribution at @xmath130 and of the diffraction function : the first accounts for the effect of the photon recoil @xmath131 on the atomic distribution due to the outcoupling process , while the diffraction function imposes energy conservation in the scattering process . equations ( [ eqnket : fcb]a ) and ( [ eqnket : fcb]b ) show thus that the contributions to the flux come from the total number of condensed and noncondensed atoms , respectively , which fulfill energy and momentum conservation of the scattering process . + we remark that according to eq . ( [ eqn : mlequiv ] ) , the flux evaluated in eqs . ( [ eqnket : fcb ] ) gives also the corresponding component of the atomic flux . the latter agrees with the corresponding expressions derived in ref . @xcite for the atomic flux outcoupled from a single bec by classical fields . in order to determine the interference contribution to the photon flux , @xmath132 in eq . , one needs the explicit form of the correlation functions @xmath133 , @xmath134 . for the considered setup , however , one can already make general statements . in absence of initial correlations , in fact , they are the product of the mean value of the field operators in each well , and thus take the form @xmath135 while @xmath136^*$ ] . from eq . ( [ g : lr ] ) one observes that these correlation functions are proportional to the mean value of atomic field operator @xmath137 . when this is zero , the interference contribution vanishes . otherwise , the amplitude of this term is proportional to the condensate fraction of the quantum gas in the right well . we remark that the noncondensed atoms in the left well do not contribute to the signal , since ( i ) there are no initial correlations between the atoms in the left and right wells , and ( ii ) the mean value @xmath138 . in the shape of eq . ( [ g : lr ] ) the analogy with homodyne detection , as it is performed with light fields , can be drawn : the condensate in the left well plays the role of the local oscillator @xcite . identifying the analon of the phase of the local oscillator is however a more delicate issue , that deserves some more analysis . for this purpose , we first assume that the mean value @xmath137 can be evaluated within a mean - field approach , such that @xmath139 , with @xmath140 real - valued function and @xmath141 real constant . the integral in eq . ( [ f : i ] ) can then be cast in the form @xmath142\ , , \label{f : i:2}\end{aligned}\ ] ] with @xmath143 and @xmath144 equation evidentiates that the interference contribution is an oscillating signal . the phase of the oscillation , however , depends on time and oscillates with a frequency determined by the difference @xmath145 between the chemical potentials , while the phase offset is determined by the relative phase between the two quasi - condensate . since the two quantum gases are independent , this relative phase will be defined only for a single experimental run @xcite . some remarks on the oscillating behaviour of the interference term are now in order . in first place , for the example considered in eq . ( [ f : i:2 ] ) the photon flux oscillates in time with frequency @xmath145 . this oscillation is observed , however , only when one chooses a time - window at the detector @xmath146 such that @xmath147 . indeed , the photon flux is composed by two components : one is constituted by the atoms extracted from the left condensate , which interfere with the atoms extracted from the right trap during the time interval @xmath146 ( first term on the rhs of eq . ) . the second component is constituted by the atoms extracted from the right trap , which interfere with the atoms extracted from the left condensate during the time interval @xmath146 ( second term on the rhs of eq . ) . this interference is analogous to the interference between two lasers at different frequencies : interference fringes are observed provided that the time window of the detector is sufficiently small so not to resolve the detuning between the lasers @xcite . as in the case of two independent lasers , there is no well defined phase offset : the relative phase is defined only for a single experimental run , while the average over a statistically significant number of runs gives no interference pattern @xcite . differing from the situation of two laser beams @xcite , however , photon scattering here creates correlations among the scattered photon , the corresponding outcoupled atom , and the quantum gas . this correlation is a form of a which - way information , that in general washes out any interference in the photon flux @xcite and can be considered as a form of photon - atoms entanglement @xcite . oscillations in the photon flux , and hence interference , can be recovered if ( i ) the beam of atoms outcoupled from one well overlap spatially with the wave function of the atoms in the second trap and ( ii ) if there are either initial correlations between the atoms in the two wells or if they both possess a non - vanishing condensate fraction . condition ( i ) corresponds to a quantum eraser @xcite : when it is fulfilled , in fact , the outcoupled atom is no more entangled with the scattered photon . we will see in the following that it can be fulfilled for certain geometries and after a certain time , corresponding to the time needed for the outcoupled atoms to travel to the second well . condition ( i ) alone is however not sufficient . the photon is not only correlated with the outcoupled atom , but also with the scattering system . therefore , only when either the atoms in left and right system are correlated ( via e.g. tunneling events which may wash out the which - way information @xcite ) or when these correlations are partly classical , as it occurs when there is a finite condensate fraction in both systems , then interference can be reestablished . this is essentially condition ( ii ) . we remark that oscillations of the photon flux as a function of time have been observed in this setup in refs . @xcite . here , they have been discussed in terms of josephson junction , taking into account the possibility that the outcoupled atoms from one condensate can be then retrapped in the second well by means of an inverse raman process . the discussion we just reported provides a quantum optical interpretation of the phenomenon , that is valid for short times , when the quantum state of the scattering atoms is not substantially perturbed by the scattering process . we now analyse the onset of oscillations in time . as we mentioned , the photon flux starts oscillating after a finite time has elapsed from the beginning of the experiment . this can be also seen in our theory , by performing the integral over @xmath15 in eq . ( [ f : i:2 ] ) . in app . [ app : int ] we show that @xmath98 can be recast in the form @xmath148 , with [ eqn : fcint_postext ] @xmath149 here , @xmath150 is the overlap between the left and the right condensate , with one being displaced by the amount @xmath151 , with the recoil velocity @xmath152 that is acquired by the outcoupled atom by scattering the photon . this overlap vanishes at time @xmath153 , i.e. , when the outcoupling lasers are switched on ( recall that the two clouds initially do not overlap ) . we note that the overlap integral is zero at all times if @xmath154 and @xmath155 are orthogonal , while it may become maximum after a certain time , when @xmath154 and @xmath155 are parallel , say , pointing along the positive @xmath156 axis . in this case , the component @xmath157 may not vanish and can be interpreted as the contribution to the interference flux @xmath158 from the outcoupled atoms that propagate from left to right ( for the term @xmath159 this is just opposite ) . interference will set in , provided that a sufficiently long time has elapsed to warrant overlap . this corresponds to times @xmath160 with @xmath161 where @xmath162 is the effective distance of the two systems taking into account their width @xmath163 . in terms of an interference experiment , for times @xmath160 the which - way information has been erased and oscillations in the photon flux can be observed . in interferometric setups the amount of visibility and which - way information are related by an inequality @xcite . it is therefore useful to determine a visibility of the oscillating signal , as it contains information on the properties of the scattering systems . a visibility can be defined for sufficiently long times , averaging over several oscillating periods after the instant @xmath164 , and will be proportional to the amplitude of the oscillations of the photon flux , hence to the product of the condensate fractions of both systems . we remark , once again , that oscillations can be observed only in a single experimental run , while they will disappear after performing an ensemble average . therefore , this behaviour can only be measured in systems , whose properties are not deeply modified by the outcoupling of atoms . when this is not verified , the presence of a condensed fraction in the right system can be revealed by performing an ensemble average over a sufficient large number of experiments , in which the signal is monitored for sufficiently short times , warranting that the properties of the quantum gas have not been significantly modified , and taking the statistical distribution of the intensity of the photon flux at a given instant of time . an amplitude can be extracted from the width of the distribution by taking into account the finite width of the diffraction function . the theory presented so far will be now applied to some specific examples . the main idea is to use the setup in fig . [ fig : homodynesetup ] in order to determine the mean value of the field operator of a quantum gas , using a bose - einstein condensate at known temperature as reference system . such setup is a matter - wave analogon of homodyne detection in quantum optics . the individual elements can be so identified : the bec acts as a local oscillator , the outcoupling procedure as beam splitter , the relative phase can be varied by changing the interwell distance . the information on the atomic gas is carried by both scattered photons and outcoupled atoms : homodyne detection of the scattered field , hence , allows one in principle to determine the mean value of the quadrature of the atomic field operator . we first discuss the case in which two bec are trapped in the left and right well , respectively , and are both illuminated by the pump and probe beams . the outcoupled atoms from both condensates propagate along the direction determined by the distance between the minima of both wells , and overlap after the time @xmath164 has elapsed . the scattered photons are revealed at a detector in the far - field . this setup has been realized in the experiment reported in refs . @xcite , where time - dependent oscillations in the atom and photon flux were measured . here , we apply the theoretical model developed so far and find it reproduces qualitatively the results of these experiments . moreover , we discuss the results in the light of question addressed in the present work . we assume two bec with equal number of atoms @xmath165 and temperature @xmath86 , that are confined either in the left or right well . the wells are described by the potential @xmath166 with @xmath167 , and @xmath168 denotes the constant offset between the two traps . for simplicity we take that both bec are at zero temperature , @xmath169 , and the atoms weakly interact , such that the contribution of the noncondensed atoms to the photon flux is small and can be neglected . in this limit the chemical potential of both condensates is equal and given by @xmath170 . with definition ( [ eqn : trappot ] ) then @xmath171 and @xmath172 . the photon flux is evaluated using the thomas - fermi approximation for the condensate wave functions @xcite , @xmath173^{1/2}\ , , \ ] ] where @xmath174 is the chemical potential at zero temperature and @xmath175 is the size of the ground state of a harmonic oscillator with frequency @xmath176 . the condensate macroscopic wave function has size @xmath177 . its fourier transform reads @xmath178 ( for the right condensate @xmath179 ) , where @xmath180 is the fourier transform of the macroscopic wave function centered at the origin and is real valued . in particular , @xmath181 where @xmath182 is a scalar and @xmath183 the bessel function of second order of the variable @xmath184 defined as @xmath185 , see @xcite . the integral for the interference contribution to the photon flux in eq . ( [ f : i:2 ] ) can be then cast in the form @xmath186\,.\ ] ] this equation can be simplified , taking that both @xmath154 and @xmath155 point along the positive @xmath156-direction . following the derivation reported in app . [ app : ketterle ] , we find that the total flux can be approximated by the expression @xmath187 \ , , \ ] ] where @xmath188 and @xmath189 $ ] is a gaussian of width @xmath190 . equation shows that the photon flux starts oscillating for times @xmath160 , with @xmath191 , namely , when the outcoupled atoms from one condensate have reached the second one . the time - dependent oscillations have frequency @xmath192 and amplitude @xmath193 the expression in eq . ( [ eqn : ftpostot ] ) has been obtained neglecting the momentum dependence of the trapped clouds ( this approximation is verified in app . [ app : ketterle ] ) and the contribution of the noncondensed fraction . we note that , although oscillations as a function of time are observed provided that @xmath194 , nevertheless their visibility ( which corresponds to @xmath195 ) is always smaller than unity , @xmath196 for @xmath194 . this can be understood considering that due to energy conservation in the scattering process , a scattered photon carries partial information about from which cloud it was scattered . we now turn our attention to the phase offset , characterising the oscillations of the photon flux as a function of time in eq . ( [ eqn : ftpostot ] ) . this phase offset is here given by the quantity @xmath197 , where @xmath198 is the relative phase between the two bec , which can take any value ( we refer to the discussion in sec . [ sec : phase ] ) , and by the quantity @xmath199which is the phase the outcoupled atoms accumulate when travelling from left to the right well ( the corresponding phase offset for the outcoupled atoms travelling from the right to the left well is given in app . [ app : ketterle ] ) . this phase can be varied by either tuning the laser frequency , and thus @xmath200 , or changing the distance between the minima of the two wells . the phase offset in eq . agrees with the one reported in ref . @xcite , that was derived from a phenomenological model . in ref . @xcite , in particular , the rate of change of the difference in phase offset between atoms outcoupled to the left and atoms outcoupled to the right was measured as a function of the bragg - frequency @xmath200 , and found in agreement with the prediction of eq . ( [ eqn : phaseoffset ] ) . we also note that our theory predicts correctly the oscillation period observed in the experiment , and also accounts for the fact why the interference current needs some time to build up . the linear response treatment we apply , however , does not predict any decrease in the visibility of the interference pattern with time , in marked contrast with the experimental results . possible reasons for the decrease in visibility in the experimental data are depletion of the condensate and heating of the sample , which could be due to spontaneous rayleigh scattering events @xcite . these effects are not taken into account in our model , but could be introduced by means of quantum langevin equations , using a formalism similar to the one developed in ref . @xcite . photon flux @xmath201 ( in units of the background contribution @xmath202 ) as a function of time ( in units of @xmath203 ) and of the angle @xmath204 at which the atoms are emitted . the photon flux is computed by numerically integrating eq . ( [ eqn : fluxesmomentumketterle ] ) . the condensates are composed by @xmath205 sodium atoms in a spherical harmomic trap of frequency @xmath206 . the scattering length is @xmath207 with @xmath208 bohr radius . the other parameters are @xmath209 , @xmath210 , @xmath211 , @xmath212 , @xmath213 and @xmath214 . ] we finally discuss the dependence of the photon flux on time and on the angle of emission of the outcoupled atoms , thus for geometries where the direction of emission @xmath155 of the outcoupled atoms forms an angle @xmath204 with the vector @xmath215 . for simplicity we assume that both vectors lie in the @xmath216 plane . the calculation is performed by numerically integrating eq . ( [ eqn : fluxesmomentumketterle ] ) . figure [ fig : fluxesmomentum ] displays the photon flux @xmath201 as a function of time and of the emission angle @xmath204 . oscillations as a function of time are observed for values of the angle about @xmath217 , corresponding to the atoms propagating in the direction parallel to @xmath218 . the period of the oscillation is @xmath219 , independent of the angle of emission . the oscillations disappear for angles in the interval @xmath220 : the photon flux is here solely given by the background contribution @xmath202 . for these angles , in fact , at all times there is no spatial overlap between the wavefunction of the outcoupled atoms and the wavefunction of the trapped atoms in the other well . indeed , from a simple geometric argument one finds that the overlap vanishes for angle @xmath221 and @xmath222 . this implies that the setup must be so constructed , that the atoms outcoupled from one well could in principle be transferred , by a raman process , into the second well . this property is the key element on which the analogy to a josephson junction has been drawn @xcite . it is also basically the way in which a quantum eraser is realised in this setup . we refer the reader to the discussion on the properties of this quantum interference process in sec . [ sec : phase ] . we now show how the setup of ref . @xcite could be used in order to determine the temperature of a bose - einstein condensate , using as reference a second bec at known temperature . for this purpose , we make the same assumptions as in the previous section , with the difference that while the left gas is a bec at temperature @xmath169 , the gas in the right well is a bec at temperature @xmath86 to be determined . we further assume that the thomas - fermi approximation can be performed also for the right condensate . hence , the chemical potential of the second condensate can be written as @xmath223 , with @xmath224 the critical temperature for the noninteracting gas in a harmonic trap , while the size of the macroscopic wave function of the right condensate scales with @xmath225 , see @xcite . using these relations the integral for the interference contribution to the photon flux can be cast in the form @xmath226\,,\end{aligned}\ ] ] with @xmath227 . the integral depends on the temperature both through a scaling factor as well as the function @xmath228 in the integrand , while the phases depend on the temperature of the second bec via the chemical potential of the right bec , which enters in the quantity @xmath229 with @xmath230 . we note that @xmath132 oscillates as a function of time with both frequency and amplitude which depend on the temperature of the second condensate . for @xmath231 , in particular , one finds that @xmath232 , where @xmath233 is the condensate fraction in the right trap . figure ( [ fig : flrtemperature]a ) displays @xmath234 as function of time and temperature @xmath86 of the right condensate , when the trap is spherical . the oscillations are visible for times @xmath160 . the oscillation frequency depends on @xmath86 , as one can clearly observe from the figure @xcite . the dependence of the amplitude of the oscillation on the temperature is visible in fig . ( [ fig : flrtemperature]a ) and is singled out in fig . ( [ fig : flrtemperature]b ) , where the amplitude @xmath235 \ , , \ ] ] evaluated at times @xmath160 , is displayed as a function of @xmath86 in units of @xmath236 . the red dashed curve represents the squared root of the condensate fraction , @xmath237 , and is reported for comparison . function @xmath238 decreases monotonically as the temperature increases from @xmath169 , and vanishes at the critical temperature @xmath239 ( which takes into account the effect of the interactions and is such that @xmath240 ) @xcite . we now assume that the quantum system in the right trap is a two - dimensional optical lattice , where the parameters can be tuned so that the ultracold atomic gas is either in a mott - insulator or superfluid state , while the left system is a bec at @xmath169 . within a mean - field treatment , in the superfluid phase a nonvanishing superfluid order parameter would give rise to a time - oscillating behaviour of the photon flux @xcite . before we start , we recall that the mott - insulator / superfluid transition in two dimensions has been experimentally characterised in ref . @xcite , and the condensate fraction has been measured in ref . @xcite , while bragg scattering by this system was recently reported in ref . @xcite . here , we argue that the photon flux may allow one to detect the superfluid order parameter , and thus monitor the quantum state of the system across the phase transition . we note , however , that a measurement based on outcoupling atoms from the optical lattice will change the phase of the atoms , if they have been prepared in the mott - insulator phase . for this specific situation , indeed , the approximation we make by treating photon scattering as a weak perturbation of the many - body state remains valid as long as the scattering is performed over a time - scale in which the creation of the local defect does not affect the rest of the system . this time scale will be inversely proportional to the sound velocity with which a perturbation propagates in the corresponding superfluid state . this restriction shows that this measurement procedure will not provide a reliable estimate of the superfluid order parameter close to the phase transition point . keeping this in mind , let us now assume that the two atomic gases are at @xmath169 and tightly confined in the @xmath156-direction by a harmonic potential , so that the motional degrees of freedom in the @xmath156-direction are frozen out and the systems are essentially two dimensional . the left potential is assumed to be a highly - anisotropic harmonic trap and right potential is a two - dimensional optical lattice . they read @xmath241 with @xmath242 , while @xmath243 is the height of the optical lattice potential and @xmath244 the spatial periodicity in the transverse directions . assuming that the left quantum gas forms a weakly - interacting bose - einstein condensate at @xmath169 , and that @xmath245 , with @xmath246 the chemical potential of the left condensate , then the macroscopic wave function reads @xmath247 , with @xmath248 is the planar density of condensed atoms in the left trap ( which is here assumed to be constant , under the condition that the laser beams illuminates the center of the condensate ) and @xmath249 is the motional ground state of the harmonic oscillator in the @xmath156-direction . the wave function for the right quantum gas is here given assuming the tight - binding limit . using the wannier decomposition , in the single - band limit one finds @xmath250 where @xmath251 is the wannier function centered at @xmath252 ( @xmath253 ) and @xmath254 annihilates an atom at the lattice site with @xmath255 and @xmath256 . the corresponding bose - hubbard model reads @xcite @xmath257 with @xmath258 the number of atoms at site @xmath259 , @xmath260 the hopping rate , @xmath261 the onsite interaction , and @xmath262 the chemical potential for the right system . we note that the two - dimensional limit is consistent when @xmath263 . in the superfluid phase the expectation value of operator @xmath264 over the lattice ground state does not vanish and is a constant , @xmath265 , and the wavefunction for the superfluid fraction reads @xmath266 . the photon flux , and in particular the interference contribution , can now be evaluated using these quantities . for simplicity , in the following we assume that the geometry of the laser is such that @xmath155 is parallel to the vector @xmath154 . an analytical expression can be determined using the gaussian ansatz for the wannier functions @xcite and shows that @xmath267 and is a convolution of signals oscillating with phase @xmath268 , weighted by the occupation of the momentum @xmath269 fulfilling energy conservation . we evaluate @xmath132 by numerically integrating eq . . the superfluid order parameter @xmath270 is determined numerically within the mean - field treatment of the bose - hubbard model @xcite . figure [ fig : flrandfa](a ) displays the interference contribution @xmath132 as a function of time and of the ratio @xmath271 between hopping strength and the onsite interaction , with @xmath272 the coordination number . this ratio is assumed to be varied by changing the onsite interaction @xmath261 , while the lattice depth @xmath243 is kept constant . for small values of @xmath271 , when the atoms in the right well are in the mott - insulator state , no interference signal is found . at a critical value @xmath273 ( unit filling ) @xcite the right system undergoes the quantum phase transition to the superfluid state and the interference photon flux @xmath132 starts to oscillate in time with finite amplitude . we study the amplitude of the oscillations as a function of @xmath274 by characterizing the quantity@xmath275 where the minimum and maximum of @xmath132 are found over a time interval large compared to the inverse oscillation frequency of @xmath132 , similarly to eq . ( [ eqn : atr ] ) . this quantity is displayed in fig . [ fig : flrandfa](b ) as a function of @xmath274 . for @xmath276 the amplitude @xmath277 is nonvanishing and increases proportional to the magnitude of the superfluid order parameter several methods have been proposed in the literature for determining the first - order correlation functions , that are based on time - of - flight techniques , see for instance refs . measurements of the first - order correlation function have been performed on bose - einstein condensates in refs . @xcite . in the following we will show how an extension of our previously considered setup may allow one to determine the spatial first - order correlation function of a quantum gas by the scattered photons . in this case , the lasers shall illuminate two spatially separated regions of a quantum gas confined in a _ single well _ potential . more specifically , the spatial dependence of the laser - atom interaction in eq . [ eqnket : v:0 ] will be characterized by the gaussian envelope @xmath278 with width @xmath279 between the two regions ( we remark that the excitation could be realized with subwavelength resolution @xcite ) . the corresponding photon flux then reads @xmath280 where @xmath281 is the first - order correlation function and @xmath282 is defined in eq . ( [ fun : f ] ) . let the pump - probe excitation be a pulse of mean duration @xmath126 such that @xmath283 , but @xmath284 , with @xmath285 the typical frequency characterising the excitation spectrum . in this limit we can approximate @xmath286 in eq . ( [ f : t : g:1 ] ) . for convenience , we denote by @xmath287 the spatial correlation function . for @xmath288 sufficiently small , so that it can be approximated with a @xmath289-like excitation , the photon flux can be recast in the form @xmath290 with @xmath291 is a constant , determined by the details of the excitation scheme , and @xmath292 is the density at @xmath293 ( assuming the system has reflection symmetry about @xmath294 ) . therefore , the photon flux exhibits oscillations with a visibility which is determined by the spatial first - order correlation function . by varying the scattering wave vector @xmath295 one would thus measure the first - order correlation function as a function of @xmath215 . realistic excitation schemes are of course characterized by finite spatial resolution @xmath288 , which results in averaging the correlation function in eq . over the finite size of the illuminated region and hence to a diminution of the contrast @xcite . this setup could be extended to measure time - dependent correlation function by applying a pair of laser pulses : assuming that the photon scattered after the first pulse can interfere with the photon scattered after the second pulse , then the photon flux at the detector will exhibit oscillations whose amplitude is proportional to the first - order correlation function . a possible realization could use a mirror placed in front of the quantum system , as realized in @xcite . in this work we have discussed a setup which allows one , to measure the condensate fraction and the first - order correlation function of a quantum gas by means of photo - detection . the photons are scattered by the quantum gas in a pump - probe type of excitation , such that the scattered photon is associated with an outcoupled atom , with which it is entangled . in addition , photon and atom are correlated with the quantum gas . this correlation is detected in the photon flux , provided that certain conditions are fulfilled , which we have identified and discussed . our analysis is based on the impulse approximation @xcite , hence on neglecting the back action of the scattering process on the quantum gas . it is therefore valid for short time transients . from the point of view of collecting a sufficient statistics , hence , time - of - flight techniques are a more convenient tool . nevertheless , one could consider to modify existing techniques , such as the one successfully demonstrated in refs . @xcite , in order to access to the same kind of information that the setup here discussed provides . an interesting outlook is to identify a setup , along the lines of the proposal ref . @xcite , which permits one to perform a quantum - non - demolition measurement of any correlations function of of the external degrees of freedom of atomic gases , and more in general , which realizes quantum - state transfer between matter and light . this would open several interesting perspectives for quantum communications @xcite . the authors acknowledge e. demler , r. fazio , j. ruoteskovki , w. schleich , and p. vignolo for stimulating discussions and helpful comments . this work was supported by the european commission ( emali , mrtn - ct-2006 - 035369 ; integrating project aqute , strep picc ) , by the spanish ministerio de ciencia y innovacin ( consolider - ingenio 2010 qoit ; fis2007 - 66944 ; ai hu2007 - 0013 , euroquam `` cmmc '' ) , and the german research foundation ( dfg , mo1845 ) . we now show that the term @xmath157 ( @xmath298 ) is the contribution due to the atoms which are outcoupled and propagate from the left to the right ( right to the left ) . for this purpose we perform the @xmath118 integral in eq . ( [ eqn : intf ] ) and obtain where @xmath300 is the relative phase of the macroscopic wavefunctions defined in eq . . with the change of variables @xmath301 , @xmath302 , we can rewrite eq . ( [ hom : propposa ] ) as @xmath303 f_l({{\bf r}}-\frac{\bar{{{\bf r}}}}{2})f_r({{\bf r}}+\frac{\bar{{{\bf r}}}}{2 } ) \ , .\end{aligned}\ ] ] the exponential in the integral over @xmath304 oscillates very fast with respect to the wave functions @xmath305 . therefore the main contribution to the integral over @xmath304 comes from @xmath306 , where the term in the exponential vanishes . by means of the saddle - point approximation we take the wave functions at the point @xmath307 out of the integral . performing the @xmath304 integration using the fresnel integral @xcite @xmath308 where @xmath309 is the recoil velocity , eq . . the calculation that leads to eq . ( [ eqn : fcint_pos ] ) is exact in the limit of homogeneous atomic systems . in such case the momentum distribution of the condensate fraction is a dirac - delta function at zero momentum and all outcoupled atoms have exactly the same momentum @xmath310 . the condensates we consider are confined by an external potential and have a finite extension in space which leads to a certain width @xmath311 in their momentum distribution , and thus to a spread in the momentum of the outcoupled atoms around the mean value @xmath310 . by taking the value of the atomic wave functions at the point of the stationary phase of the exponential in eq . ( [ bvorsaddle ] ) one neglects this momentum width : the saddle - point approximation is thus applicable if @xmath312 . equation ( [ eqn : fcint_pos ] ) agrees with the corresponding expression in eq . ( [ eqn : fcint_postext]a ) . for completeness we also give the contributions of the macroscopic wavefunctions to the background contribution , eq . ( [ f : b ] ) , making the same approximations as in eqs . ( [ eqn : fcint_pos ] ) : @xmath313 we derive here approximate expressions of the raman scattering rate for the experimental setup of @xcite , which lead to eq . ( [ eqn : ftpostot ] ) . using eq . ( [ eqn : thomasfermi ] ) for the condensate wavefunctions in eq . ( [ eqn : fcb_pos ] ) we find @xmath314 g(z ) \ , , \nonumber \\\end{aligned}\ ] ] with @xmath315 the length @xmath316 determines the typical size of the condensates and can be written as @xcite @xmath317 the function @xmath318 in eq . ( [ eqn : fposcalc ] ) reads @xmath319 a numerical evaluation of @xmath318 is shown in fig . ( [ fig : gaintvgl ] ) , and is here compared with a gaussian of the form @xmath320 showing that this function provides a good approximation of eq . . using eq . ( [ eqn : gz_gauss ] ) in eq . ( [ eqn : fposcalc ] ) one gets @xmath321 e^{- 1.25 z^2 } \ , , \nonumber \\\end{aligned}\ ] ] with @xmath322 . for times @xmath160 , with @xmath323 , such that @xmath324 we find from eq . ( [ eqn : fluxespossmalltimes ] ) @xmath325 e^{-1.25 z^2 } \nonumber \\ & \approx & \pi \gamma n_c \sqrt{\frac { t_0 ^ 2 } { 5 \pi } } e^{-\frac{1}{5}[(\omega_{{{\bf q}}}-\omega)t_0]^2 } \ , .\end{aligned}\ ] ] when performing the integral in eq . ( [ eqn : err ] ) we neglected the imaginary part in the error - functions which one gets from the exact integration . this is a good approximation for @xmath326 , as can be checked by numerical evaluation . for values @xmath327 the exponential in eq . ( [ eqn : err ] ) is negligible compared to the resonant case @xmath328 : the oscillating tails of the error - functions given by their imaginary parts will only play a role for parameters where the outcoupling efficiency vanishes and which are thus not relevant to our treatment . using the same argumentation for the other contributions to the photon flux we find [ eqn : ftc_position_final ] @xmath329 \ , , \\ f_{r \rightarrow l}(t ) & \approx & 2 \pi \gamma n_c k(\omega_{{{\bf q}}}-\omega+\delta \mu ) f_{\theta}(-q , t ) \nonumber \\ & & \times \cos \left [ \delta \mu t-\varphi_{lr}-(\omega_{{{\bf q}}}-\omega + \delta \mu)\frac{d}{v_{q } } \right ] \ , , \nonumber \\\end{aligned}\ ] ] with @xmath330 from eqs . ( [ eqn : ftc_position_final ] ) one then obtains eq . ( [ eqn : ftpostot ] ) for @xmath331 . the phase difference between @xmath157 and @xmath332 as measured in @xcite is obtained from eqs . ( [ eqn : ftc_position_final]c , d ) and reads @xmath333 ( color online ) total photon flux @xmath201 in units of background contribution @xmath202 as a function of time in units of @xmath334 . the solid black line is computed by numerically integrating eq . ( [ eqn : fluxesmomentumketterle ] ) and compared to the approximate result eq . ( [ eqn : ftpostot ] ) ( dashed red line ) . parameters are as in fig . ( [ fig : fluxesmomentum ] ) with @xmath213 . ] figure ( [ fig : fluxesmomentumalpha0 ] ) displays the total photon flux computed from eq . ( [ eqn : ftpostot ] ) ( dashed line ) and the one calculated from eq . ( [ eqn : fluxesmomentumketterle ] ) ( solid line ) for the experimental parameters of @xcite . quantitative agreement between the two solutions is found , showing that neglecting the initial momentum distribution is well justified for the considered parameters . a. imambekov , v. gritsev , and e. demler , `` fundamental noise in matter interferometers '' , in `` ultracold fermi gases '' , proceedings of the international school of physics `` enrico fermi '' , 2006 ( ios press , amsterdam , the netherlands , 2007 ) . j. f. sherson , c. weitenberg , m. endres , m. cheneau , i. bloch , and s. kuhr , nature * 467 * , 68 ( 2010 ) ; c. weitenberg , m. endres , j. f. sherson , m. cheneau , p. schau , t. fukuhara , i. bloch , and s. kuhr , nature * 471 * , 319 ( 2011 ) . k. w. murch , k. l. moore , s. gupta , and d. m. stamper - kurn , nature phys . * 4 * , 561 ( 2008 ) ; n. brahms , t.p . purdy , d.w.c . brooks , t. botter , and d.m . stamper - kurn , preprint arxiv:1012.1285 ( 2010 ) , to appear in nature phys . ( 2011 ) . s. slama , c. von cube , a. ludewig , m. kohler , c. zimmermann , and ph . w. courteille , phys . a * 72 * , 031402(r ) ( 2005 ) ; s. slama , c. von cube , b. deh , a. ludewig , c. zimmermann , and ph.w . courteille , phys . * 94 * , 193901 ( 2005 ) ; s. slama , c. von cube , m. kohler , c. zimmermann , and ph . w. courteille , phys . a * 73 * , 023424 ( 2006 ) . this commutation relation is well defined provided that @xmath335 . the mathematical subtlety related to this specific point is irrelevant to our study , being practically zero the probability that an atom in the electronic state @xmath4 can be found around this point in space . s. hofferberth , i. lesanovsky , t. schumm , a. imambekov , v. gritsev , e. demler , and j. schmiedmayer , nature physics * 4 * , 489 ( 2008 ) ; a. imambekov , i. e. mazets , d. s. petrov , v. gritsev , s. manz , s. hofferberth , t. schumm , e. demler , and j. schmiedmayer , phys . a * 80 * , 033604 ( 2009 ) . we note that the condensate depletion , following from the outcoupling , induces a variation of @xmath165 as a function of time , which also leads to a change in the chemical potential . this effect is not taken into account in our treatment , where we neglect the change of atom number in the condensates . the use of the gaussian ansatz in place of wannier functions should be handled with care , for wannier functions decay exponentially at infinity , see e.g. w. kohn , phys . rev . * 115 * , 809 ( 1959 ) . nevertheless , the precise form of the ansatz is not important for the purpose of this estimate . here , we use the critical value determined by the mean - field , @xmath340 , see @xcite . we remark that a more precise determination of @xmath341 has been reported in s. wessel , f. alet , m. troyer , and g. batrouni , phys . a * 70 * , 053615 ( 2004 ) , where it was obtained by means of quantum monte - carlo simulations .

significant progress has been made in the study of the galactic black hole binaries since the launch of the compton gamma ray observatory in april 1991 . it is now widely believed that some x - ray binary systems harbor a stellar black hole at the center of the accretion disk in each system . a variety of high energy spectra and light curves have been observed from many of them . some of them also exhibited highly relativistic jets , which are found to be in correlation with the high energy radiation . since the first mass determination of the assumed black hole in cyg x-1 about 25 years ago@xcite , the second property , i.e. , the spins , of black holes have also been inferred recently@xcite , and are considered to be the missing link in the proposed unification scheme of all types of black hole binaries@xcite . several review articles on galactic black hole binaries now exist , including those in the same proceedings . in this review article , we try to avoid any significant overlapping with other review articles , by focusing on ( a ) summarizing these new and important multiwavelength observations , and ( b ) exploring the physical connections between observations made at different wavelengths to obtain more complete pictures of these systems . all of these galactic black hole binaries were originally discovered in either x - ray ( 1 - 10 kev ) or hard x - ray ( 20 - 300 kev ) bands . their x - ray or hard x - ray luminosity dominates their total electromagnetic radiation energy output . in this section we discuss briefly their basic high energy emission characteristics , in order to explore the multiwavelength aspects of these systems . the high energy spectra of galactic ( candidate ) black hole binaries are made of primarily only two components a blackbody - like soft component and a power - law - like hard component ( see @xcite for a review of high energy continuum spectra observed in black hole and neutron star x - ray binaries ) , as shown in figure 1 for the superluminal jet and black hole source gro j1655 - 40@xcite . in some sources , the soft component is absent above 1 kev_and _ the hard power - law component becomes harder with a break between 50 - 200 kev , as shown in the same figure for groj0422 + 32 . the soft component is believed to be emitted from the inner accretion disk region and can be well described by the so - called multi - color disk blackbody spectral model@xcite , in which the total emitted spectrum is the integration of the local blackbody emission from each annulus with a temperature @xmath0 . to obtain the physical parameters of the inner accretion disk , corrections for the relativity and spectral hardening effects must be taken into account @xcite . the observed peak blackbody temperatures of the multi - color disk blackbody spectra vary between 0.1 to 2.5 kev ( see figure 5 and the corresponding section for details ) . the origin of the power - law - like hard component is still not well understood , although inverse compton up - scattering of low energy photons by high energy electrons are usually believed to be involved . the nature and origins of both the low energy photons and the high energy electrons are still not identified unambiguously yet , despite the existence of many models . there usually exists an anti - correlation between the soft component luminosity and the hardness or flatness of the power - law whose photon spectral index varies between -1.5 to -3.5 . usually the power - law component is observed to fall into one of the two states : a hard state corresponding to a minimum luminosity ( @xmath1 ergs / s)@xcite or the absence of soft x - ray component@xcite , with a photon spectral index between -1.5 to -2.0 and spectral break above 50 - 200 kev ; or a soft state corresponding to a higher soft x - ray component luminosity@xcite , with a photon spectral index between -2.5 to -3.5 and no detectable spectral break up to 300 - 600 kev@xcite . figure 1 illustrates a typical spectrum of each type ( soft or high state for groj1655 - 40 and hard or low state for groj0422 + 32 ) . however , a source may remain in a different state for a time , and the transition or evolution between the states is usually continuous , for example in cyg x-1@xcite . both the soft and the hard components have been used as signatures or indicators of black hole binary systems . similar components have , however , also been detected from neutron star binary systems , such as the type i x - ray burster 4u1608 - 52@xcite , although the absolute hard x - ray luminosity is found to be systematically higher in black hole systems than in neutron star systems @xcite . a variety of high energy light curves have been observed from galactic black hole binaries @xcite . these light curves can be placed conveniently into three classes : transients with nova - like light curves , which form the majority of the black hole binaries , such as a0620 - 00 , gs2000 + 25 , gs1124 - 68 , gs2023 + 23 , h1705 - 250 , groj0422 + 32 , groj1719 - 24 = grs1716 - 249 and grs1739 - 278 , with a fast rise ( a few days ) and slow exponential decay ( a few tens of days ) ; transients with multiple outbursts , such as the superluminal jet sources grs1915 + 105 , groj1655 - 40 and a possible jet source gx339 - 4 ; persistent sources such as cyg x-1 , lmc x-1 , lmc x-3 . we shall call them type i or nova - like , type ii or multiple - outbursts and type iii or persistent _ high energy _ light curves in the following discussion . samples of some batse hard x - ray light curves , together with some optical , radio and soft x - ray light curves are shown in figures 2 - 4 . please refer to the references where some of the original light curves were published for more details . ( note that some of the sources listed above , i.e. , groj1719 - 24 , grs1739 - 278 and gx339 - 4 , are not yet dynamically established black hole binaries with the compact object masses in excess of 3 m@xmath2 ) . these different types of light curves can not be understood with high energy observations alone . in the following , we will try to understand some of their high energy behaviors using multifrequency data . in this section we discuss briefly the application of optical and infrared observations to the understanding of the properties of inner accretion disk regions and the black holes . ellipsoidal optical and infrared light curve modulations of the companions in these systems can be used to determine their orbital periods . spectroscopic measurements of the doppler shifts of some spectral line features can be used to obtain their mass functions , i.e. , their mass lower limits . detailed modelling of their light curves allows the determination of the system inclination angles . the spectral types of the companion provide information on the nature of the companion stars . therefore in principle the complete system parameters , such as the orbital periods , orbital separations , masses of both the companions and the compact objects ( the black hole for the systems we are discussing in this article ) can be determined from optical and infrared observations@xcite . perhaps the best example and success of this technique is the precise black hole mass ( 7 m@xmath2 ) and other system parameters determination in groj1655 - 40@xcite . up to now , reliable dynamical mass determinations from optical and infrared observations have resulted in ten systems in which the compact object masses are most likely greater than 3.2 m@xmath2(see table 1 , which lists also their radio and x - ray properties discussed in the following sections ) , the observational and theoretical upper limits of a neutron star . in several systems , the lower limits to the compact object masses are already above 3.2 m@xmath2 . arguably these constitute so far the strongest evidence of the existence of black holes in binary systems . although probing only the surface of the companion star and the outer disk region , these optical and infrared studies can also be used for understanding the high energy emission ( from the inner - most region of the disk ) behaviors of these systems , especially the three types of light curves we mentioned in the previous section . the companion stars in the systems producing the type iii or persistent light curves are all high mass ( 10 - 30 m@xmath2 ) o or b stars . the type iii light curves are apparently due to the high mass transfer rates from their high mass companions , either through roche - lobe overflow or stellar wind accretion . apparently they are all accreting below the eddington rate . there are , however , some significant differences between the three sources listed above . cyg x-1 spends most of its time in a hard state , in which the total luminosity is dominated by the hard power - law . occasionally the source enters into its soft state , in which a prominent soft component ( with a higher temperature than that in the hard state ) appears , nearly simultaneous with the softening of the power - law@xcite . lmc x-1 and lmc x-3 , on the other hand , always remain in the soft state with prominent soft x - ray components and steeper power - law components , similar to the soft state of cyg x-1 . even between lmc x-1 and lmc x-3 , there exists also some major differences . for example , in lmc x-1 the soft x - ray component remains rather stable , while the power - law component can vary substantially@xcite . while in lmc x-3 , the two components may vary substantially and independently , and the intensity and spectral hardness below 13 kev show a positive correlation@xcite . it is also interesting to note that lmc x-3 and cyg x-1 are the only black hole binaries exhibiting long term periodic variations in their optical and x - ray light curves . in particular , orbital modulations in both the soft and hard x - ray bands have been seen from cyg x- 1@xcite . although the soft x - ray flux modulation may be interpreted as the modulation of the column density ( ionized or neutral ) due to stellar wind@xcite , the hard x - ray modulation is inconsistent with this interpretation@xcite . since @xmath3 30% of the total hard x - ray component is estimated to be made of the reflection of an input power - law component @xcite , one possible origin of the observed orbital modulation is that its companion star also contributes to the total reflection component . then the contribution from the companion would produce orbital modulations , similar to the optical and infrared ellipsoidal modulations . the companion stars in all these systems producing the type i or nova - like light curves are low mass ( @xmath41 m@xmath2 ) main sequence stars , and no such low mass companion black hole binaries are known to produce other types of light curves . the transient nature of such systems has been explained as due to their sufficiently low mass transfer rate and that the mass accretor is a black hole@xcite . the outburst trigger mechanism is likely due to the thermal instability in the outer part of the disk@xcite . in the quiescence state , the material is steadily transferred at a very low rate from the companion to the accretion disk via roche - lobe overflow and a significant portion of the material is accumulated in the disk . thus the quiescence state disk is quasi - steady . when an instability is developed in the disk , an outburst occurs , during which the accumulated material is transferred to the black hole with a much higher rate . in this scenario the rise time would correspond to the sound propagation time from the outer disk to the inner disk@xcite : @xmath5 ( where @xmath6 is the outer disk radius , @xmath7 the dimensionless viscosity parameter and @xmath8 the sound speed ) which is on the order of a day for @xmath9 . the decay time would correspond to the viscous time scale of the hot state disk , @xmath10 ( where @xmath11 is the vertical height of the disk and @xmath12 the angular speed of disk at radius @xmath6 ) , which is on the order of 100 days for @xmath13 . there are still some fundamental questions concerning the quiescent disk structure . the standard thin disk extending to the last stable orbit of the black hole is problematic , because the inner part of the disk can not stay in the ` cold ' branch of the ` s ' curve ( inferred from the observed quiescence x - ray emission ) , which is essential for the disk instability model@xcite . the most recent version of the advection dominated disk model@xcite with the cooling dominated inner part truncated is much more successful than the standard thin disk model , but also has some problems . first , location of the inner disk boundary , critical in the model , is not determined self - consistently , although the thermal instability condition can be applied to produce a constrain . second , it can not explain the delayed hard x - ray outburst in groj1655 - 40@xcite , with respect to the soft x - ray outburst after the initial optical on - set@xcite . therefore a more consistent quiescent state disk model is still not available yet . the companion stars in the type ii or multiple - outbursts systems may be intrinsically different from that in the type i systems ( the companion star in gx339 - 4 has not been identified yet ) . although also called a low mass system , groj1655 - 40 has an evolved companion star of 2.3 m@xmath2@xcite . usually such donor stars would have sufficiently high mass transfer rate , preventing the system from having a transient nature . this discrepancy has been reconciled by assuming that groj1655 - 40 is in a short - lived evolutionary stage where the mass transfer rate is sufficiently low to allow instabilities to occur@xcite . this is a possible way to explain the rarity of this type of systems . its multiple outbursts are not periodic and so apparently not related to its orbital motion , but most likely due to another instability after the initial outburst is triggered . perhaps the x - ray heating of the companion star induces the so - called mass transfer instability @xcite or ` echoes ' @xcite . this may be due to its peculiar type ( much massive and larger than other companion stars ) of this companion star . comparing the the k(2.2@xmath14 m ) spectra of grs1915 + 105 observed at different epochs @xcite , one finds the characteristic hei and br@xmath15 emission lines of o - be stars @xcite with time variable intensity . it has been shown that the absolute infrared magnitude , time variability , and spectral shape of grs1915 + 105 is comparable to that of ss 433 and other high mass x - ray binaries @xcite . furthermore , the infrared afterglow of an x - ray / radio outburst observed on august 1995 revealed that at that time grs1915 + 105 was enshrouded in a dusty nebula of @xmath3 10@xmath16 cm radius@xcite . in summary , the infrared observations suggest that the donor star belongs to the class of massive stars with transitional spectral classification due to the dynamically unstable stellar atmosphere or wind @xcite . if this is confirmed , grs 1915 + 105 would be a black hole with a peculiar massive companion which is losing mass at rates of 10@xmath17 - 10@xmath18 m@xmath2 yr@xmath19 . further more definitive observations are still needed to clarify its companion star type , which may be critical for understanding the nature of the system . despite that groj1655 - 40 is the only one of the type ii sources whose companion is unambiguously identified to be a peculiar star , it is worthwhile to explore the relationship between the properties of the companions and the high energy behaviors of the systems . radio fluxes from these systems originate in the incoherent synchrotron radiation of high energy electrons . the radio emission region is usually comparable or larger than the size of the accretion disk or the whole binary system . for a typical source of 0.1 jy at 3 kpc , since the surface brightness temperature can not exceed 10@xmath20 k , the minimum size of the radio emission region is about 10@xmath20 cm . the size of a typical black hole binary with an orbital period of order one day is also about 10@xmath20 cm . for those spatially resolved systems , i.e. , those jet sources , their sizes are usually between 10@xmath21 to 10@xmath16 cm . therefore the radio emission region can be significantly larger than the whole binary system . at a distance far away from the accretion disk , the strength of the magnetic field is most likely of order 10@xmath22 gauss , therefore the typical electron energy is of order 10@xmath23 ev for a characteristic radiation frequency of 10@xmath24 hz . therefore the radio emission from black hole binary systems is usually produced by high energy electrons far away from the central region of the accretion disk . corresponding to the three types of x - ray light curves , there are also three types of radio light curves , with similar morphologies . the systems producing the type i or nova - like x - ray light curves produce usually also type i radio light curves with fast rises ( although their radio flux on - sets are rarely caught ) and slow exponential decays , in good correlation with the x - ray light curves ( see ref.@xcite for a review ) . the radio emission regions in this class of systems are never spatially resolved and have been modelled as synchrotron bubble events , i.e. , expanding spherical bubbles of relativistic plasma . the origin and acceleration mechanism of the high energy electrons is still not identified , but is believed to be related to the enhanced mass accretion rate in the inner accretion disk region , inferred from the correlation between the radio and x - ray light curves . it is worthwhile to note that such mass accretion and radio emission correlation can not be firmly established in those active galactic nuclei ( agn ) , since the dynamical time scales in agn are typically larger by a factor of 10@xmath25 than the black hole binary systems . the radio light curve of cyg x-1 is very similar to its x - ray light curve , i.e. , persistent , variable and with two distinct states . its high radio flux state corresponds to its hard x - ray high and soft x - ray low flux state , i.e. , the so - called hard / low state @xcite . although the hard / low state mass accretion rate may be slightly lower than that in its soft / high state , any changes in mass accretion rate through the inner accretion disk region may not play a strong role@xcite . therefore the observed radio , soft and hard x - ray correlation suggests that the radio flux increase is not simply related to the mass accretion rate increase . instead , the radio flux seems to respond to the hard x - ray flux positively . it is therefore possible that the much higher energy ( 10@xmath23 ev ) electrons responsible for the radio flux production are somehow related to the much lower energy ( 10@xmath26 ev ) electrons , despite that the former population of electrons are distributed far away from the central region of the disk , where the hard x - ray photons are produced . perhaps the much higher energy electrons originate from the lower energy electrons and they share the same initial acceleration mechanism . exactly how these electrons are accelerated and why the acceleration process can be maintained stably still remain poorly understood . another type - iii x - ray light curve source , lmc x-3 , has never been observed at any radio frequencies , with an upper limit of @xmath4 0.3 mjy at 3 and 6 cm @xcite . this would not be surprising if its radio luminosity is similar to that from cyg x-1 , since it is located at about a factor of 20 farther away than cyg x-1 and the radio flux from cyg x-1 is only between a few and tens of mjy . the detection of a strong radio flux has been reported from the third type iii source lmc x-1 at 81 mjy at 6 cm @xcite . assuming isotropy , its absolute luminosity is about a factor of 3000 _ stronger _ than cyg x-1 , about a factor of 2 _ brighter _ than the peak luminosity of the superluminal jet sources groj1655 - 40 , and _ comparable _ to the peak luminosity of the other superluminal jet source grs1915 + 105 . we caution the readers that a later observation of lmc x-1 with the same instrument did not detect the source at an upper limit of @xmath3 1 mjy @xcite . therefore more radio observations of the source are still needed to clarify the situation . the radio light curves of the two superluminal jet sources , i.e. , groj1655 - 40 @xcite and grs1915 + 105 @xcite , exhibit close , but complicated relationships with their type ii x - ray light curves . their peak radio luminosity is also much higher than most of the type i or iii systems discussed above . in grs1915 + 105 , there seems to exist an almost one to one correspondence between the radio flares and the _ hard _ x - ray outbursts@xcite . however , there seems to be three different kinds of radio flare states@xcite . the relationship between the particular type of the radio flare and the hard x - ray emission properties is not clear . the correlation between radio intensity and the hard x - ray intensity is positive below about 100 mjy ; above this level , the correlation becomes negative@xcite . in groj1655 - 40 , a dynamically confirmed black hole binary , the initial hard x- ray outburst@xcite was accompanied with a bright radio flare resulting from superluminal jet ejection@xcite . all first three hard x - ray outbursts were well correlated with radio flares ( superluminal ejection events)@xcite , although the overall level of the radio emission followed a slow exponential decay . two subsequent hard x - ray outbursts , however , were not accompanied with any detectable radio emission above 5 mjy between 1.5 - 15 ghz@xcite . therefore , this seems to indicate that significant hard x - ray production is a necessary , but not sufficient condition for jet ejection . therefore models@xcite involving hard x - ray production from the relativistic jets seem unlikely@xcite . nevertheless , what we have learned from the radio and hard x - ray correlations are still important for understanding some problems . the positive x - ray and radio flux correlation indicates that jet ejection events are related to increased mass accretion . it is widely accepted that a magnetic field should be an essential ingredient in the jet formation according to the blandford and payne model@xcite . therefore it is very likely that rather strong magnetic fields exist in these systems . if the hard x - ray component is indeed produced by the long suspected comptonization in an optically thin ( inferred from the rather steep power law ) corona , then the correlation also supports the jet production model in which the jets are produced from the interaction between the corona and the magnetic field@xcite . the black hole ( candidate ) binary system gx339 - 4 is also a variable radio source@xcite . intense ( @xmath3 jy ) radio flares similar to those from groj1655 - 40 and grs1915 + 105 have never been observed . we , however , can not conclude that the source has never produced radio flares because of the rather sporadic observations from the southern hemisphere . because of this , the relationship between its radio and x - ray fluxes is not completely clear . the recent observation of a possible radio jet in gx339 - 4 @xcite , during an x - ray active state , may provide evidence that its radio behavior is similar to the two superluminal jet sources . we may eventually be able to conclude that all the type ii or multiple - outbursts systems share some common radio emission properties , related to their high energy , especially their hard x - ray , behavior . an essential question to be answered concerns the physical connections between the types of the companion stars , orbital separations , optical and ir light curves , x - ray and hard x - ray light curves , and radio light curves . a possible scenario follows : their companion stars and orbital separations determine the mass transfer property from the companion stars to the outer regions of the disk , this in turn controls their optical properties , at least during the initial outburst after a quiescence period . the mass flow rates through the inner disk boundary onto the black hole modulate their soft and hard x - ray light curves . the soft x - ray photons are emitted from the optically thick inner accretion disk region , and the hard x - ray photons are believed to be produced by inverse compton up - scattering of low energy photons ( probably the soft photons from the disk ) by higher energy electrons , although the origin of the electrons is still not unambiguously identified . somehow at least a portion of these electrons are accelerated to higher energies and move far away from the central disk region to produce radio fluxes via incoherent synchrotron radiation . this global picture agrees qualitatively to the three types of x - ray and radio light curves , through our investigations of the multifrequency connections in these systems . the correlation between the three types of companion stars and their corresponding light curves suggest that in the type ii or multiple - outbursts systems another instability , possibly the mass overflow instability in the companion star caused by the x - ray heating of the central soft and hard x - ray sources , may play a major role in the subsequent outbursts since the initial one . in the type iii or persistent systems , perhaps the feedback of the x - ray heating to the high mass companion stars reaches a steady state so that these systems become persistent sources . the above picture does not explain why some systems produce powerful jets , while others apparently do not do so . it is possible that in groj1655 - 40 the inferred mass transfer rate from the x - ray observations is only a portion of the total mass transfer rate from its peculiar companion star . the rest of the material is not transferred through the disk , but instead forms a disk ` wind ' , available for producing the powerful jets . in other systems the disk ` wind ' may be very weak . in grs1915 + 105 , although the exact type of its companion is unknown as we discussed above , there indeed appears to be some evidence for a significant wind from the system@xcite . this scenario , however , still does not explain why the outflows in groj1655 - 40 and grs1915 + 105 are highly collimated . this may be partially explained by the properties of the black holes , as will be discussed below . x - ray observations at kev energies can be used for probing the accretion disk very close to the black hole horizon , thus providing important tools for testing accretion disk models and some general relativistic effects . this is because the peak emission energy of the blackbody spectra from the inner disk region is between one - tenth and several kev@xcite , and that these x - ray photons are much more penetrating than euv photons emitting from the inner disk regions of agns . because of the interactions between the black holes and the accretion disks , x - ray observations can also be used for studying the properties of the black holes@xcite . in this section we will discuss three aspects of probing the accretion inner disk regions with x - ray observations . as we mentioned before , the continuum x - ray spectrum from the inner accretion disk region is well described by the multi - color disk blackbody model@xcite . the spectrum depends only upon the location of the inner accretion boundary , mass of the black hole and the mass flow rate through the inner disk region . for a steady disk in a high mass accretion rate state , the inner disk boundary is assumed to be the last stable orbit of the black hole , beyond which the material falls into the black hole with near radial trajectories . for the purpose of the spectral fitting , the free parameters are the peak blackbody temperature , absorption column density , and the normalization at a given energy . the inner disk radius and the disk inclination angle are coupled in the normalization parameter , so they can not be determined independently . in some cases the system inclination angle and/or the black hole mass can be well constrained through the optical light curve modelling and spectroscopic measurements ( to determine the companion star type ) . therefore the size of the inner disk boundary can also be constrained . the values of the black hole mass , spin and the rotation direction of the material moving around it are all coupled into the size of the inner disk boundary , since the location of the last stable orbit of a black hole depends upon all of them . if the mass of the black hole is also known from optical observations , we then have the possibility to constrain the spin of the black hole . it should be noted that various correction factors to account for spectral hardening ( due to electron scattering ) and relativity effects have to be included in this process . this has been done recently in a number of black hole systems , whose system parameters have been well constrained from optical observations@xcite . in figure 5 , the peak blackbody temperature is shown as the functions of the black hole mass , spin and mass accretion rate . it has been found that the two superluminal jet sources groj1655 - 40 and grs1915 + 105 ( by combining the interpretation of its 67 hz qpos as the @xmath27-mode oscillations@xcite ) contain most likely a black hole spinning at near the maximally allowed rate , while several other black holes with the observed soft components , but no relativistic jets , contains slowly or non - spinning black holes . it is also proposed that several ` hard ' x - ray transients , i.e. , no soft component has ever been observed from them , may contain rapidly spinning black holes with retrograde disks @xcite . therefore all types of observed black hole binary systems are naturally unified within one simple scheme . the state transitions in cyg x-1 are proposed to be due to the disk rotation direction reversal , caused by the so - called ` flip - flop ' type instability in wind accreting systems@xcite . future higher sensitivity observations , especially below 1 kev , are needed to test this scheme . the first spectral line feature from a black hole binary system was detected with the _ exosat gspc _ in cyg x-1 @xcite . since then the cyg x-1 spectral line feature has been observed with _ tenma_@xcite , _ ginga_@xcite and more recently with _ asca_@xcite . from the high resolution asca observations @xcite , the overall feature can be modelled by a narrow ( equivalent width 10 - 30 ev ) iron _ k _ emission line at 6.4 kev and an iron @xmath28 edge at 7 kev representing a reflection component , from the outer part of the accretion disk . the disk is inferred to be ionized , with a covering angle @xmath29 . due to the limited statistics , minor contribution due to the reflections from the companion star and/or fluorescence emission from the inner accretion disk region can not be excluded . the small or null contribution from the inner disk region may be due to the much higher degree of ionization of the inner disk region . however , it should be noted that these observations were made during the regular hard / low state of cyg x-1 . in the hard / low state , its inner disk radius is probably larger than 3@xmath30@xcite . therefore the inner disk contribution to the observed line profile is expected to be less important in the hard / low state . moreover , the broadening and skewness of the expected line profile from the inner disk region is less significant for a small inclination angle ( only 10@xmath31 - 40@xmath31 in cyg x-1 ) . future higher sensitivity and better resolution instruments are required for unambiguously identifying these different components in its iron line profile . recently , iron line features have also been observed from the two superluminal jet sources groj1655 - 40 and grs1915 + 105 with asca@xcite , as shown in figure 6 . except during a light curve ` dip ' in groj1655 - 40 , all other observed line profiles show strong absorption features at around 7 kev . a possible interpretation is that the continuum spectrum passes through a slab - like ionized absorption medium ( accretion disk corona ) at a distance about 10@xmath24 cm away from the original x - ray source@xcite . the degree of ionization of the corona increases for a higher continuum luminosity , indicating that the ionization is also caused by the x - ray illumination . this is perhaps the first evidence of the existence of a hot corona in a black hole binary system . it should be noted that these observed absorption features are significantly different from the apparent emission feature observed in cyg x-1 . it is not clear if this is due to the differences in their nature ( for example , jet vs. non - jet systems ) , or the inclination angle difference ( 70@xmath31 vs 10@xmath31 - 40@xmath31 ) . on the other hand , the apparent emission feature in the ` dip ' spectrum of groj1655 - 40 should not be ignored completely . this feature is very similar to the broadened and skewed line profiles observed from many seyfert i galaxies ( e.g. ref . the only difference seems to be the much higher peak energy at @xmath37.0 kev , possibly blue - shifted , if the original peak is also the 6.4 iron @xmath32 line . this is actually expected due to the much higher inclination angle of 70@xmath31 in groj1655 - 40 . it is also unlikely that the whole complex line features of other spectra ( non-`dip ' ) of the sources can be explained completely with absorptions alone . it is possible that the non-`dip ' line profiles consist of both the absorption by the disk - corona and the iron fluorescence emission ( relativistically broadened and skewed ) from the inner disk region , which is believed to be around 1 @xmath30@xcite . one possible scenario is that during the ` dip ' period the emitted x - ray luminosity from the central region of the disk becomes lower , due to perhaps the reduced mass transfer rate . the lower x - ray flux therefore changes the properties of the disk - corona ( geometry and ionization state ) , so that the resonance absorption of x - ray photons by iron ions becomes negligible . it is clear from the above discussion that rich information may be derived from iron line diagnostics of black hole binary systems . many of the current interpretations are , however , quite uncertain , due primarily to the limited number of detections and the poor statistical quality of past observations . rapid variability studies of their x - ray light curves provide essential information about many dynamic processes in the central region of the accretion disk and in the corona . many of the earlier results were reviewed by van der klis@xcite . some recent results are reviewed in a companion review article in the same proceedings@xcite . here we only review briefly the high frequency qpos detected from the two superluminal jet sources groj1655 - 40 at @xmath3300 hz@xcite and grs1915 + 105 at @xmath367 hz@xcite . an important feature in these qpos is their _ stable _ frequency , compared to the majority of other qpos observed in black hole and neutron star binaries , with the exception of the stable 34 hz qpo in the bursting pulsar groj1744 - 28 @xcite . it is thus natural to relate these qpos to the last stable orbit of the black hole . assuming that the qpo frequencies are actually the keplerian frequencies of the last stable orbit , the masses of the assumed non - spinning black holes are around 33 m@xmath2 and 7 m@xmath2 ( consistent with its dynamical mass estimate ) , in grs1915 + 105 and groj1655 - 40 , respectively . however , since the spin of the black hole in groj1655 - 40 is inferred to be near its maximal rate from x - ray spectroscopic measurements as we discussed above@xcite , the non - spinning black hole assumption is not likely to be valid and thus neither is this keplerian frequency interpretation . the other interpretation is to attribute these qpos to the @xmath27-mode oscillations due to the general relativity effects@xcite . when applied to groj1655 - 40 , the inferred black hole spin rate is remarkably consistent with that obtained from independent spectroscopic measurements@xcite . taking the same approach , the implied black hole spin in grs1915 + 105 would also be near its maximal rate and the black hole mass is around 30 m@xmath2 . in this model , however , it is not clear why the qpos seem to be associated with the hard x - ray component @xcite . we have reviewed the recent progress in the investigations of the multifrequency properties and connections of galactic black hole binaries . there is now sufficient evidence for the existence of stellar mass black holes in at least 10 systems . their high energy and radio light curves are well correlated and can be divided into three classes , which may be related to their respective types of companion stars . their x - ray spectral properties may be related closely to the spins of the black holes , in a newly proposed unification scheme of all types of black hole binaries . the correlation between the hard x - ray and radio flares suggests that the radio emission and therefore the associated outflows are related to the corona in the inner disk region . iron line studies in the two superluminal jet sources provide possible evidence of a hot disk - corona . only the black holes in the jet systems are found to be likely spinning at nearly the maximal rates , suggesting that the highly relativistic jets are related to the spin of the black hole . this supports strongly the previously suggestions that the highly relativistic jets in agns might be related to the rapid black hole spins @xcite . therefore the eventual unification of all types of black hole systems ( galactic and extra - galactic ) becomes possible using the black hole spin as one of the key parameters . however , many of the current interpretations are still quite uncertain , due to both of our limited theoretical understanding of the detailed physics involved and the limitations of the current observations . therefore more theoretical investigations and future multifrequency observations are required . nevertheless , such recent progress suggests that the current observational and theoretical studies of the galactic black hole binary systems have begun to show the premise of using these systems as the laboratories for testing some physical laws in the strongest gravitational field . compared to agns in which similar studies can also be carried out and in fact were started much earlier , the galactic black hole binary systems are more advantageous because a ) the dynamical time scales in them are much shorter ; b ) their inner disk regions are directly observable in the x - ray frequency range ; c ) the fine details of their jets are observable because they are much closer to us ; and d ) the system parameters can be measured through optical / infrared observations . we thank wan chen , ken ebisawa , jerry orosz , craig robinson and kinwah wu for kindly providing data and figures to be included in this review paper . part of the v - band light curve of groj1655 - 40 , provided by jerry orosz and obtained by charles bailyn and various yale observers and by jeff mcclintock , has not been published . snz also appreciates those stimulating discussions with many of his colleagues , including didier barret , sandip chakrabarti , wei cui , wan chen , ken ebisawa , rob fender , jerry fishman , eric grove , jean - pierre lasota , hui li , edison liang , mario livio , jeff mcclintock , ramesh narayan , bill paciesas , ron remillard , craig robinson , marco tavani , lev titarchuk , jan van paradijs and kinwah wu , etc . augusteijn , t. , kuulkers , e. & shahams , j. 1993 , _ a&a _ , * 279 * , l13 bailyn , c.d . _ et al . _ 1995 , _ nature _ , * 374 * , 701 bailyn , c.d . _ et al . _ 1995 , _ nature _ , * 378 * , 157 barr , p. , white , n.e . & page , c.g . 1985 , _ mnras _ , * 216 * , 65 barret , d. , mcclintock , j.e . & grindlay , j.e . 1997 , _ apj _ , * 473 * , 963 blandford , r. 1976 , _ mnras _ , * 176 * , 465 blandford , r. & payne , d. 1982 , _ mnras _ , * 199 * , 883 blandford , r. & znajek , r.l . 1977 , _ mnras _ , * 179 * , 433 blandford , r. & levinson , a. 1995 , _ apj _ , * 441 * , 79 bolton , c.t . 1972 , _ nature _ , * 258 * , 307 callanan , p.j . _ et al . _ 1996 , _ apj _ , * 461 * , 351 callanan , p.j . _ et al . _ 1996 , _ apj _ , * 470 * , l57 casares , j. , charles , p. a. , & naylor , t. 1992 , _ nature _ , * 355 * , 614 casares , j. , charles , & p. a. 1992 , _ mnras _ , * 255 * , 7 casares , j. _ et al . _ 1995 , _ mnras _ , * 276 * , l35 castro - tirado , a.j . , geballe , t.r . & lund , n. 1996 , _ apj _ * 461 * , l99 chaty , s. _ et al . _ 1996 _ a&a _ , * 310 * , 825 chen , w. , shrader , c. & livio , m. 1997 , _ apj _ , in press . cowley , a.p . _ et al . _ 1983 , _ apj _ , * 272 * , 118 chen , w. , livio , m. & gehrels , n. 1993 , _ apj _ , * 408 * , l5 done , c. _ et al . _ 1992 , _ apj _ , * 395 * , 275 ebisawa , k. , titarchuk , l. & chakrabarti , s.k . 1996 , _ pasj _ , * 48 * , 59 ebisawa , k. , mitsuda , k. & inoue , h. 1989 , _ pasj _ , * 41 * , 519 ebisawa , k. _ et al . _ 1992 , _ frontiers of x - ray astronomy _ y. tanaka & k. koyama ( tokyo : university academy press ) , p.301 ebisawa , k. _ et al . _ 1992 , in _ apj _ , * 367 * , 213 ebisawa , k. _ et al . _ 1996 , _ apj _ , * 467 * , 419 ebisawa , k. 1996 , in _ x - ray imaging and spectroscopy of cosmic hot plasmas _ f. makino & k. mitsuda ( ( tokyo : university academy press ) , p.427 fabian , a.c . _ et al . _ 1995 , _ mnras _ , * 277 * , l11 fender , r.p . _ et al . _ 1997 , _ mnras _ , * 286 * , l29 fender , r.p . 1997 , private communication filippenko , a.v . matheson , t. & ho , l.c . 1995 , _ apj _ , * 455 * , 614 foster , r.s . _ et al . _ 1996 , _ apj _ , * 476 * , l81 gies , d.r . & bolton , c.t . 1982 , _ apj _ , * 260 * , 240 gies , d.r . & bolton , c.t . 1986 , _ apj _ , * 304 * , 371 grove , j.e . , kroeger , r.a . & strickman , m.s . 1996 , in _ second integral workshop , the transparent universe _ , st . malo , france , esa sp-382 grove , j.e . _ et al . _ 1997 , in _ the 4th compton symposium _ , these proceedings han , x. & hjellming , r.m . 1992 , _ iauc _ , 5593 han , x. & hjellming , r.m . 1997 , _ apj _ , in press ( submitted in 1994 ) harmon , b.a . _ et al . _ 1997 , _ apj _ , * 477 * , l85 harmon , b.a . _ et al . _ 1995 , _ nature _ , * 374 * , 703 hameury , j .- m . , king , a.r . , & lasota , j .- 1986 , _ a&a _ , * 161 * , 71 hameury , j .- _ et al . _ 1997 , _ apj _ , in press . haswell , c.a . _ et al . _ 1994 , _ apj _ , * 411 * , 802 hjellming , r.m . & han , x. in _ x - ray binaries _ , eds . walter h.g . lewin , jan van paradijs & p.j . van den heuvel , cambridge university press , 1995 , ch . 7 hjellming , r.m . & rupen , m.p . _ et al . _ 1995 , _ nature _ , * 375 * , 464 hjellming , r.m . _ et al . _ 1988 , _ apj _ , * 355 * , l75 hjellming , r.m . _ et al . _ 1997 , in preparation hutchings , j.b . _ et al . _ 1987 , _ aj _ , * 94 * , 340 iwasawa , k. _ et al . _ 1997 , _ mnras _ , * 279 * , 837 kesteven , m.j . & turtle , a.j . 1991 , _ iauc _ , 5181 king , a.r . , kolb , u. & szuszkiewicz , e. 1997 , _ apj _ , * 488 * , in press kitamoto , s. _ et al . _ 1990 , in _ accretion powered compact binaries _ mauche ( cambridge university press ) , p.21 kitamoto , s. _ et al . _ 1990 , _ pasj _ , * 42 * , 85 kitamoto , s. _ et al . _ 1997 , submitted to _ apj _ kolb , u. _ et al . _ 1997 , _ apj _ , * 485 * , l33 kroeger , r.a . _ et al . _ 1996 , _ a&as _ , * 120 * , c117 lasota , j.p . 1997 , in _ accretion phenomena and related outflows _ , ed . wickramasinghe , g.v . bicknell & l. ferrario , asp conf . series vol . 121 levinson , a. & blandford , r. _ et al . _ 1996 , _ apj _ , * 456 * , l29 livio , m. 1997 , in _ accretion phenomena and related outflows _ , ed . wickramasinghe , g.v . bicknell & l. ferrario , asp conf . series vol . 121 makishima , k. _ et al . _ 1986 , _ apj _ , * 308 * , 635 martin , a.c . _ et al . _ 1995 , _ mnras _ , * 274 * , l46 mcclintock , j.e . & remillard , r.a . 1986 , _ apj _ , * 308 * , 110 meier , d. 1996 , _ apj _ , * 459 * , 185 meier , d. _ et al . _ 1997 , _ nature _ , * 388 * , 350 mineshige , s. 1996 , _ pasj _ , * 48 * , 93 mirabel , i.f . & rodriguez , l.f . 1994 , _ nature _ , * 371 * , 46 mirabel , i.f . _ et al . _ 1996 , _ apj _ , * 472 * , l111 mirabel , i.f . _ et al . _ 1997 , _ apj _ , * 477 * , l45 mitsuda , k. _ et al . _ 1989 , _ pasj _ , * 41 * , 97 mitsuda , k. _ et al . _ 1984 , _ pasj _ , * 36 * , 741 morgan , e.h . , remillard , r.a . & greiner , j. 1997 , _ apj _ , * 482 * , 993 morris , p.w . 1997 , _ apj _ , in press narayan , r. 1997 , in _ accretion phenomena and related outflows _ , ed . wickramasinghe , g.v . bicknell & l. ferrario , asp conf . series vol . 121 nowak , m.a . & wagoner , r.v . 1993 , _ apj _ , * 418 * , 187 nowak , m.a . _ et al . _ 1996 , _ apj _ , * 477 * , l91 orosz , j.a . 1996 , phd thesis , yale university orosz , j.a . , & bailyn , c.d . 1995 , _ apj _ , * 446 * , l59 orosz , j.a . & bailyn , c.d . 1997 , _ apj _ , * 477 * , 876 orosz , j.a . _ et al._1996 , _ apj _ , * 468 * , 380 orosz , j.a . _ et al . _ 1997 , _ apj _ , * 478 * , l83 owen , f.n . _ et al . _ 1976 , _ apj _ , * 203 * , l15 paciesas , w.s . _ et al . _ 1997 , in _ the 4th compton symposium _ van paradijs , j. & mcclintock , j. e .. , in _ x - ray binaries _ , eds . lewin , j. van paradijs & p.j . van den heuvel , cambridge university press , 1995 perez , c.a . _ et al . _ 1995 , _ apj _ , * 476 * , 589 priedhorsky , w.c . , _ et al . _ 1983,_apj _ * 270 * , 233 . priedhorsky , w.c . _ et al . _ a&a _ , * 300 * , 415 remillard , r.a . _ et al . _ 1994 , _ baas _ , * 26 * , 1483 remillard , r.a . _ et al . _ 1996 , _ apj _ , * 459 * , 226 remillard , r. 1996 , in _ texas in chicago , texas symposium _ , chicago robinson , c.r .. _ et al . _ 1997 , in _ the 4th compton symposium _ rodriguez , l.f . & mirabel , i.f . 1997 , _ apj _ , * 474 * , l123 rubin , b.c . _ et al . _ 1997 , _ apj _ , in press sanwal , d. , _ et al . _ 1996 , _ apj _ , * 460 * , 437 shahbaz , t. , naylor , t. , & charles , p.a . 1994 , _ mnras _ , * 268 * , 756 shahbaz , t. _ et al . _ 1994 , _ mnras _ , 271 , l10 sood , r. & campbell - wilson , d. 1994 , _ iauc _ 6006 spencer , r.e . _ et al . _ 1997 , _ vistas in astronomy _ , * 41(1 ) * , 37 tanaka , y. _ et al . _ 1995 , _ nature _ , * 375 * , 659 tanaka , y. 1991 , in _ lecture notes in physics 385 , iron line diagnostics in x - ray sources _ , ed . a. treves ( berlin : springer ) , 98 . tavani , m. _ et al . _ 1996 , _ apj _ , * 473 * , 103 tingay , s.j . _ et al . _ 1995 , _ nature _ , * 374 * , 141 ueda , y. , _ et al . _ 1997 , _ apj _ , in press van der hooft , f. _ et al . _ 1997 , _ mnras _ , * 286 * , 43 van der klis , m. , in _ x - ray binaries _ , eds . walter h.g . lewin , jan van paradijs & p.j . van den heuvel , cambridge university press , 1995 , ch . 6 , pp . 252 - 307 . van paradijs , j. 1996 , _ apj _ , * 464 * , l39 wagner , r.m . _ et al . _ 1990 , in _ accretion powered compact binaries _ mauche ( cambridge university press ) , p. wagner , r.m . _ et al . _ 1992 , _ apj _ , * 401 * , l97 webster , b.l . & murdin , p. 1972 , _ nature _ , * 235 * , 37 wu , k. & hunstead , r. w. 1997 , in _ accretion phenomena and related outflows _ wickramasinghe , g.v . bicknell & l. ferrario , asp conf . series vol . 121 zhang , s.n . , _ et al . _ 1994 , _ iauc _ 6046 zhang , s.n . 1997 , in _ accretion phenomena and related outflows _ , ed . wickramasinghe , g.v . bicknell & l. ferrario , asp conf . series vol . 121 zhang , s.n . , _ et al . _ a&a _ * 120 * , c279 zhang , s.n . , robinson , c.r . & cui , w. 1996 , _ iauc _ 6510 zhang , s.n . , _ et al . _ 1997 , _ apj _ * 477 * , l95 zhang , s.n . , _ et al . _ 1997 , _ apj _ * 479 * , 381 zhang , s.n . , cui , w. & chen , w. 1997 , * 482 * , l155 zhang , s.n . , _ et al . _ 1997 , _ apj _ , to be submitted . zhang , w. , _ et al . _ 1995 , _ apj _ * 449 * , 930

consider the isentropic compressible navier - stokes equations @xmath2 on the quarter - plane @xmath3 , where @xmath4 , @xmath5 , @xmath6 denote density , velocity , and pressure at spatial location @xmath7 and time @xmath8 , with @xmath0-law pressure function @xmath9 and noncharacteristic constant `` inflow '' or `` outflow '' boundary conditions @xmath10 or @xmath11 as discussed in @xcite . the sign of the velocity at @xmath12 determines whether characteristics of the hyperbolic transport equation @xmath13 enter the domain ( considering @xmath14 as a lower - order forcing term ) , and thus whether @xmath15 should be prescribed . the variable - coefficient parabolic equation @xmath16 requires prescription of @xmath17 in either case , with @xmath18 . by comparison , the purely hyperbolic isentropic euler equations @xmath19 have characteristic speeds @xmath20 , hence , depending on the values of @xmath21 , may have one , two , or no characteristics entering the domain , hence require one , two , or no prescribed boundary values . in particular , there is a discrepancy between the number of prescribed boundary values for and in the case of mild inflow @xmath22 small ( two for , one for ) or strong outflow @xmath23 large ( one for , none for ) , indicating the possibility of _ boundary layers _ , or asymptotically - constant stationary solutions of : @xmath24 indeed , existence of such solutions is straightforward to verify by direct computations on the ( scalar ) stationary - wave ode ; see @xcite or section [ profsec ] . these may be either of `` expansive '' type , resembling rarefaction wave solutions on the whole line , or `` compressive '' type , resembling viscous shock solutions . a fundamental question is whether or not such boundary layer solutions are _ stable _ in the sense of pde . for the expansive inflow case , it has been shown in @xcite that _ all _ boundary layers are stable , independent of amplitude , by energy estimates similar to those used to prove the corresponding result for rarefactions on the whole line . here , we concentrate on the complementary , _ compressive case _ ( though see discussion , section [ discussion ] ) . linearized and nonlinear stability of general ( expansive or compressive ) _ small - amplitude _ noncharacteristic boundary layers of have been established in @xcite . more generally , it has been shown in @xcite that linearized and nonlinear stability are equivalent to spectral stability , or nonexistence of nonstable ( nonnegative real part ) eigenvalues of the linearized operator about the layer , for boundary layers of arbitrary amplitude . however , up to now the spectral stability of _ large - amplitude compressive _ boundary layers has remained largely undetermined . we resolve this question in the present paper , carrying out a systematic , global study classifying the stability of all possible compressive boundary - layer solutions of . our method of analysis is by a combination of asymptotic ode techniques and numerical evans function computations , following a basic approach introduced recently in @xcite for the study of the closely related shock wave case . here , there are interesting complications associated with the richer class of boundary - layer solutions as compared to possible shock solutions , the delicate stability properties of the inflow case , and , in the outflow case , the nonstandard eigenvalue problem arising from reduction to lagrangian coordinates . our conclusions are , for both inflow and outflow conditions , that compressive boundary layers that are uniformly noncharacteristic in a sense to be made precise later ( specifically , @xmath25 bounded away from @xmath26 , in the terminology of section [ profsec ] ) are _ unconditionally stable _ , independent of amplitude , on the range @xmath27 $ ] considered in our numerical computations . we show by energy estimates that _ outflow boundary layers are stable also in the characteristic limit_. the omitted characteristic limit in the inflow case , analogous to the small - amplitude limit for the shock case should be treatable by the singular perturbation methods used in @xcite to treat the small - amplitude shock case ; however , we do not consider this case here . in the inflow case , our results , together with those of @xcite , completely resolve the question of stability of isentropic ( expansive or compressive ) uniformly noncharacteristic boundary layers for @xmath28 $ ] , yielding _ unconditional stability independent of amplitude or type . _ in the outflow case , we show stability of all compressive boundary layers without the assumption of uniform noncharacteristicity . the small - amplitude results obtained in @xcite are of `` general type '' , making little use of the specific structure of the equations . essentially , they all require that the difference between the boundary layer solution and its constant limit at @xmath29 be small in @xmath30 . be small in @xmath30 . ( for monotone profiles , @xmath31 . ) ] as pointed out in @xcite , this is the `` gap lemma '' regime in which standard asymptotic ode estimates show that behavior is essentially governed by the limiting constant - coefficient equations at infinity , and thus stability may be concluded immediately from stability ( computable by exact solution ) of the constant layer identically equal to the limiting state . these methods do not suffice to treat either the ( small - amplitude ) characteristic limit or the large - amplitude case , which require more refined analyses . in particular , up to now , _ there was no analysis considering boundary layers approaching a full viscous shock profile _ , _ not even a profile of vanishingly small amplitude_. our analysis of this limit indicates why : the appearance of a small eigenvalue near zero prevents uniform estimates such as would be obtained by usual types of energy estimates . by contrast , the large - amplitude results obtained here and ( for expansive layers ) in @xcite make use of the specific form of the equations . in particular , both analyses make use of the advantageous structure in lagrangian coordinates . the possibility to work in lagrangian coordinates was first pointed out by matsumura nishihara @xcite in the inflow case , for which the stationary boundary transforms to a moving boundary with constant speed . here we show how to convert the outflow problem also to lagrangian coordinates , by converting the resulting variable - speed boundary problem to a constant - speed one with modified boundary condition . this trick seems of general use . in particular , it might be possible that the energy methods of @xcite applied in this framework would yield unconditional stability of expansive boundary - layers , completing the analysis of the outflow case . alternatively , this case could be attacked by the methods of the present paper . these are two further interesting direction for future investigation . in the outflow case , a further transformation to the `` balanced flux form '' introduced in @xcite , in which the equations take the form of the integrated shock equations , allows us to establish stability in the characteristic limit by energy estimates like those of @xcite in the shock case . the treatment of the characteristic inflow limit by the methods of @xcite seems to be another extremely interesting direction for future study . finally , we point to the extension of the present methods to full ( nonisentropic ) gas dynamics and multidimensions as the two outstanding open problems in this area . new features of the present analysis as compared to the shock case considered in @xcite are the presence of two parameters , strength and displacement , indexing possible boundary layers , vs. the single parameter of strength in the shock case , and the fact that the limiting equations in several asymptotic regimes possess zero eigenvalues , making the limiting stability analysis much more delicate than in the shock case . the latter is seen , for example , in the limit as a compressive boundary layer approaches a full stationary shock solution , which we show to be spectrally equivalent to the situation of unintegrated shock equations on the whole line . as the equations on the line possess always a translational eigenvalue at @xmath32 , we may conclude existence of a zero at @xmath32 for the limiting equations and thus a zero _ near _ @xmath32 as we approach this limit , which could be stable or unstable . similarly , the evans function in the inflow case is shown to converge in the large - strength limit to a function with a zero at @xmath32 , with the same conclusions ; see section [ description ] for further details . to deal with this latter circumstance , we find it necessary to make use also of topological information provided by the stability index of @xcite , a mod - two index counting the parity of the number of unstable eigenvalues . together with the information that there is at most one unstable zero , the parity information provided by the stability index is sufficient to determine whether an unstable zero does or does not occur . remarkably , in the isentropic case we are able to compute explicitly the stability index for all parameter values , recovering results obtained by indirect argument in @xcite , and thereby completing the stability analysis in the presence of a single possibly unstable zero . we begin by carrying out a number of preliminary steps similar to those carried out in @xcite for the shock case , but complicated somewhat by the need to treat the boundary and its different conditions in the inflow and outflow case . the analyses of @xcite in the shock wave case were carried out in lagrangian coordinates , which proved to be particularly convenient . our first step , therefore , is to convert the eulerian formulation into lagrangian coordinates similar to those of the shock case . however , standard lagrangian coordinates in which the spatial variable @xmath33 is constant on particle paths are not appropriate for the boundary - value problem with inflow / outflow . we therefore introduce instead `` psuedo - lagrangian '' coordinates @xmath34 in which the physical boundary @xmath12 remains fixed at @xmath35 . straightforward calculation reveals that in these coordinates becomes @xmath36 on @xmath37 , where @xmath38 so that @xmath39 is the negative of the momentum at the boundary @xmath40 . from now on , we drop the tilde , denoting @xmath33 simply as @xmath7 . for the inflow case , @xmath22 so we may prescribe _ two _ boundary conditions on , namely @xmath41 where both @xmath42 are constant . for the outflow case , @xmath23 so we may prescribe _ only one _ boundary condition on , namely @xmath43 thus @xmath44 is an unknown in the problem , which makes the analysis of the outflow case more subtle than that of the inflow case . our next step is to rescale the equations in such a way that coefficients remain bounded in the strong boundary - layer limit . consider the change of variables @xmath45 where @xmath46 is chosen so that @xmath47 where @xmath25 is the limit as @xmath48 of the boundary layer ( stationary solution ) @xmath49 under consideration and @xmath50 is the limit as @xmath51 of its continuation into @xmath52 as a solution of the standing - wave ode ( discussed in more detail just below ) . under the rescaling , becomes @xmath53 where @xmath54 , @xmath55 , on respective domains @xmath56 stationary boundary layers @xmath57 of satisfy @xmath58 where ( d ) is imposed at @xmath59 in the inflow case , @xmath60 in the outflow case and ( imposing @xmath61 ) @xmath62 . using ( a ) we can reduce this to the study of the scalar ode , @xmath63 with the same boundary conditions at @xmath12 and @xmath64 as above . taking the antiderivative of this equation yields @xmath65 where @xmath66 is a constant of integration . noting that @xmath67 is convex , we find that there are precisely two rest points of whenever boundary - layer profiles exist , except at the single parameter value on the boundary between existence and nonexistence of solutions , for which there is a degenerate rest point ( double root of @xmath67 ) . ignoring this degenerate case , we see that boundary layers terminating at rest point @xmath25 as @xmath48 must either continue backward into @xmath52 to terminate at a second rest point @xmath50 as @xmath51 , or else blow up to infinity as @xmath51 . the first case we shall call _ compressive _ , the second _ expansive_. in the first case , the extended solution on the whole line may be recognized as a standing viscous shock wave ; that is , _ for isentropic gas dynamics , compressive boundary layers are just restrictions to the half - line @xmath68 [ resp . @xmath69 of standing shock waves . _ in the second case , as discussed in @xcite , the boundary layers are somewhat analogous to rarefaction waves on the whole line . from here on , we concentrate exclusively on the compressive case . with the choice @xmath70 , we may carry out the integration of once more , this time as a definite integral from @xmath60 to @xmath7 , to obtain @xmath71 where @xmath72 is found by letting @xmath73 , yielding @xmath74 in particular , @xmath75 in the large boundary layer limit @xmath76 . this is exactly the equation for viscous shock profiles considered in @xcite . linearizing about @xmath77 , we obtain @xmath78 where @xmath79 , @xmath80 and @xmath81 denote perturbations of @xmath82 . in the inflow case , @xmath83 , yielding @xmath84 on @xmath37 , with full dirichlet conditions @xmath85 . letting @xmath86 , @xmath87 , and denoting by @xmath88 the operator associated to the linearization about boundary - layer @xmath89 , @xmath90 where @xmath91 we have @xmath92 , with associated eigenvalue equation @xmath93 where @xmath94 . to eliminate the nonstandard inhomogeneous term on the righthand side of , we introduce a good unknown " ( c.f . @xcite ) @xmath95 since @xmath96 by differentiation of the boundary - layer equation , the system expressed in the good unknown becomes simply @xmath97 or , equivalently , with boundary conditions @xmath98 solving for @xmath99 in terms of @xmath100 and recalling that @xmath101 by , we obtain finally @xmath102 [ outrmk ] problems and are evidently equivalent for all @xmath103 , but are not equivalent for @xmath32 ( for which the change of coordinates to good unknown becomes singular ) . for , @xmath104 by inspection is a solution of , but is not a solution of . that is , we have introduced by this transformation a spurious eigenvalue at @xmath32 , which we shall have to account for later . [ @xcite ] [ profdecay ] for each @xmath105 , @xmath106 , has a unique ( up to translation ) monotone decreasing solution @xmath107 decaying to endstates @xmath108 with a uniform exponential rate for @xmath25 uniformly bounded away from @xmath70 . in particular , for @xmath109 , [ decaybd ] @xmath110 where @xmath111 is defined by @xmath112 . existence and monotonicity follow trivially by the fact that is a scalar first - order ode with convex righthand side . exponential convergence as @xmath48 follows by @xmath113 whence @xmath114 by @xmath115 for @xmath116 . exponential convergence as @xmath51 follows by a similar , but more straightforward calculation , where , in the `` centered '' coordinate @xmath117 , the constants @xmath118 are uniform with respect to @xmath119 . see @xcite for details . the following estimates are established in appendices [ basicproof ] and [ outbasicproof ] . [ hf ] nonstable eigenvalues @xmath120 of , i.e. , eigenvalues with nonnegative real part , are confined for any @xmath121 to the region @xmath122 for the inflow case , and to the region @xmath123 for the outflow case . setting @xmath124 , we may express as a first - order system @xmath125 where @xmath126 where @xmath127 with @xmath128 as in and @xmath72 as in , or , equivalently , @xmath129 [ shockrel ] the coefficient matrix @xmath130 may be recognized as a rescaled version of the coefficient matrix @xmath131 appearing in the shock case @xcite , with @xmath132 the choice of variables @xmath133 may be recognized as the modified flux form of @xcite , adapted to the hyperbolic parabolic case . eigenvalues of correspond to nontrivial solutions @xmath134 for which the boundary conditions @xmath135 are satisfied . because @xmath136 as a function of @xmath137 is asymptotically constant in @xmath7 , the behavior near @xmath138 of solutions of is governed by the limiting constant - coefficient systems @xmath139 from which we readily find on the ( nonstable ) domain @xmath140 , @xmath103 of interest that there is a one - dimensional unstable manifold @xmath141 of solutions decaying at @xmath142 and a two - dimensional stable manifold @xmath143 of solutions decaying at @xmath144 , analytic in @xmath120 , with asymptotic behavior @xmath145 as @xmath146 , where @xmath147 and @xmath148 are eigenvalues and associated analytically chosen eigenvectors of the limiting coefficient matrices @xmath149 . a standard choice of eigenvectors @xmath150 @xcite , uniquely specifying @xmath151 ( up to constant factor ) is obtained by kato s ode @xcite , a linear , analytic ode whose solution can be alternatively characterized by the property that there exist corresponding left eigenvectors @xmath152 such that @xmath153 where `` @xmath154 '' denotes @xmath155 ; for further discussion , see @xcite . in the inflow case , @xmath156 , we define the _ evans function _ @xmath157 as the analytic function @xmath158 where @xmath159 are as defined above , and @xmath160 is a solution satisfying the boundary conditions @xmath161 at @xmath12 , specifically , @xmath162 with this definition , eigenvalues of @xmath163 correspond to zeroes of @xmath157 both in location and multiplicity ; moreover , the evans function extends analytically to @xmath32 , i.e. , to all of @xmath140 . see @xcite for further details . equivalently , following @xcite , we may express the evans function as @xmath164 where @xmath165 spans the one - dimensional unstable manifold of solutions decaying at @xmath144 ( necessarily orthogonal to the span of @xmath166 and @xmath167 ) of the adjoint eigenvalue ode @xmath168 the simpler representation is the one that we shall use here . in the outflow case , @xmath169 , we define the _ evans function _ as @xmath170 where @xmath171 is as defined above , and @xmath172 are a basis of solutions of satisfying the boundary conditions , specifically , @xmath173 or , equivalently , as @xmath174 where @xmath175 is the solution of the adjoint eigenvalue ode dual to @xmath176 and @xmath177 . [ outevansrmk ] as discussed in remark [ outrmk ] , @xmath178 has a spurious zero at @xmath32 introduced by the coordinate change to `` good unknown '' . we can now state precisely our main results . taking a formal limit as @xmath76 of the rescaled equations and recalling that @xmath179 , we obtain a limiting evolution equation @xmath180 corresponding to a _ pressureless gas _ , or @xmath181 . the associated limiting profile equation @xmath182 has explicit solution @xmath183 @xmath184 ; the limiting eigenvalue system is @xmath185 @xmath186 where @xmath187 convergence of the profile and eigenvalue equations is _ uniform _ on any interval @xmath188 , or , equivalently , @xmath189 , for @xmath190 any positive constant , where the sequence of coefficient matrices is therefore a _ regular perturbation _ of its limit . following @xcite , we call @xmath191 the `` regular region '' . for @xmath192 on the other hand , or @xmath193 , the limit is less well - behaved , as may be seen by the fact that @xmath194 as @xmath195 , a consequence of the appearance of @xmath196 in the expression for @xmath197 . similarly , @xmath136 does not converge to @xmath198 as @xmath48 with uniform exponential rate independent of @xmath25 , @xmath0 , but rather as @xmath199 . as in the shock case , this makes problematic the treatment of @xmath200 . following @xcite we call @xmath200 the `` singular region '' . to put things in another way , the effects of pressure are not lost as @xmath76 , but rather pushed to @xmath144 , where they must be studied by a careful boundary - layer analysis . ( note : this is not a boundary - layer in the same sense as the background solution , nor is it a singular perturbation in the usual sense , at least as we have framed the problem here . ) [ newparam ] a significant difference from the shock case of @xcite is the appearance of the second parameter @xmath201 that survives in the @xmath76 limit . observe that the limiting coefficient matrix @xmath202 is nonhyperbolic ( in ode sense ) for all @xmath120 , having eigenvalues @xmath203 ; in particular , the stable manifold drops to dimension one in the limit @xmath76 , and so the prescription of an associated evans function is _ underdetermined_. this difficulty is resolved by a careful boundary - layer analysis in @xcite , determining a special `` slow stable '' mode @xmath204 augmenting the `` fast stable '' mode @xmath205 associated with the single stable eigenvalue @xmath206 of @xmath207 . this determines a _ limiting evans function _ @xmath208 by the prescription , of section [ evanssec ] , or alternatively via as @xmath209 with @xmath210 defined analogously as a solution of the adjoint limiting system lying asymptotically at @xmath144 in direction @xmath211 orthogonal to the span of @xmath212 and @xmath213 , where `` @xmath214 '' denotes complex conjugate , and @xmath215 defined as the solution of the limiting eigenvalue equations satisfying boundary condition , i.e. , @xmath216 . we have no such difficulties in the outflow case , since @xmath217 remains uniformly hyperbolic , and we may define a limiting evans function @xmath218 directly by , , , at least so long as @xmath201 remains bounded from zero . ( as perhaps already hinted by remark [ newparam ] , there are complications associated with the double limit @xmath219 . ) with the above definitions , we have the following main theorems characterizing the strong - layer limit @xmath76 as well as the limits @xmath220 . [ mainthm ] for @xmath221 and @xmath120 in any compact subset of @xmath140 , @xmath222 and @xmath223 converge uniformly to @xmath208 and @xmath224 as @xmath76 . [ main3 ] for @xmath120 in any compact subset of @xmath140 and @xmath25 bounded from @xmath26 , @xmath222 , appropriately renormalized by a nonvanishing analytic factor , converges uniformly as @xmath225 to the evans function for the ( unintegrated ) eigenvalue equations of the associated viscous shock wave connecting @xmath70 to @xmath25 ; likewise , @xmath224 , appropriately renormalized , converges uniformly as @xmath226 to the same limit for @xmath120 uniformly bounded away from zero . by similar computations , we obtain also the following direct result . [ main2 ] inflow boundary layers are stable for @xmath201 sufficiently small . we have also the following parity information , obtained by stability - index computations as in @xcite . [ index ] for any @xmath105 , @xmath201 , and @xmath25 , @xmath227 , hence the number of unstable roots of @xmath228 is even ; on the other hand @xmath229 and @xmath230 . likewise , @xmath231 , @xmath232 , @xmath233 , hence the number of nonzero unstable roots of @xmath234 , @xmath178 , @xmath218 is even . finally , we have the following auxiliary results established by energy estimates in appendices [ stronglimit ] , [ outnonv ] , [ char ] , and [ nonvanish - expansive - inflow ] . [ redenergy ] the limiting evans function @xmath234 is nonzero for @xmath235 on @xmath236 , for all @xmath237 . the limiting evans function @xmath218 is nonzero for @xmath235 on @xmath236 , for @xmath238 , where @xmath239 is determined by the functional equation @xmath240 . [ charsmallamp ] compressive outflow boundary layers are stable for @xmath25 sufficiently close to @xmath26 . [ expansive ] expansive inflow boundary layers are stable for all parameter values . collecting information , we have the following analytical stability results . [ v01 ] for @xmath201 or @xmath25 sufficiently small , compressive inflow boundary layers are stable . for @xmath201 sufficiently small , @xmath25 sufficiently close to @xmath26 , or @xmath241 and @xmath25 sufficiently small , compressive outflow layers are stable . expansive inflow boundary layers are stable for all parameter values . stability of inflow boundary layers in the characteristic limit @xmath242 is not treated here , but should be treatable analytically by the asymptotic ode methods used in @xcite to study the small - amplitude ( characteristic ) shock limit . this would be an interesting direction for future investigation . the characteristic limit is not accessible numerically , since the exponential decay rate of the background profile decays to zero as @xmath243 , so that the numerical domain of integration needed to resolve the eigenvalue ode becomes infinitely large as @xmath242 . [ weakreg ] stability in the noncharacteristic weak layer limit @xmath244 [ resp . @xmath26 ] in the inflow [ outflow ] case , for @xmath25 bounded away from the strong and characteristic limits @xmath246 and @xmath26 has already been established in @xcite . indeed , it is shown in @xcite that the evans function converges to that for a constant solution , and this is a _ regular _ perturbation . [ extra ] stability of @xmath234 , @xmath218 may also be determined numerically , in particular in the region @xmath247 not covered by proposition [ redenergy ] . the asymptotic results of section [ analytical ] reduce the problem of ( uniformly noncharacteristic , @xmath25 bounded away from @xmath70 ) boundary layer stability to a bounded parameter range on which the evans function may be efficiently computed numerically in a way that is uniformly well - conditioned ; see @xcite . specifically , we may map a semicircle @xmath248 enclosing @xmath249 for @xmath28 $ ] by @xmath234 , @xmath218 , @xmath228 , @xmath178 and compute the winding number of its image about the origin to determine the number of zeroes of the various evans functions within the semicircle , and thus within @xmath249 . for details of the numerical algorithm , see @xcite . in all cases , we obtain results consistent with stability ; that is , a winding number of zero or one , depending on the situation . in the case of a single nonzero root , we know from our limiting analysis that this root may be quite near @xmath32 , making delicate the direct determination of its stability ; however , in this case we do not attempt to determine the stability numerically , but rely on the analytically computed stability index to conclude stability . see section [ computations ] for further details . as in the shock case @xcite , our results indicate _ unconditional stability _ of uniformly noncharacteristic boundary - layers for isentropic navier stokes equations ( and , for outflow layer , in the characteristic limit as well ) , despite the additional complexity of the boundary - layer case . however , two additional comments are in order , perhaps related . first , we point out that the apparent symmetry of theorem [ main3 ] in the @xmath226 outflow and @xmath225 inflow limits is somewhat misleading . for , the limiting , shock evans function possesses a single zero at @xmath32 , indicating that stability of inflow boundary layers is somewhat delicate as @xmath225 : specifically , they have an eigenvalue near zero , which , though stable , is ( since vanishingly small in the shock limit ) not `` very '' stable . likewise , the limiting evans function @xmath234 as @xmath76 possesses a zero at @xmath32 , with the same conclusions . by contrast , the evans functions of outflow boundary layers possess a spurious zero at @xmath32 , so that convergence to the shock or strong - layer limit in this case implies the _ absence _ of any eigenvalues near zero , or `` uniform '' stability as @xmath76 . in this sense , strong outflow boundary layers appear to be more stable than inflow boundary layers . one may make interesting comparisons to physical attempts to stabilize laminar flow along an air- or hydro - foil by suction ( outflow ) along the boundary . see , for example , the interesting treatise @xcite . second , we point out the result of _ instability _ obtained in @xcite for inflow boundary - layers of the full ( nonisentropic ) ideal - gas equations for appropriate ratio of the coefficients of viscosity and heat conduction . this suggests that the small eigenvalues of the strong inflow - layer limit may in some cases perturb to the unstable side . it would be very interesting to make these connections more precise , as we hope to do in future work . since the structure of is essentially the same as that of the shock case , we may follow exactly the treatment in @xcite analyzing the flow of in the singular region @xmath48 . as we shall need the details for further computations ( specifically , the proof of theorem [ main2 ] ) , we repeat the analysis here in full . our starting point is the observation that @xmath250 is approximately block upper - triangular for @xmath107 sufficiently small , with diagonal blocks @xmath251 and @xmath252 that are uniformly spectrally separated on @xmath236 , as follows by @xmath253 we exploit this structure by a judicious coordinate change converting to a system in exact upper triangular form , for which the decoupled `` slow '' upper lefthand @xmath254 block undergoes a _ regular perturbation _ that can be analyzed by standard tools introduced in @xcite . meanwhile , the fast , lower righthand @xmath255 block , since scalar , may be solved exactly . we first block upper - triangularize by a static ( constant ) coordinate transformation the limiting matrix @xmath256 at @xmath144 using special block lower - triangular transformations @xmath257 where @xmath258 denotes the @xmath254 identity matrix and @xmath259 is a @xmath260 row vector . [ pretranslem ] on any compact subset of @xmath236 , for each @xmath261 sufficiently small , there exists a unique @xmath262 such that @xmath263 is upper block - triangular , @xmath264 where @xmath265 and @xmath266 , satisfying a uniform bound @xmath267 block of @xmath268 to zero , we obtain the matrix equation @xmath269 where @xmath270 , or , equivalently , the fixed - point equation @xmath271 by @xmath272 , @xmath273 is uniformly bounded on compact subsets of @xmath236 ( indeed , it is uniformly bounded on all of @xmath236 ) , whence , for @xmath274 bounded and @xmath25 sufficiently small , there exists a unique solution by the contraction mapping theorem , which , moreover , satisfies . defining now @xmath275 and @xmath276 we have converted to an asymptotically block upper - triangular system @xmath277 with @xmath278 as in . our next step is to choose a _ dynamic _ transformation of the same form @xmath279 converting to an exactly block upper - triangular system , with @xmath280 uniformly exponentially decaying at @xmath144 : that is , a _ regular perturbation _ of the identity . [ dyntranslem ] on any compact subset of @xmath236 , for @xmath190 sufficiently large and each @xmath261 sufficiently small , there exists a unique @xmath281 such that @xmath282 is upper block - triangular , @xmath283 and @xmath284 , satisfying a uniform bound @xmath285 independent of the choice of @xmath190 , @xmath25 . setting the @xmath286 block of @xmath287 to zero and computing @xmath288 we obtain the matrix equation @xmath289 where @xmath290 and the forcing term @xmath291 by derivative estimate @xmath292 together with the mean value theorem is uniformly exponentially decaying : @xmath293 initializing @xmath284 , we obtain by duhamel s principle / variation of constants the representation ( supressing the argument @xmath120 ) @xmath294 where @xmath295 is the solution operator for the homogeneous equation @xmath296 or , explicitly , @xmath297 for @xmath274 bounded and @xmath25 sufficiently small , we have by matrix perturbation theory that the eigenvalues of @xmath298 are small and the entries are bounded , hence @xmath299 for @xmath300 . recalling the uniform spectral gap @xmath301 for @xmath236 , we thus have @xmath302 . combining and , we obtain @xmath303 defining @xmath304 and recalling we thus have @xmath305 where @xmath306 is uniformly bounded , @xmath307 , and @xmath308 is contractive with arbitrarily small contraction constant @xmath309 in @xmath310 for @xmath311 for @xmath190 sufficiently large , by the calculation @xmath312 it follows by the contraction mapping principle that there exists a unique solution @xmath313 of fixed point equation with @xmath314 for @xmath315 , or , equivalently ( redefining the unspecified constant @xmath316 ) , . making now the further change of coordinates @xmath317 and computing @xmath318 we find that we have converted to a block - triangular system @xmath319 related to the original eigenvalue system by @xmath320 where @xmath321 since it is triangular , may be solved completely if we can solve the component systems associated with its diagonal blocks . the _ fast system _ @xmath322 associated to the lower righthand block features rapidly - varying coefficients . however , because it is scalar , it can be solved explicitly by exponentiation . the _ slow system _ @xmath323 associated to the upper lefthand block , on the other hand , by , is an exponentially decaying perturbation of a constant - coefficient system @xmath324 that can be explicitly solved by exponentiation , and thus can be well - estimated by comparison with . a rigorous version of this statement is given by the _ conjugation lemma _ of @xcite : [ conjugation ] let @xmath325 , with @xmath326 continuous in @xmath120 and @xmath327 , for @xmath120 in some compact set @xmath249 . then , there exists a globally invertible matrix @xmath328 such that the coordinate change @xmath329 converts the variable - coefficient ode @xmath330 to a constant - coefficient equation @xmath331 satisfying for any @xmath190 , @xmath332 a uniform bound @xmath333 see @xcite , or appendix c , @xcite . by proposition [ conjugation ] , the solution operator for is given by @xmath334 where @xmath335 is a uniformly small perturbation of the identity for @xmath315 and @xmath336 sufficiently large . with these preparations , we turn now to the proofs of the main theorems . we begin by recalling the following estimates established in @xcite on @xmath337 , that is , the value of the dual mode @xmath338 appearing in at the boundary @xmath339 between regular and singular regions . for completeness , and because we shall need the details in further computations , we repeat the proof in full . [ matching ] for @xmath120 on any compact subset of @xmath236 , and @xmath336 sufficiently large , with @xmath338 normalized as in @xcite , latexmath:[\[\label{wcon } @xmath76 , uniformly in @xmath120 , where @xmath66 , @xmath341 are independent of @xmath190 and @xmath342 is the limiting direction vector appearing in the definition of @xmath234 . [ matching2 ] under the hypotheses of lemma [ matching ] , latexmath:[\[\label{limcon } latexmath:[\[\label{wcon2 } as @xmath76 , uniformly in @xmath120 , where @xmath66 , @xmath341 are independent of @xmath190 and @xmath344 is the solution of the limiting adjoint eigenvalue system appearing in definition of @xmath345 . first , make the independent coordinate change @xmath346 normalizing the background wave to match the shock - wave case . making the dependent coordinate - change @xmath347 @xmath348 as in , reduces the adjoint equation @xmath349 to block lower - triangular form , @xmath350 with `` @xmath214 '' denoting complex conjugate . denoting by @xmath351 a suitably normalized element of the one - dimensional ( slow ) stable subspace of @xmath352 , we find readily ( see @xcite for further discussion ) that , without loss of generality , @xmath353 as @xmath76 , while the associated eigenvalue @xmath354 uniformly for @xmath120 on an compact subset of @xmath355 . the dual mode @xmath356 is uniquely determined by the property that it is asymptotic as @xmath48 to the corresponding constant - coefficient solution @xmath357 ( the standard normalization of @xcite ) . by lower block - triangular form , the equations for the slow variable @xmath358 decouples as a slow system @xmath359 dual to , with solution operator @xmath360 dual to , i.e. ( fixing @xmath361 , say ) , solutions of general form @xmath362 @xmath363 arbitrary . denoting by @xmath364 therefore , the unique ( up to constant factor ) decaying solution at @xmath59 , and @xmath365 , we thus have evidently @xmath366 which , as @xmath76 , is uniformly bounded by latexmath:[\[\label{weakexp } @xmath309 and , by , converges for @xmath7 less than or equal to @xmath368 for any fixed @xmath369 simply to @xmath370 defining by @xmath371 the fast coordinate of @xmath372 , we have , by , @xmath373 whence , by duhamel s principle , any decaying solution is given by @xmath374 where @xmath375 recalling , for @xmath236 , that @xmath376 , combining and , and noting that @xmath72 converges uniformly on @xmath377 as @xmath76 for any @xmath378 to @xmath379 we obtain by the lebesgue dominated convergence theorem that @xmath380 recalling , finally , , and the fact that @xmath381 for @xmath25 sufficiently small , we obtain as claimed . again , make the coordinate change @xmath346 normalizing the background wave to match the shock - wave case . applying proposition [ conjugation ] to the limiting adjoint system @xmath382 we find that , up to an @xmath383 coordinate change , @xmath384 is given by the exact solution @xmath385 of the limiting , constant - coefficient system @xmath386 this yields immediately , which , together with , yields . the rest of our analysis is standard . [ regconj ] on @xmath387 for any fixed @xmath336 , there exists a coordinate - change @xmath388 conjugating to the limiting equations , @xmath389 , satisfying a uniform bound latexmath:[\[\label{tbd } make the coordinate change @xmath346 normalizing the background profile . for @xmath391 $ ] , this is a consequence of the _ convergence lemma _ of @xcite , a variation on proposition [ conjugation ] , together with uniform convergence of the profile and eigenvalue equations . for @xmath392 $ ] , it is essentially continuous dependence ; more precisely , observing that @xmath393 for @xmath392 $ ] , setting @xmath394 , and writing the homological equation expressing conjugacy of and , we obtain @xmath395 which , considered as an inhomogeneous linear matrix - valued equation , yields an exponential growth bound @xmath396 for some @xmath118 , giving the result . make the coordinate change @xmath346 normalizing the background profile . lemma [ regconj ] , together with convergence as @xmath76 of the unstable subspace of @xmath397 to the unstable subspace of @xmath398 at the same rate @xmath399 ( as follows by spectral separation of the unstable eigenvalue of @xmath400 and standard matrix perturbation theory ) yields @xmath401 gives @xmath402 where @xmath403 denotes the solution operator of the limiting adjoint eigenvalue equation @xmath404 . applying proposition [ conjugation ] to the limiting system , we obtain @xmath405 by direct computation of @xmath406 , where @xmath407 is independent of @xmath336 . together with and , this gives latexmath:[\[|\tilde w_1^+(0,\lambda)- \tilde w_1^{0+}(0,\lambda)|\le c(l)v_+ @xmath274 bounded , @xmath409 taking first @xmath410 and then @xmath76 , we obtain therefore convergence of @xmath411 and @xmath412 to @xmath413 and @xmath414 , yielding the result by definitions and . straightforward , following the previous argument in the regular region only . first make the coordinate change @xmath346 normalizing the background profile location to that of the shock wave case , where @xmath415 as @xmath225 . by standard duality properties , @xmath416 is independent of @xmath417 , so we may evaluate at @xmath12 as in the shock case . denote by @xmath418 , @xmath419 the corresponding modes in the shock case , and @xmath420 the resulting evans function . noting that @xmath421 and @xmath422 are asymptotic to the unique stable mode at @xmath59 of the ( same ) adjoint eigenvalue equation , but with translated decay rates , we see immediately that @xmath423 @xmath424 on the other hand is initialized at @xmath425 ( in the new coordinates @xmath426 ) as @xmath427 whereas @xmath418 is the unique unstable mode at @xmath60 decaying as @xmath428 , where @xmath429 is the unstable right eigenvector of @xmath430 denote by @xmath431 the associated dual unstable left eigenvector and @xmath432 the eigenprojection onto the stable vector @xmath433 . by direct computation , @xmath434 yielding @xmath435 for @xmath436 , on which @xmath437 . once we know , we may finish by a standard argument , concluding by exponential attraction in the positive @xmath7-direction of the unstable mode that other modes decay exponentially as @xmath438 , leaving the contribution from @xmath439 plus a negligible @xmath440 error , @xmath341 , from which we may conclude that @xmath441 collecting information , we find that @xmath442 @xmath341 , yielding the claimed convergence as @xmath225 , @xmath443 . for @xmath120 uniformly bounded from zero , @xmath444 converges uniformly as @xmath226 to @xmath445 whereas the shock evans function @xmath446 is initiated by @xmath447 proportional to @xmath448 agreeing in the first two coordinates with @xmath449 . by the boundary - layer analysis of section [ estw1 ] , the backward ( i.e. , decreasing @xmath7 ) evolution of the adjoint eigenvalue ode reduces in the asymptotic limit @xmath76 ( forced by @xmath226 ) to a decoupled slow flow @xmath450 in the first two coordinates , driving an exponentially slaved fast flow in the third coordinate . from this , we may conclude that solutions agreeing in the first two coordinates converge exponentially as @xmath7 decreases . performing an appropriate normalization , as in the inflow case just treated , we thus obtain the result . we omit the details , which follow what has already been done in previous cases . following @xcite , we note that @xmath222 is real for real @xmath120 , and nonvanishing for real @xmath120 sufficiently large , hence @xmath451 is well - defined and constant on the entire ( connected ) parameter range . the number of roots of @xmath228 on @xmath140 is therefore even or odd depending on the _ stability index _ @xmath452.\ ] ] similarly , recalling that @xmath453 , we find that the number of roots of @xmath178 on @xmath140 is even or odd depending on @xmath454.\ ] ] examining the adjoint equation at @xmath32 , @xmath455 @xmath456 , we find by explicit computation that the only solutions that are bounded as @xmath48 are the _ constant solutions _ @xmath457 . taking the limit @xmath458 as @xmath459 along the real axis of the unique stable eigenvector of @xmath460 , we find ( see , e.g. , @xcite ) that it lies in the direction @xmath461 , where @xmath462 is the positive characteristic speed of the hyperbolic convection matrix @xmath463 , i.e. , @xmath464 , @xmath465 . thus , @xmath466 as claimed . on the other hand , the same computation carried out for @xmath467 yields @xmath468 . ( note : @xmath469 as @xmath76 . ) similarly , as @xmath226 , @xmath470 finally , note @xmath227 implies that the stability index , since continuously varying so long as it does nt vanish and taking discrete values @xmath471 , must be constant on the connected set of parameter values . since inflow boundary layers are known to be stable on some part of the parameter regime by energy estimates ( theorem [ main2 ] ) , we may conclude that the stability index is identically one and therefore there are an even number of unstable roots for all @xmath472 . to establish that @xmath473 , we compute @xmath474 since @xmath475 is independent of @xmath120 , we need only show that the first component of @xmath476 is nonzero . note that @xmath477 solves the limiting adjoint variational equations @xmath478 with @xmath479 by , and the fact that @xmath480 , @xmath481 is chosen so that asymptotically at @xmath144 it lies in the direction of @xmath482 . . then the third component solves @xmath484 which has solution @xmath485 where @xmath486 integrating the equation for the first component of @xmath487 yields @xmath488 using the condition @xmath489 we have @xmath490 so that @xmath491 finally , note that by using we have @xmath492 so that for all @xmath493 , @xmath494 which implies @xmath495 [ apparent ] the result @xmath227 at first sight appears to contradict that of theorem [ main3 ] , since @xmath496 for the shock wave case . this apparent contradiction is explained by the fact that the normalizing factor @xmath497 is exponentially decaying in @xmath111 for @xmath32 , since @xmath498 , while @xmath499 . recalling that @xmath443 as @xmath225 , we recover the result of theorem [ main3 ] . similarly , we compute @xmath500 where @xmath501 satisfies the variational equation @xmath502 , or , written as a first - order system , @xmath503 which may be solved exactly for the unique solution decaying at @xmath60 of @xmath504 recalling from that @xmath505 , hence @xmath506 we thus find that @xmath507 as claimed . the proof that @xmath233 goes similarly . finally , as in the proof of the inflow case , we note that nonvanishing implies that the stability index is constant across the entire ( connected ) parameter range , hence we may conclude that it is identically one by existence of a stable case ( corollary [ v01 ] ) , and therefore that the number of nonzero unstable roots is even , as claimed . by proposition [ redenergy ] we find that @xmath228 has at most a single zero in @xmath140 . however , by our stability index results , theorem [ index ] , the number of eigenvalues in @xmath140 is even . thus , it must be zero , giving the result . by theorem [ main3 ] , @xmath178 , suitably renormalized , converges as @xmath226 to the evans function for the ( unintegrated ) shock wave case . but , the shock evans function by the results of @xcite has just a single zero at @xmath32 on @xmath436 , already accounted for in @xmath178 by the spurious root at @xmath32 introduced by recoordinatization to `` good unknown '' . finally , we treat the remaining , `` corner case '' as @xmath25 , @xmath201 simultaneously approach zero . the fact ( lemma [ index ] ) that @xmath508 shows that this limit is quite delicate ; indeed , this is the most delicate part of our analysis . consider again the adjoint system @xmath509 by the boundary analysis of section [ estw1 ] , @xmath510 where @xmath511 , and @xmath512 is the unique stable eigenvalue of @xmath513 , satisfying ( by matrix perturbation calculation ) @xmath514 and thus @xmath515 as @xmath226 ( hence @xmath76 ) on bounded subsets of @xmath140 . combining these expansions , we have @xmath516 for @xmath201 sufficiently small . from the @xmath517 equation @xmath518 , we thus obtain @xmath519 observing , finally , that , for @xmath140 , the ratio of real to imaginary parts of @xmath520 is uniformly positive , we find that @xmath521 for @xmath201 sufficiently small , which yields nonvanishing of @xmath222 on @xmath140 as claimed . in this section , we show , through a systematic numerical evans function study , that there are no unstable eigenvalues for @xmath522\times(0,1],\ ] ] in either inflow or outflow cases . as defined in section [ evanssec ] , the evans function is analytic in the right - half plane and reports a value of zero precisely at the eigenvalues of the linearized operator . hence we can use the argument principle to determine if there are any unstable eigenvalues for this system . our approach closely follows that of @xcite for the shock case with only two major differences . first , our shooting algorithm is only one sided as we have the boundary conditions and for the inflow and outflow cases , respectfully . second , we `` correct '' for the displacement in the boundary layer when @xmath523 in the inflow case and @xmath524 in the outflow case so that the evans function converges to the shock case as studied in @xcite ( see discussion in section [ sectionshocklimit ] ) . @xmath525{inflow1 } & \includegraphics[width=6.4cm , height=4.5cm]{inflow2 } \\ \mbox{\bf ( a ) } & \mbox{\bf ( b)}\\ \includegraphics[width=5.75cm]{inflow3 } & \includegraphics[width=5.6cm]{inflow4 } \\ \mbox{\bf ( c ) } & \mbox{\bf ( d ) } \end{array}$ ] the profiles were generated using matlab s bvp4c routine , which is an adaptive lobatto quadrature scheme . the shooting portion of the evans function computation was performed using matlab s ode45 package , which is the standard 4th order adaptive runge - kutta - fehlberg method ( rkf45 ) . the error tolerances for both the profiles and the shooting were set to abstol=1e-6 and reltol=1e-8 . we remark that kato s ode ( see section [ evanssec ] and @xcite for details ) is used to analytically choose the initial eigenbasis for the stable / unstable manifolds at the numerical values of infinity at @xmath526 . finally in section [ errorstudy ] , we carry out a numerical convergence study similar to that in @xcite . @xmath527{outflow1 } & \includegraphics[width=5.75cm]{outflow2 } \\ \mbox{\bf ( a ) } & \mbox{\bf ( b)}\\ \includegraphics[width=5.75cm]{outflow3 } & \includegraphics[width=5.3cm]{outflow4 } \\ \mbox{\bf ( c ) } & \mbox{\bf ( d ) } \end{array}$ ] the high - frequency estimates in proposition [ hf ] restrict the set of admissible unstable eigenvalues to a fixed compact triangle @xmath249 in the right - half plane ( see and for the inflow and outflow cases , respectively ) . we reiterate the remarkable property that @xmath249 does not depend on the choice of @xmath25 or @xmath201 . hence , to demonstrate stability for a given @xmath0 , @xmath25 and @xmath201 , it suffices to show that the winding number of the evans function along a contour containing @xmath249 is zero . note that in our region of interest , @xmath528 $ ] , the semi - circular contour given by @xmath529 contains @xmath249 in both the inflow and outflow cases . hence , for consistency we use this same semicircle for all of our winding number computations . a remarkable feature of the evans function for this system , and one that is shared with the shock case in @xcite , is that the evans function has limiting behavior as the amplitude increases , section [ analytical ] . for the inflow case , we see in figure [ first ] , the mapping of the contour @xmath530 for the monatomic case ( @xmath531 ) , for several different choices of @xmath201 , as @xmath532 . we remark that the winding numbers for @xmath533 are all zero , and the limiting contour touches zero due to the emergence of a zero root in the limit . note that the limiting case contains the contours of all other amplitudes . hence , we have spectral stability for all amplitudes . the outflow case likewise has a limiting behavior , however , all contours cross through zero due to the eigenvalue at the origin . nonetheless , since the contours only wind around once , we can likewise conclude that these profiles are spectrally stable . we remark that the outflow case converges to the limiting case faster than the inflow case as is clear from figure [ second1 ] . indeed , @xmath534 and the limiting case @xmath535 , as well as all of the values of @xmath25 in between , are virtually indistinguishable . @xmath527{index1 } & \includegraphics[width=5.75cm]{index2 } \\ \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] in our study , we systematically varied @xmath201 in the interval @xmath536 $ ] and took the @xmath532 limit at each step , starting from a @xmath537 ( or some other appropriate value , for example when @xmath538 ) on the small - amplitude end and decreased @xmath25 steadily to @xmath539 for @xmath540 , followed by evaluation at @xmath535 . for both inflow and outflow cases , over @xmath541 contours were computed . we remark that in the @xmath532 limit , the system becomes pressureless , and thus all of the contours in the large - amplitude limit look the same regardless of the value of @xmath0 chosen . as an additional verification of stability , we computed the evans function along the unstable real axis on the interval @xmath542 $ ] for varying parameters to show that there are no real unstable eigenvalues . since the evans function has a root at the origin in the limiting system for the inflow case , and for all values of @xmath25 in the outflow case , we can perform in these cases a sort of _ numerical stability index analysis _ to verify that the evans function cuts transversely through the origin and is otherwise nonzero , indicating that there are no unstable real eigenvalues as expected . in figure [ second2 ] , we see a typical example of @xmath543 the inflow and @xmath544 outflow cases . note that in both images , the evans function cuts transversally through the origin and is otherwise nonzero as @xmath120 increases . @xmath545{limit1 } & \includegraphics[width=5.9cm]{limit2 } \\ \mbox{\bf ( a ) } & \mbox{\bf ( b ) } \end{array}$ ] when @xmath201 is far from the midpoint @xmath546 of the end states , the the evans function of the boundary layer is similar to the evans function of the shock case evaluated at the displacement point @xmath417 . hence , when we compute the boundary layer evans function near the shock limits , @xmath523 for the inflow case and @xmath524 for the outflow case , we multiply for the correction factor @xmath547 so that our output looks close to that of the shock case studied in @xcite . the correction factors are @xmath548 for the inflow case and @xmath549 for the outflow case , where @xmath550 is the growth mode of @xmath551 and @xmath552 is the decay mode of @xmath198 . in figure [ second3 ] , we see that these highly displaced profiles appear to be very similar to the shock cases with one notable difference . these images have a small dimple near @xmath32 to account for the eigenvalue there , whereas those in the shock case @xcite were computed in integrated coordinates and thus have no root at the origin . ccccccc + @xmath190 & @xmath553 & @xmath554 & @xmath555 & @xmath556 & @xmath557 & @xmath558 + 8 & 7.8(-1 ) & 8.4(-1 ) & 9.2(-1 ) & 1.0(0 ) & 1.2(0 ) & 1.3(0 ) + 10 & 1.4(-1 ) & 1.2(-1 ) & 9.2(-2 ) & 6.8(-2 ) & 4.4(-2 ) & 2.8(-2 ) + 12 & 1.4(-2 ) & 7.9(-3 ) & 3.6(-3 ) & 1.3(-3 ) & 3.1(-4 ) & 7.3(-5 ) + 14 & 1.3(-3 ) & 4.9(-4 ) & 1.3(-4 ) & 2.4(-5 ) & 8.7(-6 ) & 8.2(-6 ) + 16 & 1.2(-4 ) & 3.0(-5 ) & 4.7(-6 ) & 2.8(-6 ) & 2.7(-6 ) & 2.6(-6 ) + 18 & 1.1(-5 ) & 5.8(-6 ) & 8.0(-6 ) & 8.1(-6 ) & 8.0(-6 ) & 8.0(-6 ) + + @xmath190 & @xmath553 & @xmath554 & @xmath555 & @xmath556 & @xmath557 & @xmath558 + 8 & 5.4(-3 ) & 5.4(-3 ) & 5.4(-3 ) & 5.4(-3 ) & 5.4(-3 ) & 5.4(-3 ) + 10 & 9.2(-4 ) & 9.1(-4 ) & 9.1(-4 ) & 9.1(-4 ) & 9.1(-4 ) & 9.1(-4 ) + 12 & 1.5(-4 ) & 1.5(-4 ) & 1.5(-4 ) & 1.5(-4 ) & 1.5(-4 ) & 1.5(-4 ) + 14 & 2.5(-5 ) & 2.7(-5 ) & 2.0(-5 ) & 2.0(-5 ) & 2.0(-5 ) & 2.0(-5 ) + 16 & 2.3(-6 ) & 2.6(-6 ) & 2.6(-6 ) & 2.5(-6 ) & 2.5(-6 ) & 2.5(-6 ) + 18 & 6.6(-6 ) & 3.6(-6 ) & 8.7(-6 ) & 8.7(-6 ) & 8.7(-6 ) & 8.7(-6 ) + ccccccc + abs / rel & @xmath553 & @xmath554 & @xmath555 & @xmath556 & @xmath557 & @xmath558 + @xmath559 & 5.4(-4 ) & 4.1(-4 ) & 4.0(-4 ) & 5.0(-4 ) & 3.4(-4 ) & 8.6(-4 ) + @xmath560 & 3.1(-5 ) & 4.6(-5 ) & 3.4(-5 ) & 3.3(-5 ) & 3.3(-5 ) & 3.2(-5 ) + @xmath561 & 2.9(-6 ) & 3.6(-6 ) & 3.9(-6 ) & 6.8(-6 ) & 2.7(-6 ) & 2.5(-6 ) + @xmath562 & 4.6(-7 ) & 9.9(-7 ) & 1.1(-6 ) & 6.0(-7 ) & 2.9(-7 ) & 3.2(-7 ) + + abs / rel & @xmath553 & @xmath554 & @xmath555 & @xmath556 & @xmath557 & @xmath558 + @xmath559 & 9.2(-4 ) & 9.2(-4 ) & 9.1(-4 ) & 9.1(-4 ) & 9.1(-4 ) & 9.2(-4 ) + @xmath560 & 5.3(-5 ) & 4.9(-5 ) & 5.3(-5 ) & 5.3(-5 ) & 5.3(-5 ) & 5.3(-5 ) + @xmath561 & 6.7(-5 ) & 6.7(-5 ) & 6.7(-5 ) & 6.7(-5 ) & 6.7(-5 ) & 6.7(-5 ) + @xmath562 & 2.9(-6 ) & 2.9(-6 ) & 2.9(-6 ) & 2.9(-6 ) & 2.9(-6 ) & 2.9(-6 ) + as in @xcite , we carry out a numerical convergence study to show that our results are accurate . we varied the absolute and relative error tolerances , as well as the length of the numerical domain @xmath563 $ ] . in tables 12 , we demonstrate that our choices of @xmath564 , abstol=1e-6 and reltol=1e-8 provide accurate results . our starting point is remark [ shockrel ] , in which we observed that the first - order eigensystem in variable @xmath565 may be converted by the rescaling @xmath566 to a system identical to that of the integrated equations in the shock case ; see @xcite . artificially defining @xmath567 , we obtain a system [ ep ] @xmath568 identical to that in the integrated shock case @xcite , but with boundary conditions @xmath569 imposed at @xmath12 . this new eigenvalue problem differs spectrally from only at @xmath32 , hence spectral stability of is implied by spectral stability of . hereafter , we drop the tildes , and refer simply to @xmath5 , @xmath570 . with these coordinates , we may establish by exactly the same argument used in the shock case in @xcite , for completeness reproduced here . the following identity holds for @xmath571 : @xmath572 we multiply by @xmath573 and integrate along @xmath7 . this yields @xmath574 we get by taking the real and imaginary parts and adding them together , and noting that @xmath575 . [ kawashima ] the following identity holds for @xmath571 : @xmath576 |v'|^2\ ] ] we multiply by @xmath577 and integrate along @xmath7 . this yields @xmath578 using on the right - hand side , integrating by parts , and taking the real part gives @xmath579 = { \int_{\mathbb{r}^+}}\left[\frac{h(\bv)}{\bv^{\gamma+1 } } + \frac{\bv_x}{2 \bv^2 } \right ] |v'|^2 + \r(\lambda){\int_{\mathbb{r}^+}}\frac{|v'|^2}{\bv}.\ ] ] the right hand side can be rewritten as @xmath580 = \frac{1}{2 } { \int_{\mathbb{r}^+}}\left[\frac{h(\bv)}{\bv^{\gamma+1 } } + \frac{a\gamma}{\bv^{\gamma+1 } } \right ] |v'|^2 + \r(\lambda){\int_{\mathbb{r}^+}}\frac{|v'|^2}{\bv}.\ ] ] now we manipulate the left - hand side . note that @xmath581 hence , by taking the real part we get @xmath579 = { \int_{\mathbb{r}^+}}|u'|^2 - 2\r(\lambda)^2 { \int_{\mathbb{r}^+}}|v|^2.\ ] ] this combines with to give . [ hlem ] for @xmath582 as in , we have @xmath583 where @xmath584 is the profile solution to . defining @xmath585 we have @xmath586 for @xmath587 , hence the maximum of @xmath588 on @xmath589 $ ] is achieved at @xmath590 . substituting into and simplifying yields . using young s inequality twice on right - hand side of together with , we get @xmath591 { \int_{\mathbb{r}^+}}\bv |u|^2.\end{aligned}\ ] ] assuming that @xmath592 and @xmath593 , this simplifies to @xmath594 { \int_{\mathbb{r}^+}}\bv @xmath595 { \int_{\mathbb{r}^+}}\bv |u|^2,\ ] ] or equivalently , @xmath596 setting @xmath597 gives . similarly as in the inflow case , we can convert the eigenvalue equations into the integrated equations as in the shock case ; see @xcite . artificially defining @xmath598 , we obtain a system [ outep ] @xmath599 identical to that in the integrated shock case @xcite , but with boundary conditions @xmath600 imposed at @xmath12 . we shall write @xmath601 for @xmath602 , for any function @xmath603 . this new eigenvalue problem differs spectrally from only at @xmath32 , hence spectral stability of is implied by spectral stability of . hereafter , we drop the tildes , and refer simply to @xmath5 , @xmath570 . the following identity holds for @xmath571 : @xmath604 we multiply by @xmath573 and integrate along @xmath7 . this yields @xmath605 we get by taking the real and imaginary parts and adding them together , and noting that @xmath606 . [ outkawashima ] the following inequality holds for @xmath607 : @xmath608 |v'|^2 & + \r(\lambda){\int_{\mathbb{r}^-}}\frac{|v'|^2}{\bv}+ \frac { |v'_0|^2}{4\bv_0 } + 2\r\lambda ^2{\int_{\mathbb{r}^-}}|v|^2\notag \\&\le { \int_{\mathbb{r}^-}}|u'|^2 + \bv_0|u_0|^2.\end{aligned}\ ] ] we multiply by @xmath577 and integrate along @xmath7 . this yields @xmath609 using on the right - hand side , integrating by parts , and taking the real part gives @xmath610 = { \int_{\mathbb{r}^-}}\left[\frac{h(\bv)}{\bv^{\gamma+1 } } + \frac{\bv_x}{2 \bv^2 } \right ] { |v'_0|^2}{2\bv_0}.\ ] ] the right hand side can be rewritten as @xmath611 \notag\\&= \frac{1}{2 } { \int_{\mathbb{r}^-}}\left[\frac{h(\bv)}{\bv^{\gamma+1 } } + \frac{a\gamma}{\bv^{\gamma+1 } } \right ] |v'|^2 + \r(\lambda){\int_{\mathbb{r}^-}}\frac{|v'|^2}{\bv}+ \frac { |v'_0|^2}{2\bv_0}.\end{aligned}\ ] ] now we manipulate the left - hand side . note that @xmath612 hence , by taking the real part and noting that @xmath613 we get @xmath610 = { \int_{\mathbb{r}^-}}|u'|^2 - 2\r\lambda ^2{\int_{\mathbb{r}^-}}|v|^2 - \r\lambda |v_0|^2 + \r(\lambda u_0\bar v_0).\ ] ] this combines with to give @xmath614 |v'|^2 & + \r(\lambda){\int_{\mathbb{r}^-}}\frac{|v'|^2}{\bv}+ \frac { |v'_0|^2}{2\bv_0 } + 2\r\lambda ^2{\int_{\mathbb{r}^-}}|v|^2\\ & + \r\lambda we get by observing that and young s inequality yield @xmath615 here we used @xmath616 . note that @xmath617 and @xmath618 . using young s inequality twice on right - hand side of together with , and denoting the boundary term on the right by @xmath619 , we get @xmath620 { \int_{\mathbb{r}^-}}\bv |u|^2+i_b.\end{aligned}\ ] ] here we treat the boundary term by @xmath621 therefore using , we simply obtain from the above estimates @xmath622 { \int_{\mathbb{r}^-}}\bv |u|^2+\frac 1\theta |\alpha|^2 \bv_0|u_0|^2\\ & < 2\theta { \int_{\mathbb{r}^-}}|u'|^2 + \left[\frac{\gamma}{2\theta } + \frac{1}{4 \epsilon}\right ] { \int_{\mathbb{r}^-}}\bv |u|^2+j_b\end{aligned}\ ] ] where @xmath623 assuming that @xmath624 , this simplifies to @xmath625 { \int_{\mathbb{r}^-}}\bv |u|^2+j_b.\end{aligned}\ ] ] note that @xmath626 . therefore for @xmath627 , we get @xmath628 and @xmath629 for sake of simplicity , choose @xmath630 and @xmath631 . this shows that @xmath632 can be absorbed into the left by the term @xmath633 and thus we get @xmath634 { \int_{\mathbb{r}^-}}\bv |u|^2 = \left [ 3\gamma+ \frac{3}{8}\right ] { \int_{\mathbb{r}^-}}\bv |u|^2,\end{aligned}\ ] ] provided that @xmath635 . this shows @xmath636 working in @xmath637 variables as in , the limiting eigenvalue system and boundary conditions take the form [ limitep ] @xmath638 corresponding to a pressureless gas , @xmath181 , with @xmath639 hereafter , we drop the tildes . multiplying by @xmath640 and integrating on @xmath641\subset\mathbb{r}^+$ ] , we obtain @xmath642 integrating the third and fourth terms by parts yields @xmath643 u'\bar{u } dx \\ & \quad+ \int^b_0 \frac{|u'|^2}{1-\bv } dx + \int^b_0 v ( \overline{\lambda v + v ' } ) dx \\ & = \left [ v \bar{u } + \frac{u'\bar{u}}{1-\bv } \right ] \big|^b_0.\\\end{aligned}\ ] ] taking the real part , we have @xmath644 |u|^2 - \frac{|v|^2 } { 2 } \right ] \big|^b_0,\label{mnen}\end{aligned}\ ] ] where @xmath645.\ ] ] note that @xmath646 thus , @xmath647 and the third term on the right - hand side vanishes , leaving @xmath648 \big|^b_0\\ & \quad = \left [ \r(v \bar{u } ) + \frac{\r(u'\bar{u})}{1-\bv } - \frac{|v|^2 } { 2 } \right](b).\\\end{aligned}\ ] ] we show finally that the right - hand side goes to zero in the limit as @xmath649 . by proposition [ conjugation ] , the behavior of @xmath5 , @xmath570 near @xmath650 is governed by the limiting constant coefficient systems @xmath651 , where @xmath652 and @xmath653 . in particular , solutions @xmath134 asymptotic to @xmath654 at @xmath144 decay exponentially in @xmath655 and are bounded in coordinate @xmath5 as @xmath48 . observing that @xmath656 as @xmath48 , we thus see immediately that the boundary contribution at @xmath657 vanishes as @xmath658 . thus , in the limit as @xmath658 , @xmath659 but , for @xmath355 , this implies @xmath660 , or @xmath661 , which , by @xmath662 , implies @xmath663 . this reduces to @xmath664 , yielding the explicit solution @xmath665 . by @xmath666 , therefore , @xmath667 for @xmath355 . substituting into , we obtain @xmath668 . it follows that there are no nontrivial solutions of , for @xmath355 except at @xmath32 . [ symmetrizers ] the above energy estimate is essentially identical to that used in @xcite to treat the limiting shock case . working in @xmath637 variables as in , the limiting eigenvalue system and boundary conditions take the form [ outlimitep ] @xmath669 corresponding to a pressureless gas , @xmath181 , with @xmath670 @xmath671 in particular , @xmath672 hereafter , we drop the tildes . multiplying by @xmath640 and integrating on @xmath673\subset\mathbb{r}^-$ ] , we obtain @xmath674 integrating the third and fourth terms by parts yields @xmath675 u'\bar{u } dx \\ & \quad+ \int^0_a \frac{|u'|^2}{1-\bv } dx + \int^0_a v ( \overline{\lambda v + v ' } ) dx \\ & = \left [ v \bar{u } + \frac{u'\bar{u}}{1-\bv } \right ] \big|^0_a.\\\end{aligned}\ ] ] taking the real part , we have @xmath676 |u|^2 - \frac{|v|^2 } { 2 } \right ] \big|^0_a,\label{mnen2}\end{aligned}\ ] ] where @xmath645\equiv 0\ ] ] and the third term on the right - hand side vanishes , as shown in section [ stronglimit ] , leaving @xmath677 \big|^0_a.\\\end{aligned}\ ] ] a boundary analysis similar to that of section [ stronglimit ] shows that the contribution at @xmath72 on the righthand side vanishes as @xmath678 ; see @xcite for details . thus , in the limit as @xmath678 we obtain @xmath679 ( 0)\\ & \quad = \left[(1-v_0)\r(v \bar{u } ) - \frac{|v|^2 } { 2 } \right ] ( 0),\\ & \quad \le \left[(1- v_0 ) |v||u| - \frac{|v|^2 } { 2}\right](0)\\ & \quad \le ( 1- v_0 ) ^2\frac{|u(0)|^2}{2},\end{aligned}\ ] ] where the second equality follows by and the final line by young s inequality . next , observe the sobolev - type bound @xmath680 together with @xmath681 hence @xmath682 for @xmath683 , where @xmath684 is the unique solution of @xmath685 thus , for @xmath683 , @xmath686 for @xmath687 . for @xmath355 , this implies @xmath660 , or @xmath661 , which , by @xmath688 , implies @xmath689 . this reduces to @xmath664 , yielding the explicit solution @xmath665 . by @xmath666 , therefore , @xmath667 for @xmath355 . it follows that there are no nontrivial solutions of , for @xmath355 except at @xmath32 . by iteration , starting with @xmath690 , we obtain first @xmath691 then @xmath692 , then @xmath693 , then @xmath694 , then @xmath695 and @xmath696 , terminating with @xmath697 . [ smallcase ] our evans function results show that the case @xmath201 small not treated corresponds to the shock limit for which stability is already known by @xcite . this suggests that a more sophisticated energy estimate combining the above with a boundary - layer analysis from @xmath12 back to @xmath339 might yield nonvanishing for all @xmath237 . we now show stability of compressive outflow boundary layers in the characteristic limit @xmath242 , by essentially the same energy estimate used in @xcite to show stability of small - amplitude shock waves . as in the above section on the outflow case , we obtain a system [ charep ] @xmath698 identical to that in the integrated shock case @xcite , but with boundary conditions @xmath671 in particular , @xmath699 this new eigenvalue problem differs spectrally from only at @xmath32 , hence spectral stability of is implied by spectral stability of . hereafter , we drop the tildes , and refer simply to @xmath5 , @xmath570 . [ proof of proposition [ charsmallamp ] ] we note that @xmath700 . by multiplying by both the conjugate @xmath701 and @xmath702 and integrating along @xmath7 from @xmath60 to @xmath246 , we have @xmath703 integrating the last three terms by parts and appropriately using to substitute for @xmath704 in the third term gives us @xmath705\big|_{x=0}.\end{aligned}\ ] ] we take the real part and appropriately integrate by parts to get @xmath706 } dx } + { \int_{-\infty}^0 { g(\bv ) |u|^2 } dx } + { \int_{-\infty}^0 { \frac{\bv^\gamma}{h(\bv)}|u'|^2 } dx}= g(0),\end{aligned}\ ] ] where @xmath707\ ] ] and @xmath708|u|^2 + \r\left[v\bar u+\frac{v^\gamma u'\bar u}{h(\bv)}\right ] - \frac{|v|^2}{2}\ ] ] evaluated at @xmath12 . here , the boundary term appearing on the righthand side is the only difference from the corresponding estimate appearing in the treatment of the shock case in @xcite . we shall show that as @xmath709 , the boundary term @xmath710 is nonpositive . observe that boundary conditions yield @xmath711\big|_{x=0 } = \r(v(0)\bar u(0))\left[1+\frac{\bv^\gamma \bv'}{h(\bv)}\right]\big|_{x=0}.\ ] ] we first note , as established in @xcite , that @xmath712 on @xmath713 $ ] , under certain conditions including the case @xmath709 . straightforward computation gives identities : @xmath714 using and , we abbreviate a few intermediate steps below : @xmath715\right]\notag\\ & = -\frac{\bv_x}{2}\left [ \frac{\bv^\gamma\left((\gamma+1)h(\bv ) - \bv h'(\bv)\right)}{h(\bv)^2 } + \frac{d}{d\bv}\left [ \frac{\gamma h(\bv)-\bv h'(\bv)}{h(\bv)^2 } ( a\gamma - h(\bv ) ) \right]\right]\notag\\ & = -\frac{a\bv_x\bv^{\gamma-1}}{2 h(\bv)^3 } \times\notag\\ & \qquad\left [ \gamma^2(\gamma+1)\bv^{\gamma+2 } - 2 ( a+1)\gamma(\gamma^2 - 1)\bv^{\gamma+1}+(a+1)^2\gamma^2(\gamma-1)\bv^\gamma\right.\notag\\ & \qquad\qquad + \left . a\gamma(\gamma+2)(\gamma^2 - 1)\bv - a ( a+1 ) \gamma^2 ( \gamma^2 - 1 ) \right]\notag\\ & = -\frac{a\bv_x\bv^{\gamma-1}}{2 h(\bv)^3}[(\gamma+1)\bv^{\gamma+2}+\bv^\gamma(\gamma-1)\left((\gamma+1)\bv-(a+1)\gamma\right)^2 \label{charpremn}\\ & \qquad + a\gamma(\gamma^2 - 1)(\gamma+2)\bv - a ( a+1)\gamma^2(\gamma^2 - 1)]\notag\\ & \geq -\frac{a\bv_x\bv^{\gamma-1}}{2 h(\bv)^3}[(\gamma+1)\bv^{\gamma+2}+ a\gamma(\gamma^2 - 1)(\gamma+2)\bv - a ( a+1)\gamma^2(\gamma^2 - 1)]\notag\\ & \geq-\frac{\gamma^2 a^3 \bv_x ( \gamma+1)}{2 h(\bv)^3 v_+}\left[\left(\frac{v_+^{\gamma+1}}{a\gamma}\right)^2 + 2(\gamma-1)\left(\frac{v_+^{\gamma+1}}{a\gamma}\right)-(\gamma-1)\right].\label{charmn}\end{aligned}\ ] ] this verifies @xmath716 as @xmath709 . second , examine @xmath717|u(0)|^2+\left[1+\frac{\bv^\gamma \bv'}{h(\bv)}\right]\r(v(0)\bar u(0))- \frac{|v(0)|^2}{2}.\ ] ] applying young s inequality to the middle term , we easily get @xmath718|u(0)|^2 = : -\frac 12 i|u(0)|^2.\ ] ] now observe that @xmath258 can be written as @xmath719\bv ' - \frac{\bv^\gamma h'(\bv)}{h^2(\bv)}.\end{aligned}\ ] ] using and , we get @xmath720 and thus @xmath721\bv ' - \frac{\bv^\gamma h'(\bv)}{h^2(\bv)}.\end{aligned}\ ] ] now since @xmath722 , as @xmath723 , @xmath724 . therefore , as @xmath725 is close to @xmath26 , @xmath726 . this , @xmath727 , and give , as @xmath725 is close enough to @xmath26 , @xmath728 } dx } & + { \int_{-\infty}^0 { \frac{\bv^\gamma}{h(\bv)}|u'|^2 } dx}\le 0,\end{aligned}\ ] ] which evidently gives stability as claimed . for completeness , we recall the argument of @xcite in the expansive inflow case . * profile equation . * note that , in the expansive inflow case , we assume @xmath729 . therefore we can still follow the scaling to get @xmath730 l. q. brin and k. zumbrun . analytically varying eigenvectors and the stability of viscous shock waves . , 22:1932 , 2002 . seventh workshop on partial differential equations , part i ( rio de janeiro , 2001 ) . k. zumbrun . stability of large - amplitude shock waves of compressible navier - stokes equations . in _ handbook of mathematical fluid dynamics . iii _ , pages 311533 . north - holland , amsterdam , 2004 . with an appendix by helge kristian jenssen and gregory lyng .

last years showed much progress in the experimental determination of the @xmath0-factors of hydrogen - like ions @xcite . measurements , accurate up to a few parts in @xmath1 @xcite , were performed by studying a single ion confined in a penning trap . these experiments provided the stringent tests of bound - state quantum electrodynamics ( qed ) theory and yielded the best determination of the electron mass @xcite . in order to match the experimental accuracy , many sophisticated calculations were performed during the past years , in particular those of the one - loop self - energy @xcite , the one - loop vacuum - polarization @xcite , the nuclear recoil @xcite , and the two - loop qed effects @xcite . these calculations made it possible to determine the electron mass from the experimental values of the bound - electron @xmath0 factor @xcite . the theoretical accuracy of the bound - electron @xmath0 factor in hydrogen - like ions is presently at the @xmath2 level @xcite for light elements up to carbon but deteriorates quickly for heavier elements because of the unknown higher - order two - loop qed effects scaling with the nuclear charge number @xmath3 as @xmath4 . the experimental investigations of the @xmath0-factors of hydrogen - like ions were so far performed for ions with spinless nuclei . however , when applied to the ions with a non - zero nuclear spin , such investigations can be useful as a new method for the determination of the magnetic dipole moments of the nuclei . this method has important advantages over the more traditional approaches , such as nuclear magnetic resonance ( nmr ) , atomic beam magnetic resonance , collinear laser spectroscopy , and optical pumping ( op ) . these advantages are that ( i ) the simplicity of the system under investigation ( a hydrogen - like ion ) allows for an _ ab initio _ theoretical description with a reliable estimation of uncalculated effects and ( ii ) the influence of the nuclear - structure effects ( which are the main limiting factors for theory ) is relatively weak . this is in contrast to the existing methods , in which the experimental data should be corrected for several physical effects , which are difficult to calculate . among such effects is the diamagnetic shielding of the external magnetic field by the electrons in the atom . the nmr results should be also corrected for the paramagnetic chemical shift caused by the chemical environment @xcite and the op data are sensitive to the hyperfine mixing of the energy levels @xcite . significant ( and generally unknown ) uncertainties of calculations of these effects often lead to ambiguities in the published values of nuclear magnetic moments @xcite . the goal of the present investigation is to perform an _ ab initio _ calculation of the @xmath0-factor of a hydrogen - like ion with a non - zero nuclear spin . it can be demonstrated that the nuclear - spin dependent part of the atomic @xmath0-factor can be parameterized in terms of the nuclear magnetic shielding constant @xmath5 , which describes the effective reduction of the coupling of the nuclear magnetic moment @xmath6 to an external magnetic field @xmath7 caused by the shell electron(s ) , @xmath8 the relativistic theory of the @xmath0-factor of a hydrogen - like ion with a non - zero nuclear spin ( and , thus , the theory of the nuclear magnetic shielding ) was examined in detail in ref . @xcite . in the present work , we go beyond the relativistic description of the nuclear magnetic shielding and calculate the dominant corrections to it , namely , the self - energy , the vacuum - polarization , and the nuclear magnetization distribution corrections . as a result , we bring the theory to the point where the uncertainty due to nuclear - structure effects impedes further progress . the main challenge of the present work is the calculation of the self - energy correction . to the best of our knowledge , the only previous attempt to address it was made in ref . @xcite . in that work , the self - energy contribution to the shielding constant was estimated by the leading logarithm of its @xmath9 expansion ( where @xmath10 is the fine - structure constant ) . in our work , we calculate the self - energy correction rigorously to all orders in the binding nuclear strength parameter @xmath9 and , independently , perform an analytical calculation to the leading order in the expansion in this parameter ( including both the logarithm and constant terms ) . first results of this work were reported in ref . @xcite . the rest of the paper is organized as follows . in sec . [ sec:1 ] we summarize the relativistic theory of the @xmath0-factor of an ion with a non - zero nuclear spin and the theory of the nuclear magnetic shielding . our calculation of the self - energy and vacuum - polarization corrections to all orders in @xmath9 is described in sec . [ sec:2 ] . in sec . [ sec : zaexp ] we report the calculation of the qed correction to the nuclear magnetic shielding to the leading order in the @xmath9 expansion . [ sec:3 ] deals with the other effects , namely , the nuclear magnetization distribution , the nuclear recoil , and the quadrupole interaction . numerical results and discussion are given in sec . [ sec:4 ] . the paper ends with conclusion in sec . [ sec:5 ] . the relativistic units ( @xmath11 = @xmath12 ) and the charge units @xmath13 are used throughout this paper . we consider a hydrogen - like ion with a non - zero spin nucleus placed in a weak homogenous magnetic field @xmath7 directed along the @xmath14 axis . assuming that the energy shift due to the interaction with the @xmath7 field ( the zeeman shift ) is much smaller than the hyperfine - structure splitting ( hfs ) , the energy shift can be expressed in terms of the @xmath0 factor of the atomic system @xmath15 , @xmath16 where @xmath17 , @xmath18 is the bohr magneton , @xmath19 and @xmath11 are the elementary charge and the electron mass , respectively , and @xmath20 is the @xmath14 projection of the total angular momentum of the system @xmath21 . to the leading order , the energy shift is given by @xmath22|fm_f{\rangle}\,,\end{aligned}\ ] ] where @xmath23 is the wave function of the ion , @xmath24 and @xmath25 are the angular momentum quantum numbers of the electron and nucleus , respectively ; @xmath6 is the operator of the magnetic moment of the nucleus . the matrix element ( [ eq2 ] ) can easily be evaluated , yielding the well - known leading - order relativistic result @xcite , @xmath26 where @xmath27 is the dirac bound - electron @xmath0 factor @xcite , @xmath28 is the nuclear @xmath0 factor , @xmath29 is the nuclear magnetic moment , @xmath30 is the nuclear magneton , @xmath31 is the proton mass , and @xmath32 and @xmath33 are the angular - momentum recoupling coefficients , @xmath34/2\,,\\ { \langle}\vec{i}\cdot\vec{f } { \rangle}&= & [ f(f+1)+i(i+1)-j(j+1)]/2\,.\end{aligned}\ ] ] generalizing the leading - order result to include the higher - order correction , we write the atomic @xmath0 factor @xmath15 as @xmath35 in the above equation , the bound - electron @xmath0 factor @xmath36 incorporates all corrections that do not depend on the nuclear spin , whereas the nuclear shielding constant @xmath5 parameterizes the nuclear - spin dependent corrections . the bound - electron @xmath0 factor @xmath37 has been extensively studied during the last years , both theoretically @xcite and experimentally @xcite . the goal of the present work is the _ ab initio _ theoretical description of the nuclear shielding parameter @xmath5 . it can be seen from eq . ( [ eq4a ] ) that the nuclear - spin dependent contribution to the atomic @xmath0 factor @xmath15 is suppressed by the electron - to - proton mass ratio and thus is by about three orders of magnitude smaller than the nuclear - spin independent part proportional to @xmath37 . it is , however , possible to form a combination of the atomic @xmath0 factors which is free of the nuclear - spin independent contributions . so , for ions with the nuclear spin @xmath38 , we introduce the sum of the atomic @xmath0 factors @xmath39 that is directly proportional to the nuclear magnetic moment , @xmath40 this combination of the @xmath0 factors is particularly convenient for the determination of the nuclear magnetic dipole moments from experiment . indeed , if both the @xmath41 and @xmath42 @xmath0-factors are measured and @xmath5 is known from theory , eq . ( [ eq4b ] ) determines the nuclear magnetic moment @xmath43 . for the ions with a nuclear spin @xmath44 , eq . ( [ eq4b ] ) is not applicable and the nuclear magnetic moment should be determined from eq . ( [ eq4a ] ) . contributions to the nuclear magnetic shielding are described by feynman diagrams with two external interactions , one with the external magnetic field ( the zeeman interaction ) , @xmath45 and another , with the magnetic dipole field of the nucleus ( the hfs interaction ) @xmath46 the leading - order contribution to the magnetic shielding comes from the following energy shift @xmath47 where the summation runs over the whole dirac spectrum with the reference state excluded . as follows from eqs . ( [ eq1 ] ) and ( [ eq4a ] ) , contributions to the shielding constant @xmath48 are obtained from the corresponding energy shifts @xmath49 by @xmath50 for the electronic states with @xmath51 , all nuclear quantum numbers in eq . ( [ eq5 ] ) can be factorized out . the expression for the shielding constant is then obtained from that formula by using the following substitutions @xmath52 where the zero subscript refers to the zero spherical component of the vector . here and in what follows , @xmath53 denotes the dirac state with the relativistic angular quantum number @xmath54 and the fixed momentum projection @xmath55 . so , for @xmath51 , the leading contribution to the magnetic shielding is given by a simple expression @xmath56 performing the angular integrations with help of eq . ( [ app : eq3 ] ) , one gets for the @xmath57 reference state @xmath58 where @xmath59 are the radial integrals of the form @xmath60\,,\end{aligned}\ ] ] @xmath61 and @xmath62 are the upper and lower radial components of the dirac wave function , respectively , and the angular prefactors @xmath63 are given by @xmath64 , @xmath65 . for the point nuclear model , the sum over the dirac spectrum in the above expression can be evaluated analytically @xcite , see also more recent studies @xcite , @xmath66 where @xmath67 . for the extended nucleus , the calculation is easily performed numerically @xcite . the self - energy correction in the presence of the external magnetic field and the magnetic dipole field of the nucleus is graphically represented by six topologically non - equivalent feynman diagrams shown in fig . [ fig:1 ] . general expressions for these diagrams are conveniently obtained by using the two - time green s function method @xcite . the resulting formulas are summarized below . the irreducible contributions of diagrams on figs . [ fig:1](a)-(c ) can be represented in terms of the one - loop self - energy operator @xmath68 . this part will be termed as the perturbed - orbital contribution . the corresponding energy shift is @xmath69 where we used the short - hand notations for the reduced green s function , @xmath70 and its derivative @xmath71 the ( unrenormalized ) self - energy operator @xmath72 is defined by its matrix elements as follows @xmath73 where @xmath74 is the operator of the electron - electron interaction , @xmath75 , @xmath76 is the photon propagator , @xmath77 are the dirac matrices , and @xmath78 , where @xmath79 is a small imaginary addition that defines the positions of the poles of the electron propagators with respect to the integration contour . the summation over @xmath80 runs over the complete dirac spectrum . renormalization of the one - loop self - energy operator is well described in the literature , see , e.g. , ref . @xcite . in this work , the renormalized part of the self - energy operator is defined as @xmath81 where @xmath82 is the mass counterterm , @xmath83 is the dirac @xmath83 matrix , @xmath84 is the one - loop renormalization constant , @xmath85 is the binding coulomb potential of the nucleus , and the renormalization is to be performed in momentum space with a covariant regularization of ultraviolet ( uv ) divergencies . details on the renormalization procedure and explicit formulas for @xmath86 can be found in refs . @xcite . the irreducible contribution of the diagram shown in fig . [ fig:1](d ) , together with the corresponding derivative term , is referred to as the _ hfs - vertex _ contribution . it is given by @xmath87 where @xmath88 is the renormalized part of the 3-point vertex representing the interaction with the hfs field and @xmath89 is the derivative of the renormalized self - energy operator over the energy argument , @xmath90 . the unrenormalized 3-point hfs vertex operator is defined by its matrix elements as @xmath91 the renormalized part of the operator is obtained as @xmath92 where @xmath93 is the one - loop renormalization constant and the renormalization is to be performed in momentum space with a covariant regularization of uv divergencies , see ref . @xcite for details . the irreducible contribution of the diagram shown in fig . [ fig:1](e ) , together with the corresponding derivative term , is referred to as the _ zeeman - vertex _ contribution . it is given by @xmath94 where the 3-point zeeman vertex is defined analogously to the hfs one . the contribution of the diagram shown on fig . [ fig:1](f ) , together with the corresponding derivative terms , will be termed as the _ double - vertex _ contribution . it is defined by @xmath95 where @xmath96 denotes the 4-point vertex representing the interaction with both the zeeman and hfs interactions , @xmath97 @xmath98 denotes the second derivative of the self - energy operator over the energy argument , @xmath99 , @xmath100 denotes the derivative of the vertex operator over the energy argument , and @xmath101 denotes the intermediate electron states with the energy @xmath102 . the last term in eq . ( [ dvr ] ) is added artifically , in order to make the whole expression for @xmath103 infrared ( ir ) finite . the same term will be subtracted from the derivative contribution defined below , see eq . ( [ der ] ) . we note that all terms in eq . ( [ dvr ] ) are uv finite , so that there is no need for any uv regularization . there are , however , ir divergences , which appear when the energy of the intermediate electron states in the electron propagators coincides with the energy of the reference state . the divergences cancel out in the sum of individual terms in eq . ( [ dvr ] ) . finally , the remaining contribution will be termed as the _ derivative _ term . it is given by @xmath104 \nonumber \\ & & \times { \langle}v_{\rm zee } \,\frac1{({\varepsilon}_a - h)^{{\prime}}}\ , v_{\rm hfs}{\rangle}\,.\end{aligned}\ ] ] the second term in the brackets is added artifically , in order to compensate the ir reference - state divergence present in the first term , making the total expression for @xmath105 ir finite . note that this term is exactly the same as the one added to eq . ( [ dvr ] ) but has the opposite sign . finally , the total self - energy correction is given by the sum of the contributions discussed above , @xmath106 which are given by eqs . ( [ po1 ] ) , ( [ vrhfs ] ) , ( [ vrzee ] ) , ( [ dvr ] ) , and ( [ der ] ) , respectively . the general formulas reported so far represent contributions to the energy shift . we now have to separate out the nuclear degrees of freedom and convert the corrections to the energy into corrections to the shielding constant . in most cases , this is achieved simply by using the substitutions ( [ eqsubst ] ) . the double - vertex contribution , however , requires an explicit angular - momentum algebra calculation for the separation of the nuclear variables . we will see that most of the corrections to the shielding constant can be regarded as generalizations of the corrections already discussed in the literature . so , our present calculation will be largely based on the previous investigations of the self - energy correction to the lamb shift @xcite , to the hyperfine structure @xcite , and to the @xmath0 factor @xcite . the double - vertex correction of the kind similar to that in the present work appeared in the evaluation of the self - energy correction to the parity - nonconserving transitions in refs . @xcite . however , in this work we develop a different scheme for the evaluation of the double - vertex contribution based on the analytical representation of the dirac - coulomb green function . the matrix elements of the self - energy operator are diagonal in the relativistic angular momentum quantum number @xmath107 and the momentum projection @xmath43 . because of this , the angular reduction of the perturbed - orbital contribution is achieved by the same set of substitutions ( [ eqsubst ] ) as for the leading - order magnetic shielding . the resulting contribution to the shielding constant is conveniently represented as @xmath108 where the first - order perturbations of the reference - state wave function are given by @xmath109 and @xmath110 and @xmath111 is the standard second - order perturbation @xcite of the reference - state wave function induced by _ both _ interactions , @xmath112 and @xmath113 . note that only the diagonal in @xmath107 part of the perturbed wave functions contributes to @xmath114 . the calculation of the non - diagonal matrix elements of the self - energy operator is performed by a straightforward generalization of the method developed in ref . @xcite for the first - order self - energy correction to the lamb shift . the hfs - vertex correction to the energy shift ( [ vrhfs ] ) for the reference state with @xmath115 can be converted to the correction to the magnetic shielding by the substitution ( [ eqsubst ] ) . the result is @xmath116 where the covariant regularization of the ultraviolet ( uv ) divergences is implicitly assumed . the right - hand - side of eq . ( [ eqhfs ] ) differs from the vertex and reducible parts of the self - energy correction to the hyperfine structure only by the perturbed wave function @xmath117 in place of one of the reference - state wave functions @xmath118 . the main complication brought by this difference is that the perturbed wave function contains components with different values of the relativistic angular quantum number @xmath107 . so , for the reference state with @xmath119 , the perturbed wave function has components with @xmath120 and @xmath121 , both of which contribute to the first term in eq . ( [ eqhfs ] ) , denoted in the following as @xmath122 . the second ( reducible ) term contains only the @xmath123 component of the perturbed wave function and its calculation is done exactly as described in ref . below , we present the generalization of formulas derived in ref . @xcite needed for the evaluation of @xmath122 . as explained in ref . @xcite , the covariant separation of the uv divergences is conveniently performed by dividing the vertex contribution into the zero- and many - potential parts , according to the number of interactions with the binding coulomb field in the electron propagator , @xmath124 the zero - potential part is calculated in momentum space with the dimensional regularization of the uv divergences . the fourier transform of the @xmath112 is done by @xmath125 where @xmath126 is the transferred momentum . the contribution of the zero - potential hfs vertex part to the shielding constant is @xmath127_0}{{\vec{q}}^2}\ , \psi_{\delta a_{\nicefrac12}}({\vec{p}}_2)\,,\end{aligned}\ ] ] where @xmath128 and @xmath129 are 4-vectors with the fixed time component @xmath130 , @xmath131 , @xmath132 and @xmath133 are the reference - state and the perturbed wave functions , respectively , @xmath134 is the dirac conjugation , and @xmath135 is the renormalized one - loop vertex operator @xcite . for evaluating the integrals over the angular variables , it is convenient to use the following representation of the vertex operator sandwiched between the dirac wave functions @xmath136 \,,\end{aligned}\ ] ] where @xmath137 , @xmath138 are the spin - angular dirac spinors @xcite , and the scalar functions @xmath139 are given by eqs . ( a7)(a12 ) of ref . integration over the angular variables yields ( cf . ( 30 ) of ref . @xcite ) @xmath140 + { \cal r}_2 \ , [ p_{1r}k_1(-\kappa_a,\kappa_{\delta a})-p_{2r}k_1^{\prime}(-\kappa_a,\kappa_{\delta a } ) ] \nonumber \\ & + p_{1r}p_{2r}({\cal r}_3+{\cal r}_4)\ , k_2(\kappa_a,\kappa_{\delta a})\ , + p_{1r}p_{2r}({\cal r}_5+{\cal r}_6)\ , k_2(-\kappa_a,-\kappa_{\delta a } ) \bigr\}\,,\end{aligned}\ ] ] where @xmath141 , @xmath142 , @xmath143 is the relativistic angular quantum number of the perturbed wave function , @xmath144 , @xmath145 , @xmath146 , and the basic angular integrals @xmath147 are defined and evaluated in appendix [ app : angular ] . the many - potential vertex contribution is free from uv divergences and thus can be calculated in coordinate space . the result after the integration over the angular variables is @xmath148 where @xmath149 is the radial integral of the hfs type given by eq . ( [ eqrad ] ) , @xmath150 is the angular coefficient , @xmath151 @xmath152 is the reduced matrix element of the normalized spherical harmonics given by eq . ( c10 ) of ref . @xcite , @xmath153 is the relativistic generalization of the slater radial integral given by eqs . ( c1)(c9 ) of ref . @xcite , and the subtraction in the last line of eq . ( [ eq00 ] ) means that the contribution of the free propagators ( already accounted for by the zero - potential term ) needs to be subtracted . the zeeman - vertex correction to the energy shift ( [ vrzee ] ) for the reference state with @xmath115 can be converted to the correction to the magnetic shielding by the substitution ( [ eqsubst ] ) . the result has the form analogous to that for the hfs - vertex contribution , @xmath154 where the covariant regularization of the uv divergences is implicitly assumed . the above expression looks very similar to eq . ( [ eqhfs ] ) and can be evaluated almost in the same way , except for the zero - potential vertex contribution . the expression for the zero - potential zeeman vertex contribution is different from the hfs case because the momentum representation of the interaction with a constant magnetic field involves a @xmath155-function . the diagonal matrix element of the zeeman vertex operator was evaluated previously in ref . @xcite ; here we present the generalization of the formulas required for the non - diagonal case . the fourier transform of the interaction with the external magnetic field is given by @xmath156 the contribution of the zero - potential vertex to the shielding constant is @xmath157_0 \psi_{\delta a_{\nicefrac12}}({\vec{p}^{\prime}})\ , , \nonumber \\\end{aligned}\ ] ] where the gradient @xmath158 acts on the @xmath155-function only . this expression is transformed by integrating by parts and carrying out the integration with the @xmath155-function analytically . the result after the angular integration consists of two parts ( cf . ( 27 ) and ( 36 ) of ref . @xcite ) , @xmath159 where @xmath160\,,\end{aligned}\ ] ] and @xmath161 \nonumber \\ & + \frac1p\ , b_1(\rho)\ , \bigl [ g_af_{\delta a } a_3(\kappa_a,-\kappa_{\delta a } ) + f_ag_{\delta a } a_3(-\kappa_a,\kappa_{\delta a})\bigr ] \nonumber \\ & + b_2(\rho)\ , \bigl [ \tilde{g}_ag_{\delta a } a_4(\kappa_a,\kappa_{\delta a } ) + \tilde{f}_af_{\delta a } a_4(-\kappa_a,-\kappa_{\delta a})\bigr ] \nonumber \\ & + b_3(\rho)\ , \bigl [ g_ag_{\delta a } a_4(\kappa_a,\kappa_{\delta a } ) - f_af_{\delta a } a_4(-\kappa_a,-\kappa_{\delta a})\bigr]\biggr\}\,,\end{aligned}\ ] ] where @xmath162 , @xmath163 , @xmath164 , @xmath165 , @xmath166 and @xmath167 , the scalar functions @xmath168 , @xmath169 are given by eqs . ( 24 ) , ( 30)-(32 ) of ref . @xcite and @xmath170 are the basic angular integrals defined and evaluated in appendix [ app : angular ] . note that the expression for @xmath171 is non - symmetric with respect to @xmath172 and @xmath173 , but the result does not change when the two wave functions are interchanged . the double - vertex correction is defined by eq . ( [ dvr ] ) . all parts of it are uv finite and thus can be evaluated in coordinate space . we denote the individual terms in the right - hand - side of eq . ( [ dvr ] ) by @xmath174 and consider each of them separately , @xmath175 the first term is @xmath176 after separating the nuclear variables and integrating over the angles , the contribution to the magnetic shielding is ( for @xmath177 ) @xmath178 where @xmath179 note that the expression for @xmath180 is ir divergent when any two or all three of the intermediate states have the same energy as the reference state : @xmath181 , @xmath182 , @xmath183 , @xmath184 . the ir divergence cancels when all parts of eq . ( [ dvr ] ) are added together . the contribution of the second term to the shielding is @xmath185 where @xmath186 is the normalization integral , @xmath187 the expression for @xmath188 is ir divergent when @xmath184 . the third term is given by @xmath189 \,,\end{aligned}\ ] ] where @xmath190 the fourth term @xmath191 is obtained from @xmath192 by an obvious substitution @xmath193 . the fifth term is given by @xmath194 finally , the total double - vertex contribution is given by the sum of the five terms discussed above , @xmath195 despite the fact that the individual terms @xmath196 are ir divergent , the sum of them is finite and can be evaluated without any explicit regularization , provided that ( i ) the integration over the frequency of the virtual photon in the self - energy loop @xmath197 is performed after all five terms are added together and ( ii ) the contour of the @xmath197 integration is suitably chosen . one can show that if the @xmath197 integration is performed along the contour consisting of the low - energy and high - energy parts , as e.g. , in refs . @xcite , the integrand becomes regular at @xmath198 and , therefore , can be directly evaluated numerically . the numerical evaluation of the double - vertex correction @xmath199 was the most time consuming part of our calculation , due to a large number of the partial waves involved and a four - dimensional radial integration in eq . ( [ delta1 ] ) . the radial integration was carried out with help of the numerical approach developed in our calculation of the two - loop self - energy and described in detail in ref . @xcite . because of numerical cancellations between the five terms in eq . ( [ sum ] ) , especially in the region of small values of @xmath197 , we took care to treat all five terms exactly in the same way . in particular , the normalization integrals @xmath186 in eqs . ( [ delta2 ] ) and ( [ delta3 ] ) were evaluated numerically , in order to be consistent with the evaluation of eq . ( [ delta1 ] ) , whereas the corresponding contributions could have been evaluated more easily by taking the derivative of the dirac - coulomb green function . vacuum polarization corrections to the magnetic shielding calculated in the present work are shown in fig . [ fig:2 ] . the diagrams in fig . [ fig:2](a)-(c ) come from the insertion of the uehling potential into the electron line of the leading - order magnetic shielding , whereas the diagram in fig . [ fig:2](d ) respresents the uehling potential insertion into the hfs interaction . we note that the diagram with the vacuum polarization insertion into the zeeman interaction vanishes in the uehling - potential approximation . in our treatment , we neglect contributions with additional coulomb interactions in the vacuum - polarization loop ( which correspond to the wichmann - kroll part of the one - loop vacuum polarization ) and the additional diagram of the wichmann - kroll type with both hfs and zeeman interactions attached to the vacuum - polarization loop . we expect that the part accounted for in the present work yields the dominant contribution to the vacuum polarization correction . the contribution of the diagrams in fig . [ fig:2 ] ( a)-(c ) is an analogue of the perturbed - orbital self - energy contribution and is given by a similar expression , @xmath200 where @xmath201 is the uehling potential , @xmath202\,,\end{aligned}\ ] ] and @xmath203 and the nuclear - charge density @xmath204 is normalized by the condition @xmath205 . we note that this contribution can also be calculated by incorporating the uehling potential into the dirac equation and re - evaluating the leading - order magnetic shielding , in this way accounting for the uehling potential to all orders . we performed calculations in both ways , which ensured that the perturbations of the reference - state wave function @xmath206 and @xmath207 are computed correctly . the contribution of the diagram in fig . [ fig:2](d ) to the magnetic shielding can be expressed as @xmath208 where @xmath209 is the hfs interaction modified by the vacuum - polarization insertion . in the previous section , we calculated the qed corrections to the magnetic shielding without any expansion in the nuclear binding strength parameter @xmath9 . now we turn to the evaluation of this corrections within the expansion in this parameter . we will derive the complete expression for the leading term of the @xmath9 expansion , which enters in the relative order @xmath210 . the results obtained in this section will be applicable for light hydrogen - like ions , where no all - order calculations were possible because of large numerical cancellation . they will also provide an important cross - check with the all - order calculations described in the previous section . for the derivation , it is convenient to use the formalism of the non - relativistic quantum electrodynamics ( nrqed ) . in this approach , all qed effects are calculated by an expansion in powers of @xmath10 and @xmath9 and represented as expectation values of various effective operators on the nonrelativistic reference - state wave function . let us start with the effective hamiltonian @xmath211 , which includes leading one - loop radiative corrections : @xmath212 where @xmath213 denotes anticommutator , @xmath214 . the qed effects in the above hamiltonian are parameterized in terms of the constants @xmath107 , @xmath215 , and @xmath216 , which are interpreted as the electron magnetic moment anomaly , the charge radius and the magnetic polarizability , respectively , @xmath217 where @xmath218 is the photon momentum cutoff . in the non - qed limit , all qed constants vanish , @xmath219 , and the effective hamiltonian @xmath211 turns into the schrdinger - pauli hamiltonian . the qed constants @xmath220 and @xmath216 depend explicitly on the the photon momentum cutoff parameter @xmath218 . this dependence cancels out with contributions coming from the emission and absorption of virtual photons with the energy smaller than @xmath218 . the complete expression for any physical quantity , e.g. , the lamb shift and the shielding constant , does not depend on the artificial photon cutoff parameter . the above formula for @xmath221 is obtained from the well known expressions for the one - loop radiative corrections to electromagnetic formfactors @xmath222 and @xmath223 @xcite . the formula for @xmath216 has not appeared in the literature previously . we have derived it by a method similar to that used in ref . @xcite for the electric polarizability , denoted by @xmath224 in that work . we will start with rederiving the known nonrelativistic expression for the shielding constant within our approach . it comes from the @xmath225 term in the electron kinetic energy in eq . ( [ t02 ] ) . the electromagnetic potential @xmath226 is the sum of the external magnetic potential @xmath227 , @xmath228 and the potential induced by the nuclear magnetic moment , @xmath229 the corresponding energy shift is @xmath230 where matrix elements are calculated with the nonrelativistic wave function . the shielding constant @xmath5 is obtained from the energy shift @xmath49 by @xmath231 . for the ground @xmath232 hydrogenic state , the nonrelativistic results is @xmath233 before considering the qed effects to the shielding constant , it is convenient to first recalculate the leading qed correction to the energy levels . the total contribution is split into two parts induced by the virtual photons of low ( @xmath234 ) and high ( @xmath235 ) energy , @xmath236 the high - energy part is the expectation value of the @xmath237 term in the effective hamiltonian in eq . ( [ t02 ] ) @xmath238 where @xmath239 and @xmath240 . the vacuum polarization can be incorporated in the above expression by adding @xmath241 to @xmath221 in eq . ( [ t03 ] ) . the low - energy part @xmath242 is induced by the emission and the absorption of the virtual photons of low @xmath243 energy , @xmath244 -\ln\biggl[\frac{2\,(h - e)}{m\,(z\,\alpha)^2}\biggr]\biggr\ } \frac{\vec p}{m}\biggr\rangle\,.\end{aligned}\ ] ] the terms containing @xmath245 cancel out in the sum of the low- and high - energy parts , as expected . the total leading - order lamb shift contribution is @xmath246\,\delta_{l0 } \nonumber \\ & & + \biggl(\frac{10}{9}-\frac{4}{15}\biggr)\,\delta_{l0 } -\frac{4}{3}\,\ln k_0(n , l)\biggr\}\,,\end{aligned}\ ] ] where @xmath247 is the bethe logarithm , @xmath248 \vec p\,\bigr\rangle\,.\ ] ] we now turn to the calculation of the qed correction to the magnetic shielding , which is performed similarly to that for the lamb shift . the total contribution is split into the low and the high - energy parts : @xmath249 where @xmath250 and @xmath251 the high - energy part can be conveniently rewritten in the form @xmath252 in the low - energy part , we separate out @xmath245 and then perform an expansion in the magnetic fields , @xmath253 using the identity @xmath254\biggr]\biggr\rangle \nonumber \\ & = \biggl\langle4\,\pi\,z\,\alpha\,\delta^3(r ) + \frac{e}{2\,m^3}\bigl\{\vec \pi\,,\,\vec\nabla\times\vec b\bigr\ } + \frac{2\,e^2}{m^3}\,b^2 \biggr\rangle\,,\end{aligned}\ ] ] we transform @xmath255 to the form @xmath256 \nonumber \\ & & \biggl[\,2\,\biggl\langle \frac{\alpha}{3\,r}\,\frac{1}{(e - h ) ' } \,4\,\pi\,z\,\alpha\,\delta^3(r)\biggr\rangle\,\vec\mu\cdot\vec b\\\nonumber & & -4\,\pi\,\alpha\,\bigl\langle\vec a_e\cdot\vec\nabla\times\vec b_i\bigr\rangle + 16\,\pi\,\alpha\,\bigl\langle\vec b\cdot\vec b_i\bigr\rangle\biggr]\,.\end{aligned}\ ] ] the artificial parameter @xmath245 cancels out in the sum @xmath257 separately for each type of the matrix elements , @xmath258\ , \vec\mu\cdot\vec b\ , \biggl\langle 4\,\pi\,\delta^3(r)\,\frac{1}{e - h)'}\,\frac{1}{r}\biggr\rangle \nonumber \\ & -\frac{4}{3}\,\alpha^2\,\biggl [ \ln(z\,\alpha)^{-2}+\frac{31}{48}-\frac{1}{5}\biggr]\ , \bigl\langle\vec a_e\cdot\vec\nabla\times\vec b_i\bigr\rangle \nonumber \\ & + \frac{16}{3}\,\alpha^2\ , \biggl[\ln(z\,\alpha)^{-2}-\frac{13}{24}\biggr]\,\bigl\langle\vec b\cdot\vec b_i\bigr\rangle.\end{aligned}\ ] ] using the following results for the matrix elements with @xmath259 states @xmath260 we obtain @xmath261\,\vec\mu\cdot\vec b\,.\ ] ] the calculation of the remaining low - energy contribution @xmath262 is slightly more complicated . we first return to the integral form of @xmath242 , derive an expression for @xmath262 , and then drop all terms with @xmath245 , as they are already accounted for by @xmath255 . the integral form of @xmath242 is @xmath263 it can be rewritten , using the identity @xmath264 $ ] , in the form @xmath265 all terms with positive powers of @xmath218 are discarded , since one assumes that the limit @xmath266 is taken after the expansion in @xmath10 is done . we shall now expand the integrand in eq . ( [ t34 ] ) in the magnetic fields . the hamiltonian @xmath235 is @xmath267 where @xmath268 . the first term with @xmath269 in eq . ( [ t35 ] ) can be absorbed into @xmath270 . so , the correction due to this term is @xmath271-\ln k_0\biggr\ } \\ & = & \frac{16\,\alpha^2}{9\,\pi}\,\frac{(z\,\alpha)^3}{n^3}\,\vec\mu\cdot\vec b\ , \biggl\{\ln k_0 + \frac{1}{2}-\ln\biggl[\frac{2\,\epsilon}{m\,(z\,\alpha)^2}\biggr]\biggr\}\ , . \nonumber\end{aligned}\ ] ] the correction due to the two last terms in eq . ( [ t35 ] ) is @xmath272 integrating by parts and using the results and the notation of @xmath273 from ref . @xcite @xmath274-\ln k_3\biggr\}\,,\end{aligned}\ ] ] we obtain @xmath275-\ln k_3-\frac{1}{3}\biggr\}\,.\ ] ] the sum of these two contributions @xmath276 , after dropping the @xmath245 terms , is @xmath277 finally , the total correction to energy is @xmath278 expressing the energy shift in terms of the shielding constant ( @xmath279 ) , we obtain the complete result for the leading qed correction to the nuclear magnetic shielding in hydrogen - like ions , valid for the @xmath259 states , @xmath280 + 2\,\ln k_0(n)-3\,\ln k_3(n)-\frac{421}{96}+\frac{3}{5}\biggr\}\ , . \label{eq : za}\end{aligned}\ ] ] the term of @xmath281 in the brackets is the contribution of the vacuum polarization . the numerical results for the bethe logarithm @xmath282 and the @xmath283 bethe - logarithm - type correction @xmath273 @xcite for the @xmath284 state are @xmath285 we note that the numerical value of the constant term in eq . ( [ eq : za ] ) , @xmath286 , is comparable in magnitude but of the opposite sign to the logarithmic term at @xmath287 , @xmath288 . this entails a significant numerical cancellation between these two terms for hydrogen and light hydrogen - like ions . as a result , the total qed correction turns out to be much smaller in magnitude than could be anticipated from the leading logarithm alone . we now turn to the effect induced by the spatial distribution of the nuclear magnetic moment , also known as the bohr - weisskopf ( bw ) correction . our treatment of the bw effect is based on the effective single - particle ( sp ) model of the nuclear magnetic moment . within this model , the magnetic moment is assumed to be induced by the odd nucleon ( proton , when @xmath3 and @xmath289 are odd and neutron , when @xmath3 is even and @xmath289 is odd ) with an effective @xmath0-factor , which is adjusted to yield the experimental value of the nuclear magnetic moment . the treatment of the magnetization distribution effect on hfs within the sp model was originally developed in refs . @xcite and later in ref . our present treatment closely follows the procedure described in refs . @xcite . the wave function of the odd nucleon is assumed to satisfy the schrdinger equation with the central potential of woods - saxon form and the spin - orbital term included ( see , e.g. , ref . @xcite ) @xmath290 where @xmath291 @xmath292^{-1}\,,\ ] ] and @xmath85 is the coulomb part of the interaction ( absent for neutron ) , with the uniform distribution of the charge ( @xmath293 ) over the nuclear sphere . the parameters @xmath294 , @xmath295 , @xmath296 , and @xmath172 were taken from refs . @xcite . the nuclear magnetic moment can be evaluated within the sp model to yield @xcite @xmath297 g_l\ , , \ \ \ \ \ \ \ \ \ \ \ & \\ \ \mbox{for } \ \ i = l+\frac12\ , , & \\ \displaystyle -\frac{i}{2(i+1)}\ , g_s+ \left[\frac{i(2i+3)}{2(i+1 ) } -\frac{2i+1}{4(i+1)}{\langle}\phi_{so}r^2{\rangle}\right ] & \ g_l\,,\ \\ \ \mbox{for } \ \ i = l-\frac12\ , , & \\ \end{aligned } \right.\end{aligned}\ ] ] where @xmath25 and @xmath234 are the total and the orbital angular momentum of the nucleus , respectively , @xmath298 is the @xmath0 factor associated with the orbital motion of the nucleon ( @xmath298 = 1 for proton and @xmath299 for neutron ) and @xmath300 is the effective nucleon @xmath0 factor , determined by the condition that eq . ( [ 5eq11 ] ) yields the experimental value of the magnetic moment . it was demonstrated in ref . @xcite that , within the sp model , the bw effect can be accounted for by adding a multiplicative magnetization - distribution function @xmath301 to the standard point - dipole hfs interaction . the distribution function is given by @xcite @xmath302 \nonumber \\ & & + \frac{\mu_n}{\mu } \int_r^{\infty}dr{^{\prime}}\ , { r{^{\prime}}}^2 |u(r{^{\prime}})|^2\,\frac{{r}^3}{{r{^{\prime}}}^3}\ , \left [ -\frac{2i-1}{8(i+1)}\ , g_s+ \left(i-\frac12+\frac{2i+1}{4(i+1)}\,r^2\phi_{so}(r ) \right ) g_l \right]\,,\end{aligned}\ ] ] for @xmath303 and @xmath304 \nonumber \\ & & + \frac{\mu_n}{\mu } \int_r^{\infty}dr{^{\prime}}\ , { r{^{\prime}}}^2 |u(r{^{\prime}})|^2\,\frac{{r}^3}{{r{^{\prime}}}^3}\ , \left [ \frac{2i+3}{8(i+1)}\ , g_s+ \left(\frac{i(2i+3)}{2(i+1)}-\frac{2i+1}{4(i+1)}\,r^2\phi_{so}(r ) \right ) g_l \right]\,,\end{aligned}\ ] ] for @xmath305 . in the above formulas , @xmath306 is the wave function of the odd nucleon . it can easily be seen that @xmath307 outside the nucleus . the recoil correction to the magnetic shielding was obtained in ref . @xcite in the nonrelativistic approximation , @xmath308 where @xmath309 is the nuclear mass and @xmath310 the electric - quadrupole correction to the magnetic shielding is @xmath311 where @xmath312 is the quadrupole moment of the nucleus and @xmath313 is the quadrupole correction to the @xmath0 factor calculated in ref . @xcite , @xmath314 } { 135\,\gamma ( 1+\gamma)^2\left[15 - 16({{z\alpha}})^2\right ] } \nonumber \\ & = \alpha\,({{z\alpha}})^3 \,\left[\frac{11}{135}+ \frac{43}{405}({{z\alpha}})^2+\ldots\right]\,.\end{aligned}\ ] ] numerical results for the self - energy correction to the nuclear magnetic shielding can be conveniently parameterized in terms of the dimensionless function @xmath315 defined as @xmath316 our numerical all - order ( in @xmath9 ) results for the self - energy correction to the magnetic shielding are summarized in table [ tab : se ] . we note significant numerical cancellation between the individual contributions , which is particularly strong for low values of @xmath3 . the fact that the resulting sum is a smooth function of @xmath3 and demonstrates the expected @xmath3 scaling serves as a consistency check of our calculations . besides of numerical cancellation , the additional complication arising in the low-@xmath3 region is that the convergence of the partial - wave expansions becomes slower when @xmath3 decreases . because of these two complications , the accuracy of our results worsens for smaller values of @xmath3 and no all - order results were obtained for @xmath317 . the all - order numerical results can be compared with the @xmath9-expansion results obtained in sec . [ sec : zaexp ] . as follows from eq . ( [ eq : za ] ) , the @xmath9 expansion of the function @xmath318 for the @xmath284 state reads @xmath319\,,\end{aligned}\ ] ] where @xmath320 denotes the higher - order terms . in fig . [ fig : se ] , the numerical all - order results for the function @xmath318 are plotted together with the contribution of the leading logarithmic term in eq . ( [ seza ] ) ( dashed line , red ) and the contribution of both terms in eq . ( [ seza ] ) ( dashed - dot line , blue ) . we observe that the leading logarithm alone gives a large contribution that disagrees strongly with the all - order results . however , when the constant term is added , the total @xmath9-expansion contribution shrinks significantly and even changes its sign for @xmath321 . only after the constant term is accounted for , we observe reasonable agreement between the all - order and @xmath9 expansion results . ] ] and 5 . [ fig : total ] ] the vacuum - polarization correction to the nuclear magnetic shielding is parameterized as @xmath322 our numerical results for the vacuum - polarization correction are presented in table [ tab : vp ] . the calculation was performed for the extended nucleus and the uehling potential was included to all orders . note that our present treatment is not complete , as the wichmann - kroll part of the correction is still missing . we estimate the uncertainty due to omitted terms to be within 30% of the calculated contribution . the @xmath9 expansion of the function @xmath323 reads @xmath324 where the kronecker symbol @xmath325 indicates that the correction vanishes for the reference states with @xmath326 . comparison of the all - order numerical results with the leading term of the @xmath9 expansion is given in fig . [ fig : vp ] . it is remarkable that the all - order results grow fast when @xmath3 is increased and eventually become more than an order of magnitude larger than the leading - order result . in the high-@xmath3 region , the self - energy and vacuum - polarization corrections are ( as usual ) of the opposite sign , the self energy being about twice larger than the vacuum polarization . in the low-@xmath3 region , however , the self - energy correction changes its sign and the total qed contribution becomes positive for very light ions . the summary of our calculations of the nuclear magnetic shielding constant @xmath5 for several hydrogen - like ions is given in table [ tab : total ] . the first line of the table presents results for the leading - order nuclear shielding , including the finite nuclear size effect . the leading - order contribution was calculated using the fermi model for the nuclear charge distribution and the point - dipole approximation for the interaction with the nuclear magnetic moment . the nuclear charge radii were taken from ref . the results are in good agreement with those reported in ref . the qed correction presented in the second line of table [ tab : total ] is the sum of the all - order results for the self energy and the vacuum polarization . its error comes from the numerical uncertainty of the self - energy part and the estimate of uncalculated vacuum - polarization diagrams . since there were no all - order calculations performed for oxygen , we used an extrapolation of our all - order results ( taking into account the derived values of the @xmath9 expansion coefficients ) . the bohr - weisskopf correction , presented in the third line of table [ tab : total ] , was calculated by reevaluating the leading - order contribution with the point - dipole hfs interaction modified by the extended - distribution function @xmath301 given by eqs . ( [ 5eq12 ] ) and ( [ 5eq13 ] ) . because the effective single - particle model of the nuclear magnetic moment is ( of course ) a rather crude approximation , we estimate the uncertainty of the bohr - weisskopf correction to be 30% , which is consistent with previous estimates of the uncertainty of this effect @xcite . this uncertainty includes also the error due to the nuclear polarization effect , which is not considered in the present work . the quadrupole and the recoil corrections , given in the fourth and fifth lines of table [ tab : total ] , respectively , were evaluated according to eqs . ( [ eqq ] ) and ( [ eqrec ] ) . the error of the quadrupole correction comes from the nuclear quadrupole moments . the magnetic dipole and electric quadrupole moments of nuclei were taken from refs . @xcite . the @xmath3 dependence of individual contributions to the magnetic shielding constant @xmath5 is shown in fig . [ fig : total ] . the leading - order contribution is separated into three parts , the point - nucleus nonrelativistic part ( solid line , black ) , the point - nucleus relativistic part ( dashed line , green ) , and the finite nuclear size correction ( dashed - dotted line , blue ) . we observe that the finite nuclear size correction , as well as the other effects calculated in this work , become increasingly important in the region of large nuclear charges . the only exception is the nuclear recoil correction . it practically does not depend on the nuclear charge number @xmath3 , since the linear @xmath3 scaling in eq . ( [ eqrec ] ) is compensated by the increase of the nuclear mass @xmath309 with @xmath3 . as a consequence , the recoil effect is completely negligible for high- and medium-@xmath3 ions , but turns into one of the dominant corrections for @xmath317 . for most ions , the uncertainty of the theoretical prediction is roughly 30% of the bohr - weisskopf effect ; it can be immediately estimated from fig . [ fig : total ] . for very light ions , however , there is an additional uncertainty due to the unknown relativistic recoil effect . its contribution can be estimated by multiplying the nonrelativistic recoil correction plotted on fig . [ fig : total ] by the factor of @xmath327 . we observe that for very light ions , the relativistic recoil is the dominant source of error in theoretical predictions . in this work we have performed an _ ab initio _ calculation of the nuclear magnetic shielding in hydrogen - like ions with inclusion of relativistic , nuclear , and qed effects . the uncertainty of our theoretical predictions for the nuclear magnetic shielding constant defines [ according to eq . ( [ eq4b ] ) ] the precision to which the nuclear magnetic dipole moments can be determined from experiments on the @xmath0-factors of hydrogen - like ions . it can be concluded from table [ tab : total ] and fig . [ fig : total ] that the present theory permits determination of nuclear magnetic moments with fractional accuracy ranging from @xmath328 in the case of @xmath329o@xmath330 to @xmath331 for @xmath332bi@xmath333 . for most hydrogen - like ions , the dominant source of error in the theoretical predictions is the bohr - weisskopf effect . since this effect can not be accurately calculated at present , we conclude that the theory of the nuclear magnetic shielding has reached the point where the uncertainty due to nuclear - structure effects impedes further progress . for very light ions , however , the dominant theoretical error comes from the unknown relativistic recoil effect , whose calculation might be a subject of future work . stimulating discussions with k. blaum and g. werth are gratefully acknowledged . v. a. y. and z. h. were supported by the alliance program of the helmholtz association ( ha216/emmi ) . k.p . acknowledges support by nist precision measurement grant pmg 60nanb7d6153 . c ...... @xmath3 & & & & & & + + 10 & -9.x584 & 6.x095 & 9.x180 & -49.x722 & 43.x523 & -0.x508(100 ) + 14 & -4.x741 & 1.x711 & 4.x321 & -21.x048 & 19.x047 & -0.x710(15 ) + 16 & -3.x571 & 0.x705 & 3.x140 & -14.x718 & 13.x655 & -0.x789(9 ) + 20 & -2.x217 & -0.x405 & 1.x761 & -7.x848 & 7.x782 & -0.x927(4 ) + 26 & -1.x277 & -1.x115 & 0.x772 & -3.x473 & 3.x983 & -1.x110(2 ) + 32 & -0.x858 & -1.x407 & 0.x292 & -1.x644 & 2.x333 & -1.x283(1 ) + 40 & -0.x624 & -1.x580 & -0.x043 & -0.x586 & 1.x315 & -1.x519(1 ) + 45 & -0.x575 & -1.x643 & -0.x171 & -0.x267 & 0.x975 & -1.x681 + 54 & -0.x595 & -1.x744 & -0.x335 & 0.x023 & 0.x622 & -2.x029 + 60 & -0.x672 & -1.x825 & -0.x423 & 0.x112 & 0.x487 & -2.x321 + 70 & -0.x929 & -2.x028 & -0.x571 & 0.x175 & 0.x355 & -2.x999 + 82 & -1.x612 & -2.x508 & -0.x812 & 0.x191 & 0.x284 & -4.x457(2 ) + 83 & -1.x701 & -2.x569 & -0.x839 & 0.x192 & 0.x281 & -4.x636(1 ) + 92 & -3.x011 & -3.x400 & -1.x174 & 0.x198 & 0.x280 & -7.x107(2 ) + cccc @xmath3 & & & + + 10 & 0.118 & 0.110 & 0.228 + 14 & 0.135 & 0.121 & 0.256 + 16 & 0.144 & 0.126 & 0.271 + 20 & 0.164 & 0.137 & 0.302 + 26 & 0.200 & 0.155 & 0.355 + 32 & 0.242 & 0.175 & 0.417 + 40 & 0.314 & 0.206 & 0.520 + 45 & 0.369 & 0.228 & 0.597 + 54 & 0.500 & 0.275 & 0.775 + 60 & 0.618 & 0.315 & 0.933 + 70 & 0.891 & 0.398 & 1.289 + 82 & 1.449 & 0.546 & 1.996 + 83 & 1.512 & 0.562 & 2.074 + 92 & 2.227 & 0.727 & 2.954 l ..... & & & & & + + leading & 143.3x127 & 375.9x60 & 657.x93 & 1461.x6 & 4112x + qed & -0.0x026(2 ) & -0.1x03(15 ) & -0.x59(8 ) & -4.x1(0.8 ) & -30x(7 ) + bohr - weisskopf & -0.0x013(4 ) & -0.0x61(18 ) & -0.x54(16 ) & -8.x2(2.5 ) & -42x(13 ) + quadrupole & -0.0x007(1 ) & -0.0x18 & -0.x42 & 6.x9(0.1 ) & 7x + recoil & -0.0x120 & -0.0x15 & -0.x02 & 0.x0 & 0x + total & 143.2x960(5 ) & 375.7x63(24 ) & 656.x36(18 ) & 1456.x3(2.6 ) & 4046x(15 ) + throughout this work , we repeatedly used the following result for the matrix element of the hfs and zeeman interaction @xmath334 where @xmath335 , @xmath336 is the clebsch - gordan coefficient , and @xmath337 the radial integral @xmath59 is defined by eq . ( [ eqrad ] ) and the angular coefficient @xmath152 is the reduced matrix element of the normalized spherical harmonics , see , e.g. , eq . ( c10 ) of ref . @xcite . an important particular case is @xmath338 and @xmath339 , in which case the above formulas reduce to @xmath340 @xmath342_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}}_2 ) = \int_{-1}^1 d\xi\,f(\xi)\ , k_1(\kappa_a,\kappa_b)\ , , \\ \frac{3i}{4\pi } \int d{\hat{{\vec{p}}}}_1\ , d{\hat{{\vec{p}}}}_2\ , & \ f(\xi)\ , \chi^{\dag}_{\kappa_a { \nicefrac12}}({\hat{{\vec{p}}}}_1)\ , [ { \hat{{\vec{p}}}}_2\times { \vec{\sigma}}]_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}}_2 ) = \int_{-1}^1 d\xi\,f(\xi)\ , k_1^{\prime}(\kappa_a,\kappa_b)\ , , \\ \frac{3i}{4\pi } \int d{\hat{{\vec{p}}}}_1\ , d{\hat{{\vec{p}}}}_2\ , & \ f(\xi)\ , \chi^{\dag}_{\kappa_a { \nicefrac12}}({\hat{{\vec{p}}}}_1)\ , [ { \hat{{\vec{p}}}}_1\times { \hat{{\vec{p}}}}_2]_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}}_2 ) = \int_{-1}^1 d\xi\,f(\xi)\ , k_2(\kappa_a,\kappa_b)\,,\end{aligned}\ ] ] where @xmath343 is an arbitrary function and @xmath344 . the integrals over all angles except for @xmath345 in the above equations are evaluated analytically , as described in appendix of ref . the results relevant for this work are @xmath346 k_1^{\prime}(-1,1 ) = -1 , & k_1^{\prime}(1,-1 ) = \xi , & k_1^{\prime}(-1,-2 ) = -\frac{\displaystyle 1}{\displaystyle \sqrt{2 } } , & k_1^{\prime}(1,2 ) = \frac{\displaystyle \xi}{\displaystyle \sqrt{2 } } , \\[2pt ] k_2(-1,-1 ) = 0 , & k_2(1,1 ) = \frac{\displaystyle -1+\xi^2}{\displaystyle 2 } , & k_2(-1,2 ) = 0 , & k_2(1,-2 ) = \frac{\displaystyle -1+\xi^2}{\displaystyle 2\sqrt{2}}. \end{array}\end{aligned}\ ] ] the basic angular integrals @xmath347 needed for the evaluation of the zeeman vertex contribution are given by @xmath348_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}})\ , , \\ a_3(\kappa_a,\kappa_b ) & \ = \int d{\hat{{\vec{p}}}}\ , \chi^{\dagger}_{\kappa_a { \nicefrac12}}({\hat{{\vec{p}}}})\ , i\,[{\vec{\sigma}}\times \vec{\nabla}_{\omega}]_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}})\ , , \\ a_4(\kappa_a,\kappa_b ) & \ = \int d{\hat{{\vec{p}}}}\ , \chi^{\dagger}_{\kappa_a { \nicefrac12}}({\hat{{\vec{p}}}})\ , i\,[{\hat{{\vec{p}}}}\times \vec{\nabla}_{\omega}]_0\ , \chi_{\kappa_b { \nicefrac12}}({\hat{{\vec{p}}}})\,.\end{aligned}\ ] ] these integrals are evaluated by using the standard racah algebra . the results relevant for this work are @xmath349 a_2(-1,1 ) = 2/3 , & a_2(1,-1)=-2/3 , & a_2(-1,-2 ) = \sqrt{2}/3 , & a_2(1,2)=-\sqrt{2}/3 , \\[2pt ] a_3(-1,1 ) = 4/3 , & a_3(1,-1 ) = 0 , & a_3(-1,-2 ) = 2\sqrt{2}/3 , & a_3(1,2)=-\sqrt{2},\\[2pt ] a_4(-1,-1 ) = 0 , & a_4(1,1 ) = -2/3 , & a_4(-1,2 ) = 0 , & a_4(1,-2 ) = -\sqrt{2}/3 . \end{array}\end{aligned}\ ] ]

in high energy gamma - ray astrophysics observations , it is vital to reduce the background effectively in order to achieve a high sensitivity , for the source intensity is quite low . most of the background events are generated via an interaction between cosmic - rays and the instrument , and an accurate cosmic - ray flux model and computer simulation are necessary to develop the background rejection algorithm and evaluate the remaining background level . the model needs to cover the entire solid angle ( @xmath6 ) and the sensitive energy range of the instrument . while there are many measurements on cosmic - rays reported in literature , they are fragmentary and need to be compiled , reviewed and validated . the large area telescope ( lat ) is the high - energy gamma - ray detector onboard the gamma - ray large area space telescope ( glast ) satellite to be launched to low - earth orbit ( altitude of @xmath7 ) in 2007 @xcite . it consists of 16 towers in a 4 x 4 array . each tower has a set of high - z foil and si strip pair conversion trackers ( tkr ) and a csi(tl ) hodoscopic calorimeter ( cal ) . the towers are surrounded on the top and sides by a set of plastic scintillator anticoincidence detectors ( acd ) . because the lat will have a large active volume ( @xmath8 ) , the data acquisition trigger , based on signals in three consecutive x - y pair layers of the tkr , is expected to be about 10 khz , whereas the gamma - ray signal from a strong extraterrestrial source ( e.g. , crab pulsar ) is only about one per minute . in order to achieve the required sensitivity , we have to develop a `` cosmic - ray proof '' instrument by an optimized on - line / off - line event filtering scheme . all possible background types need to be studied in detail by beam tests and computer simulations . to validate the basic design of lat in a space - like environment , and collect particle incidences to be used as a background event database for lat , a balloon flight engineering model ( bfem ) , representing one of 16 towers that compose the lat , was built and launched on august 4 , 2001 @xcite . the instrument operated successfully and took data for about two hours during the ascent and three hours during the level flight . a geant4-based @xcite simulator of bfem was constructed for this balloon experiment . we also have developed a cosmic - ray background flux model based on existing measurements and theoretical predictions , and validated the model by comparing with the data of bfem . the model covers the entire solid angle ( @xmath9 ) , a wide energy range between 10 mev to 100 gev and abundant components ( proton , alpha , @xmath10 , @xmath11 and gamma ray ) . in this paper , we present the model functions as `` a working model of cosmic - ray background fluxes '' for high energy astrophysical missions . although the models are intended to be used primarily for glast bfem , they are expressed in analytic functions in which modulations due to solar activity and earth geomagnetism are parameterized , and they are easy to be applied to other balloon and satellite experiments . the models in low - earth orbit are being used to simulate the glast lat in the orbital environment and to develop background rejection algorithms . the glast lat bfem was a reconfigured version of the beam test engineering model ( btem ) that represents one of 16 towers of the lat and consists of acd , tkr and cal . the bfem differed from btem in that 6 silicon tracker layers were removed and the data acquisition system was reconfigured . the btem was tested with @xmath2 , @xmath3 and gamma beams at the stanford linear accelerator center ( slac ) , and the results were published ( @xcite ) . the bfem acd consisted of 13 segmented plastic anticoincidence scintillators to help identify the background events due to charged particles . among them , 4 tiles were placed above the tkr and 8 on the side . in addition , one big tile was placed above 4 top tiles to cover gaps between tiles . the tkr consisted of 13 x - y layers of si strip detectors ( ssds ) of @xmath12 thickness to measure particle tracks and 11 lead foils to convert gamma - rays . the area of ssds was @xmath13 for each layer , but the top 5 x - y pairs had smaller area . among 11 lead foils , the top 8 were 3.6% radiation length thick , the following three were 28% radiation length thick , and the last two x - y layers had no lead converter . the instrument trigger was provided by hits in three x - y pairs in - a - row ( six - fold coincidence ) . the cal consisted of 80 csi crystals hodoscopically laid out ( eight layers of 10 logs in x or y directions ) and measured energy deposit and shower profile . a set of four plastic scintillators called external gamma - ray target ( xgt ) was mounted above the instrument to tag cosmic - ray interactions in the scintillators and constrain the gamma - ray origin . the four detector components , readout electronics and support structures were mounted in a pressure vessel ( pv ) because not all of the electronics and data recorders were designed to be operated in vacuum . we developed a monte - carlo simulation program of the bfem with a geant4 toolkit . the geometry of the bfem implemented in the simulator is shown in figure 1 : support structures and the pressure vessel as well as detectors were implemented . standard electromagnetic processes , hadronic interactions and decay processes were simulated . cutoff length ( range threshold of secondary particles to be generated ) was set at @xmath14 in tkr and 1 mm for other regions . we utilized geant4 ver 5.1 with patch 01 . cosmic - rays in or near the earth atmosphere consist of the primary and the secondary components . the primary cosmic - rays are generated in and propagate through extraterrestrial space . the main component is known to be protons . when primary particles penetrate into the air and interact with molecules , they produce relatively low - energy particles , i.e. , secondary cosmic - rays . the first step in building a working model of cosmic - ray fluxes is to represent the existing measurements and theoretical predictions by analytic functions . below we describe the model functions particle by particle . data are very scarce or absent for large zenith angle and low energy , and expected uncertainties are also mentioned . primary particles that reach the top of the atmosphere are believed to come predominantly from outside the solar system . their spectra are known to be affected by the solar activity and earth geomagnetic field . the spectra of the primaries in interstellar space can be modeled by a power - law in rigidity ( or momentum ) as @xmath15 where @xmath16 and @xmath17 are the kinetic energy and rigidity of the particle , respectively . these particles are decelerated by solar wind as they enter the solar system and hence their flux shows anti correlation with the solar activity . to model this solar modulation , we adopted the formula given by @xcite : the modulation is expressed by an effective shift of energy as @xmath18 where @xmath19 is the magnitude of electron charge , @xmath20 the atomic number of particle , @xmath21 the particle mass and @xmath22 the speed of light . a parameter @xmath23 is introduced to represent the solar modulation which varies from @xmath24 for solar activity minimum to @xmath25 for solar activity maximum . since the glast balloon experiment took place in 2001 august , we fixed @xmath23 at 1100 mv , a typical value for solar activity maximum . the second and much bigger modulation is introduced as the low energy cutoff due to the earth geomagnetism . we modeled this effect based on the ams measurements @xcite in which the proton flux was measured over a wide geomagnetic latitude range . we defined the reduction factor as @xmath26 here , @xmath27 is the cutoff rigidity calculated by assuming the earth magnetic field to be of dipole shape , as @xmath28 where @xmath29 is the altitude , @xmath30 the earth radius and @xmath31 the geomagnetic latitude @xcite . for the glast balloon experiment they were @xmath32 km , @xmath33 rad , and @xmath34 . we also referred to @xmath2 measurements by ams @xcite and found that electrons are less affected by earth geomagnetism . although the geomagnetic cutoff is expected to be the same regardless of the particle type , we gave priority to the experimental data and used @xmath35 instead of equation ( 3 ) for @xmath2/@xmath3 . the reduction factor of equations ( 3 ) and ( 5 ) reproduces the cutoff seen in ams data in @xmath36 , but gives lower flux in high geomagnetic latitude region . the primary spectrum is then expressed as @xmath37 where @xmath38 and 6.0 for proton / alpha and @xmath2/@xmath3 , respectively . this model function can be used for the entire solar cycle and entire low - earth orbit by adjusting @xmath23 and @xmath27 . we assume the angular distribution is uniform for @xmath39 and to be 0 for @xmath40 , where @xmath41 is the zenith angle from vertical . the cutoff angle @xmath42 is introduced to represent the earth horizon and is @xmath43 at low - earth orbit ( @xmath44 ) . the primary flux at balloon altitude is attenuated by air and the effect is described in the following sections particle by particle . note that the east - west effect is not taken into account for simplicity : we aim to construct models which provide the correct flux integrated over the azimuth angle . recent measurements by bess @xcite and ams @xcite provided us with an accurate spectrum of primary protons . these data were taken in similar parts of the solar cycle ( june / july , 1998 ) , and we expect that the solar modulation parameters are similar between them . bess measured the proton flux at an altitude of 37 km near the geomagnetic north pole and evaluated the spectrum at the top of the atmosphere , which was not much affected by the geomagnetic cutoff . ams measured the flux at various positions on earth in energies below and above the geomagnetic cutoff at an altitude close to low - earth orbit ( 380 km ) . above the geomagnetic cutoff the two measurements agree up to about 50 gev and differ less than 10% between 50 and 100 gev @xcite . their data can be modeled well by equation ( 2 ) with @xmath45 , @xmath46 and @xmath47 . other experiments @xcite can also be represented by the model if the solar modulation parameter @xmath23 is adjusted to an appropriate solar cycle ( see figure 2 ) . the flux at a balloon altitude suffers a small attenuation due to interactions with air . the nuclear interaction length in air is @xmath48 , and the probability for vertically - downward protons to reach the altitude of glast bfem ( atmospheric depth was @xmath49 ) is 95.8% . for protons of oblique direction , we calculated the atmospheric depth in line - of - sight by assuming the trajectory to be straight . then the effective atmospheric depth scales as @xmath50 down to @xmath51 and is @xmath52 for the horizontal direction , by assuming the scale height of the air above the altitude of the payload ( at @xmath53 ) to be 7.6 km , a typical value in the u.s . standard atmosphere . below the horizon the flux is assumed to be 0 . our primary proton spectral model function ( equation 6 with 4% air attenuation ) with adjustment for the geomagnetic cutoff is shown in figure 3 with ams data taken at similar geomagnetic latitude region . in energy ranges below the geomagnetic cutoff , there are secondary particles produced in the earth s atmosphere by interactions of primaries . the secondary proton spectra at satellite altitude was measured by ams in a wide geomagnetic latitude region , showing strong dependence on the geomagnetic cutoff . we represent their data , region by region , by a broken power - law or cutoff power - law model . the formula of the broken power - law is , @xmath54 and that of the cutoff power - law model is @xmath55 below 100 mev , we found no data and assume a power - law spectrum with index of -1 : @xmath56 below a few @xmath57 10 mev , there is no probability for protons to trigger bfem / lat , hence the spectral model below 100 mev ( equation 9 ) does not affect the data very much . we tabulated the model parameters to describe the ams data in table 1 . in low geomagnetic latitude regions ( @xmath58 ) the downward and upward spectra are identical , indicating that protons are confined locally by the geomagnetic field . the particle direction is expected to be randomized , and we adopt a uniform angular distribution . for @xmath59 , we assume a uniform zenith angle dependence in the upper and lower hemispheres , for we lack experimental data about angular dependence . flux in the horizontal direction is uncertain , and we leave it discontinuous . analysis of the bfem data ( @xmath60 ) showed that horizontal particles ( @xmath61 ) have little chance to trigger the bfem . the 3-in - a - row trigger is improbable for the narrow geometry of this single tower . the secondary proton spectrum at balloon altitude is not necessarily the same as that in low - earth orbit , for the larmor radius of a typical secondary proton is much smaller than the satellite altitude : for example , the radius of a 1 gev proton is 100 km for average earth magnetic field of 0.3 gauss . hence we must refer to existing balloon measurements . the spectra measured at @xmath62 gv at atmospheric depth similar to our experiment ( @xmath63 ) are collected in figure 3 @xcite . although the flux differs from experiment to experiment , vertically - downward flux at @xmath64 is much higher than that measured by ams ( at 380 km ) : they could be represented as @xmath65 in 10 mev4 gev . above the geomagnetic cutoff , we assume that the secondary proton flux follows a power - law function of @xmath66 , i.e. , that of the primary . note that the downward flux is known to be proportional to the atmospheric depth @xcite , and balloon data in figure 3 were already scaled to @xmath49 . on the other hand , upward flux does not depend on the atmospheric depth very much ( e.g. , verma 1967 ) and we modeled the data with the following formulas . they are @xmath67 for energy above 100 mev and @xmath68 for energy below 100 mev . as can be seen from figure 3 , secondary fluxes measured by balloon experiments are mutually inconsistent within a factor of 2 . hence our model inherits this much uncertainty . the angular distribution of the secondary cosmic - ray flux has been measured in several balloon and rocket experiments . while substantial build - up of the flux near the earth horizon has been reported in several experiments @xcite , its functional form is not well determined . for small zenith angles , the secondary proton downward flux is expected to be proportional to the atmospheric depth in line - of - sight , and we multiply the flux by @xmath50 . the flux is expected to be saturated near the horizon and we use a constant factor of 2 for @xmath69 . for upward flux , we assume the same angular dependence as a function of nadir angle instead of zenith angle . again the flux disconnects at horizon due to uncertainty . we adopted the ams and bess data to model the alpha particle spectrum . the two experiments are consistent with each other to @xmath7010 % @xcite , giving @xmath71 , @xmath72 and @xmath45 for equation 2 . other experiments @xcite can also be represented well by adjusting the solar modulation parameter @xmath23 ( figure 4 ) . we did not take the secondary component into account , for the flux is more than three orders of magnitude lower than the primary , as shown by @xcite . for the glast balloon experiment , we used equation 3 with @xmath73 and @xmath74 . the zenith angle dependence and the attenuation by air are assumed to be the same as that of proton primaries . the fluxes of cosmic - ray electrons and positrons are modeled in 3 components : the primary , the secondary downward and the secondary upward fluxes . unlike proton and alpha primaries , no single experiment has measured cosmic primary electron and positron fluxes over a wide energy range . we referred to a compilation of measurements by webber @xcite and used the power - law fit given there , i.e. , @xmath75 a recent review of measurements by @xcite gave a consistent spectrum above 10 gev . the positively charged fraction , interpreted as @xmath76 , has been measured by several experiments . among them , @xcite obtained the fraction to be a constant , @xmath77 , between 5 gev and 50 gev . we adopted their results and modeled the interstellar electron spectrum to be equation 1 with @xmath78 and @xmath79 . the spectrum of the primary positrons takes the same form except for the normalization factor . like proton and alpha primaries , the angular distribution is assumed to be uniform and independent of zenith angle @xmath41 for @xmath80 and zero for @xmath81 , where @xmath82 corresponds to particle going toward the nadir . the solar and geomagnetic modulation parameters for the glast balloon experiment are @xmath83 and @xmath74 . to model the @xmath2/@xmath3 primary flux for our experiment , we also need to take into account the energy loss due to the ionization and bremsstrahlung in the atmosphere . for vertically downward particles it is calculated to be @xmath84 , where @xmath85 is the radiation length of air . this energy loss gives the same spectral shape but smaller normalization by a factor of 1.42 . we calculated the atmospheric depth in the line - of - sight as a function of zenith angle in the same way as that for proton and alpha primaries , and took the energy loss into account . secondary cosmic ray electrons and positrons are generated in two parts , a downward - moving component and an upward - moving one . to model them at a satellite altitude , we referred to the ams data and represented them with a power - law model : @xmath86 or a broken power - law model ( equation 7 ) , or a power - law model with a hump : @xmath87 below 100 mev we found no data and use the power - law model with a spectral index of -1 , i.e. , equation 9 . again there is no probability for @xmath88 to cause trigger below 10 mev . the model functions are tabulated in table 2 : in near geomagnetic equator , @xmath3 flux is much higher than that of @xmath2 . we use the same flux models for downward and upward component , since ams did not detect significant difference between them . as for proton secondaries , a uniform angular distribution is assumed in upper and lower hemispheres . most of the existing balloon measurements of secondary electron flux at around @xmath89 @xcite were rather old and inconsistent with each other , as shown in figure 5 . however , they consistently show much higher and flatter spectra than that of ams . we therefore referred to balloon measurements and adopted a higher and flatter spectral model for the glast balloon experiment . the vertically downward spectrum is @xmath90 for 100 mev4 gev and @xmath91 in energy 10100 mev . the flux above the geomagnetic cutoff is expected to follow the functional form of proton primaries ( @xmath92 ) . upward flux has been measured by @xcite in 1.02.4 gev and is similar to that of downward , and we used the same spectral model for both downward and upward for simplicity . we did not find data at large zenith angle , and assume the same angular distribution as that for proton secondaries . although ams @xcite observed an overabundance of positrons relative to electrons in the low geomagnetic latitude region , fluxes of these two particles are almost identical in the region of @xmath59 , as shown in table 2 . a recent balloon measurement by @xcite also showed that the positron fraction is nearly 50% at @xmath89 . we therefore used the same spectral shape and flux for positron secondaries . note that the data points of balloon experiments referenced , where only the flux of @xmath93 is given , have been divided by 2 to convert to the @xmath2 flux in figure 5 . as for particles at balloon altitudes , the gamma - ray flux consists of primary and secondary components . the primary component is of extraterrestrial origin and the intensity is attenuated by air at balloon altitude . secondaries are produced by the interaction between cosmic - ray particles and the atmospheric molecules , and the intensity of the downward component depends on the residual atmospheric depth . we modeled the extragalactic diffuse gamma - rays measured by several gamma - ray satellites , as compiled by @xcite . above 1 mev , the spectrum can be expressed by a power - law function of @xmath94 the zenith - angle dependence is the same as those of proton / alpha/@xmath10 primaries . for energies where glast bfem has sensitivity for gamma rays ( @xmath95 ) , the primary flux is lower than that of secondary downward gamma by more than an order of magnitude . we therefore do not take into account the gamma primary component in the following analysis of bfem data ( 5 ) . secondary gamma - rays are produced either by interactions of cosmic - ray hadrons or bremsstrahlung of @xmath10 with nuclei in the atmosphere . the spectrum has been measured by several satellite and balloon experiments @xcite . for residual atmosphere less than @xmath96 , the vertically - downward gamma - ray flux is known to be proportional to the atmospheric depth , whereas the upward flux remains almost constant @xcite . observations at palestine of the vertically downward spectrum at the atmospheric depth of @xmath97 are compiled in @xcite and @xcite , showing consistent results . we multiplied their data by @xmath98 to scale them to the residual atmosphere of @xmath63 , the value for the glast balloon experiment , and modeled the spectrum as follows : for 1 - 1000 mev , it is @xmath99 where @xmath100 . for 1100 gev , the spectrum is @xmath101 the composite spectrum is shown with reference data in figure 6 . note that we do not expect this component in satellite orbit . for the vertically - upward component , only a few data are available @xcite as collected in figure 7 . we found data taken at palestine , texas ( @xmath102 ) down to a few mev , whereas we found none below a few mev : so we substituted with data by @xcite at @xmath103 . the flux depends on cut - off rigidity , as discussed in @xcite : they measured the flux of vertically - upward gammas above 80 mev , for which glast lat / bfem has high sensitivity , in @xmath27 of 417.5 gev , and showed that the flux anticorrelates the cutoff rigidity as @xmath104 . @xcite confirmed the dependence by comparing the gamma - ray intensity above 35 mev at equator and at @xmath105 . we adopted their results and scaled the flux at @xmath103 by a factor of @xmath106 to take the dependence into account . the composite spectrum for vertically upward flux seems to be represented well by two power - law functions : for 120 mev , it is @xmath107 and for 201000 mev , @xmath108 above 1 gev we could not find data and adopted the same spectral index as that of downward gamma rays ( @xmath109 ) . then , the spectrum in 11000 gev is expressed as @xmath110 the spectral model with reference data points is given in figure 7 . as can be seen from the figure , observations are very scarce and give inconsistent results with each other . we regard that the uncertainty is a factor of 3 . angular dependence of the secondary gamma - ray flux has been measured by some authors @xcite and is known to depend strongly on zenith angle . we constructed the model function referring to @xcite , where they measured the flux in 1.510 mev region at @xmath111 . we multiplied their data between @xmath112 ( i.e. , downward component ) by 3.8/2.5 to correct the atmospheric depth , and expressed the relative flux as @xmath113 here , @xmath41 in model functions is given in radians . true angular dependence of the flux and spectral shape could be correlated with each other . for simplicity , we assumed that the functions above represents the zenith - angle dependence at 3 mev and the spectral shape is independent of zenith angle except for the difference between downward and upward . our model function and referenced data are given in figure 8 with a typical measurement error , indicating that the uncertainty is large , up to a factor of 2 . due to its short lifetime , we can neglect any primary muon component . the spectrum of atmospheric muons is reported by @xcite , where they measured the flux of downward @xmath11 in 0.1 - 10 gev band . at float altitude ( residual atmospheric depth is @xmath114 ) , the @xmath4 spectrum can be expressed as @xmath115 for 300 mev20 gev , and as @xmath116 for 10300 mev . we could not find data for the upward spectrum and assumed the same function for simplicity . data of @xcite were obtained at geomagnetic polar region , whereas glast balloon experiment was conducted at @xmath62 gv . they also measured muon spectra at @xmath62 gv in energy above 2 gev @xcite and obtained a consistent spectrum . a recent measurement by @xcite at @xmath105 also gave a consistent spectrum for energies above 0.5 gev . we therefore used the model functions above ( equation 25 and 26 ) for the glast balloon experiment . the ratio of @xmath5 to @xmath4 , defined as @xmath117 , is found to be 1.6 in the data of @xcite : so we modeled the @xmath5 spectrum with the same spectral shape but 1.6 times higher normalization . as shown by @xcite and @xcite , the downward @xmath11 flux is proportional to the atmospheric depth . we therefore assumed the same angular dependence as that of proton secondaries . our model functions and reference data are given in figure 9 . note that we do not expect muons in satellite orbit : even for 10 gev muons the life time is @xmath118 in the laboratory frame and they can travel only 50 km . after the integration and test at stanford linear accelerator center and goddard space flight center , the instrument was shipped to the national scientific balloon facility ( nsbf ) in palestine , texas . on august 4 , 2001 , the bfem was successfully launched . the balloon carried the instrument to an altitude of 38 km in about two hours , and achieved about three hours of level flight until the balloon reached the limit of telemetry from the nsbf . the atmospheric depth at the float altitude was about @xmath119 . due to a small leak of the pressure vessel , the internal pressure went down to 0.14 atmosphere and we failed to record all the data to hard disk . however , a sample of triggered data was continuously obtained via telemetry with about 12 hz out of a total trigger rate of 500 hz . more than @xmath120 events were recorded during the level flight and showed that all detectors and daq system kept working in a space - like high - counting environment . the data obtained allow us to study the cosmic - ray background environment . more details of the preparation and the flight of the bfem were described in @xcite . in the next section , we compare the simulation prediction with the observed data . after the flight we found two issues to be taken into account in data analysis . first , three layers out of 26 , i.e. , 7th layer , 15th layer and 16th layer , were found not to participate in the trigger , although hits in these layers were recorded normally when the trigger condition was satisfied . we took this effect into account in the geant4 simulation . fortunately , three or more consecutive pairs of x and y layer ( which is the minimum to cause a trigger ) worked in the lower part of the tkr ( 1st to 6th layer ) , in the middle part of the tkr ( 9th to 14th layer ) , and in the upper part of the tkr ( 17th to 26th layer ) , and the loss of trigger efficiency due to this failure was only @xmath70 30% . the second issue was related to the event sampling in telemetry . the sampling unintentionally had a bias toward events with smaller size ( primarily number of hits in the tkr ) in some data packets . to select unbiased events we required that the packet contain exactly 10 events ; a detailed study showed such packets to have an unbiased sample within 10% . this selection reduces the number of events , but we still have 36k events to be used for data analysis . in order to validate the cosmic - ray flux model and the monte - carlo simulator described in 2 and 3 , we compared the simulation prediction with the data . the observed trigger rate was about 500 hz during the level flight and is in good agreement with the simulation prediction . for further study , we divided events into two classes , `` charged events '' and `` neutral events . '' a charged event is one in which any one of 13 acd tiles shows energy deposition above 0.38 mev ( corresponds to 0.2 of the mean energy deposited in an acd scintillator by a minimum ionizing particle , a mip ) . the rates of charged and neutral events observed were 439 hz and 58 hz , respectively , and they are reproduced by the simulation within 5% . the contribution of each particle type to the trigger is summarized in table 3 , implying that about 70% of neutral events are from atmospheric gamma - rays . we also compared the count rate of each tkr layer and summarized the results in figure 10 . there , 26 tkr layers are numbered from 0 to 25 in which the layer with number 0 is the closest to cal . we see jumps in rate between layers 15 and 16 , and layers 17 and 18 in charged events . these jumps correspond to the difference of area covered by ssds described in 2 ( area of 3/5 are covered by si in layers 1825 , 4/5 are covered in layers 16 and 17 , and fully covered in layers 015 ) . in neutral events we can see three big rate changes above layers 5 , 7 and 9 , corresponding to the thick lead converter . the simulation reproduces not only the trigger rate but also these structures very well , validating our instrument simulator and background flux model . the differences between the observations and model are all less than 10% , with the largest difference seen in lower layers ( layers 05 ) , where the model overpredicts the rates . in this region with no converter layers , the uncertainties of the low - energy electron , positron and gamma - ray models may affect the results . the difference could also be because of the uncertainty of passive materials under the acd implemented in the simulation . we also applied the event reconstruction program that recognizes the pattern to find the tracks in tkr @xcite on both real data and simulation data . the distribution of the number of reconstructed tracks is given in figure 11 , where the number is two for single track events ( one in x - layer and the other in y - layer ) and is four for double track events . the distribution is well reproduced by simulation for both charged and neutral events . we also selected the single track events out of the charged events and compared the goodness - of - fit ( @xmath121 ) which evaluates the straightness of the track between data and simulation , as shown in figure 12 . the value of @xmath121 depends on a particle energy and we set it constant ( 30 mev ) to obtain a consistent value regardless of whether the particle hits the cal or not . the two peaks in the figure correspond to tracks with a curved trajectory ( @xmath122 ) and those with a straight one ( @xmath123 ) . the former consists of electrons , positrons , and gamma rays , whereas the latter is composed of protons , alphas and high energy muons . the small difference of the lower peak position between data and simulation could be due to the misalignment of ssds . except for this , the data are well reproduced by simulation , providing another evidence that the model flux of each particle type is correct . finally we examined the zenith angle distribution of cosmic - ray particles . we first selected single track events with a straight track by requiring @xmath121 to be less than 0.1 , and compared the zenith angle distribution between data and simulation prediction , as shown in figure 13 . in the reconstruction , particles are assumed to go downward , therefore @xmath124 ranges from 1.0 ( vertically downward ) to 0.0 ( horizontal ) . our flux model reproduces the data , not only for the vertical direction but also for large zenith angle down to @xmath125 , within 10% . we also applied event selection criteria and studied the gamma - ray zenith - angle distribution . the cuts we used here are not optimized for background reduction , for the number of events is limited . instead , we implemented a set of loose cuts to filter out charged particles and save a sufficient number of gammas . to select downward gammas , we required events with no hit in the acd , double tracks in the tkr , and gaps between two tracks increasing as they go downward ( downward - opening v - shape trajectory ) . for upward gammas , we did not require that the event be neutral , since the generated electron / positron pair is expected to hit the acd . instead , we required that the two bottommost tkr layers did not detect a hit and the tkr showed double tracks with upward - opening v - shape trajectory . we also set cal energy threshold at 6 mev , half of the energy deposition by mip crossing one csi log , to eliminate events with interaction in cal . the results are shown in figure 14 in which the bisector of two tracks are assumed to be the direction of gamma . data are well reproduced by simulation down to @xmath126 for downward and up to @xmath127 for upward , validating the spectral and angular distribution models for gammas . the background flux model used to compare with the bfem data involves many parameters , some of which are poorly defined . in the course of this investigation , we tried a number of alternate assumptions in cases where no well - defined observations were available . the extent to which such alternatives affected the agreement with the data provides an indication of how sensitive the model is to such parameters . below we list some of these possibilities and describe their influence on the comparison . @xmath128 fluxes of secondary proton , electron , and positron were at one time derived directly from the ams measurements . models based on balloon measurements increased the trigger rate by about 15% and improved the agreement of charged events . @xmath128 the model function of the secondary upward - moving gamma - ray flux were at one time simpler : in 1 mev to 1 gev , it was expressed by a single power - law function of index of -1.7 , with 1.6 times higher normalization at 1 mev . the new model increased the neutral event rate by about 15% and improved the match between bfem data and simulation prediction . @xmath128 we once expressed the angular dependence of the secondary charged particles by @xmath129 , an empirical model given in old rocket measurements @xcite . the new model did not alter the trigger rate nor the angular distribution shown by figure 13 very much . we adopted our model to take into account the fact that proton and muon secondary fluxes are proportional to the atmospheric depth , although the bfem data alone could not determine which model is better . @xmath128 as described in 3 , fluxes of secondary @xmath88 and upward gamma are poorly known , although these particles contributes to the trigger significantly . detailed study showed that the main component of secondary @xmath88 which causes the trigger is of energy of about 100 mev and that of upward gamma is of energy of 10100 mev , where the particle fluxes are uncertain by a factor of 23 . however , if we change the @xmath88 flux by 50% , the trigger rate of charged event is changed by @xmath130 and disagrees with data . similar disagreement will arise in neutral event rate by adjusting the flux of upward gamma . with a substantial number of parameters in this model , we can not be certain that all of them are simultaneously optimized . what we have found is that specific variations in model parameters tend to affect only one aspect of the comparison between the model and the bfem observations . the fact that we have obtained reasonable agreement with this range of observables - total rate , rate per layer , charged component , neutral component , straight tracks , curved tracks , and angular distribution - suggests that the model is a good approximation to reality . we have constructed the glast bfem instrument simulator with a geant4 toolkit ( 2 ) and the background flux models of abundant components ( proton , alpha , electron , positron , gamma ray and @xmath131 ) based on previous measurements and theoretical predictions ( 3 ) . the simulation successfully reproduces the observed data within 10% , indicating that the flux model is a good representation of the cosmic - ray background at balloon altitude at palestine , texas . models for the entire low - earth orbit are also given in 3 , where the effect of cutoff rigidity and solar modulation potential are parameterized . they are being used for the simulation of glast lat in orbit and the study of the background rejection . the authors would like to thank assistance given by m. sjgren , k. hirano and h. mizushima in various phases of this work . they are also grateful to a. tylka , r. battiston , y. komori , a. yamamoto and t. yoshida for helpful discussions about the cosmic - ray background flux . they would also like to thank e. do couto e silva , h. tajima , t. kotani , r. c. hartman , l. rochester , t. burnett , r. dubois , j. wallace and glast bfem team and glast lat collaboration for their support and discussions . they also would like to thank the staff of national scientific balloon facility for their excellent support of launch , flight and recovery . abe , k. , et al . 2003 , physics letter b , 564 , 8 agostinelli , s. , et al . , 2003 , nuclear instruments and methods a , 506 , 250 alcaraz , j. , et al . 2000a , physics letter b , 472 , 215 alcaraz , j. , et al . 2000b , physics letter b , 484 , 10 alcaraz , j. , et al . 2000c , physics letter b , 490 , 27 alcaraz , j. , et al . 2000d , physics letter b , 494 , 193 barwick , s. w. , et al . 1998 , journal of geophysical research , 103 , 4817 boezio , m. , et al . 1999a , apj , 518 , 457 boezio , m. , et al . 1999b , physical review letters , 82 , 4757 boezio , m. , et al . 2000 , physical review d , 62 , 032007 boezio , m. , et al . 2003 , physical review d , 67 , 072003 burnett , t. h. , et al . 2002 , ieee transactions on nuclear science , 49 , 1904 couto e silva , e. do , et al . 2001 , nuclear instruments and methods a , 474 , 19 daniel , r. r. , & stephens , s. a. 1974 , rev . space phys . 12 , 233 duvernois m. a. , et al . , 2001 , proceeding of 27th international cosmic ray conference , 3 , 61 gehrels , n. , & michelson , p. 1999 , astroparticle physics , 11 , 277 gleeson , l. j. , & axford , w. i. 1968 , apj , 154 , 1011 golden , r. l. , et al . 1994 , apj , 436 , 769 imhof , w. l. , nakano , g. h. , & reagan , j. b. 1976 , journal of geophysical research , 81 , 2835 israel , m. h. 1969 , journal of geophysical research , 74 , 4701 kuryan , y. a. , mazets , y. p. , proskura , m. p. , & sokolov , i. a. 1979 , geomagnetism and aeronomy , 19 , 6 longair , m. s. 1992 , high energy astrophysics , vol . 1 ( 2nd edition ; cambridge : cambridge university press ) menn , w. , et al . , 2000 , apj 533 , 281 moskalenko , i. v. , & strong , a. w. 1998 , apj , 493 , 694 pennypacker , c. r. , smoot , g. f. , buffington , a. , muller , r. a. , & smith , l. t. 1973 , journal of geophysical research , 78 , 1515 ryan , j. m. , jennings , m. c. , radwin , m. d. , zych , a. d. , & white , r. s. 1979 , journal of geophysical research , 84 , 5279 sanuki , t. , et al . , 2000 , apj , 545 , 1135 seo , e. s. , ormes , j. f. , streitmatter , r. e. , stochaj , s. j. , jones , w. v. , stephens , s. a. , & bowen , t. 1991 , apj , 378 , 763 shnfelder , v. , graser , u. , & daugherty , j. 1977 , apj , 217 , 306 shnfelder , v. , graml , f. , & penningsfeld , f. -p . 1980 , apj , 240 , 350 singer , s. f. 1950 , physical review letters , 77 , 729 sreekumar , p. , et al . , 1998 , apj , 494 , 523 thompson , d. j. , et al . , 2002 , ieee transactions on nuclear science , 49 , 1898 van allen , j. a. , & gangnes , a. v. 1950a , physical review , 78 , 50 van allen , j. a. , & gangnes , a. v. 1950b , physical review , 79 , 51 verma , s. d. 1967 , journal of geophysical research , 72 , 915 webber , w. r. 1983 , in composition and origin of cosmic rays , ed . m. m. shapiro , ( dordrecht : d. reidel publishing co. ) , 83 zombeck m. v. 1990 , handbook of space astronomy and astrophysics ( 2nd edition ; cambridge : cambridge university press ) cccc region & direction & model & parameters + @xmath132 & down / upward & cutoff pl & 0.136/0.123/0.155/0.51 + @xmath133 & down / upward & broken pl & 0.1/0.87/600/2.53 + @xmath134 & down / upward & broken pl & 0.1/1.09/600/2.40 + @xmath135 & down / upward & broken pl & 0.1/1.19/600/2.54 + @xmath136 & down / upward & broken pl & 0.1/1.18/400/2.31 + @xmath137 & downward & broken pl & 0.13/1.1/300/2.25 + & upward & & 0.13/1.1/300/2.95 + @xmath138 & downward & broken pl & 0.2/1.5/400/1.85 + & upward & & 0.2/1.5/400/4.16 + @xmath139 & downward & cutoff pl & 0.23/0.017/1.83/0.177 + & upward & broken pl & 0.23/1.53/400/4.68 + @xmath140 & downward & cutoff pl & 0.44/0.037/1.98/0.21 + & upward & broken pl & 0.44/2.25/400/3.09 + cccc region & @xmath141 ratio & model & parameters for @xmath2 spectrum + @xmath142 & 3.33 & broken pl & 0.3/2.2/3/4.0 + @xmath143 & 1.66 & pl & 0.3/2.7 + @xmath144 & 1.0 & pl+hump & 0.3/3.3/@xmath145/1.5/2.3 + @xmath139 & 1.0 & pl+hump & 0.3/3.5/@xmath146/2.0/1.6 + @xmath140 & 1.0 & pl & 0.3/2.5 + lllllllll particle type & proton & alpha & @xmath2/@xmath3 & @xmath147 2ndary & @xmath147 2ndary & @xmath11 & total & data + & & & & downward & upward & & & + charged events & 216 & 19 & 110 & 12 & 42 & 39 & 438 & 439 + neutral events & 4.8 & 0.1 & 12.3 & 16.8 & 26.8 & 0.5 & 61.3 & 58.2 +

quantum process tomography ( qpt ) @xcite is a way to completely characterize a quantum process . it is the main tool for experimental characterization of quantum gates , being developed for potential use in a quantum computer . qpt has been realized in numerous experiments , practically in all types of qubit systems , including , e.g. , nmr @xcite , linear optics @xcite , ion traps @xcite , and superconducting qubits @xcite . in this paper we mainly focus on qpt with superconducting qubits , even though our discussion is applicable to other systems as well . unfortunately , qpt requires resources , which scale exponentially with the number of qubits . for @xmath1 ( superconducting ) qubits the number of initial states is usually @xmath2 ( or sometimes @xmath3 ) , and the number of `` measurement directions '' ( state tomography `` rotations '' ) for each initial state is typically @xmath4 ( or @xmath3 ) . each such setup typically requires a few hundred or a few thousand experimental runs . from this scaling it is easy to estimate that the qpt of a 1-qubit or a 2-qubit quantum gate requires a manageable number of experimental runs , while a 3-qubit qpt is rather difficult to realize , and the full qpt with 4 and more qubits seems to be impractical . the problem of exponential scaling of qpt resources with the number of qubits can be mitigated if we do not need full information about a quantum gate operation , but instead need only some information . thus a partial characterization of a multi - qubit operation is an important area of theoretical research @xcite . this includes randomized benchmarking @xcite , which typically provides only one number : the gate fidelity . randomized benchmarking becomes increasingly preferable for superconducting qubits @xcite . another promising way to solve or a least alleviate the problem of exponential scaling is to use a compressed - sensing implementation of qpt @xcite . one more problem with qpt is its sensitivity to state preparation and measurement errors ( spam in the terminology of ref.@xcite ) . randomized benchmarking does not suffer from spam - errors , so this is another reason why this technique is increasingly popular . however , the obvious drawback of randomized benchmarking is that it gives only the total ( average ) error and does not give any information about particular kinds of error . hence , it does not tell us about the origin of a quantum gate imperfection . in this paper we consider standard qpt , which gives full information about the quantum process . because of the scaling problem , we are essentially talking about quantum gates with less than 4 qubits , for which full qpt is a very useful tool . the standard way of representing qpt results is via the process matrix @xmath0 @xcite in the pauli basis ( see the next section ) . unfortunately , this matrix is a rather inconvenient object to work with . even though in principle it contains full information about the process , it does not show useful information in a straightforward way . thus the important problem of converting experimental qpt data into a useful characterization of particular decoherence processes @xcite is not quite simple . besides the standard matrix @xmath0 , there are other ways to represent qpt results . for example , they can be represented via the so - called pauli transfer matrix @xmath5 @xcite . the advantage of using @xmath5 is that it contains only real elements , from @xmath5 it is simple to see whether the quantum operation is trace preserving , also simple to see whether the process is unital , and for any clifford operation there is exactly one non - zero element in each row and column of @xmath5 with unit magnitude . in this paper we discuss one more way of representing the experimental qpt results @xcite . it is essentially the standard @xmath0-matrix representation in the pauli basis , with the only difference being that we factor out the desired unitary operation @xmath6 , so that the error matrix @xmath7 describes only the imperfections of the experimental quantum gate . there are two natural ways to define such an error matrix ( @xmath7 and @xmath8 ) : we can assume that the @xmath6-operation is either before of after the error process ( see fig . 1 below ) . even though in theoretical analyses it is rather usual to separate the error channel and unitary operation , we are not aware of any detailed discussion of the qpt representation by the error matrices . as discussed in this paper , the error matrix @xmath7 ( as well as @xmath8 ) has a number of convenient properties . in particular , its main element is equal to the process fidelity @xmath9 , while other non - zero elements correspond to imperfections . unitary imperfections are directly given by the imaginary parts of the elements along the left column and/or top row . decoherence produces other elements , which have a relatively simple relation to kraus operators characterizing decoherence ( which are the operators in the lindblad equation ) . since the elements of the error matrix are small , most calculations can be approximated to first order , thus making them relatively simple . this includes a relatively simple rule for the composition of quantum operations and accumulation of the error - matrix elements due to the lindblad - form decoherence . the error - matrix representation has already been used in the experimental qpt @xcite . our paper is organized in the following way . in sec.[sec - standard ] we briefly review some properties of the standard process matrix @xmath0 . in sec . [ sec - chi - err ] we introduce the error matrices @xmath7 and @xmath8 , and then in sec . [ sec - properties ] some of their properties are discussed , including physical interpretation . in sec.[sec - comp ] we consider composition of the error processes . then in sec . [ sec - corr ] we discuss the use of the error matrix to find the necessary unitary correction to an experimental quantum gate . in sec . [ sec - lindblad - error ] we consider the contribution to the error matrix from decoherence described by the lindblad - form master equation . a possible identification of spam errors and their subtraction from the error matrix are discussed in sec.[sec - spam ] . finally , sec . [ sec - concl ] is the conclusion . two appendices discuss topics that are somewhat different from the main text . in appendix a we consider several simple examples of the @xmath0-matrix calculation , and in appendix b we discuss an intuitive interpretation of decoherence described by the lindblad - form master equation , unraveling the quantum dynamics into the `` jump '' and `` no jump '' scenarios . a linear quantum operation @xmath10 ( transforming initial density matrix @xmath11 into the final state @xmath12 ) is usually described via the process matrix @xmath0 ( which is hermitian , with non - negative eigenvalues ) , defined as @xcite _ fin= _ m , n _ mn e_m _ ine_n^ , [ chi - def]where the matrices @xmath13 form a basis in the space of complex @xmath14 matrices , which are the linear operators in the @xmath15-dimensional hilbert space of the problem . for example , for @xmath1 qubits @xmath16 ; therefore there are @xmath17 matrices @xmath13 , and the matrix @xmath0 has dimensions @xmath18 . ( note that @xmath13 are operators in the space of wavefunctions , and these operators have the same dimension as density matrices . ) even though in principle any basis @xmath19 ( not necessarily orthogonal @xcite ) can be used in eq . ( [ chi - def ] ) , the most usual choice for a system of qubits is the use of pauli matrices . in this case for one qubit the basis @xmath19 consists of 4 matrices : i , x_x , y_y , z_z , and for several qubits the kronecker product of these matrices is used ; for example for two qubits @xmath20 . note that sometimes a different definition for @xmath21 is used : @xmath22 . also note that the basis of pauli matrices is orthogonal ( under the hilbert - schmidt inner product ) but not normalized @xcite , so that e_m|e_n ( e_m^e_n ) = _ mnd , d=2^n ; [ emen ] this is why normalization by @xmath15 in often needed in the qpt . in this paper we will always assume the pauli basis @xmath19 and use @xmath23 . therefore @xmath24 , but for clarity we will still write @xmath25 in formulas when appropriate . it is simple to find the matrix @xmath0 for a multi - qubit unitary operation @xmath6 . we first need to find its represention in the pauli basis , @xmath26 , and then compare @xmath27 with eq.([chi - def ] ) , which gives _ mn= u_m u_n^ * , u_n = tr(ue_n^ ) , [ chi - u]where the star notation means complex conjugation and the expression for @xmath28 follows from the orthogonality property ( [ emen ] ) . calculation of the matrix @xmath0 for a quantum process with decoherence is usually significantly more cumbersome . some examples are considered in appendix a ( see also , e.g. , @xcite ) . a process is called `` trace - preserving '' if the transformation ( [ chi - def ] ) preserves the trace of the density matrix , i.e. if @xmath29 when @xmath30 . in this case the matrix @xmath0 should satisfy the condition @xmath31 ( which gives @xmath2 real equations ) , and therefore @xmath0 is characterized by @xmath32 real parameters instead of @xmath33 parameters in the general ( non - trace - preserving ) case . in this paper we always assume a trace - preserving operation . the fidelity of a trace - preserving quantum operation compared with a desired unitary operation is usually defined as f_=tr ( ^des ) , [ fidelity]where @xmath34 is for the desired unitary operation and @xmath0 is for the actual operation ( it is easy to show that @xmath35 ) . this definition has direct relation @xcite 1-f_=(1-f_av ) to the average state fidelity @xmath36 , which assumes uniform averaging over all pure initial states . sometimes @xcite @xmath37 is called `` gate fidelity '' while @xmath9 is called `` process fidelity '' . characterization by @xmath38 is usually used in randomized benchmarking @xcite ; it is easy to see that @xmath39 . note that the fidelity definition ( [ fidelity ] ) requires unitary desired operation and trace - preserving actual operation . for a general ( non - unitary ) desired operation ( still assuming trace - preserving operations ) it should be replaced with the `` uhlmann fidelity '' f_= ( tr)^2=(tr)^2 , which is essentially the same definition as for the fidelity between two density matrices @xcite . the process matrix @xmath0 for a non - trivial unitary operation typically has many non - zero elements ( e.g. , 16 elements for the two - qubit controlled-@xmath41 , controlled - not , and @xmath42 operations ) , and it is nice - looking on the standard bar ( `` cityscape '' ) chart used for visualization . however , a large number of non - zero elements ( we will call them `` peaks '' ) creates a problem in visual comparison between the desired and experimental @xmath0-matrices , especially for high - fidelity experiments . a natural way to make it easier to compare between the actual and desired operations is to show the difference between them , i.e. , to show the error . for example , it is possible to calculate and display the difference @xmath43 in the pauli basis . however , such element - by - element difference does not make much sense mathematically . instead , let us represent the actual quantum process as a _ composition _ of the desired unitary @xmath6 and the error process ( fig . 1 ) , and find the process matrix @xmath7 for this error operation . this essentially reduces the comparison between @xmath0 and desired @xmath6 ( we use a loose language here ) to the comparison between @xmath7 and the memory ( identity ) operation . and @xmath8 via their relation to @xmath0 and the desired unitary operation @xmath6 . the time runs from left to right.,width=321 ] so , the general idea is to convert the desired unitary @xmath6 into the memory operation , and this converts @xmath0 into the error process matrix @xmath7 . there are two ways of defining such error matrix : we can place the error process either before or after the desired unitary @xmath6 ( fig . 1 ) . thus , we define two error matrices : @xmath7 and @xmath8 using the relations [ see eq.([chi - def ] ) ] @xmath44 from fig . 1 it is obvious that the process @xmath7 can be represented as the composition of the inverse ideal unitary ( @xmath45 ) and the actual process @xmath0 after that . similarly , @xmath8 is the composition of @xmath0 and inverse unitary @xmath46 after that . hence , @xmath7 and @xmath8 represent legitimate quantum processes and therefore they satisfy all the properties of the usual matrix @xmath0 ( see the previous section ) ; in particular , @xmath7 and @xmath8 are positive hermitian matrices . we will use the standard pauli basis for the error matrices @xmath7 and @xmath8 . using the composition relation , the error matrices can be calculated from @xmath0 as @xcite @xmath47 in an experiment the error process matrices can be calculated either from @xmath0 using these equations or by directly applying the definitions ( [ chi - err - def ] ) and ( [ tilde - chi - err - def ] ) to the experimental data . for example , to find @xmath8 the measured final density matrices can be transformed numerically as @xmath48 and then the usual procedure of @xmath0 calculation can be applied . note that the matrix @xmath7 can be thought of as the @xmath0-matrix in the rotated basis @xmath49 instead of @xmath19 ; similarly , @xmath8 is formally the @xmath0-matrix in the basis @xmath50 ; however , we will not use this language to avoid possible confusion . in the ideal ( desired ) case both error processes @xmath7 and @xmath8 are equal to the perfect memory ( identity ) operation , @xmath51 , for which ^*i*_mn=_m0_n0 , [ memory]where with index 0 we denote the left column and/or the top row , which correspond to the unity basis element ( in the usual notation @xmath52 for one qubit , @xmath53 for two qubits , etc . ) ; the process matrix ( [ memory ] ) is given by eq . ( [ chi - u ] ) with @xmath54 . so , in the ideal case the error matrices have only one non - zero element at the top left corner : @xmath55 . therefore , _ any other non - zero element in @xmath7 ( or @xmath8 ) indicates an imperfection of the quantum operation_. this is the main advantage in working with @xmath7 or @xmath56 instead of the usual matrix @xmath0 . there are also some other advantages discussed below . the standard process fidelity ( [ fidelity ] ) for a trace - preserving operation has a very simple form for the error matrices . since @xmath0 , @xmath7 , and @xmath8 are essentially the same operator in different bases , we have @xmath57 . therefore f_= _ 00^err= _ 00^err , [ f - chi - error]i.e . the process fidelity is just the height of the main ( top left ) element of the error matrix @xmath7 or @xmath56 . a systematic unitary error can be easily detected ( to first order ) in the error matrix @xmath7 or @xmath8 because it appears at a special location : it produces non - zero imaginary elements along the top row and left column of the matrix , i.e. the elements @xmath58 with @xmath59 ( similarly for @xmath8 ) . to see this fact , let us assume that instead of the desired unitary operation @xmath6 , the quantum gate actually realizes a slightly imperfect unitary @xmath60 . then @xmath7 corresponds to the unitary @xmath61 , which can be expanded in the pauli basis as @xmath62 . since @xmath63 , we have @xmath64 and @xmath65 . note that @xmath66 can always be chosen real because @xmath67 is defined up to an overall phase . now let us show that to first order all @xmath68 are purely imaginary . this follows from the first - order expansion of the equation @xmath69 , which gives @xmath70 . hence , @xmath71 for @xmath59 ( purely imaginary @xmath72 ) , and the difference @xmath73 is only of second order ( for a real @xmath66 ) . another way of showing that in the first order @xmath68 are imaginary is by using representation @xmath74 ( neglecting the overall phase ) with a hermitian matrix @xmath75 , so that the expansion @xmath76 contains all real coefficients , while to first order @xmath77 and @xmath78 . now using eq . ( [ chi - u ] ) we see that to first order the only non - zero elements ( except @xmath79 ) are ( ^err_n0)=-im(^ err_0n ) -i u^err_n , n0 . [ imag - unitary]note that the diagonal elements @xmath80 ( as well as the change of @xmath79 ) in this case are of second order . in particular , in this approximation @xmath81 . as discussed later , if decoherence causes a significant decrease of the fidelity @xmath9 , then a better approximation of the systematic unitary error effect is eq . ( [ imag - unitary ] ) multiplied by the fidelity , ( ^err_n0)-i f_u^err_n . [ imag - unitary-2 ] using this equation it is possible to find @xmath82 from an experimental matrix @xmath7 and therefore estimate the systematic unitary error @xmath67 in the experiment . the same property ( [ imag - unitary ] ) [ and its version ( [ imag - unitary-2 ] ) corrected for @xmath9 ] can be shown for @xmath8 using the similar derivation for the unitary imperfection @xmath84 . the elements @xmath85 are different compared with the matrix elements of @xmath7 ; they are related via the equation @xmath86 or equivalent equation @xmath87 . decoherence produces additional small peaks in the error matrix @xmath7 ( and/or @xmath8 ) . as discussed later , to first order these peaks are linear in the decoherence strength and simply additive for different decoherence mechanisms . therefore , to first order we have a weighted sum of different patterns in @xmath7 for different mechanisms . if these patterns for the most common decoherence mechanisms are relatively simple , then there is a rather straightforward way of extracting information on decoherence from experimental @xmath7 matrix . in sec . [ sec - lindblad - error ] we will discuss the general way to calculate the first - order pattern for a particular decoherence mechanism ; for a practical quantum gate this pattern may contain many elements . a special role is played by the real elements along the top row and left column of @xmath7 ( or @xmath8 ) : they correspond to the gradual non - unitary ( `` bayesian '' ) evolution in the absence of the `` jumps '' due to decoherence ( in contrast to the imaginary elements , which correspond to the unitary imperfection ) see discussion later . note that the diagonal matrix elements of @xmath7 are the error probabilities in the so - called pauli twirling approximation @xcite . therefore these elements can be used in simulation codes , which use the pauli twirling approximation for the analysis of quantum algorithms in multi - qubit systems . concluding this section , we emphasize that the error matrix @xmath7 ( and/or @xmath8 ) is just a minor modification of the standard @xmath0 matrix ; they are related by a linear transformation and therefore equivalent to each other . however , in the error matrix only one peak ( at the top left corner ) corresponds to the desired operation , while other peaks indicate imperfections . this makes the error matrix more convenient to work with , when we analyze deviations of an experimental quantum operation from a desired unitary and try to extract information about the main decoherence mechanisms . in this paper we always assume high - fidelity operations , 1-f_1 , [ high - fid]so that the first - order approximation of imperfections is quite accurate . since the error matrix @xmath7 is positive , its off - diagonal elements have the upper bound |^err_mn| . ( the same is true for @xmath8 , but for brevity we discuss here only @xmath7 . ) therefore for a high - fidelity operation ( [ high - fid ] ) only the elements in the left column and top row can be relatively large , @xmath88 , while other off - diagonal element have a smaller upper bound , @xmath89 , because all diagonal elements except @xmath90 are small . actually , it is possible to show ( see below ) that @xmath91 are also small , @xmath92 , so only @xmath93 can be relatively large . the elements @xmath93 and @xmath94 play a special role because a unitary imperfection produce them in the first order , while other elements are of second order [ see eq . ( [ chi - u ] ) and discussion in the previous section ] . it is easy to see that @xmath95 and @xmath94 can approach @xmath96 if the error is dominated by the unitary imperfection . the physical intuition in analyzing the error of a quantum gate is that the ( small ) infidelity @xmath97 comes from small unitary imperfections and from rare but `` strong '' decoherence processes , which cause `` jumps '' that significantly change the state . as discussed later , this picture should also be complemented by small non - unitary state change in the case when no jump occurred ( this change is essentially the quantum bayesian update @xcite due to the information that there was no jump ) . it is not easy to formalize this intuition mathematically ; however , there is a closely related procedure . let us diagonalize the error matrix @xmath7 , so that ^err= tdt^-1 , [ diag - chi - err]where @xmath98 is a unitary @xmath99 matrix and @xmath100 is the diagonal matrix containing @xmath101 real eigenvalues of @xmath7 . these eigenvalues are non - negative , and we can always choose @xmath98 so that in @xmath102 they are ordered as @xmath103 . the sum of the eigenvalues is equal 1 ( since we consider trace - preserving operations ) and the largest eigenvalue @xmath104 is close to 1 because @xmath105 . [ here @xmath106 because any diagonal element of a hermitian matrix should be in between the largest and smallest eigenvalues , as follows from expanding the corresponding basis vector in the eigenbasis . ] all other eigenvalues are small because @xmath107 . the diagonalization ( [ diag - chi - err ] ) directly gives the evolution representation via the kraus operators @xcite , @xmath108 in which the operators @xmath109 form an orthogonal basis , @xmath110 [ see eq . ( [ emen ] ) ] because @xmath98 is unitary . recall that @xmath6 is the desired unitary gate see eq.([chi - err - def ] ) . the main term in the kraus - operator representation ( [ kraus - representation ] ) is the term with @xmath111 ; let us show that @xmath112 ( up to an overall phase , which can always be eliminated ) ; this means @xmath113 and @xmath114 for @xmath59 [ note that @xmath115 because @xmath98 is unitary ; therefore it is sufficient to show that @xmath113 ] . this can be done in the following way . since ^err_mn=_k _ k a_m^(k)(a_n^(k))^ * , [ chi - via - a]the fidelity is f_=_k_k |a_0^(k)|^2 . [ f - sum - k ] the contribution of the terms with @xmath117 to the fidelity is at least smaller than @xmath97 because this is the bound for @xmath118 and @xmath119 for a unitary @xmath98 ( this contribution is actually even much smaller , as discussed below ) . therefore @xmath120 , and so @xmath121 . choosing real @xmath122 , we obtain @xmath113 , and therefore @xmath123 . since @xmath124 is close to @xmath125 and other @xmath109 are orthogonal to @xmath124 , the components @xmath126 are small for @xmath117 . using the relation @xmath127 ( since @xmath98 is unitary ) , we find the bound @xmath128 . now using @xmath129 , @xmath130 ( see the previous paragraph ) , we obtain @xmath131 , neglecting the terms of the order @xmath132 . this means that the contribution to the fidelity ( [ f - sum - k ] ) from the kraus operators with @xmath133 is limited at least by @xmath134 [ neglecting the order @xmath135 , and therefore a good approximation for the fidelity is f__0 |a_0^(0)|^2 , 1-f_(1-|a_0^(0)|^2 ) + ( 1-_0 ) , [ f - separation]which corresponds to the intuitive separation of the infidelity @xmath97 into the `` coherent '' error @xmath136 and the error @xmath137 due to rare but strong decoherence `` jumps '' . thus the kraus - operator representation ( [ kraus - representation ] ) can be crudely interpreted in the following way . after the desired unitary @xmath6 , we apply the kraus operator @xmath112 with the probability @xmath111 , while with small remaining probabilities @xmath138 we apply very different ( orthogonal to @xmath124 ) kraus operators @xmath109 . imperfection of @xmath124 ( compared with @xmath139 ) leads to the `` coherent '' error @xmath136 with @xmath140 . other kraus operators are practically orthogonal to @xmath141 , so they correspond to `` strong decoherence '' and practically always lead to an error , which happens with the total probability @xmath142 , thus explaining eq . ( [ f - separation ] ) . while this interpretation seems to be quite intuitive , there are two caveats . first , it is incorrect to say that @xmath109 is applied with the probability @xmath138 . instead , we should say @xcite that the evolution scenario |_in , _ in occurs with the probability @xmath143 , which depends on the initial state and is equal to @xmath138 only on average , @xmath144 , after averaging over pure initial states . [ this can be proven by using @xmath145 and @xmath146 . ] note that the operators @xmath109 can violate the inequality @xmath147 , even though @xmath148 is always satisfied . the second caveat is that @xmath124 is not necessarily unitary , as would be naively expected for the separation into a coherent operation and decoherence . the non - unitary part of @xmath124 can be interpreted as due to the absence of jumps @xmath149 , similar to the null - result evolution @xcite ( see below and also discussions in sec . [ sec - lindblad - error ] and appendix b ) . we may include both contributions ( imperfect unitary part and non - unitary part of @xmath124 ) into what we call the `` coherent error '' , so it is characterized by the difference between @xmath124 and @xmath139 ( probably it is better to call it the `` gradual error '' ) . with understanding of these two caveats , the discussed above interpretation of the kraus representation ( [ kraus - representation ] ) can be useful for gaining some intuition . note that the contribution to @xmath7 [ see eq.([chi - via - a ] ) ] from the imperfection of the main kraus operator @xmath124 mainly causes the elements @xmath150 in the top row and left column of the error matrix ( other elements are of the second order in the imperfection ) . in contrast , other ( `` decoherence jump '' ) operators @xmath109 mainly produce other elements of @xmath151 ; their contribution to @xmath152 is limited by @xmath153 , as follows from the positivity of @xmath154 , which gives @xmath155 , and the derived above inequality @xmath156 . significant contributions to the diagonal elements of @xmath151 may come from both @xmath124 and @xmath149 . therefore , we can apply the following approximate procedure to crudely separate the error matrix , ^err=^coh + ^dec , ^coh_mn= _ 0 a_m^(0)(a_n^(0))^ * , [ chi - coh - dec]into the `` coherent '' ( or `` gradual '' ) part @xmath157 and `` strong decoherence '' @xmath158 [ see eq . ( [ chi - via - a ] ) ] . we first estimate the `` coherent probability '' @xmath104 as _ 0 f_/ ( 1-_n>0 estimation @xmath159 . using @xmath104 and @xmath160 we can construct @xmath161 , while the remaining part of @xmath7 is @xmath158 . the diagonal elements @xmath80 with @xmath59 ( error probabilities ) are thus separated into the contributions @xmath162 due to `` strong decoherence '' . a further simplification of this procedure is to approximate the coherent part as @xmath163 for @xmath164 , @xmath165 , and @xmath166 . one more useful approach for the intuitive understanding of @xmath7 elements along the left column and top row is the following . using the completeness relation ( [ kraus - representation ] ) we can write the main ( `` coherent '' ) kraus operator in the polar decomposition representation as @xmath167 where @xmath168 is some unitary , which corresponds to the unitary imperfection [ since the overall phase is arbitrary , we can choose @xmath169 . then let us expand the operators in the pauli basis , @xmath170 where @xmath82 and @xmath171 are the expansion components , and we introduced the naturally defined unitary error @xmath172 and decoherence error @xmath173 ( average probability of `` jumps '' ) . now using eq . ( [ f - separation ] ) and neglecting the second - order products @xmath174 , we find the intuitively expected formula for fidelity , f_1 - e_u -e_d . similarly , taking into account only the contribution from the main kraus operator , for the elements @xmath175 with @xmath59 we find ^err_n0 ( 1-_u- e_d ) u^err_n + ( 1-e_u-_d ) . [ chi - n0-approx]here @xmath176 are purely imaginary ( in the first order ) because @xmath67 is unitary , while @xmath171 are real because eq.([expansion - dec ] ) is the expansion of a hermitian operator . therefore , we see that the imaginary parts of the elements @xmath177 are due to unitary imperfection , while their real parts come from the absence of `` jumps '' ( described by kraus operators with @xmath133 ) via eq . ( [ a0-pd ] ) . it is easy to see that the evolution @xmath178 is essentially the bayesian update of the quantum state @xcite due the absence of jumps . in experiments with superconducting qubits the quantum gate infidelity is usually dominated by decoherence , @xmath179 , unless the unitary part is very inaccurate . we will often assume this situation implicitly . in this case case @xmath180 , and from eq . ( [ chi - n0-approx ] ) we obtain eq . ( [ imag - unitary-2 ] ) . using eqs . ( [ expansion - dec ] ) and ( [ chi - n0-approx ] ) we can show the bound @xmath181 . the starting point is to see that all components of @xmath182 in the pauli basis are not larger than 1 , i.e. @xmath183 for any @xmath184 . this is because @xmath185 and the product of two pauli operators is a pauli operator ( with a phase factor ) ; therefore a particular ( say , @xmath184th ) component is essentially a sum of pairwise products of the `` vector coordinates '' @xmath186 and the same coordinates @xmath187 in a different order ( also , with phase factors ) . therefore , the sum of products is limited by the norms of the two vectors , which is 1 for both vectors ( @xmath188 are limited by 1 , the sum of @xmath189th contributions to @xmath171 in eq . ( [ expansion - dec ] ) is limited by @xmath190 . this gives @xmath191 via eq . ( [ chi - n0-approx ] ) . since the elements @xmath192 are small , so that their second - order contributions to the diagonal elements are practically negligible , @xmath193 ^ 2\leq ( 1-f_\chi)^2/4 $ ] , it does not matter much whether we include the evolution due to the `` no - jump '' bayesian update into the `` coherent part '' ( [ chi - coh - dec ] ) or not . note that this bayesian evolution does not produce an additional error since @xmath194/d=1-\frac{1}{2}\sum_{k>0}\lambda_k=1-(1-\lambda_0)/2 $ ] and therefore from eq . ( [ a0-pd ] ) we see that @xmath195 , i.e. the `` no - jump '' scenario brings only the unitary error . in this section we discussed only the error matrix @xmath7 ; however , the same analysis can be also applied to @xmath8 . the relation between the error matrices is @xcite ^err = w_(u)^errw_(u)^ , ^err = w_(u)^^err w_(u ) , [ w - transf]where the unitary matrix @xmath196 corresponds to the effective change of the basis @xmath197 due to @xmath6 , w_(u),mn= tr ( e_m^u e_n u^)/d . [ w - def]note that @xmath198 . it is convenient to think that the @xmath199-transformation ( [ w - transf ] ) is due to the error matrix `` jumping over '' the unitary @xmath6 ( see fig . 1 ) . it is easy to see from eq . ( [ w - def ] ) that @xmath200 ; therefore for an ideal memory @xmath201 and eq . ( [ w - transf ] ) essentially transforms only the difference from the ideal memory . the @xmath202-transformation @xmath203 ( [ w - transf ] ) obviously corresponds to the unitary transformation of kraus operators @xmath109 , `` jumping over '' @xmath6 to the left ( fig . 1 ) , _ k= u^a_k u . these @xmath204 form the kraus - operator representation of @xmath8 with the same `` probabilities '' @xmath138 . let us calculate the error matrix @xmath7 for the composition of two quantum operations : desired unitary @xmath205 with error process @xmath206 and after that the desired unitary @xmath207 with error @xmath208 ( fig . 2 ) . it is obvious that the resulting desired unitary is @xmath209 ( note that the matrix multiplication is from right to left , while on the quantum circuit diagrams the time runs from left to right ) . we assume sufficiently high fidelity of both operations , @xmath210 , @xmath211 [ for brevity we omit the subscript @xmath0 in the notation ( [ f - chi - error ] ) for fidelity ] . over the unitary @xmath207.,width=321 ] let us start with the simple case when there are no unitaries , @xmath212 . then the relation between the initial state @xmath11 and the final state @xmath12 is _ fin = _ m , n , p , q ^err_1 , mn ^err_2 , pq e_p e_m _ in e_n^e_q^ , which leads to the usual lengthy expression for the composition of two operations : ^err_ab= _ m , n , p , q ^err_1 , mn ^err_2 , pq ( e_p e_m e_a^ ) tr ( e_q e_n e_b^)^ * , [ comp - gen]where we used relation e_p e_m= _ a ( e_p e_m e_a^ ) e_a ( actually there is only one non - zero term in this relation , but there is no simple way to write it ) . equation ( [ comp - gen ] ) is valid for any two operations ( not necessarily error processes ) and is very inconvenient to use . fortunately , it is greatly simplified for the error processes , since @xmath213 and @xmath214 contain only one large ( close to one ) element , @xmath215 and @xmath216 , while all other elements are small . therefore we can use the first order approximation of eq . ( [ comp - gen ] ) , which gives the simple additive relation @xmath217 for all elements except the main element @xmath79 . the approximation ( [ comp - err-2 ] ) can also be written as ^err _ 1^err+ _ 2^err-^*i*. [ comp - err - matrix ] obviously , it is much easier to deal with the composition of error processes than with the composition of general quantum processes . note that eq . ( [ comp - err ] ) can also be naturally understood using the kraus - operator representation ( [ kraus - representation ] ) in the case when infidelity is dominated by decoherence . the approximation ( [ comp - err ] ) neglects the second - order corrections due to possibly significant coherent errors . while this is good enough for the off - diagonal elements of @xmath7 , let us use a better approximation for the more important diagonal elements ( except @xmath90 ) , explicitly taking into account the top rows and left columns of @xmath218 and @xmath208 in eq . ( [ comp - gen ] ) , _ nn^err f_2 _ 1,nn^err+ f_1 _ 2,nn^err + 2 im(^err_1 , 0n ) im(^err_2 , 0n ) , [ comp - diag ] where the formally similar term @xmath219 is neglected , because the elements @xmath220 can not be relatively large , in contrast to @xmath221 . for the fidelity the exact and approximate results are @xmath222 where the similar term @xmath223 is neglected again . note that @xmath224 calculated using approximations ( [ comp - diag ] ) and ( [ comp - err - fid2 ] ) is slightly smaller than 1 , but the difference is of second order in infidelity . the reason for taking a special care of the elements @xmath225 and @xmath226 becomes clear if we consider the composition of small unitaries @xmath227 and @xmath228 . then the approximate addition ( [ comp - err-2 ] ) is valid for the first - order off - diagonal elements of @xmath7 ( which are in the left column and top row ) ; however , the diagonal elements are of second order , and for them the errors add up `` coherently '' , generating the `` interference '' term @xmath229 in eq . ( [ comp - diag ] ) . note that the neglected terms @xmath230 have a somewhat similar origin , considering the `` coherent '' composition of the main ( `` no - jump '' ) kraus operators in the representation ( [ kraus - representation ] ) . we emphasize that eqs . ( [ comp - err])([comp - err - fid2 ] ) do not change if we exchange the sequence of @xmath218 and @xmath208 . so in this approximation small imperfections of quantum `` memory '' operations commute with each other , as intuitively expected . note that for the fidelity ( [ comp - err - fid1 ] ) this result is exact : commutation of arbitrary error processes does not change @xmath9 . so far we assumed @xmath212 . now let us consider arbitrary desired unitaries @xmath205 and @xmath207 . then @xmath7 for the composition can be calculated in two steps ( see fig . 2 ) : we first exchange the sequence of @xmath218 and @xmath207 , thus producing the effective error process @xmath231 , and then use the discussed above rule for the composition of two `` memory '' operations @xmath231 and @xmath232 . the transformation of @xmath218 when it `` jumps over '' @xmath207 ( fig . 2 ) is essentially the same as the transformation between @xmath8 and @xmath7 [ see fig . 1 ] and is given by the equation ^err_1(u_2 ) = w_(u_2 ) ^err_1w_(u_2)^ , [ chi - err - jump]where @xmath233 is given by eq . ( [ w - def ] ) . this transformation can also be written as ^err_1(u_2 ) = ^*i*+ w_(u_2 ) ( ^err_1-^*i * ) w^_(u_2 ) , so that only the small difference from the ideal memory operation @xmath234 is being transformed . note that this transformation does not change fidelity , @xmath235 . thus we have a relatively simple procedure to find @xmath7 for the composition of two quantum operations : we first apply the transformation ( [ chi - err - jump ] ) to move the two error processes together ( fig . 2 ) and then apply approximate rules ( [ comp - err])([comp - err - fid2 ] ) to the matrices @xmath236 and @xmath208 . the similar procedure can be used to calculate @xmath8 for the composition of two quantum operations . we should first move @xmath237 to the left by jumping it over @xmath205 , ^err_2(u_1)= w^_(u_1 ) ^ err_2w_(u_1 ) , [ chi-2-u1]and then use the approximate composition rules ( [ comp - err])([comp - err - fid2 ] ) for the error matrices @xmath238 and @xmath239 . for the composition of several quantum operations @xmath240 , the error processes should be first moved to the end of the sequence by jumping them over the desired unitaries ( or moved to the beginning if we consider the language of @xmath8 ) and then we use the composition rules ( [ comp - err])([comp - err - fid2 ] ) . the procedure further simplifies if we can neglect `` coherent '' errors and use the simple additive rule ( [ comp - err-2 ] ) for all elements ( except @xmath241 ) . in experiments it is often useful to check how large the inaccuracy is of the unitary part of a realized quantum gate and find the necessary unitary corrections to improve fidelity of the gate . the error matrix @xmath7 ( or @xmath8 ) gives us a simple way to do this , because small unitary imperfections directly show up as the imaginary parts of the elements @xmath152 and @xmath177 . let us assume that we apply a small unitary correction @xmath242 after an operation characterized by the desired @xmath6 and error matrix @xmath7 with fidelity @xmath9 . then the process matrix @xmath243 for the correction operation mainly consists of the element @xmath244 and the imaginary first - order elements @xmath245 [ see eq . ( [ imag - unitary ] ) ] , while all other elements are of second order . using eq . ( [ comp - diag ] ) we see that after the correction the elements in the left column of the error matrix approximately change as _ n0^err _ n0^err + f_u_n^corr , [ corr - change]so to correct the unitary imperfection we need to choose u_n^corr - i i m ( _ n0^err)/f _ , [ corr - unitary]which cancels the imaginary part of the left - column elements ( here the factor @xmath246 needs an implicit assumption that the infidelity is dominated by decoherence ) . the increase of the gate fidelity @xmath247 due to this correction procedure can be estimated using eq . ( [ comp - err - fid2 ] ) , f__n0 ( im^ err_n0)^2/f_. [ delta - f]in this derivation the factor of 2 in the second term of eq.([comp - err - fid2 ] ) is compensated by the fidelity decrease due to the first term . the result ( [ delta - f ] ) in the case @xmath248 coincides with what we would expect from the unitary correction in absence of decoherence . the factor @xmath246 in eq . ( [ delta - f ] ) implicitly assumes that the infidelity is dominated by decoherence . note that the fidelity increase @xmath247 is of second order , so in an experiment we should not expect a significant improvement of fidelity due to unitary correction , unless the unitary imperfection is quite big . also note that in an experiment it may be easy to apply a unitary correction only in some `` directions '' , for example , by applying single - qubit pulses , while other corrections may be very difficult . in this case only some of the elements @xmath94 can be compensated , and then the fidelity improvement is given by eq . ( [ delta - f ] ) , in which summation is only over the elements , compensated by the correction procedure . the above analysis of the compensation procedure assumes a small compensation . if the unitary error is large , then to find the optimal correction @xmath249 we can use an iterative procedure , in which we first estimate the correction via eq.([corr - unitary ] ) , then use the exact composition relation ( [ comp - gen ] ) , and then again adjust the correction via eq.([corr - unitary ] ) . our analysis of the unitary compensation procedure assumed application of @xmath249 after the quantum gate . if the compensation is applied before the quantum gate , then it is more natural to use the language of @xmath8 ( see fig . 1 ) ; in this case eqs . ( [ corr - change])([delta - f ] ) remain valid , with @xmath7 replaced by @xmath8 . applying corrections both before and after the gate may in some cases increase the number of correctable `` directions '' in the space of unitary operators . to illustrate analysis of the unitary corrections , let us consider the two - qubit controlled - z ( cz ) gate in the `` quantum von neumann architecture '' @xcite , in which single - qubit @xmath41-rotations are realizable very easily ( without an additional cost ) , and so such corrections can be easily applied . application of @xmath41-rotation over the small angle @xmath250 to the first qubit and @xmath41-rotation over the small angle @xmath251 to the second qubit produces the correction unitary @xmath252 in the basis @xmath253 . here @xmath254 , but if we can also introduce correction @xmath255 of the cz angle , then it will be @xmath256 . expansion of @xmath249 in the pauli basis gives four non - zero elements : @xmath257 where we used the standard notation for the two - qubit operator basis @xmath258 ; note that the index 0 in the notations of our paper corresponds to @xmath259 . it is important to emphasize that this expansion of @xmath249 is not exactly what we used in eqs.([corr - change ] ) and ( [ corr - unitary ] ) because we assumed real @xmath260 , while @xmath261 in eq.([u - corr - ii ] ) is not real . we therefore need to adjust the overall phase of @xmath249 to make @xmath261 real . this can be easily done by replacing @xmath249 with @xmath262 . however , for small angles @xmath263 this would produce only a small change in eqs.([u - corr - iz])([u - corr - zz ] ) . the left column of @xmath264 therefore contains the same non - zero elements , @xmath265 if we correct only @xmath250 and @xmath251 ( so that @xmath266 ) , then we should choose them [ see eq.([corr - unitary ] ) ] as _ 1 2 i m ( ^err_iz , ii)/f _ , _ 2 2 i m ( ^err_zi , ii)/f _ , [ correct - phi12]where @xmath7 and @xmath267 are measured experimentally . this will cancel the left - column elements @xmath268 and @xmath269 in the corrected quantum gate and produce fidelity improvement @xmath270/f_\chi $ ] . if we can also correct @xmath255 , then we should choose corrections @xmath271 this will cancel the left - column elements @xmath272 , @xmath273 , and @xmath274 in the corrected quantum gate and produce fidelity improvement @xmath275/f_\chi $ ] . note that @xmath249 commutes with the cz gate , @xmath276 , so it does not matter if the single - qubit corrections are applied before or after the gate ( the @xmath255 correction is obviously a correction of the gate itself ) . similarly , it does not matter if we use @xmath7 or @xmath8 in this correction procedure . let us consider a quantum evolution described by the lindblad - form master equation = - [ h , ] + _ j _ j ( b_j b_j^- b^_jb_j -b^_jb_j ) , [ lindblad - evol]where @xmath277 is the hamiltonian ( which has a significant time dependence for a multi - stage quantum gate ) and @xmath278th decoherence mechanism is described by the kraus operator @xmath279 and the rate @xmath280 . mathematically it is natural to work with the combination @xmath281 ; however , we prefer to keep @xmath282 and @xmath279 separate because they both have clear physical meanings ( see appendix b ) . the imperfection of a quantum gate comes from imperfect control of the hamiltonian @xmath277 and from decoherence . if both imperfections are small , we can consider them separately . so , in this section we assume perfect @xmath277 and analyze the process error matrix due to decoherence only . moreover , we will consider only one decoherence mechanism , since summation over them is simple and can be done later . therefore we will drop the index @xmath278 in eq.([lindblad - evol ] ) and characterize the decoherence process by @xmath283 and @xmath284 . to find the error matrix @xmath7 ( or @xmath8 ) of such operation , we can divide the total gate duration @xmath285 into small timesteps @xmath286 , for each of them representing the evolution as the desired unitary @xmath287 $ ] and the error process @xmath288 . then using the same idea as in section [ sec - comp ] , we can jump the error processes over the unitaries , moving them to the very end ( for @xmath7 ) or to the very beginning of the gate ( for @xmath8 ) . finally , we can add up the error processes for all @xmath286 using approximations ( [ comp - err])([comp - err - fid2 ] ) . for small @xmath286 the error matrix @xmath289 can be found by expanding @xmath283 and @xmath290 in the pauli basis , @xmath291 , @xmath292 , and then comparing the decoherence terms in eq.([lindblad - evol ] ) with eq . ( [ chi - def ] ) , @xmath293 it is easy to see that @xmath294 since @xmath24 for the pauli basis , so the left - column and top - row contributions due to the @xmath295-terms in eq . ( [ b - mn ] ) are real . note that we can always use the transformation @xmath296 in the lindblad equation , which compensates ( zeroes ) the component @xmath297 but changes the hamiltonian , @xmath298 with @xmath299 ( there is no change , @xmath300 , if @xmath283 is hermitian ) . therefore we can use @xmath301 in eq . ( [ b - mn ] ) , and then the left - column and top - row elements come only from the @xmath295-terms , which correspond to the the terms @xmath302 of the lindblad equation and so correspond to the `` no - jump '' evolution ( see appendix b ) . thus we have a clear physical picture of where the components of @xmath303 come from : the imaginary parts of the left - column and top - row elements come from the unitary imperfection ( which may also be related to the decoherence - induced change of the hamiltonian ) , the real parts of the left - column and top - row elements come from the `` no - jump '' evolution ( see appendix b ) , and other elements come from the decoherence `` jumps '' , which `` strongly '' change the state ( recall that @xmath283 have only components orthogonal to @xmath139 ) . a similar interpretation has been used in sec . [ sec - properties ] . note that @xmath304 for small @xmath286 because there is practically no unitary evolution . now let us use the language of @xmath8 , which relates the error process to the beginning of the gate ( @xmath305 ) . we can find @xmath8 by moving the error processes @xmath306 to the start of the gate using the transformation relation ( [ chi-2-u1 ] ) and then summing up the error contributions using the approximate additive rule ( [ comp - err - matrix ] ) . in this way we obtain @xmath307 , \\ & & u(t)=\exp [ \frac{-i}{\hbar } \int_0^t h(t)\ , dt ] , \label{u(t ) } \end{aligned}\ ] ] where @xmath308 is given by eq . ( [ b - mn ] ) , @xmath309 is the unitary evolution occurring within the interval @xmath310 , and eq.([u(t ) ] ) assumes the time - ordering of operators . similar procedure can be used to find @xmath7 ; then we should move the errors to the end of the gate , by jumping them over the remaining unitary @xmath311 , @xmath312 . \qquad \label{lindblad - w}\end{aligned}\ ] ] note that the @xmath313-transformation of @xmath308 in eq.([lindblad - chi - err ] ) , @xmath314 , is equivalent to the @xmath315-transformation of the kraus operator @xmath283 , which relates it to the end of the gate , b(t)=u_rem(t ) b u_rem^ ( t ) . [ b(t ) ] therefore instead of eq . ( [ lindblad - chi - err ] ) we can use ^err^*i*+(t ) dt , [ lindblad - chi - err-2]in which @xmath316 is given by eqs . ( [ b - mn])([c - n ] ) using @xmath317 instead of @xmath283 . similarly , instead of eq . ( [ lindblad - chi - err - tilde ] ) we can use ^err^*i*+(t ) dt , in which @xmath318 is given by eqs.([b - mn])([c - n ] ) with ( t)u^(t ) b u(t ) [ tilde - b ] instead of @xmath283 . note that if @xmath283 is orthogonal to @xmath139 ( @xmath301 , see discussion above ) , then @xmath317 and @xmath319 are also orthogonal to @xmath139 , so they still describe `` strong '' error jumps . since the element @xmath320 does not change in the transformation @xmath321 , in the approximation ( [ lindblad - chi - err ] ) the fidelity is f _ 1-_0^t_g _ n0 |b_n|^2 dt , [ lindblad - f]so that if @xmath322 does not depend on time , then @xmath323 decays linearly with the gate time @xmath285 . ( as discussed above , the element @xmath297 is equivalent to a unitary imperfection , and therefore brings infidelity , which scales quadratically with time . ) an interesting observation is that in this approximation the fidelity does not depend on the desired unitary evolution @xmath309 . therefore , for example , for _ any two - qubit gate _ ( which does not involve higher physical levels in the qubits ) the contribution to the infidelity due to the energy relaxation and ( markovian non - correlated ) pure dephasing of the qubits is 1-f_= + + + , [ lindbad - t1-t2]where @xmath324 is the energy relaxation time , @xmath325 is the pure dephasing time , and the qubits are labeled by superscripts @xmath326 and @xmath327 ( see appendix a ) . the independence of the fidelity ( [ lindblad - f ] ) on the unitary evolution can be understood in the following way . the process fidelity @xmath9 is related to the state fidelity , uniformly averaged over all pure initial states . a unitary evolution does not change the uniform distribution of pure states ; therefore , the average rate of rare `` decoherence jumps '' does not depend on the unitary part . the approximation ( [ lindblad - chi - err - tilde])([lindblad - f ] ) neglects the second - order corrections . in the case of significant coherent errors ( which is not typical when only lindblad - equation decoherence is discussed ) , the natural second - order correction is to add @xmath328 to the elements @xmath329 with @xmath164 and @xmath59 ( we assume @xmath330 ) . since this correction is small , it may be important only for the diagonal elements @xmath80 , because it affects the resulting fidelity @xmath331 . this will introduce correction into eq.([lindblad - f ] ) , which scales quadratically with the gate time @xmath285 . the similar second - order correction @xmath332 can be introduced to the elements of @xmath333 . note that in eqs . ( [ lindblad - chi - err ] ) and ( [ lindblad - chi - err-2 ] ) the elements of @xmath7 are linear in the decoherence rate @xmath322 . therefore the `` pattern '' of @xmath7 elements is determined by the decoherence mechanism characterized by the operator @xmath308 and its transformation @xmath334 , and this pattern is multiplied by the decoherence rate @xmath322 . ( the experimental @xmath151 may need subtraction of the discussed above second - order correction to become linear in @xmath322 . ) if there are several decoherence mechanisms in eq.([lindblad - evol ] ) , then in the first order their contributions to @xmath7 simply add up . therefore , if the patterns for the different decoherence mechanisms @xmath279 are sufficiently simple and distinguishable from each other , then the decoherence rates @xmath282 can be found directly from the experimentally measured @xmath7 ( again , subtraction of the second - order correction may be useful in the case of significant coherent errors ) . a very important difficulty in experimental implementation of the qpt is due to spam errors : imperfect preparation of the initial states and state tomography errors , which include imperfect tomographic single - qubit rotations and imperfect measurement of qubits . in this section we discuss a way , which may help solving this problem . first , let us assume that the imperfect state preparation can be represented as an error channel , which acts on the ideal initial state . if we use @xmath2 initial states of @xmath1 qubits , then the transformation between @xmath2 ideal and real density matrices of the initial states can always be described by a linear transformation , characterized by @xmath33 parameters . so by the number of parameters it seems that the representation of the preparation error by an error channel is always possible . the problem , however , is that this transformation may happen to be non - positive . also , if more than @xmath2 initial states are used in an experiment , then an error - channel representation may be impossible by the number of parameters . nevertheless , we will use this representation , arguing that it can somehow be introduced phenomenologically . similarly , we assume that the imperfections of the tomographic single - qubit rotations and measurement can also be represented as an error channel . and @xmath335 . , width=302 ] using these two assumptions , we describe the preparation errors by the error matrix @xmath335 ( which is close to the ideal memory @xmath234 ) , and the tomography / measurement errors are described by @xmath336 ( also close to @xmath234 ) see fig . thus the experimentally measured error matrix @xmath337 is due to @xmath335 , @xmath7 ( which we need to find ) and @xmath336 . the general idea is to measure @xmath335 and @xmath336 by doing the process tomography without the gate ( doing the tomography immediately on the initial states ) and then subtract this spam error from @xmath338 to obtain @xmath7 . so , the procedure which first comes to mind is to use @xmath339 , where @xmath340 is the experimentally measured @xmath0 without the gate . however , in general this would be wrong . the reason is that without the gate we measure the simple sum of the spam - error components [ see eq.([comp - err - matrix ] ) ] , ^err , i^prep+ ^meas -^*i * , but in the presence of the gate @xmath6 the preparation error @xmath341 changes because it is `` jumped over '' @xmath6 [ see sec.[sec - comp ] and eq . ( [ chi - err - jump ] ) ] , so that ^err , exp ^err + w_(u)(^prep -^*i*)w_(u)^+ ( ^meas -^*i * ) , [ spam - exp]where @xmath196 is given by eq . ( [ w - def ] ) . therefore to find the actual error matrix @xmath7 from the experimental @xmath338 , it is insufficient to know @xmath342 . we can still find @xmath7 from eq.([spam - exp ] ) , but we need to know @xmath335 and @xmath336 separately . the idea how to find both @xmath335 and @xmath336 is to do a calibration qpt with a set of gates with very high fidelity , for which the error is negligible , and then use eq . ( [ spam - exp ] ) to separate the changing contribution from @xmath335 and non - changing contribution from @xmath343 . for example , in experiments with superconducting qubits the one - qubit @xmath344 and @xmath345 rotations about @xmath346 and @xmath21 axes usually have much better fidelity than two - qubit or multi - qubit gates . therefore , for the qpt of a two - qubit or multi - qubit gate we can rely on the spam - error identification using these one - qubit gates . let us start with considering a single qubit and discussing the change @xmath347 of the preparation error contribution , when we apply a high - fidelity @xmath346 gate , @xmath348 . by using eq . ( [ spam - exp ] ) or by simply comparing the terms in the equation @xmath349 , it is easy to find that this transformation flips the signs of the off - diagonal elements @xmath350 , @xmath351 , @xmath352 , and @xmath353 of @xmath341 ( the same for the symmetric , complex - conjugate elements ) , while the off - diagonal elements @xmath354 and @xmath355 ( and symmetric elements ) do not change . the diagonal elements do not change , as expected for a pauli twirling . similarly , if we apply the high - fidelity @xmath21 gate , then the off - diagonal elements @xmath354 , @xmath351 , @xmath352 , and @xmath355 flip the sign , while the elements @xmath350 and @xmath353 do not change . recall that the contribution from @xmath336 does not change . therefore , by comparing @xmath338 for the gates @xmath346 , @xmath21 , and @xmath356 ( no gate ) , we can find separately all the off - diagonal elements of @xmath335 and @xmath336 . to find diagonal elements of @xmath335 and @xmath336 , we can use high - fidelity gates @xmath357 ( @xmath345 rotation over @xmath346 axis ) and @xmath358 ( @xmath345 rotation over @xmath21 axis ) . the gate @xmath357 exchanges diagonal elements @xmath359 and @xmath360 of @xmath335 ( it also exchanges and flips signs of some off - diagonal elements , but for simplicity we focus on the diagonal elements only ) . similarly , the operation @xmath358 exchanges the elements @xmath361 and @xmath360 of @xmath335 . since the elements of @xmath336 do not change , we can find the diagonal elements of @xmath335 and @xmath336 separately . actually , this can not be done in the unique way , because the contributions proportional to @xmath139 in @xmath362 and in @xmath363 are indistinguishable from each other ( these are the depolarization - channel contributions ) . however , this non - uniqueness is not important because any choice gives the same spam - error contribution to @xmath338 . in this way , by doing qpt of the gates @xmath346 , @xmath21 , @xmath357 , @xmath358 and @xmath356 for a single qubit , we can find @xmath335 and @xmath336 ( assuming that these gate are nearly perfect in comparison with preparation and measurement errors ) . for two or more qubits we can do the similar procedure , applying the combinations of these 5 single - qubit gates , and thus identifying all the elements of the multi - qubit matrices @xmath335 and @xmath336 . in fact , the system of equations for this identification is overdetermined , so we can either use an ad - hoc way of calculating the elements or use the numerically efficient least - square method ( via the pseudo - inverse ) . note that the described procedure has an obvious relation to the pauli and clifford twirling , but for us it is sufficient to use only a small subset of operations , and we do not average the result . the described procedure of the spam - error identification [ which is then subtracted from @xmath338 using eq.([spam - exp ] ) ] is surely very cumbersome . however , this is at least some way to deal with the spam problem , which does not seem to have a simple solution . moreover , there are several ways to make experimental procedure less cumbersome , which are discussed next . the situation is greatly simplified if the spam - error is dominated by only one component : either @xmath335 or @xmath336 . if @xmath335 is negligible , then from eq . ( [ spam - exp ] ) we see that the error matrix of the analyzed multi - qubit gate @xmath6 can be estimated as ^err^err , exp-(^err , i-^*i * ) , so besides the standard qpt of the @xmath6 gate we only need the qpt of no operation ( @xmath356 gate ) . in the opposite limit when @xmath343 is negligible , it is easier to use the language of @xmath364 , because we can neglect the change of @xmath336 when it is jumped over @xmath6 to the left . in this case ^err^err , exp-(^err , i-^*i * ) ; recall that @xmath365 . when both @xmath335 and @xmath336 are significant in the spam - error , it is still possible to simplify the described above procedure by using the idea of compressed - sensing qpt @xcite . in contrast to the usual application of the compressed - sensing idea to the qpt of the gate @xmath6 , we can apply it to find @xmath335 and @xmath343 by using a small random subset of @xmath366 combinations of single - qubit gates in the procedure . it is also possible to combine the random choice of the gates with the random choice of initial states and measurement directions , thus further reducing the amount of experimental work . it is important to emphasize that we do not need to know the matrices @xmath335 and @xmath336 very precisely , so their compressed - sensing estimate should be sufficient . one more idea , which may be practically useful , is to measure @xmath367 and select only few significant peaks in it . for each of these peaks we identify which contribution to it comes from @xmath368 and from @xmath336 by applying just one or a few single - qubit rotations , which change this particular peak . it is beneficial to choose the rotations , which affect more than one significant peak of @xmath367 . in this way a relatively small number of qpts is sufficient to find the significant peaks of @xmath335 and @xmath336 . then by using eq . ( [ spam - exp ] ) we estimate the spam contribution for the multi - qubit gate @xmath6 and subtract it from the experimental error matrix @xmath338 to find `` actual '' @xmath7 . note that we do not need a complicated procedure to find the `` actual '' process fidelity @xmath9 of the gate @xmath6 . if the spam errors can be represented by the error channels ( fig . 3 ) and if there are no significant coherent spam errors , then f_f_^exp / f_^i , [ spam - fid]where @xmath369 and @xmath370 are the experimentally measured fidelities for the gate @xmath6 and no gate , respectively [ see eq . ( [ comp - err - fid2 ] ) ] . in this paper we have discussed representation of quantum operations via the error matrices . instead of characterizing an operation by the standard process matrix @xmath0 , we separate the desired unitary operation @xmath6 and the error process , which is placed either after or before @xmath6 ( fig . 1 ) . this defines two error matrices : @xmath7 and @xmath8 [ eqs.([chi - err - def ] ) and ( [ tilde - chi - err - def ] ) ] . we use the standard pauli basis @xmath197 for all process matrices . the error matrices @xmath7 and @xmath8 are related to @xmath0 via unitary transformations [ eqs . ( [ chi - err - chi ] ) and ( [ chi - tilde - err - chi ] ) ] , as well as to each other [ eqs.([w - transf ] ) and ( [ w - def ] ) ] . therefore the error matrices are equivalent to @xmath0 ; however , they are more convenient to use than @xmath0 . the error matrices have only one large element , which is located at the top left corner and is equal to the process fidelity @xmath9 . any other non - zero element corresponds to an imperfection of the quantum gate . therefore , the bar chart of @xmath7 ( or @xmath8 ) is a visually convenient way of representing the imperfections of an experimental quantum gate . the elements of @xmath7 ( or @xmath8 ) have more intuitive physical meaning related to the operation imperfection , than the elements of @xmath0 ( even though the meaning of most of the elements is still not as intuitive as we would wish ) . it is important that since the error - matrix elements are small for a high - fidelity gate , the first - order approximation is typically sufficient . the imaginary parts of the elements in the left column and top row correspond to the unitary imperfection , @xmath371 where the correction factor @xmath372 can be taken seriously only if most of infidelity comes from decoherence . the real parts of the elements in the left column and top row correspond to the small non - unitary change of the quantum state in the case when no `` jumps '' due to decoherence occur ( this change is due to the bayesian update , see appendix b ) . it is natural to combine this non - unitary change with the unitary imperfection into a `` coherent '' ( or `` gradual '' ) state change , which happens in absence of `` jumps '' . finally , other elements of @xmath329 , with @xmath164 and @xmath59 , correspond to the strong `` jumps '' of the quantum state due to decoherence . these jumps are characterized by kraus operators practically orthogonal to @xmath139 and therefore always bring an error ( see discussions in secs.[sec - properties ] and [ sec - lindblad - error ] ) . the diagonal elements @xmath373 with @xmath59 ( probabilities of @xmath13-type errors in the pauli twirling approximation @xcite ) have contributions from both the `` coherent '' imperfection and decoherence `` jump '' processes ; however , the `` coherent '' contribution to @xmath373 is of second order ( crudely , @xmath374 or @xmath375 ) , so typically the main contribution is expected to be from decoherence , unless the unitary part is very inaccurate . we mainly discuss @xmath7 , but everything is practically the same in the language of @xmath8 . the composition of two error processes in the absence of desired unitary operations can be represented in the first order as a simple addition of the corresponding error matrices [ eqs.([comp - err])([comp - err - matrix ] ) ] . however , if for @xmath376 sequential error processes the `` coherent '' elements @xmath377 add up with the same phases and thus the sum grows linearly with @xmath376 , then the second - order contribution ( in particular , to the diagonal elements @xmath373 ) grows as @xmath378 , and for large @xmath376 it can become significant in comparison with the first - order decoherence contribution , which grows linearly with @xmath376 . for a composition of two quantum gates with non - trivial desired unitaries we need first to `` jump '' the error process over the unitary ( see fig . 2 ) , that is described by the transformation ( [ chi - err - jump ] ) , and then add the error matrices . essentially the same procedure can be done to calculate the error matrix contribution due to the lindblad - form decoherence in a quantum gate , which has finite duration and non - trivial evolution in time . for each short time step @xmath286 the decoherence produces a contribution to @xmath7 [ eq . ( [ chi - err - b - mn ] ) ] , but this contribution should be `` jumped over '' the unitary evolution to the beginning or the end of the gate before being summed up [ eqs.([lindblad - chi - err - tilde ] ) and ( [ lindblad - chi - err ] ) ] . the equivalent language is to `` jump '' the decoherence kraus operators over the desired unitaries , before the summation of error matrices [ eqs . ( [ b(t)])([tilde - b ] ) ] . it is interesting that in the leading order the contribution to the infidelity @xmath97 from the decoherence ( if it occurs within the same hilbert space ) does not depend on the desired unitary evolution [ eqs . ( [ lindblad - f ] ) and ( [ lindbad - t1-t2 ] ) ] . since the elements @xmath94 directly tell us about the unitary imperfection , it is easy to find the needed unitary correction [ eq . ( [ corr - unitary ] ) ] and the corresponding fidelity improvement [ eq . ( [ delta - f ] ) ] . however , the fidelity improvement is of second order and therefore is typically not expected to be significant . we have considered a particular example of correcting a cz gate using single - qubit @xmath41-rotations and cz - phase corrections [ eqs.([correct - phi12])([correct - czphase ] ) ] . the qpt suffers from errors in preparing the initial states and tomography measurement ( spam errors ) . while this problem does not seem to have a simple solution , in sec . [ sec - spam ] we have discussed a way , which may be helpful in alleviating this problem . a natural idea is to measure the error matrix @xmath367 in the absence of the gate and then subtract it from the measured error matrix @xmath338 of the characterized gate @xmath6 . however , this idea works only if the spam is dominated by one type of error : either at the preparation or at the tomography measurement . in general we need to know the contributions @xmath335 and @xmath336 from both errors separately because their addition depends on the gate @xmath6 [ eq . ( [ spam - exp ] ) ] . this can be done if some high - fidelity single - qubit gates are available ; then analyzing the change of @xmath338 with application of different high - fidelity gates , we can separate the contributions from @xmath335 and @xmath336 . note that this method assumes that the spam - errors can be represented as error processes at the preparation and tomography stages ; the accuracy of this assumption is questionable . one of the ways to check this assumption is to check one of its predictions : eq . ( [ spam - fid ] ) says that the `` actual '' fidelity of a quantum gate is the ratio of its qpt - measured fidelity and fidelity of the no - gate operation . the gate fidelity calculated in this way can then be compared with the fidelity obtained from the randomized benchmarking . the appendices of this paper are to a significant extent separated from the main text . in appendix a we consider several simple examples of @xmath0-matrices for unitary evolution and decoherence , including the energy relaxation and pure dephasing ( markovian and non - markovian ) . in appendix b we discuss unraveling of the lindblad - form evolution into the `` jump '' and `` no - jump '' scenarios , which can bring useful intuition in the analysis of decoherence ; several examples are considered to illustrate the technique . reiterating the main point of this paper , we think that characterization of quantum gates by error matrices in the pauli basis is a convenient way of presenting experimental qpt results . the author thanks yuri bogdanov , abraham kofman , justin dressel , andrzej veitia , jay gambetta , michael geller , and john martinis for useful discussions . the research was funded by the office of the director of national intelligence ( odni ) , intelligence advanced research projects activity ( iarpa ) , through the army research office grant no . w911nf-10 - 1 - 0334 . all statements of fact , opinion , or conclusions contained herein are those of the authors and should not be construed as representing the official views or policies of iarpa , the odni , or the u.s . government . we also acknowledge support from the aro muri grant no . w911nf-11 - 1 - 0268 . in this appendix we consider several simple examples of the quantum processes , for which we calculate the standard process matrix @xmath0 of the qpt . the matrix @xmath0 is defined via eq . ( [ chi - def ] ) , which is copied here for convenience , _ _ mn e_m _ ine_n^. [ chi - def-2]for the operator basis @xmath19 we use the pauli basis , so that for one qubit it consists of four pauli matrices , @xmath379 while for several qubits the kronecker ( direct , outer , tensor ) product of these matrices is used . we use the notation @xcite in which @xmath380 . as a very simple example , let us calculate the matrix @xmath0 for a @xmath41-rotation of a qubit over the angle @xmath381 . this realizes the unitary operator @xmath6 , which acts as @xmath382 , we have in general @xmath383 . let us represent @xmath6 in the pauli basis , u = ( cc 1&0 + 0&e^i ) = e^i/2 ( ( ) i -i ( ) z ) , [ u - z]where the unimportant overall phase factor @xmath384 does not affect the density matrix evolution , so that _ fin = [ ( ) i -i ( ) z ] _ in [ ( ) i + i ( ) z ] [ rho - for - z](note that the pauli matrices are herimitian , so we do not need to conjugate them ) . now by comparing eq . ( [ rho - for - z ] ) with eq . ( [ chi - def-2 ] ) we immediately find @xmath385 other elements are zero . it is easy to see that the method of finding @xmath0 by comparing eq . ( [ rho - for - z ] ) with eq.([chi - def-2 ] ) is equivalent to using eq . ( [ chi - u ] ) . in a similar way we can calculate the matrix @xmath0 for a one - qubit @xmath346-rotation over angle @xmath381 . the result is obviously the same as eqs . ( [ chi - z-1])([chi - z-2 ] ) , with index @xmath41 replaced by @xmath346 : @xmath386 , @xmath387 , @xmath388 . similarly , for @xmath21-rotations we replace @xmath41 in eqs.([chi - z-1])([chi - z-2 ] ) with @xmath21 . a qubit rotation about an axis @xmath389 on the bloch sphere over an angle @xmath381 corresponds to the unitary operator @xcite u== ( ) i - i ( ) ( ) , where @xmath390 . note that this formula neglects the overall phase factor ( which does not exist in the bloch - sphere space ) , as seen by comparing it with eq . ( [ u - z ] ) . using eq . ( [ chi - u ] ) or , alternatively , comparing equation @xmath383 with the definition ( [ chi - def-2 ] ) , we still can easily find the elements of the @xmath0-matrix ; now all 16 elements are in general non - zero , though they are determined by only 3 real parameters . let us start with the case when the first qubit is @xmath41-rotated over the angle @xmath381 , while the second qubit is `` idling '' . then the unitary operator is @xmath391 \otimes i$ ] , and eq . ( [ rho - for - z ] ) becomes _ fin = [ ( i -i z)i ] _ in [ ( i + i z)i ] , [ rho - for - z-2]where for brevity we omit the argument @xmath392 of sines and cosines . comparing this equation with ( [ chi - def-2 ] ) , we have to use double - letter combinations for both indices @xmath393 and @xmath184 ; however , we see that the second - qubit letter is always @xmath356 , so we essentially obtain the single - qubit result ( [ chi - z-1])([chi - z-2 ] ) with added index @xmath356 for the second qubit : @xmath394 as another example , let us consider the controlled - phase operation ( note that our use of the name `` controlled - phase '' is different from the terminology of ref . @xcite ) , @xmath395 where in this ( rather sloppy ) notation we omit the kronecker product sign `` @xmath396 '' . by comparing @xmath397 with ( [ chi - def-2 ] ) we obtain 16 non - zero elements : @xmath398 the controlled - phase gate becomes the cz gate at @xmath399 . then @xmath400 , and all 16 elements of @xmath0 in eqs.([c - p-1])([c - p-5 ] ) become @xmath401 . note that if in an experimental realization the phase @xmath402 fluctuates symmetrically around @xmath344 with a small variance @xmath403 , then the element @xmath404 increases by @xmath405 , while other 15 elements decrease in absolute value by @xmath406 . for the perfect cnot gate ( with the first qubit being the control ) the unitary can be represented as u = i + x , [ u - cnot]where we used the relations @xmath407 and @xmath408 ( here we again use the notation `` @xmath396 '' for more clarity ) . then non - zero elements of @xmath0 are @xmath409 as directly follows from the combinations of 4 terms in eq.([u - cnot ] ) . for the perfect @xmath42 gate the unitary is @xmath410 so the non - zero elements of the matrix @xmath0 are given by the pairwise products of these 4 terms , @xmath411 it is very easy to find the matrix @xmath0 for one qubit with pure dephasing ( assuming no other evolution ) . after waiting for time @xmath412 , the qubit is @xmath41-rotated over a random angle @xmath381 . from the definition @xmath413 we see that for a random evolution we simply need to average the @xmath0-matrix over the possible evolution realizations . therefore the @xmath0-matrix for pure dephasing is given by averaging eqs . ( [ chi - z-1])([chi - z-2 ] ) : @xmath414 where @xmath415 denotes averaging over realizations . for a symmetric probability density distribution of @xmath381 we get @xmath416 and therefore @xmath417 , so the only non - zero elements are @xmath418 and @xmath419 . it is important to emphasize that the result ( [ deph-1 ] ) does not assume exponential dephasing ; it remains valid for an `` inhomogeneous '' contribution to the dephasing ( slightly different qubit frequencies in different experimental runs ) and/or the `` 1/f '' contribution ( when the qubit frequency fluctuation has a broad range of timescales ) . it is also important that the value @xmath420 which determines @xmath419 can be directly obtained from the ramsey - fringes data . let us consider the ramsey protocol : start with @xmath421 , apply @xmath345 @xmath346-rotation , wait time @xmath412 , apply the second @xmath345 rotation about the axis , which is shifted from @xmath346 by an angle @xmath422 in the equatorial plane of the bloch sphere , and finally measure the probability @xmath423 of the state @xmath424 . it is easy to find that for a pure dephasing ( including non - exponential case ) p(|1 ) = + = + , where for the second equation we assumed @xmath425 . in the case of exponential dephasing characterized by the dephasing time @xmath325 , we have @xmath426 . in the general case the time dependence of @xmath427 is arbitrary ; however , it can be found experimentally from the amplitude of the ramsey oscillations and then can be used in eq . ( [ deph-1 ] ) to obtain @xmath419 . it is important to mention that experimentally @xmath325 is often defined as the time at which @xmath428 . if the qubit dephasing is due to fast ( `` white noise '' ) fluctuations of the qubit energy , then @xmath429 and correspondingly @xmath430 , so that at short time , @xmath431 , there is a linear dependence , _ zzt/2t_. however , if the pure dephasing is dominated by the very slow fluctuations of the qubit energy , then @xmath432 $ ] and the ramsey - fringes dependence has a gaussian shape . in this case at @xmath433 the dephasing error is quadratic in time , _ zzt^2/2t_^2 . in the presence of both mechanisms = ( -t / t _ , fast ) ; this formula can also be used as an approximation in the case of a broad range of the fluctuation timesclales . the corresponding @xmath419 is still given by eq . ( [ deph-1 ] ) . note that in the presence of energy relaxation ( discussed later ) the value of @xmath427 can still be directly extracted from the ramsey - fringes data see eq.([ramsey - t1-deph ] ) below . now let us calculate the matrix @xmath0 taking into account the qubit energy relaxation , but assuming the absence of pure dephasing . let us start with the zero - temperature case ( relaxation to the state @xmath421 only ) . using `` unraveling '' of the energy relaxation in the same way as in refs . @xcite and @xcite ( see also appendix b ) , we may think about two probabilistic scenarios : |0+|1\ { c |0 with prob . p_r=||^2 ( 1-e^-t / t_1 ) + with prob . ( in the case of no relaxation the state evolves due to the bayesian update ) . this corresponds to the technique of kraus operators and for the density matrix gives _ in a_r _ in a_r^+ a_no _ in a_no^ , [ rel - kraus - eq]where the kraus operators @xmath434 ( for the scenario with relaxation ) and @xmath435 ( for the scenario with no relaxation ) are a_r =( cc0 & + 0 & 0 ) , a_no =( cc 1 & 0 + 0 & e^-t/2t_1 ) . note that @xmath436 ( the completeness relation ) . here we use the standard kraus - operator representation @xcite , in contrast to the somewhat modified representation ( [ kraus - representation ] ) . to find the @xmath0-matrix , we expand the kraus operators in the pauli basis , @xmath437 now comparing evolution ( [ rel - kraus - eq ] ) with the form ( [ chi - def-2 ] ) , we collect the @xmath0-matrix elements ( the relaxation term brings elements involving @xmath346 and @xmath21 , while the no - relaxation term brings elements with @xmath356 and @xmath41 ) : @xmath438 note that at small @xmath439 the element @xmath419 is quadratic in time ( very small ) , while other elements ( except @xmath418 ) are linear in time ( in this case @xmath440 , as expected from the discussion in sec . [ sec - properties ] ) . also note that the non - zero elements in the left column and top row ( @xmath441 and @xmath442 ) are real and come from the no - relaxation scenario ( see sec . [ sec - properties ] ) . for a non - zero temperature there are two kinds of the relaxation processes ( up and down ) with the rates @xmath443 and @xmath444 satisfying the standard relations @xmath445 and @xmath446 , where @xmath98 is temperature and @xmath447 is the energy difference between the qubit states ; this gives @xmath448 . correspondingly , there are three scenarios with the kraus operators @xmath449 then in a similar way as above we find the @xmath0-matrix elements : @xmath450 phase evolution commutes with the energy relaxation , therefore we may apply @xmath41-rotation over a random angle @xmath381 after the energy relaxation . the @xmath41-rotation does not affect scenario(s ) with relaxation , so we need to change only the kraus operator for the no - relaxation scenario : @xmath451 , then calculate the corresponding elements of the @xmath0-matrix in the same way as above and average over @xmath381 . therefore , the elements @xmath452 , @xmath453 , @xmath454 , @xmath455 due to energy relaxation will not be affected by the pure dephasing ( since they come from the relaxation scenario ) . the calculation shows that the elements @xmath441 and @xmath442 are also not affected when @xmath456 ( satisfied for a symmetric noise ) , so the only affected elements are @xmath419 and @xmath418 , which are non - zero for both decoherence mechanisms . using again the condition @xmath457 , it is easy to find @xmath458 where @xmath459 and @xmath460 correspond to pure dephasing and energy relaxation , and were calculated above [ eqs.([deph-1 ] ) and ( [ rel - temp-1])([rel - temp-5 ] ) ] . for completeness let us also show the unaffected elements : @xmath461 another way of deriving eqs.([deph+rel - zz])([deph+rel - iz ] ) is the following . let us apply pure dephasing after energy relaxation and write the composition of quantum operations as _ m , n _ mn _ mn^rel _ mn^deph e_m e_m _ ine_n^e_n^. [ chi - composition]even though a product of pauli matrices is a pauli matrix ( possibly with a phase factor ) and therefore @xmath0-matrix of a composition of operations can in principle be calculated in a straightforward way , usually this is a very cumbersome procedure . however , in our case the matrix @xmath459 has only two non - zero elements ( @xmath462 and @xmath463 ) , so the calculation is not very long and leads to eqs.([deph+rel - zz])([deph+rel - iz ] ) . now when we have explicit formulas for the @xmath0-matrix elements , which depend on @xmath439 , temperature , and @xmath464 , let us discuss again how to extract @xmath464 from the ramsey - fringes data . in the presence of energy relaxation ( at arbitrary temperature ) the ramsey oscillations are @xmath465 therefore , if the energy relaxation time @xmath324 is measured separately , the ramsey data give the value of @xmath466 at any time @xmath412 . this value can be used to calculate the @xmath0-matrix even in the case of arbitrary non - exponential dephasing . now let us discuss the @xmath0-matrix at a relatively short time @xmath412 and neglect the terms quadratic in time . then we obtain @xmath467 in the case of exponential pure dephasing @xmath468 , and then @xmath469 , where @xmath470 and @xmath471 is the dephasing time . in this appendix we discuss the technique of kraus operators applied to the evolution described by the standard lindblad - form master equation . we show that each lindblad term describes two evolutions : a `` jump '' process with some rate and a continuous evolution between the jumps caused by the absence of jumps . such interpretation can be useful in intuitive analysis of decoherence processes . a markovian evolution of a quantum system is usually described by the lindblad - form master equation [ for convenience we copy eq.([lindblad - evol ] ) here ] = - [ h , ] + _ n _ n ( b_nb_n^- b_n^b_n- b_n^b_n ) , [ lindblad]where the first term describes the unitary evolution due to the hamiltonian @xmath472 , while the @xmath184th decoherence mechanism is described by the `` rate '' @xmath473 ( real number with dimension s@xmath474 ) and dimensionless operator @xmath475 . note that mathematically @xmath473 can be absorbed by redefining @xmath476 ; however , we do not do this because @xmath473 and @xmath475 have separate physical meanings . our goal here is to discuss a simple physical interpretation of the decoherence terms in the lindblad equation . for simplicity let us neglect the unitary evolution ( @xmath477 ) and consider first only one decoherence mechanism ; then we can omit the index @xmath184 ( summation over @xmath184 is simple ) . it is easy to check that the term @xmath478 corresponds to the abrupt change ( `` jump '' ) of the state | = , = , [ jump]which occurs with the rate ( jump probability per second ) = || b|||^2 , = tr ( b b^ ) = tr ( b^b ) . [ rate](here we show the formulas in both the wavefunction and density matrix languages ; the wavefunction language is usually more convenient to use . ) note a possible confusion in terminology : both @xmath322 and @xmath479 are rates ; to distinguish them let us call @xmath322 the `` process rate '' or just `` rate '' , while @xmath479 will be called `` jump rate '' . the remaining term @xmath480 in the lindblad form corresponds to the jump process ( [ jump ] ) _ not happening_. the physical reason of this evolution `` when nothing happens '' is the same as the partial collapse in the null - result measurement @xcite : this is essentially the bayesian update @xcite , which accounts for the information that the jump did not happen . therefore , _ the physical meaning of the lindblad form is the description of two scenarios : the jump process either happening or not happening . _ to see this mathematically , let us use the technique of kraus operators @xcite , in which the evolution @xmath481 is unraveled into the probabilistic mixture of `` scenarios '' described by kraus operators @xmath109 , @xmath482 [ in the wavefunction language @xmath483 . the sum of the probabilities should be equal to 1 , which leads to the completeness relation _ k a_k^a_k = . [ completeness]note the change of notations compared with eq.([kraus - representation ] ) . during a short time @xmath286 the evolution ( [ lindblad ] ) can be described by two scenarios : the jump either occurs or not , with the corresponding kraus operators @xmath484 and @xmath435 . for the jump scenario a_jump= b , so that @xmath485 gives the same contribution to @xmath486 as the term @xmath487 in eq.([lindblad ] ) . the no - jump kraus operator @xmath435 should satisfy the completeness relation ( [ completeness ] ) , @xmath488 . using the bayesian - update approach @xcite in which @xmath489 , we find @xmath490 note that @xmath491 is a positive hermitian operator and @xmath492 is also a positive operator . a square root of a positive operator is defined via taking square roots of its eigenvalues in the diagonalizing basis . this is why in eq . ( [ a - no ] ) we deal with operators essentially as with numbers . using eq . ( [ a - no-2 ] ) for @xmath435 , we find in the linear order a_no a_no^- t b^b - t b^b , [ a - no - evol]with this linear - order approximation becoming exact at @xmath493 . it is easy to see that eq . ( [ a - no - evol ] ) gives the evolution described by the term @xmath494 in the lindblad form ( [ lindblad ] ) . thus we have shown that the lindblad - form master equation ( [ lindblad ] ) with one decoherence term describes a jump process @xmath283 [ see eq . ( [ jump ] ) ] occurring with the jump rate ( probability per second ) @xmath495 . in the case of several decoherence mechanisms there are several kraus operators @xmath496 describing the jumps @xmath497 during a short duration @xmath286 . the no - jump kraus operator in this case is @xmath498 , which contributes to all terms @xmath499 in eq . ( [ lindblad ] ) . note that eq . ( [ a - no ] ) and its generalization for several processes is not the unique form for @xmath435 , which follows from the completeness relation ( [ completeness ] ) . it is formally possible to add an arbitrary unitary rotation @xmath6 , so that @xmath500 , which does not affect @xmath501 . using a natural assumption that @xmath502 at @xmath503 , we can expand @xmath6 in the linear order as @xmath504 , where @xmath505 should be a hermitian matrix . then @xmath506 acquires the extra term @xmath507\delta t$ ] , from which we see that @xmath505 is essentially an addition to the hamiltonian @xmath472 . so the formalism permits a change of the hamiltonian due to decoherence , and therefore @xmath472 in eq.([lindblad ] ) should be considered as the effective hamiltonian ( which may include the `` lamb shift '' mechanism ) . for a qubit relaxation from the excited state @xmath424 to the ground state @xmath421 with the rate @xmath508 , the jump operator is @xmath509 , where the ground state corresponds to the upper line , so that @xmath510 , @xmath511 . in this case @xmath512 , so that the no - jump evolution during time @xmath286 changes the qubit state as |0+|1 , [ qubit - relax - no - jump]while in the case of jump obviously @xmath513 can be taken into account in the similar way using an additional lindblad - form term . the physical mechanism of the markovian pure dephasing in superconducting qubits is the fast ( `` white noise '' ) fluctuations of the qubit energy , which leads to random fluctuations of the qubit phase @xmath514 , so that the random phase shift @xmath515 accumulated during a short time @xmath286 has the gaussian probability distribution with zero mean and variance @xmath516 , where @xmath325 is the dephasing time . it is easy to check that the correct evolution due to pure dephasing [ @xmath517 , @xmath518 , @xmath519 can be reproduced using the lindblad equation ( [ lindblad ] ) with @xmath520 and @xmath521 . this means that instead of the physically correct picture of continuous change of @xmath381 , we may use a completely different picture : random jumps of the phase @xmath381 by @xmath344 , |0+|1|0-|1 , occuring with the jump rate [ see eq . ( [ rate ] ) ] @xmath522 , which in this case is independent of the qubit state and equal to @xmath322 . note that @xmath523 , so in this case there is no no - jump evolution . since both pictures lead to the same evolution of @xmath524 , we can use any of them , depending on convenience in a particular problem . one more picture which can be used for pure dephasing is the `` random measurement of state @xmath421 '' , for which @xmath525 and @xmath526 . it is easy to see that it leads to the same lindblad equation , but has different jump process and non - trivial no - jump evolution . obviously , we can also use @xmath527 and @xmath526 for the same master equation ( see the brief discussion in sec.[sec - lindblad - error ] of the transformation @xmath528 in the lindblad equation ) . the decay of a resonator state is usually characterized by the energy decay rate @xmath529 , so that in the qubit terminology @xmath530 . the standard lindblad form ( [ lindblad ] ) in this case has @xmath531 and @xmath532 , where @xmath533 is the annihilation operator , @xmath534 . the jump evolution is then | , [ res - jump]with the jump rate [ see eq . ( [ rate ] ) ] @xmath535 , where @xmath536 is the average number of photons . the no - jump evolution during time @xmath286 is then _ n _ n |n , [ res - no - jump]which is similar to the no - jump evolution ( [ qubit - relax - no - jump ] ) for the qubit . note that the probability of the no - jump evolution is given by the squared norm , @xmath537 , so that the sum of the jump and no - jump probabilities is 1 . it is interesting to analyze the special case : evolution of a coherent state , @xmath538 . since it is the eigenstate of the operator @xmath533 , the jump evolution ( [ res - jump ] ) does not change the state , while the continuous no - jump evolution essentially changes the parameter @xmath539 in time , @xmath540 . therefore the energy decay is due to the no - jump evolution only . in this ( very unusual ) case a pure initial state remains pure because the jump does not change the state and therefore it is essentially a one - scenario ( no - jump ) evolution . the standard lindblad - form evolution for the resonator with @xmath531 and @xmath532 can be understood in the following way . let us assume that @xmath529 is the coupling with the outside modes , and let us imagine ( gedankenexperiment ) that we use an ideal photon detector , which clicks every time when a photon escapes from the resonator . when the detector clicks , we know that there is one photon less in the resonator . however , all transitions @xmath541 are indistinguishable ( because the energy levels are equidistant ) , and therefore the bayesian update of the quantum state @xcite involves only the rates , which are proportional to @xmath184 , _ it is easy to see that this bayesian update coincides with the jump process ( [ res - jump ] ) , and therefore the lindblad - form evolution can be interpreted as being due to an outside measurement with a single - photon detector . [ note that the lindblad - form master equation describes the ensemble - averaged evolution ; this is why we are free to choose any measurement model , in contrast to the case of individual ( selective ) evolution , which depends on what is actually measured . ] the energy relaxation ( at zero temperature ) in a slightly ahnarmonic superconducting 3-level qubit is sometimes described by essentially the same lindblad equation as for the resonator , with @xmath542 and @xmath532 , so that latexmath:[$b|2\rangle = \sqrt{2 } \ , @xmath511 . this is actually incorrect . the reason is that the qubit is anharmonic , and therefore in the described above gedankenexperiment it is in principle possible to distinguish transitions @xmath544 where the top row corresponds to the state @xmath421 and the bottom row corresponds to @xmath545 . the formula @xmath546 is not exact because of anharmonicity . if for more accuracy we want to take into account direct transitions @xmath547)([3-level-2 ] ) , the ( incorrect ) one - process description ( with @xmath532 ) adds the extra term @xmath548 into @xmath549 , while other terms in these two approaches coincide . concluding this appendix , we emphasize that unraveling of the lindblad - form evolution into the `` jump '' and `` no - jump '' processes is often useful for gaining intuition in the analysis of decoherence , and it may also be useful in analytical and numerical calculations . t. yamamoto , m. neeley , e. lucero , r. c. bialczak , j. kelly , m. lenander , m. mariantoni , a. d. oconnell , d. sank , h. wang , m. weides , j. wenner , y. yin , a. n. cleland , and j. m. martinis , phys . b * 82 * , 184515 ( 2010 ) . e. magesan , j. m. gambetta , b. r. johnson , c. a. ryan , j. m. chow , s. t. merkel , m. p. da silva , g. a. keefe , m. b. rothwell , t. a. ohki , m. b. ketchen , and m. steffen , phys . rev . lett . * 109 * , 080505 ( 2012 ) . j. kelly , r. barends , j. bochmann , b. campbell , y. chen , b. chiaro , e. jeffrey , m. mariantoni , a. megrant , j. mutus , c. neill , p. omalley , s. ohya , p. roushan , d. sank , a. vainsencher , j. wenner , t. white , a. n. cleland , and j. m. martinis , bull . . soc . * 58 * , g26.00005 ( 2013 ) . j. m. chow , j. m. gambetta , a. d. corcoles , s. t. merkel , j. a. smolin , c. rigetti , s. poletto , g. a. keefe , m. b. rothwell , j. r. rozen , m. b. ketchen , and m. steffen , phys . rev 109 * , 060501 ( 2012 ) . to discuss orthogonality and normalization of matrices , we need to define the inner product of matrices @xmath551 and @xmath283 of equal size . this can be done by direct generalization of the innner product of two vectors : the matrices are treated as vectors , and the inner product is defined element - by - element as @xmath552 . this is called frobenius or hilbert - schmidt inner product ; we use the notation similar to the notation for vectors in quantum mechanics , while in mathematics notations @xmath553 or @xmath554 are often used . it is easy to see that @xmath555 . this definition leads to the frobenius ( hilbert - schmidt ) norm for a matrix : @xmath556 . m. mariantoni , h. wang , t. yamamoto , m. neeley , r.c . bialczak , y. chen , m. lenander , e. lucero , a.d . oconnell , d. sank , m. weides , j. wenner , y. yin , j. zhao , a.n . korotkov , a.n . cleland , and j.m . martinis , science * 334 * , 61 ( 2011 ) . a. galiautdinov , a. n. korotkov , and j. m. martinis , phys . a * 85 * , 042321 ( 2012 ) ; j. ghosh , a. galiautdinov , z. zhou , a. n. korotkov , j. m. martinis , and m. r. geller , phys . rev . a * 87 * , 022309 ( 2013 ) .

there is no currently accepted scenario for the formation of the parent bodies of primitive meteorites which accounts for the most obvious of their properties . these properties ( reviewed by scott and krot 2005 and discussed in more detail in section 2.1 ) include ( a ) dominance by aerodynamically well - sorted mineral particles of sub - mm size ; ( b ) class - to - class variation in well - defined physical , chemical , and isotopic properties ; ( c ) a spread of 1 myr or so between the formation times of the oldest and youngest objects found in the same meteorite ; ( d ) a spread of 1 - 3 myr in radiometric ages of different meteorite types ; and ( e ) a dearth of melted asteroids , with model results for even some melted asteroids which imply myr delays in formation relative to ancient minerals . in recent years , meteoritic evidence has appeared for some very early - formed planetesimals , which represent a minority of both meteorites and asteroids . this implies that primary accretion started early and continued for several million years - thus , it was fairly inefficient and did not run quickly to completion . most current models for this primary accretion " stage ( reviewed by cuzzi and weidenschilling 2006 and dominik et al 2007 ) can be classified as either ( a ) incremental growth , where large particles sweep up smaller ones by inelastic collisions involving porous surfaces , and growth proceeds hierarchically ; or ( b ) instability , where physical sticking is irrelevant and collective effects drive collapse to km - sizes or larger on very short timescales . those who favor instability models , most of which rely on gravity and occur in a particle - rich nebula midplane , are concerned by the poorly understood sticking of mineral particle aggregates and the apparent difficulty of growing beyond meter size due to rapid inward migration and collisional disruption . those who favor incremental growth have noted that midplane instability models are precluded by even very weak global turbulence , and that , in the dense midplane layers that form when turbulence is absent , incremental accretion is at low relative velocity and the meter - size barrier is not a problem . however , the most sophisticated models of incremental accretion in nonturbulent nebulae find it to be _ so _ efficient that large planetesimals grow in only @xmath0 years throughout the asteroid belt region ( weidenschilling 2000 ) , a short time which is difficult to reconcile with constraints ( a - e ) above . a third class of scenario suggests that a complex interplay between several nonlinear processes - turbulence , pressure gradients , and gravity - may concentrate appropriately sized particles and lead to planetesimal growth ( cuzzi et al 2001 , 2005 , 2007 ; johansen et al 2006 , 2007 ) . scenarios which involve preferentially meter - sized objects encounter concerns about whether meter - sized objects can survive their own high - velocity mutual collisions in this sort of turbulent environment ( _ cf . _ sirono 2000 , langkowski et al 2007 , ormel et al 2007 ) . however , in principle , such boulder - concentration scenarios can proceed independently in the same environment as discussed in this paper , where we focus on mm - size particles . most of the above scenarios provide no natural explanation for observed meteorite properties ( a ) and ( b ) above . in particular , the evidence for the h chondrite class ( and perhaps all the ordinary chondrites ) suggests that entire asteroids of 10 - 100 km diameter formed directly from a physically , chemically , and isotopically homogeneous mix of dust - rimmed particles of similar size ( section 2.1 ) . in this paper , we outline a possible path by which entire batches of mm - size , aerodynamically sorted particles might proceed directly in turbulence ( even if sporadically ) to planetesimals having the properties outlined above . we find that sufficiently large and dense clumps of mm - size particles can form by turbulent concentration such that , even if classical gravitational instability ca nt operate , their self - gravity may still allow them to survive disruptive forces and slowly sediment into a sandpile " planetesimal . this simple analytical model is backed up by some numerical simulations that support the basic idea . in sections 2.2 and 2.3 , we review the most relevant physics that determines the properties of dense , particle - rich zones or clumps in turbulent nebulae . in section 3 , we address the fate of these dense clumps using analytical and numerical models of their evolution . we derive the combination of size and mass density a clump must have to evolve into a sandpile " having some degree of internal strength . in subsequent stages not modeled here , we imagine that collisions and thermal sintering transform these sandpiles into the cohesive rocky parent bodies we see today . however , in the outer solar system , lower energy collisions and weaker thermal processing might well allow planetesimals to retain their initial low - strength states , as seen for some cometary objects ( _ eg . , _ asphaug and benz 1996 ) . one can extract a number of clues from primitive ( unmelted ) meteorites regarding the primary accretion process by which their parent bodies first formed ( scott and krot 2005 , taylor 2005 ) . for the best evidence , one must look back through extensive subsequent evolutionary stages . even unmelted bodies in the 100 km size range have incurred extensive collisional evolution ( bischoff et al . 2006 ) , producing compaction , fragmentation , and physical grinding and mixing on and beneath their surfaces , which may obscure the record of primary accretion . model studies ( _ e.g. _ , petit et al 2001 , kenyon and bromley 2004 , 2006 ; bottke et al 2005 ; chambers 2004,2006 ; weidenschilling and cuzzi 2006 ) suggest that the collisional stage occurred after dispersal of the nebula gas allowed the orbital eccentricities of primitive bodies to grow . we are concerned with an earlier stage , when the still - abundant nebula gas led to a more benign environment with fewer and gentler collisions . the direct products of primary accretion might be most clearly visible in the rare primary texture " seen in some cm ( metzler et al 1992 ) and co ( brearley 1993 ) chondrites . in these objects , or more specifically in unbroken fragments within them , the texture consists of similarly - sized , dust - rimmed particles packed next to each other as if gently brought together and compressed , with no evidence for local fracturing or grinding . even after collisional effects associated with subsequent stages of growth have blurred this signal , perhaps even mixing material from different parent bodies , evidence remains in the bulk properties of all chondrite classes . chondrite classes are defined by their distinctive mineral , chemical , and isotopic properties ( _ e.g. _ grossman et al 1988 , scott and krot 2005 , weisberg et al 2006 ) . large samples of material with a quite well defined nature were accumulated at one place and/or time , and material of a quite different , but equally well - defined , nature was accumulated at another place and/or time into a different parent body . amongst the most obvious aspects of these class properties is a dominance within chondrites of sub - mm size mineral particles ( generically chondrules " ; grossman 1988 , jones et al 2000 , 2005 ; connolly et al 2006 ) which are aerodynamically well - sorted . aerodynamic sorting of chondrite components was first emphasized by dodd ( 1976 ) , hughes ( 1978 , 1980 ) , and skinner and leenhouts ( 1993 ) , and has since been discussed by a variety of authors ( see cuzzi and weidenschilling 2006 for a summary of the observations and arguments supporting aerodynamics ) . chondrules have a size which varies from class to class , but is distinctive and narrowly defined within a given class or chondrite . the evidence suggests that chondrules do not comprise merely a thin surface layer swept up by a large object ( scott 2006 ) . in the case which we suggest as an archetype , the h - type ordinary chondrite parent body is widely believed to be a 100 km radius object ( perhaps the asteroid hebe ) , initially composed entirely of a physically , chemically , and isotopically homogeneous mix of chondrules and associated material , which was thermally metamorphosed by accreted @xmath1al to a degree which varied with depth , into an onion - shell structure , and subsequently broken up in several stages ( taylor et al 1987 , trieloff et al 2003 , mcsween et al 2002 , grimm et al 2005 ) . bottke et al ( 2005 ) conclude from models of collisional evolution in the primordial ( massive ) and the current ( depleted ) asteroid belt , that the primordial asteroid mass distribution was dominated by objects having diameter of around 100 km , rather than having a powerlaw size distribution somewhat like that of the current population . we see the essential challenge as understanding how such a large object can be assembled from such a homogeneous mixture of mm - size constituents , while other objects ( arguably the parents of the l and ll type ordinary chondrites as well , and logically then the parents of the chondrites of all classes ) are assembled from distinct , yet comparably homogeneous , ensembles of qualitatively similar particles . moreover this assembly , or primary accretion , phase of nebula evolution must persist for a duration of 1myr or more between the formation times of the oldest and youngest meteorites ( wadhwa et al 2007 ) and even of objects found in the same meteorite . the @xmath2 1 myr age spread between ancient refractory inclusions ( cais ) and chondrules is well known ( russell et al 2006 ) , but there may even be a comparably extended age spread between different chondrules in a given meteorite ( kita et al 2005 ) . in recent years , radiometric dating of achondrites ( melted meteorites ) has revealed some early - formed ( and early melted ) planetesimals ( kleine et al 2005 , 2006 ) . one expects early formation and melting to go together , because objects larger than only 10 km or so , which accreted at the same time that refractory inclusions formed ( 4.567 byr ago ) , would have incorporated their full complement of @xmath1al and would have melted ( latourette and wasserburg 1998 ; woolum and cassen 1999 , hevey and sanders 2006 ) . thus , early accretion implies melting , and lack of melting requires late accretion . model results imply that , to avoid complete melting , 100 km bodies must accrete only after a typically 1.5 - 2 myr delay relative to ancient cai minerals ( reviewed by mcsween et al 2002 and ghosh et al 2006 ) . while unknown selection and sampling effects may influence the limited meteorite data record , spectral analysis of the entire asteroid database also suggests that melted asteroids are rare . that is , while debate continues as to whether small amounts of melt might be found on the surfaces of many asteroids , which might be caused by impacts ( _ e.g. _ gaffey et al 1993 , 2002 ) , only vesta and a handful of other examples of fully melted basaltic objects have been found in spite of extensive searches ( binzel et al 2002 ; moskovitz et al 2007 ) . put together , this body of information suggests that primary accretion started early but continued for several million years - thus , it was inefficient , or at least sporadic . by comparison , midplane incremental accretion in a nonturbulent nebula accretes numerous ceres size bodies and tens of thousands of 10 - 100 km size objects in only @xmath3 years ( weidenschilling 2000 ) ; thus , it is highly efficient . models suggest that the density , temperature , and composition of the nebula evolved significantly over several million years ( bell et al 1995 , dalessio et al . 2005 , ciesla and cuzzi 2006 ) . thus one expects the properties of forming planetesimals to change with _ time _ even though the chemical , isotopic , and mineralogical properties of the mixture of solids from which they formed might have been fairly uniform in _ space _ at any _ given _ time ( see cuzzi et al 2003 and ciesla 2007 for more detailed discussions ) . reality might have been even more complicated - there is some evidence that chondrites of very distinct chemical and isotopic properties might have formed at roughly the same time ( kita et al 2005 ) , suggesting a combination of spatial _ and _ temporal gradients . in this paper we present the overview of a scenario by which accumulation of a well - defined particle mix into a large planetesimal - making it a snapshot " of the local nebula mix - might occur sporadically or inefficiently in weak turbulence , over an extended period of time and throughout the terrestrial planet region . the challenge for the future is to show that the accretion processes discussed here can _ quantitatively _ account for the mass needed to build planetesimals of the mass needed to create planetary embryos and planets during the age of the nebula ( chambers 2004 , 2006 ; cuzzi et al 2007 ) . a variety of astronomical and meteoritical evidence supports weak , but widespread and sustained , nebula turbulence . the observed abundance of small particles at high altitudes above the midplane in many visible protoplanetary nebulae for millions of years is most easily explained by ongoing turbulence - both to regenerate the small grains in collisions and to redistribute them to the observed altitudes ( dullemond and dominik 2005 ) . the survival of ancient , high - temperature mineral inclusions ( cais ) in chondrites , in spite of their tendency to drift inwards towards the sun , can also be explained by turbulent diffusion ( cuzzi et al 2003 , 2005 ) . moreover , the recent discovery by stardust of both high- and moderate - temperature crystalline silicates in comet wild2 ( brownlee et al 2007 , zolensky et al 2007 ) can be easily understood in this way ( ciesla 2007 ) . it remains a subject of active theoretical debate as to how , or even whether , the nebula can maintain itself in a turbulent state . ongoing dissipation by molecular viscosity @xmath4 on the smallest , or kolmogorov , lengthscale requires a production mechanism to continually generate turbulent kinetic energy . although the very existence of ongoing disk accretion into the star , with the ensuing gravitational energy release , provides an ample energy source , the vehicle for transferring that energy into turbulent gas motions is undetermined . baroclinic instabilities require substantial opacity at thermal wavelengths , but might be suitable to generate turbulence during the fairly opaque chondrule - cai epoch in the inner solar system ( klahr and bodenheimer 2003 ) . the widely accepted magnetorotational instability ( mri ) ( stone et al 2000 ) may or may not be precluded at the high gas densities of the inner solar system ( turner et al 2007 ) . the role of pure hydrodynamics operating on radial shear remains unclear , because keplerian disks are nominally stable to linear ( small ) perturbations ( balbus et al 1996 , stone et al 2000 ) . however , a number of current theoretical studies suggest turbulence might indeed be present ( _ eg . _ arldt and urpin 2004 , busse 2004 , umurhan and regev 2004 , dubrulle et al 2005 , afshordi et al 2005 , mukhopadhyay et al 2005 , mukhopadhyay et al 2006 ) . much of this recent work focusses on nonlinear ( finite amplitude ) instabilities occurring at the very high reynolds numbers characterizing the nebula , which are well beyond the capability of current numerical and even laboratory models . , of particles having local concentration factor larger than some value of @xmath5 ( from cuzzi et al 2001 ; see also section 2.3 ) . the curves with symbols represent comparison of the predictive theory described in cuzzi et al ( 2001 ) with 3d numerical simulations at moderate reynolds numbers ; the curves without symbols are extension of the same theory to nebula conditions denoted by their values of @xmath6 . concentration factors @xmath7 of chondrule precursors ( leading to @xmath8 ) are fairly common for a range of nebula turbulent intensities ( denoted by @xmath6 ) , and help explain lack of chondrule isotopic fractionation ( cuzzi and alexander 2006 ) . in this paper we focus on the much less common , but much denser , regions towards the bottom right side of the plot ( @xmath9 or @xmath10 ) . more general pdfs which separate out the role of vorticity and incorporate the effect of particle mass loading are discussed in section 2.3 and appendix a. , width=297,height=297 ] in this paper , we simply assume the presence of weak nebula turbulence . the intensity of turbulence may be characterized by a nondimensional parameter @xmath11 , where @xmath12 is the velocity of the most energetic eddies and @xmath13 is the sound speed . values of @xmath6 in the range @xmath14 are consistent with observed nebula lifetimes , and can be sustained by only a few percent of the gravitational energy released as the nebula disk flows into the sun . one then estimates the most energetic ( largest ) eddy size as @xmath15 , where @xmath16 is the nebula vertical scale height , or disk thickness . a cascade to smaller scales ensues , where the smallest ( kolmogorov ) eddy scale is @xmath17 1 km and @xmath18 for @xmath19 is the nebula reynolds number ( cuzzi et al 2001 ) . it has been found both numerically and experimentally that particles of a well defined size and density are concentrated into very dense zones in realistic 3d turbulence ( eaton and fessler 1994 ) . the optimally concentrated particle has a gas drag stopping time @xmath20 which is equal to the overturn time of the kolmogorov scale eddies . in the nebula , @xmath21 where @xmath22 and @xmath23 are the particle radius and material density , and @xmath13 and @xmath24 are the gas sound speed and density . the kolmogorov eddy frequency @xmath25 scales with the orbital frequency @xmath26 as @xmath27 , and particles are most effectively concentrated when @xmath28 ( eaton and fessler 1994 ) . chondrule - size silicate particles are concentrated for canonical nebula properties if @xmath29 ; moreover the _ shape _ of the concentrated particle size distribution is parameter - independent , and agrees very well with that of chondrules ( paque and cuzzi 1997 , cuzzi et al 2001 ) . we define the concentration factor @xmath30 , where we denote the local volume mass density of particles as @xmath31 and @xmath32 is its nebula average value , and the particle mass loading @xmath33 . for instance , figure 1 ( cuzzi et al 2001 ) shows the probability distribution function ( pdf ) for @xmath5 determined from numerical simulations at moderate reynolds number ( curves with symbols ) and predicted for plausible nebula reynolds numbers ( curves without symbols ) . note that the volume fraction having @xmath34 is quite substantial , which helps explain some mineralogical and isotopic properties of chondrules ( cuzzi and alexander 2006 ) . is different from the definition in cuzzi and alexander ( 2006 ) , which is equivalent to @xmath35 , where @xmath36 is the fractional abundance of solids to hydrogen gas by mass . thus values of @xmath37 in cuzzi and alexander ( 2006 ) are the equivalent of @xmath38 - a much more common occurrence than the situation we study here . ] not surprisingly , the volume fraction decreases for larger @xmath5 . in this work we focus on the less abundant , but higher mass density , concentrations towards the lower - right hand side of the pdf . the turbulent concentration pdfs of figure 1 did not allow for the feedback of the concentrated particles on the gas . this mass loading " would be expected to damp gas turbulent motions once the particle mass density significantly exceeds that of the gas , and lead to a saturation of concentration at some value . recent work which _ includes _ the feedback effects , through drag , of particle density on the damping of turbulence , has shown that values of @xmath39 can indeed be achieved - although less commonly than in the unloaded models such as shown in figure 1 . this _ cascade model _ of turbulent concentration ( hogan and cuzzi 2007 ) is summarized in appendix a. based on these results , we adopt @xmath40 as an upper bound in all stability and evolution modeling . the ability of gravity to overwhelm opposing forces is a well - known theme in astrophysics , dating back 80 years to early studies of star formation by jeans . over 30 years ago , gravitational instability ( gi ) of solids in a dense layer near the nebula midplane was proposed to lead directly to solid planetesimals on a dynamical timescale ( the orbit time ) ( safronov 1969 , goldreich and ward 1973 ) . traditional gi thresholds require the dynamical ( collapse ) time of a dense region @xmath41 , where @xmath42 is the gravitational constant , to be shorter than both the transit time due to random velocity and the local shear ( vorticity ) timescale ( toomre 1964 ) . these arguments imply that gi occurs when the local particle density exceeds a few times the roche density @xmath43 , where @xmath44 is the solar mass and @xmath45 is the distance from the sun ( safronov 1991 , cuzzi et al 2001 ) . however , this scenario is frustrated by self - generated midplane turbulence even if the nebula is not globally turbulent ( _ e.g. _ weidenschilling 1980,1995 ; cuzzi et al 1993,1994 ; johansen et al 2007 ) . that is , a dense layer of cm - size and larger particles orbits at a different velocity than the pressure - supported gas , leading to a vertical velocity shear which becomes turbulent and stirs the particle layer . a more recent flavor of gi requires the particles to be small enough that the particle - rich midplane acts as a single fluid , like fog , so that its vertical density stratification stabilizes it against self - generated turbulence ( sekiya 1998 , sekiya and ishitsu 2001 , youdin and shu 2002 , sekiya and takeda 2003 , youdin and chiang 2004 ) . because the particles must be very small to satisfy this _ one - phase _ requirement ( mm - size and smaller ) , their settling towards the midplane is frustrated by even the faintest breath of nebula turbulence ( @xmath46 ) . if the particles are prevented from settling into a sufficiently dense layer that their mass density can stratify the total density ( eg , dubrulle et al 1995 , cuzzi and weidenschilling 2006 ) , the conditions needed to trigger the instability are never achieved . in the following subsections , we study the evolution of a dense clump such as might form towards the lower right domain of figure 1 , albeit in a simplified way . we treat a clump as an isolated object when in reality it exists in a context of surrounding material of varying density . in section 4 , we discuss the implications of more realistic conditions . the role of the _ gas pressure _ in gi for nebula particles has been almost universally ignored , even though its importance was pointed out decades ago by sekiya ( 1983 ) . we encountered it ourselves independently . to test our gi thresholds , we initially ran numerical models of static , but very dense , clumps ( @xmath47 ) , expecting them to collapse with gravitational free - fall or dynamical collapse times @xmath48 and velocities @xmath49 , where @xmath50 is the clump diameter . we assumed a spherically symmetrical dense particle clump with density @xmath31 , embedded in gas with density @xmath24 , where @xmath51 . instead , only very slow shrinkage ensued . we suspected that the collapse was being artificially blocked by our incompressible code . we then developed a fully compressible 1d model which confirmed our incompressible calculations and clearly demonstrated a much higher , gas - pressure - dependent @xmath37 threshold for gi in the limit when @xmath52 . results from the model are shown in figure 2 . is like a mach number . nebula values of @xmath13 and @xmath24 lie in the top set of nearly linear curves , allowing only slow inward sedimentation on the timescale @xmath53 ( for @xmath54 ) . that is , under nebula conditions of gas density and temperature , and particle loading , the gas behaves as if it were incompressible and prevents dynamical timescale gi for particles with stopping time much less than the dynamical collapse time.,width=297,height=240 ] * figure 2 * shows results from this fully compressible model , illustrating how @xmath55 is required for traditional ( dynamical timescale ) instability to occur . the initial condition was a gaussian blob having density distribution @xmath56 . slow shrinkage of the particles through the gas is seen in the stable , incompressible " regime where nebula parameters lie , well within the flat curves at the top of the plot with @xmath57 for @xmath58 km , @xmath59cm / s , and @xmath60 g @xmath61 . the high value of @xmath62 required for true dynamical timescale collapse ( given the chosen value of @xmath40 ) would require smaller gas sound speed or larger clump size @xmath50 and gas density @xmath24 , by orders of magnitude . the slow shrinkage in the stable regime results from particles sedimenting inwards towards their mutual center under their own self - gravity at their terminal velocity @xmath63 , where @xmath64 is the local gravitational acceleration due to the clump s mass , and @xmath65 is the particle stopping time . thus @xmath66 for @xmath52 , and the sedimentation " timescale is @xmath67 roughly 30 - 300 orbit periods at 2.5 au for 300@xmath68 radius chondrules and @xmath69 ; that is , much longer than @xmath70 . it seems inappropriate to describe this slow ongoing evolution as an instability - certainly in midplane scenarios where it was slow vertical sedimentation by individual particles toward the midplane that produced the unstable " situation in the first place , and where only further slow sedimentation transpires . the situation is more akin to star formation mediated by ambipolar diffusion than by traditional gravitational instability . gas pressure even precludes the safronov - goldreich - ward gi in the form it was originally proposed , where cm - sized particles were envisioned ( although @xmath53 would be faster for literally cm - sized particles , and more closely approaches @xmath70 ) . the results of sekiya ( 1983 ) have apparently been overlooked by all subsequent workers , who have invariably assumed a toomre - safronov - goldreich - ward type _ particle _ ensemble which is decoupled from the gas , and stabilized ( on small scales ) only by particle random velocities . this gas pressure constraint is , however , crucial for particles in the chondrule size regime where @xmath71 an hour and @xmath72 a year , and must be applied to all gi models involving chondrule - sized particles . the good news is that for the interesting range of nebula gas density and particle concentration values , the gas _ does _ actually behave as if it were incompressible ( validating the use of an incompressible code in numerical models such as shown in section 3.4 ) . the physics is easily understood in the one - phase regime @xmath73 . the particles feel no direct pressure force , but are firmly trapped to the gas which does . the ( inward ) gravitational force is dominated by particles for @xmath74 . the particles drag and compress the gas with them as they begin to collapse under their self - gravity , producing a radial gas density gradient @xmath75 ; this translates into a large outward _ pressure _ gradient @xmath76 which acts on the gas , and thereby also on the strongly coupled particles . the gravitational force per unit volume is @xmath77 , where @xmath78 is the clump mass and @xmath79 its radius . after some incipient shrinkage and compression occurs , the outward pressure gradient force per unit volume becomes roughly @xmath80 , where @xmath13 is the gas sound speed . then for gi to be possible , @xmath81 , giving the criterion for gi as @xmath82 , where @xmath83 . @xmath84 is like an inverse mach number @xmath85 , where @xmath86 . an alternate ( perturbation ) approach compares the incremental changes @xmath87 and @xmath88 associated with a small gravitationally induced shrinkage characterized by @xmath89 . it is straightforward to show ( for @xmath90 ) that @xmath91 under nebula conditions . thus the clump is stiffened " by pressure and even a small incremental shrinkage is self - limiting . inspection of * figure 2 * shows small oscillations which suggest that a formal linear stability analysis might lead to an improved stability criterion ; this is , however , beyond the scope of the present paper . for @xmath92 a few @xmath93 km and @xmath60 g @xmath61 ( desch et al 2005 ) , @xmath94 . this criterion for gi can be compared with the traditional criterion for ( marginal , or maximal ) instability which is widely used as _ the _ criterion for gravitational instability , roughly twice the roche density @xmath43 ( goldreich and ward 1973 , safronov 1991 , cuzzi et al 1993 ) . the ratio of the two thresholds is : @xmath95 for @xmath96 to @xmath97 . for the midplane gis which have been widely discussed in the past ( reviewed by cuzzi and weidenschilling 2006 ) , @xmath98 km , so the traditional gi criterion falls short by a factor of more than @xmath99 ( _ cf . _ sekiya 1983 ) ! this degree of mass concentration is unlikely under any plausible circumstances , especially for particles which are already well trapped to the gas . for the ubiquitous dense clumps we discuss here , which form at all elevations , @xmath100 km . thus the pressure - supported gas phase prevents the tightly coupled particles from undergoing gi until the particle mass loading is hundreds of times larger than the traditional gi criterion ( which is on the order of the roche density @xmath101 ) . nevertheless , sekiya ( 1983 ) also noted that when @xmath31 is in the range usually cited for gi ( a few times @xmath101 , or @xmath102 a few hundred ) , a 3d incompressible mode " arises which , while not explicitly stated , we interpret as retaining an identifiable cohort of particles . an incompressible mode " might resemble a blob of water oscillating in zero - gravity . dense zones which result from turbulent concentration are candidates for such incompressible modes , even at mass loadings too low to induce actual gi . however , the fate of such slowly evolving entities in the presence of likely perturbations has never been explored . below we discuss the least avoidable , most pervasive disruptive perturbation which such clumps would encounter once they form , which we believe to be ram pressure disruption due to systematic velocity differences between the dense zones and the gas , and then derive conditions under which these clumps can survive to become planetesimals . several kinds of perturbation by the enveloping nebula gas might disrupt a dense clump before the slow sedimentation of its constituent particles towards their mutual center ( on the timescale @xmath53 ) can produce a moderately compact object . perhaps the simplest to discuss and dismiss are turbulent pressure , velocity , and/or vorticity _ fluctuations _ encountered by the clump . turbulent pressure fluctuations on the scale @xmath50 have typical intensity @xmath103 , where @xmath104 is certainly smaller than the fluctuating velocity @xmath105 of the largest eddy ( the dominant energy - containing eddy ) . the ratio of even the largest eddy pressure fluctuations to the steady ram pressure from the headwind @xmath106 ( @xmath45 is the semimajor axis ) is @xmath107 . since @xmath108 and @xmath109 , @xmath110 unless @xmath111 . nebula evolutionary timescales , and parameters leading to turbulent concentration of chondrule - sized particles in the asteroid belt region , imply @xmath112 depending on gas density and other properties ( cuzzi et al 2001 ) . this suggests that turbulent velocity fluctuations on clump lengthscales play a small role . we have run simulations of clumps settling in gravity with and without enveloping turbulence , and the typical evolutions are qualitatively similar . we next address local _ average _ vorticities on the lengthscale of a clump . it has long been known that local vorticity is a factor in gravitational instability ( toomre 1964 , goldreich and ward 1973 ) . because eddies much smaller than the largest scale of turbulence typically have larger vorticity than the keplerian shear commonly explored ( _ eg . _ toomre 1964 ) , this is a concern in principle . for example , in the inertial range , @xmath113 ( tennekes and lumley 1972 ) . then for @xmath100 km , @xmath15 and @xmath114 , @xmath115 . however , turbulent concentration has the property that dense particle zones preferentially lie in zones of locally _ low _ vorticity compared to the average at their lengthscale ( eaton and fessler 1994 ) . statistical studies ( hogan and cuzzi 2007 ) show that dense zones in high @xmath116 environments can form in regions with local vorticity 1 - 2 orders of magnitude smaller than the average value expected for that lengthscale ( see also figure 5 ) . below we sketch how such a constraint is derived and applied . a simplified requirement for gravitational binding of a clump of lengthscale @xmath50 and mass density @xmath31 , in the presence of local rotation at angular velocity @xmath117 , is @xmath118 . recalling that the local vorticity @xmath117 can vary by orders of magnitude from its average value @xmath119 , we normalize both sides of the above relationship between @xmath31 and @xmath120 by the average inertial range enstrophy @xmath121 , and define the normalized enstrophy @xmath122 . this quantity provides the horizontal axis on figure 5 . then the simple expression above can be rewritten as @xmath123 . inertial range relationships can be used to express @xmath124 where @xmath125 is the kolmogorov scale and @xmath126 is some scaling factor . making use of @xmath127 as discussed earlier , and substituting nebula parameters at 2.5 au , we find that gravitational binding requires @xmath128 . this constraint yields diagonal lines on plots of the sort shown in figure 5 ( for an example see cuzzi et al 2007 ) . interesting regions of parameter space remain accessible to a degree that depends on quantitative modeling of the pdf @xmath129 . obviously , more detailed numerical models of clumps in realistic turbulence are needed to assess these perturbations . we will assume , for the purposes of this study , that turbulent pressure ( and associated vorticity ) fluctuations are a minor influence and that regions which are stable to their own local vorticity exist ( see cuzzi et al 2007 ) . this leaves the dominant disruptive process , unavoidably shared by turbulent and laminar nebulae , as ram pressure disruption due to _ systematic _ velocity differences between the clump and the gas , as described below . incidentally , survival against ram pressure gives the horizontal stability boundaries in figure 3 of cuzzi et al ( 2007 ) . emplacing these constraints on pdfs obtained from cascade models is somewhat involved , and a full quantitative description of the situation is deferred to a future paper . here , for simplicity , we envision the evolution of a single , isolated dense clump with @xmath74 , merely as an example of one of the many dense clumps formed near the lower right hand corner of figure 1 . the clump exerts a strong collective influence on the entrained gas , and the mixture experiences solar gravity as a unit . it attempts to move as an individual large object relative to the nebula gas , dragging the entrained gas along . for instance , a clump forming at some distance above the nebula midplane settles downward under the acceleration of the vertical component of solar gravity , whereas the surrounding gas remains under vertical hydrostatic balance . also , nebula gas generally orbits at a slower velocity than the keplerian orbit velocity @xmath130 obeyed by massive particles . this is because of its generally outward radial pressure gradient force @xmath131 , where @xmath45 is the distance from the sun , and @xmath132 ) . thus , particles experience a headwind with magnitude @xmath133 a few @xmath93 cm / sec . these headwinds result in a ram pressure @xmath134 which can disrupt a strengthless clump . for example , a dense drop of ink settling in a glass of water is quickly disrupted by the rayleigh - taylor instability and mixes with its surroundings ( thomson and newhall 1885 ) . however , a dense drop of fluid _ with surface tension _ can avoid disruption and settle indefinitely at terminal velocity ( as do raindrops , or water in oil ) . in this situation , the criterion for stability of a fluid droplet , which determines its terminal velocity and thence its size , is given by the weber number @xmath135 , where @xmath136 , and @xmath137 is the surface tension coefficient . drops are disrupted when @xmath138 exceeds some critical value @xmath139 ( pruppacher and klett 1997 ) . dense particle clumps in the nebula have no surface tension , but they do have self - gravity . by direct analogy with the droplet surface tension criterion , we define a _ gravitational weber number _ @xmath140 as the ratio of the ram pressure force per unit area to the self - gravitational force per unit area for a flattened disk , which is the initial stage of a strengthless spherical clump of diameter @xmath50 and particle density @xmath141 upon encountering a headwind @xmath142 ( thomson and newhall 1885 ) . for a disk of surface mass density @xmath143 , the gravitational force per unit area is @xmath144 . then @xmath145 where @xmath146 is an effective drag coefficient for the clump , on the order of unity ( see appendix b ) . @xmath147 can be written in other ways as well . the orbit frequency at @xmath45 provides the useful relationship @xmath148 , where @xmath149 g @xmath61 at 2.5 au is different from safronov s ( 1991 ) roche density @xmath150 discussed earlier ] ; then @xmath151 by analogy with the more familiar surface tension case , there will be some critical _ gravitational _ weber number for stability @xmath152 , that is probably on the order of unity ( pruppacher and klett 1997 ) ; its value must be constrained by numerical experiments as described below . then we require for stability against headwind disruption that @xmath153 or @xmath154 these expressions determine the combination of clump diameter @xmath50 and particle loading @xmath37 that will stabilize it against a headwind of magnitude @xmath155 . the headwind due to nebula radial pressure gradient will occur even if a clump is at the midplane and has zero settling velocity ; this gives the lower limit on ram pressure that the clump must be stable against . neglecting vertical settling restricts potentially stable clumps to lying within about 0.01 gas scale heights of the midplane , where the headwind @xmath156 is comparable to the vertical settling velocity . this restriction must be factored into statistical estimates of planetesimal production ( see , e.g. cuzzi et al 2007 ) . for typical nebula parameters , we assume gas density to be in the range @xmath157 g @xmath61 ( a nominal minimum mass value ) to @xmath97 g @xmath61 ( a more recent value supported by nebula evolution and chondrule formation ; _ cf . _ desch et al 2005 ) , semimajor axis @xmath158 cm ( 2.5 au ) , pressure gradient / headwind parameter @xmath132 ( see , _ eg . _ , nakagawa et al 1986 , cuzzi et al 1993 ) , and solar mass @xmath159 g . in this regime , @xmath160 km . recall from section 2.3 and appendix a that mass loading limits the maximum achievable mass loading to @xmath10 . a clump satisfying the above constraint , with @xmath161 km , would have the mass of a 10 - 100 km radius body of unit density - thus , these precursors can lead to quite sizeable objects . this characteristic size range is intriguingly close to the roughly 50 km radius primordial building block size of bottke et al ( 2005 ) . the very existence of _ any _ preferred size for the primordial population , if true , is an intriguing result . to obtain a sanity check on the concepts of section 3.3 , and to constrain the ( unknown ) value of @xmath162 for this problem , we developed a very simplified numerical model of a clump experiencing a steady nebula headwind from a more slowly orbiting , pressure - supported gas , using hill s approximation which transforms a cylindrical system into a cartesian system rotating at some mean rate @xmath163 - useful if the domain covers a sufficiently narrow radial and angular region that the curvature is negligible . the overall orbital motion of particles ( the keplerian velocity ) is thus subtracted out , and the gas exhibits a small differential velocity because of its deviation from keplerian . a particle clump under influence of the gas will slowly lag behind in the @xmath164 ( orbital ) direction , and slowly drift to smaller @xmath165 ( radial ) values , due to the gas headwind . the model is not intended to be a high - fidelity representation of a realistic nebula situation , but merely to check the basic weber number model of section 3.3 . the code evolves the velocity field of a two - phase system . the gas is perturbed by a clump of particles , which are mutually attracted by gravitational forces and are themselves dragged by , and exchange momentum with , the gas . the hill frame , with cartesian coordinates @xmath166 , represents a radially narrow region corresponding to some range of radius @xmath45 , orbital angle @xmath167 , and vertical distance @xmath168 , respectively . periodic boundary conditions are used in all three dimensions . the instantaneous navier - stokes ( eulerian ) equations describing the conservation of mass and momentum for the incompressible gas are expressed in the hill frame as , @xmath169 and @xmath170 where @xmath171 is fluid velocity , @xmath24 is gas mass density , @xmath172 is gas viscosity , @xmath173 is local pressure ( which fluctuates along with gas velocity and vorticity variations ) , @xmath174 represents the ( constant ) global nebula radial pressure gradient , and @xmath175 is the kepler frequency vector . this setup causes the gas to flow towards the clump in the @xmath164 direction at a uniform speed @xmath142 . we are currently neglecting the keplerian radial shear term @xmath176 , where @xmath177 and @xmath178 , in the gas and particle equations to get a better comparison with our analytical stability model . the shear rate associated with this term @xmath179 , whereas a typical ( fluctuating ) local shear or strain due only to turbulent motions is @xmath180 . because the turbulent shear is so widely varying , and likely to be larger than the keplerian shear , in this study we neglect both of these complications , though they are both suitable avenues for future work . in our code , particles are followed in the lagrangian sense . the term @xmath181 represents the force per unit mass imparted to the gas by the particles . it has the general form @xmath182 where @xmath183 is the mean weighted particle velocity at a grid point , @xmath20 is the particle gas drag stopping time , and @xmath31 is the particle mass density . to obtain @xmath183 , a weighted sum is carried out over particles in the eight cell volumes adjacent to each fluid grid point . weighting functions are used which vary inversely with the distance of the particle from the grid point ( squires 1990 , his section 5.1 ) . because of the periodic boundary conditions , the wake of a clump can impinge artificially on the clump from the upwind direction . to avoid this , we include the term @xmath184 in equation ( 6 ) as a `` sponge force '' per unit mass to restore the gas velocities downstream of the clump at the outflow boundary plane to their initial values @xmath185 ; this ensures constant inflow values at the upstream boundary plane . the sponge force has the form @xmath186 where @xmath187 is some time scale . the function @xmath188 is modeled as a sigmoid in the @xmath164 direction very near the outflow boundary . the parameters of the sponge were determined through test runs with the goals of minimizing the sponge s spatial extent and maximizing its effectiveness without introducing numerical instabilities . once determined , the same values were used for all production runs . the ( lagrangian ) equation describing the motion of particle @xmath189 subject to the forces of gas drag and mutual gravity is ( again neglecting the keplerian shear term ) @xmath190 the term @xmath191 describing the drag force per unit mass on particle @xmath189 by the gas takes the form @xmath192 where @xmath193 is the gas velocity interpolated to the particle s position @xmath194 at time @xmath195 . finally , the mutual gravity force per unit mass on each particle is @xmath196 where @xmath42 is the gravitational constant , and @xmath197 is the number of particles . @xmath198 is the unit vector associated with the distance @xmath199 between particles @xmath200 and @xmath189 , and @xmath201 is a constant that softens the force at small separations to prevent numerical singularities , chosen to be comparable to a grid cell in extent . ( 6 ) and ( 9 ) are solved using psuedo - spectral methods commonly used to solve navier - stokes equations for a turbulent fluid ( canuto et al 1987 ) . periodic boundary conditions are assumed and the number of nodes used in each direction is generally 128 , 256 , 128 with a spacing of @xmath202 . a fast fourier transform ( fft ) algorithm is used to evaluate the dynamical variables @xmath171 at the computational nodes . a second - order runge - kutta scheme is used to time - advance the gas and particle velocities . a third - order taylor series interpolation scheme is used to determine gas velocities at the particle positions from values at the eight nearest neighbor nodes . the mutual gravity calculation is done in a brute force fashion with a @xmath203 algorithm to evaluate the force contributions from all particles . the code is written in fortran 77 and is parallelized using openmp directives . the fft calculations are done simultaneously on the planes perpendicular to the @xmath164 and @xmath204 axes , and all loops involving particle indices are parallelized . typically , 3000 superparticles " are used , but some runs used more than 20000 . each superparticle " represents the dynamical effects , and typical response , of a large number of actual chondrules . initially the particles are arranged in a uniform spherical clump . the initial velocities for the gas and the particles inside and outside the clump are determined from the expressions in nakagawa et al . ( 1986 ) : @xmath205 where @xmath206 . the initial velocity of the gas outside the clump is found by setting @xmath207 in equation ( 12 ) : @xmath208 ; inside the clump the specific case value of @xmath37 is used in equations ( 12 ) and ( 13 ) to initialize the gas and particle velocities . runs with single particles of different @xmath20 were made to verify that the nakagawa et al . ( 1986 ) initial conditions are steady state solutions for the particle equations . the wallclock time to evolve one integration step is 0.7 sec using 256 intel itanium 2 processors running at 1.5 gighertz . the wallclock time to evolve 1 orbital period is 121 hours . * figure 3 * shows how , as predicted by the simple theory of section 3.3 , certain combinations of clump mass loading @xmath37 and dimension @xmath50 remain stable against ram pressure disruption for nebula headwinds produced by a pressure gradient characterized by @xmath209 . @xmath135 is dimensionless , and the second expression in eqn . ( 2 ) can be used to obtain the critical value of @xmath140 which separates stable from unstable configurations of the clump , @xmath210 , directly from measured values in code units ( see appendix b ) . we expect that @xmath162 will be in the range 1 - 10 . the combination of parameters in the stable case shown ( right panels ) gives @xmath211 ; if self gravity is turned off , the clump is disrupted in @xmath212 orbits , whereas it will sediment into itself on a timescale @xmath213 times as long ( see appendix b ) . in the stable cases , a dense core is seen to continually shrink and become denser throughout the run , even as material is shed from the periphery of the clump , such that the value of @xmath214 for the core continues to increase ( see appendix c , * figure 7 * ) . ) ; bottom row : projected onto the @xmath215 plane ( @xmath216 projections are similar ) , with the gas inflow into the page . in the top left panel the open circle shows the initial size and position of the clump . after less than an orbit , ram pressure disrupts the gravity - free clump ( left ) but the gravitating clump is stable ( right ) and continuing to shrink inexorably , even while being eroded . movies showing this evolution are available online ( see appendix c).,width=326,height=288 ] the numerical clump models exhibit considerable erosion over the duration of the runs from a viscously stirred surface layer , and this gradual but inexorable mass loss might lead to their disruption if arbitrarily long runs were practical at this time . however , as discussed in appendices b and c , this large amount of erosion is a gross overestimate of what would happen in the nebula , because numerical viscosity ( and consequently shear around the periphery of the clump ) plays a far larger role in the current numerical models that it ever would in the actual nebula . that is , the viscously stirred and eroded surface layer in the numerical clumps contains orders of magnitude more mass than would be the case for a nebula clump of the size and density required to survive ram pressure disruption ( see appendix b ) . in appendix c , we describe the movies from which * figure 3 * is taken and also show how the protected inner regions of the clumps are behaving exactly as predicted , both in stable and unstable regimes . for more realistic , larger clump @xmath217 in the nebula , a far larger fraction of the clump mass will show this behavior rather than being artificially eroded . the primary purpose of the numerical models is merely to provide an independent sanity check on the physics of our weber number model and obtain some idea of @xmath162 , which gives us an order - of - magniture estimate for the product @xmath214 which nebula clumps require to survive headwinds of a particular @xmath218 . future coding improvements are needed to allow larger ( higher @xmath217 ) clumps which would more closely approach nebula conditions in terms of their balance between pressure forces and viscous forces ( see appendices b and c ) ; these improvements will probably include an implicit time advance for the drag terms and perhaps also a tree or particle - in - mesh code for the particles . we have shown how self - gravity can stabilize dense clumps of mm - sized particles , which form naturally in 3d turbulence , against disruptive gas ram pressure on timescales which are sufficiently long for their constituent mineral grains to sediment towards their mutual centers and form physically cohesive sandpiles " of order 10 - 100 km in size . the essence of the result is a critical gravitational weber number " on the order of unity , in which self - gravity plays the role of surface tension in more familiar situations such as raindrops . characteristic mass densities and lengthscales are determined which meet this requirement . we show numerical results which are in general agreement with the predictions of the simple theory , for isolated , spherical initial clumps . the scenario we have sketched out leads from aerodynamically size - sorted nebula particles , having the properties of chondrules , to sizeable planetesimals formed entirely from such particles which contain a snapshot or grab - sample of the local particle mixture . turbulent concentration first produces the dense zones of size - sorted particles . the more common , less dense of these regions may provide typical chondrule melting environments ( cuzzi and alexander 2006 ) . some of the less common , very dense zones , which we have shown elsewhere can achieve mass densities 100 times larger than that of the gas , have the potential to become planetesimals depending on their lengthscales , nebula location , and local vorticity . it is intriguing that our characteristic stabilized clump masses are not far from the mass inferred by bottke et al ( 2005 ) to represent a typical primordial object in the pre - dispersal , pre - erosional asteroid belt . ultimately , the sandpiles resulting from completed sedimentation will become compacted further by inevitable collisions with other sandpiles , leading to today s fairly dense asteroids , while retaining a physical and chemical memory of their parent particle clumps . the mechanism is easily extended to the unmelted aggregate particles of the outer solar system , where , however , if chondrules are absent , the size - sorting fingerprints of the process ( cuzzi et al 2001 ) might be less evident . the conclusions of this paper differ from the suggestions in cuzzi et al ( 2001 ) , who focussed on the possible role of ultra - dense clumps with size comparable to a kolmogorov scale ( 0.1 -1 km ) . since that time , hogan and cuzzi ( 2007 ) found that particle mass loading saturates the value of @xmath51 at a value of about 100 , redirecting our attention to larger clumps in the @xmath219 km size range . clumps and fluid structures of these large sizes are much more accessible with numerical codes , both of the standard direct simulation type and the cascade type ( appendix a ) . in this scenario , primitive bodies may not primarily represent spatial , but perhaps temporal , samples of the particulate contents of the nebula as its chemical , physical , and isotopic properties evolve over several myr . the temporal , as well as the spatial , variation in the physical , chemical and isotopic properties of the concentrated particles , which are being continually altered by thermal events and mineralogical alteration in the nebula gas , can then help account for class - to - class variations between the chondrite groups . an implication of this drawn - out , inefficient process is that younger chondrite types should contain evidence of leftovers " or refugees " from earlier times , which escaped primary accretion . several aspects of chondrite makeup are compatible with this implication . of course , it is well known that ancient cai minerals and less - ancient , less refractory amoeboid olivine aggregates are found alongside much younger objects in the same chondrites ( _ eg . _ scott and krot 2005 ) . also , the oldest chondrites ( cvs ) contain primarily type i chondrules ( which have an age observationally indistinguishable to that of cais ) and no ( more oxidized , heavier o - isotope ) type ii chondrules , while the younger ordinary chondrites contain a mixture of type i and type ii chondrules , some with hints of age variation even within a given chondrite ( kita et al 2005 ) . the oldest and first - formed objects are all likely to have melted from their abundant @xmath1al ( kleine et al 2005 , 2006 ; hevey and sanders 2006 ) , producing differentiated achondrites and metallic objects , so it is no longer possible to dissect their primordial components . of course , the real world is more complicated . some complications are meteoritic : as only one example , co chondrites and ordinary chondrites appear to be about the same age ( younger than cv chondrites ) , but have very different chemical and isotopic properties ( kita et al 2005 ) , which would , in the context of this scenario , argue for some degree of spatial ( radial ? ) variation in the makeup of nebula particulates at that time anyway . on the theoretical side , of course , self - consistent numerical models which follow clump evolution in realistic turbulence with headwinds and vertical settling need to be pursued , to check these preliminary assessments which make simplistic assumptions about the local headwind and assume isolated clumps . several recent studies have found that transient pressure ridges " arise in the gas in association with spiral density waves or vortical structures ( haghighipour and boss 2003 , rice et al 2004 , johansen et al 2007 ) ; in such regions , the local radial pressure gradient and headwind diminish ( other regions will have unusually _ large _ headwinds ) . moreover , clumps do not exist in isolation , but more realistically as a dense core within a larger , less dense envelope which weakens the headwind felt by the dense clump core relative to the nebula average we characterize by @xmath209 . both of these conditions would allow some clumps to survive with smaller @xmath214 . other complications include interactions between different strengthless objects , leading to mergers or disruptions . our stability criteria are most likely to be satisfied in a region fairly close to the nebula midplane ( @xmath220 , see section 3.3 ) , because at higher altitudes the vertical component of solar gravity leads to vertical settling of dense clumps at higher terminal velocities than are easily stabilized against . however , this vertical settling , even if it leads quickly to disruption of individual clumps , enhances the downward transport rate of small particles and increases the solid / gas ratio closer to the midplane from that generally predicted by simple 1d diffusion models ( _ eg . _ , bosse et al 2006 ) . all these effects have implications for the occurrence frequency of clumps of the requisite @xmath214 for stability . quantitative statistical determinations of the volume fraction of stable clumps which can become planetesimals at any given time will use , for instance , approaches such as the cascade models described in appendix a. simplified , preliminary work exploring these factors indicates that , while the volume fraction of stable clumps is low at any given time ( so the process is clearly not an efficient one ) , accretion rates are roughly an earth mass per myr in the asteroid belt region ( cuzzi et al 2007 ) . however , considerable work remains to establish the statistical formation rate of appropriate clumps capable of following this evolutionary path and to study the evolution of clumps in realistic turbulence , with variable headwind velocity and simultaneous vertical settling . we thank j. eaton , m. gaffey , p. goldreich , w. hartmann , g. laughlin , k. sreenivasan , k. squires , and a. wray for very helpful conversations on various aspects of this research , and a. wray , a. dobrovolskis , p. garaud , and s. weidenschilling for thorough reviews of the original manuscript and a number of useful comments which have been included . n. turner and a. carballido also provided useful comments on an earlier version . this work was supported by a grant to jnc from nasa s planetary geology and geophysics program . generous grants of cpu time from nasa s hec program were essential to the progress of this research ; we thank e. tu , k. schulbach , c. niggley and r. pesta in particular for their help along these lines . we also thank j. chang for coding optimization assistance . this aspect of our work incorporates two related and important concepts : that of _ intermittency _ , and that of a statistical cascade process . for instance , it is widely known that dissipation of turbulent kinetic energy occurs at the kolmogorov scale ; it is less widely known that the spatial distribution of dissipation , like that of the particle concentration factor @xmath5 , is highly _ intermittent_. intermittent quantities are spatially and temporally unpredictable , and fluctuate increasingly with increasing reynolds number @xmath116 . however , the statistical properties of intermittent quantities like @xmath5 ( their probability distribution functions or pdfs ) are well determined on any lengthscale . in what follows , it will be necessary to distinguish between two closely related quantities : the concentration @xmath5 and the mass loading @xmath221 . this is because the emergence of particle feedback on the gas due to mass loading ( @xmath37 ) , depends both on @xmath5 _ and _ on the initial value of @xmath32 . it is ultimately @xmath5 that is evolved in the cascade , but _ how _ it evolves depends on @xmath37 . so both quantities will be alluded to in parallel . ; the parameter @xmath222 depends on the local mass loading @xmath37 ; larger @xmath222 indicates a narrower pdf @xmath223 . open symbols : for particle concentration @xmath5 ; filled symbols : for enstrophy ( @xmath224 ) . the small insets show @xmath223 at several values of @xmath37 . from hogan and cuzzi ( 2007).,width=297,height=240 ] from the conditioned cascade model , for a 24-level case with initial @xmath225 . the curves of figure 1 are the @xmath226averaged equivalent of this plot ( and show @xmath5 instead of @xmath37 ) , but do not include the effects of mass loading . note the flattening or saturation at @xmath10 , reflecting the near - vertical asymptote of the conditioned-@xmath222 curve of figure 4 at @xmath10 . a 24-level cascade is in the plausible range for clumps of size @xmath227 km in a nebula with @xmath228 . in this figure , the enstrophy @xmath229 is normalized to its mean value for the binning lengthscale ( from hogan and cuzzi 2007).,width=297,height=230 ] in its inertial range , which is extensive at high @xmath116 , turbulence is a scale - free process that is often referred to as a cascade and inspired the approach of cascade modeling ( meneveau and sreenivasan 1991 ) . the physics of transport of kinetic energy , vorticity , and dissipation from their sources at large scales is independent of scale until viscous processes enter at the smallest ( kolmogorov ) scale . scale - independence and @xmath230independence are connected because @xmath116 determines the depth of the inertial range : @xmath231 ( tennekes and lumley 1972 ) . cascade models simplify this complex , nonlinear , 3d scale - free process into a set of partition rules which are independent of scale , or of _ level _ in the cascade . larger @xmath116 means a larger number of eddy spatial bifurcations ( or levels ) between @xmath232 and @xmath125 , and stronger fluctuations in intermittent properties ( meneveau and sreenivasan 1991 ) . cascade models do nt preserve all the information of full 3d models ( such as the tubelike spatial structures characterizing vorticity ) but they do reproduce the statistical properties ( the pdfs ) , which are of primary importance for our problem . the successful use of such models to describe the intermittency of dissipation , and other quantities , in turbulence ( juneja et al 1994 , sreenivasan and stolovitsky 1995 ) led us to pursue a cascade model for turbulent concentration of particles . a cascade model consists of a set of multipliers " - fractions which define how a quantity of interest is unequally partitioned from a fluid parcel at some level , into its equal - sized subdivisions at the next level . the pdf of particle concentration factor @xmath5 is determined over all the numerous end - level outcomes of applying a multiplier @xmath233 ( chosen from its own pdf @xmath223 ) to each volume element as it bifurcates at each level of the cascade . it has been observed that @xmath223 is independent of level in the turbulent inertial range ( sreenivasan and stolovitsky 1995 ) . hence , multiplier pdfs determined from a direct numerical simulation ( dns ) at low @xmath116 ( with a relatively small number of levels ) , merely applied repeatedly over additional levels , can _ predict _ the properties at higher @xmath116 ( if nothing changes in the physics ) . our cascade model ( hogan and cuzzi 2007 ) uses a common form for @xmath223 : @xmath234 , where the parameter @xmath222 determines the width of @xmath223 ( sreenivasan and stolovitsky 1995 ) . both the concentration @xmath5 and the vorticity @xmath120 are described by their own @xmath222-distributions . multiplier pdfs with small @xmath235 are broad , and in them multipliers much different from @xmath236 occur with higher probability , producing a grainier , more intermittent spatial distribution . pdfs with large @xmath237 are narrow and centered on @xmath236 ; because each eddy bifurcation " then involves a nearly 50 - 50 partitioning , the resulting spatial distribution is nearly uniform and subsequent growth of @xmath37 becomes more difficult . the cascade model can then be used to address the issue of how particle mass loading can affect the cascade - that is , change the physics as the cascade progresses . this dependence is known as _ conditioning _ of a cascade . we found that the process can be represented quite well as two separate _ one - phase _ conditioned cascades for @xmath5 and @xmath120 in which multiplier distributions @xmath223 for both @xmath5 and @xmath120 depend only on the local particle mass loading @xmath37 . we used our full 3d numerical simulations to establish how @xmath222 for both these properties depends on local particle or fluid properties . * figure 4 * shows conditioning curves for both @xmath5 and @xmath120 ( the latter expressed as enstrophy @xmath224 ) , as extracted from multiple 3d dns simulations at various @xmath116 as large as 2000 ( hogan and cuzzi 2007 ) . note how the multiplier pdfs extracted from the dns results ( shown using insets ) get narrower ( have larger @xmath222 ) as mass loading increases , choking off intermittency . in these runs , @xmath238 , so @xmath239 . these results establish the upper limit @xmath10 which can be obtained by turbulent concentration , regardless of the initial value of @xmath240 . in the nebula , with a smaller initial @xmath240 , higher @xmath116 or deeper cascades ( and larger ensuing @xmath5 values ) are needed to reach this saturation point . to determine the joint pdf of concentration @xmath5 and vorticity @xmath120 , we require not only the conditioned multipliers for @xmath241 and @xmath5 ( * figure 4 * ) , but also their spatial correlation . hogan and cuzzi ( 2007 ) found , on average , a 70 - 30 preference for anticorrelation at each partitioning , consistent with previous observations that particle concentration zones avoid zones of high fluid vorticity ( squires and eaton 1990 , 1991 ; eaton and fessler 1994 ; ahmed and elghobashi 2000 , 2001 ) . including this partitioning asymmetry factor as a weighted coin gives us a two - phase ( particle - gas ) , conditioned cascade model which shows very good agreement with the joint pdf @xmath242 directly determined from our full 3d mass loaded simulations ( see hogan and cuzzi 2007 ) . however , to match the full 3d dns simulations at @xmath243 , the cascade model needs only about 15 levels , taking about 10 cpu - hours ( for 1024 realizations ) compared to over 90000 cpu hours to converge our full 3d simulation . using these conditioned multipliers and asymmetry factor , entire pdfs of particle concentration ( mass loading ) and vorticity can be generated for arbitrary numbers of levels ( arbitrary @xmath116 ) . for example , the 24 level model of * figure 5 * ( from hogan and cuzzi 2007 ) is of direct relevance for @xmath219 km scale dense zones of interest for the nebula . * figure 5 * illustrates the saturation of the particle concentration at @xmath244 and the fact that high @xmath37 occurs preferentially at low enstrophy or vorticity . the upper left quadrant is of most interest for determining the _ numbers _ of clumps which can survive to become planetesimals ( see cuzzi et al 2007 for more details ) . while the primary result of this paper - the existence of some stability regime determined by a critical gravitational weber number @xmath162 - can be expressed in a nondimensional way , some aspects of the results ( specifically the numerical results and the inferred value of @xmath214 for the nebula ) require us to pursue the relationship between nebula parameters and code units more deeply . values in code units are denoted by primes below . the fundamental quantities in the problem are ( 1 ) the gas and particle densities @xmath24 and @xmath141 ; ( 2 ) the clump diameter @xmath50 ; ( 3 ) the local orbit frequency @xmath245 where @xmath45 is the distance from the sun ; ( 4 ) the local radial pressure gradient @xmath246 , where @xmath165 is in the hill frame ; ( 5 ) the particle stopping time @xmath20 ; ( 6 ) the velocity of material in the hill frame @xmath247 , that is , relative to keplerian , where @xmath247 represents either the particle or gas velocity ; and ( 7 ) the gravitational constant @xmath42 . the principal forces are the coriolis force due to the rotating frame @xmath248 , the pressure force @xmath249 , and the gas drag forces that couple the gas and particles ( section 3.3 ) . the code unit of length is radians ( the domain is 2@xmath250 radians on the short dimension ( @xmath165 or @xmath204 ) and the grid cell size is therefore @xmath251 where @xmath252 is the number of grid points along that axis ) . the code time unit ( ctu ) is @xmath253 where @xmath254 is the code value of rotation frequency ; thus the orbit period is @xmath255 in code time units or ctu . we treat as equivalent @xmath256 . as noted in section 3.3 , @xmath257 where @xmath149 g @xmath61 at 2.5 au ( note this is different from safronov s ( 1991 ) roche density @xmath150 ) . to normalize mass and mass density ( and ultimately establish the code value of the gravitational constant @xmath258 ) we assume a nebula gas mass density @xmath259 g @xmath61 ( minimum mass nebula ; cuzzi et al 1993 ) at 2.5 au ; then code quantities @xmath260 and @xmath261 , and @xmath262 . also , in code units @xmath263 where @xmath264 is the actual orbit frequency at the region of interest . then the gravitational constant in code units , @xmath258 , can simply be written as @xmath265 . finally , the pressure force is written as @xmath266 where @xmath267 is typical ( nakagawa et al 1986 ; cuzzi et al 1993 ) , @xmath268 . then the dimensionless parameters @xmath37 and @xmath269 are set to satisfy various constraints of the code ( resolvable clump , adequate particle statistics , reasonable timescale , etc ) and @xmath270 is allowed to vary , to determine empirically the critical value of @xmath271 as the largest value of @xmath140 that remains stable . that is , we want @xmath269 to be large enough such that the clump is well resolved , but not so large that it fills the entire cross section of the computational box . typically @xmath269 is only 10 - 20 grid cells so far , and as noted below , this exaggerates the viscous stresses and surficial erosion of material . we want @xmath37 to be large enough that the clump is much denser than the gas , and we want a large enough number of particles to act like a continuum . the value of @xmath20 should ensure that the clump sedimentation time @xmath53 is larger than its ram pressure disruption time @xmath272 ( see below ) , to provide a real test that self - gravity is preserving the clump rather than simple collapse . the dynamical collapse time of the clump under its own self - gravity , and in the absence of gas pressure , is @xmath273 , or in code units @xmath274 . the mass loading which produces a clump dynamical time comparable to the orbit time is then obtained using @xmath275 or @xmath276 , again in fair agreement with prior expectations . however , gas drag prevents clump collapse on this timescale , as described in section 3.1 , and actual shrinkage takes a time @xmath53 ( eqn . the number of orbits over which we need to follow the clump to ensure it really survives is roughly @xmath277 where we have used @xmath278 . this expression can be assessed using real quantities , and is 30 - 300 orbits for particle radius 300@xmath68 and mass loading @xmath279 . this behavior ( @xmath280 ) can be controlled in the code , once the other parameters are established as above , by adjusting the particle stopping time @xmath20 . because of long code run times , at present we are limited to stipulating only that the sedimentation time significantly exceeds the nominal ( self - gravity - free ) disruption time ; that is , @xmath281 , where @xmath282 this expression for the disruption time @xmath272 is derived from the ram pressure force ( sect 2.3 ) by setting the distance traveled in time @xmath195 by the windward half of the clump , under acceleration by the ram pressure force , equal to the clump diameter ( assuming the leeward half of the clump does nt get accelerated ) , and solving for @xmath195 . the value of @xmath146 for our clumps is on the order of unity , as we verify below . the code calls for some input viscosity @xmath283 ; with it , the reynolds number of the clump having diameter @xmath269 radians , in a headwind of speed @xmath284 ( in code units of radians per ctu ) , is @xmath285 where @xmath286 is the defined code viscosity ( radians@xmath287/ctu ) . we assume an inertial range expression within the wake of the clump to determine the wake s kolmogorov scale : @xmath288 , and thus @xmath289 radians . our highest resolution runs to date had a gas relative velocity @xmath290 38 radians / ctu and used a clump size @xmath291 radians . the nominal kolmogorov scale associated with this flow is @xmath292 0.02 radians ( compared to the grid cell size of 0.05 - 0.1 radians ) ; the wake turbulence is thus under - resolved and the code s true viscosity is numerical : @xmath293 ( at the boundary of the clump ) . while we are not concerned with the fine - scale details of the wake , this marginal resolution introduces some caveats . the ratio of ram pressure force to viscous force ( both per unit area ) is , in general , @xmath294 where @xmath217 is the reynolds number of a clump , and we have approximated @xmath295 . because we are dominated by numerical viscosity , @xmath293 , thus in code units @xmath296 in many cases so far . for comparison , the reynolds number for _ nebula _ clumps of interesting sizes is @xmath297 . this means that viscous stresses around the periphery of the clump , and the fractional depth of the viscous boundary layer , are grossly exaggerated in the numerical model relative to the actual nebula regime of interest . the anomalously large role of ( numerical ) viscosity leads to anomalously large erosion " from the surface of our numerical clumps , and degrades our results in the sense that potentially stable clumps _ might _ appear less stable , due to the erosive mass loss experienced over their sedimentation time . going to higher resolution cases ( @xmath298 ) significantly ameliorated this effect but the problem has certainly not been entirely removed . another way to characterize the degree of artificial erosion is to estimate the _ physical _ kolmogorov scale for the actual nebula / clump situation : for @xmath299 cm sec@xmath300 , @xmath301 km , and @xmath142=3800 cm / s , we find @xmath302 100 m , or @xmath303 of the size of a @xmath99 km clump , compared to a fractional size of perhaps @xmath304 or so in the code , as determined by numerical viscosity . clearly , surface erosion is a much smaller effect in the nebula case than in our crude models . while even at this resolution apparently stable configurations can be found , it would be desirable for future studies to improve on this situation . _ clump drag coefficient @xmath146 : _ the drag coefficient of the clump is reynolds - number - dependent and must also be scaled to nebula values to obtain an estimate of @xmath214 . weidenschilling ( 1977 ) gave expressions for the drag force per unit area ( giving the pressure force per unit area ) , experienced by solid spheres of different @xmath217 in the stokes drag regime of interest here , as @xmath305 . for @xmath306 , @xmath307 , or @xmath308 for @xmath309 , as in our code . the large drag coefficient and ram pressure is partly due to low - pressure zones set up immediately behind particles in this range of @xmath217 . by comparison , for a solid sphere with @xmath310 as under nebula conditions , @xmath311 . however , neither of these values might be entirely applicable to our clumps , which are not rigid spheres . in practice , we have determined @xmath146 empirically from observed values of @xmath272 with self gravity turned off . recalling that @xmath312 , we simply plot our estimate of @xmath272 , the point at which sizeable vortex - driven internal voids appear in the clump , against the combined parameter @xmath313 , giving @xmath314 as the slope of the best fit straight line ( * figure 6 * ) . it seems that @xmath315 is a good assumption , but this calculation should be redone with higher resolution ( lower numerical viscosity ) codes . at nebula @xmath217 , @xmath146 could reach its theoretical high-@xmath316 value of 0.4 and this is what we assume in deriving @xmath214 . finally then , the several stable cases we have run ( _ eg . _ figure 3 ) can be characterized by @xmath317= 38 rad / ctu , @xmath318 = 380 ( which corresponds to a minimum mass nebula local density @xmath319 at distance 2.5 au ) , @xmath320 , and @xmath321 radian , and while we have not as yet accurately determined the transition value of @xmath140 , a value @xmath322 is clearly stable . solving equation ( 4 ) of section 3.3 for @xmath214 under _ nebula _ conditions , _ assuming _ @xmath323 , taking @xmath324 for a high-@xmath217 situation and assuming a nominal @xmath325 , we find @xmath326 , independent of @xmath24 and neglecting a questionable coefficient of order unity ( because of the contribution of viscous stresses in our low-@xmath217 code ) . foreseeable refinements ( specifically , lower numerical viscosity ) may increase @xmath162 ( the surface tension analogue gives @xmath327 ) and thus decrease the value of @xmath214 . _ vs. _ the quantity @xmath328 ; the slope is then @xmath314 . this determination is slightly qualitative as to when we judge the clump to be disrupted ; we decided to use the time at which vortex - driven voids first appear in a @xmath215 planar cross - section through the nominal center of the clump . , width=297,height=268 ] quicktime movies showing our time evolutions are available online at with filenames as given below . after publication , the movie files will also be available on the apj web site as mpgs . each shows a dense clump in the hill frame , rotating at local keplerian velocity , with more slowly - orbiting gas impinging on the clump from its leading side . the two left - hand panes show the view from along the radial axis ( in the @xmath329 plane ; top ) and along the vertical axis ( in the @xmath216 plane ; bottom ) . the large right - hand pane shows the view from the orbitally leading direction ( in the @xmath330 plane ) . the wake formed by the clump is easily visible in the two left - hand panes . cases @xmath331 and @xmath332 are sampled at a similar time near their end - point in figure 3 . file @xmath331 shows how a clump with no self - gravity is disrupted in about 1.5 code time units for the initial conditions chosen ( the orbit period is @xmath255 code time units ) , in agreement with our simple estimates of @xmath272 . files @xmath332 and @xmath333 are intended to illustrate the same ( stable ) weber number case ( @xmath322 ) but we vary two constituent parameters of @xmath140 ( we decrease both the gas velocity and mass loading by a factor of three ) to illustrate similarity ; because of the specific parameters chosen , @xmath333 runs for a longer time , but is suffering considerable erosion towards the end of the run . nevertheless , as discussed below , it retains a dense core that continues to contract . as discussed in appendix b , more realistic cases , with higher numerical resolution , would have smaller numerical viscosity and incur less erosion . case @xmath332 , while not run as long as @xmath333 , appears solidly stable with a dense core that is continuing to shrink at the end of the run . note some interesting oscillatory behavior , such as hinted at in our 1d compressible runs ( figure 2 ) . the simulations which created the movies referred to above can be used to assess the behavior of the central , dense regions of each clump in a quantitative manner . * figure 7 * has three panels , showing the time variation of some properties of the particles which lie within the central regions of the clump ( its densest fraction , as sampled out to some cumulative mass threshold ) . in the top panel , we show how much mass is being eroded from the clump overall ( particles moving faster than some comoving velocity threshold are assumed to have left the clump ) . all three clumps are losing mass , naturally , but the non - gravitating clump 26 is losing it much more rapidly than the gravitating clumps 29a and 31a . in the central panel we show the mean mass density as measured over some fraction @xmath334 of the clump mass lying at the highest densities ; in clumps 26 and 29a , we chose @xmath335 , and in clump 31a , we chose @xmath336 . in the nongravitating clump , the core density never increases and it quickly blows up , while the central regions of the gravitating clumps ( removed from the artificial mass erosion at their peripheries ) inexorably get denser at a steady and perhaps even slightly increasing rate ( an increasing rate is predicted by the simple model ) . the lower panel shows the effective average radius of the region being sampled . the non - gravitating clump never shrinks at all , but both gravitating clumps do . moreover , as the gravitating clumps shrink , their density increases faster than their linear dimension decreases , so the product @xmath214 increases - causing them to become _ more _ stable as time goes on . this is an argument for stability even in the presence of erosion . as expected , the density growth timescale ( the sedimentation time @xmath53 of equation 1 ) is three times smaller for the denser clump of case 29a ( middle panel ) . meanwhile , the apparent agreement of the variation of the fractional radius containing half of the mass seen between cases 29a and 31a in the lower panel is fortuitous , because the total masses of the clumps are changing ( and eroding ) at different rates and from different depths in the two cases . nevertheless , the cores are both shrinking and becoming denser in the two stable cases . the amount of erosion that is incurred from the margins of each clump is related to ( perhaps proportional to ) the thickness of the viscous boundary layer relative to the radius of the clump , which can be expressed in terms of the reynolds number . as we have argued above in this appendix , actual nebula clumps capable of remaining stable would have reynolds numbers five orders of magnitude larger - impossible to study with numerical models . the relevance of all viscous boundary layer effects , including the erosion that inflicts our current numerical runs , will decrease by that factor in the real situation . ; bottom panel : effective radius containing mass fraction @xmath334 . these results show that , away from the artificially perturbed perimeter regions , @xmath147-stable clumps are behaving as expected by the simple model , and @xmath147-unstable clumps do not show this behavior.,width=316,height=316 ]

infrastructure as a service ( iaas ) systems @xcite build on virtualisation technologies to allow automated infrastructure provisioning . virtual machine ( vm ) based provisioning gives users two major benefits : they do not need to be experts in physical infrastructure maintenance , and they can easily follow their demand patterns and scale their virtual computing infrastructure ( composed of several vms ) with tools built on top of iaas systems . these two benefits led to the wide and rapid adoption of such infrastructure offerings . unfortunately , their rapid adoption has led to infrastructures that still have plenty of open research issues ( e.g. , energy aware vm placement , service level objective specifications ) . however , even iaas systems operated by academia are used in production nowadays . as production level systems are used by a multitude of users on a daily basis , changing the internal behaviour of such systems might hinder their user experience ( such as reliability and usability ) . thus , research focused on improving the internals of iaas systems ( e.g. , introducing a new experimental virtual machine placement algorithm ) can not be done on such production systems directly . consequently , to analyse new and novel ideas for internal behaviour , researchers are either restricted to severely limited iaas deployments ( e.g. , rarely utilising more than a few hosts ) or should resort to theoretical modelling of expected internal behaviour . however , results based on such research is often questioned by the operators of production clouds because their applicability to large scale systems is often not proven . some researchers use simulators to further evaluate their models @xcite . these simulators allow researchers the evaluation of new ideas in life - like scenarios and as a result such simulators could pave the way for the new research results allowing their wide - spread adoption . although a plethora of iaas related simulators exist even today , these simulators have very different focuses . some are designed completely from the user s point of view and hide the cloud s internals so users can make decisions on how and what services should be moved to the clouds @xcite . because of their user orientation , in these simulators it is frequently problematic to introduce changes in iaas behaviour . some others @xcite emphasise the need for precision for such simulations . despite their extensibility , these simulators not only scale very poorly ( making it problematic to evaluate more elaborate iaas scenarios where sometimes thousands of physical machines collaborate ) , but they also require complex setup procedures to be precise ( e.g. , one should model every possible application in the system to receive realistic results ) . finally , there are simulators that introduce some assumptions in the system that reduce the precision of the simulations but reach unprecedented speeds @xcite . unfortunately , despite having clear advantages , they are too specific to allow investigations on internal iaas changes ( e.g. , groudsim only models external interfaces of clouds , simgrid merely focuses on virtualisation , and cloudsim has conflicting extensions e.g. , power modelling is not available while using networking ) . in this article , a new versatile simulation framework is presented ( called discrete event based energy consumption simulator for clouds and federations dissect - cf ) . compared to the previously mentioned simulators , dissect - cf offers two major benefits : a unified resource sharing model , and a more complete iaas stack simulation ( including for example virtual machine image repositories , storage and in - data - centre networking ) . the benefits of the sharing model are threefold : @xmath0 it allows a single framework to model resource bottlenecks ( e.g. , cpu , network ) , @xmath1 generic resource sharing performance optimisations immediately improve entire simulations , @xmath2 it provides a unified view on resource usage counters ( i.e. , allows resource type independent , generic monitoring ) . finally , dissect - cf also opens up possibilities for more fine - grained energy consumption modelling by allowing the user to derive energy consumption from multiple resource usage counters . as a result of these new advancements , the new simulator could foster research on schedulers that could either have better insight into internal iaas behaviour or collaborate with internal schedulers of iaas systems in order to achieve previously unprecedented flexibility , adaptability and elasticity in future cloud systems . unfortunately , dissect - cf s focus on supporting research on infrastructure cloud schedulers introduces several limitations to its applicability . first of all , for performance reasons the simulator represents networks with a simple flow model , which has already been shown by several studies ( e.g. , @xcite ) to be inaccurate for smaller - sized network transfers . fortunately , smaller - sized network transfers have a negligible influence on scheduling decisions in most cloud related schedulers . also , because scheduler focused research usually uses task or virtual machine instantiation / termination traces for behavioural studies , dissect - cf uses the black box philosophy for applications . thus , the simulator will not provide accurate results on resource utilisation if a particular application s behaviour can not be approximated with simple resource consumption metrics ( e.g. , when there is unstable cpu utilisation for extended periods of time ) . in fact , these limitations are present in most simulators ( except those that have packet level network simulations or employ more complex flow models see @xcite ) . finally , as the new simulator is aimed at providing a framework for researchers to experiment with the internals of infrastructure clouds , the included scheduling mechanisms themselves are present only as examples for future work and they do not extend the scheduling related state - of - the art themselves . the behaviour of dissect - cf was analysed by first validating it against the behaviour of a small - scale infrastructure cloud at the university of innsbruck . according to the findings of this article , the system s simulated behaviour matches real - life experiments with negligible error ( in terms of application execution time , larger scale network transfers and energy consumption ) . for larger scale experiments , dissect - cf was validated with proven results from two other simulators that are close to the new simulator s functionality ( namely cloudsim @xcite and groudsim @xcite ) . then , performance of these two simulators was compared to the newly proposed one . comparisons were executed with both real - world ( using the grid workloads archive @xcite ) and synthetic traces . the use of real - world traces also revealed that dissect - cf based simulations allow 1.5 - 32@xmath3 faster behavioural analysis of simple cloud schedulers or vm placement strategies . the performance differences were further investigated through synthetic traces and it is shown that dissect - cf scales significantly better in complex resource sharing situations with the help of its unified resource sharing model ( one can observe an improvement of even over 2800@xmath3 in execution time in some cases ) . the rest of this article is organised as follows . section [ sec - relworks ] presents the related research results . then , section [ sec - design ] reveals the architecture of the newly proposed simulator and discusses its internal behaviour and extensibility options . section [ sec - eval ] analyses the properties of dissect - cf by comparing its behaviour to real - life systems and by comparing its performance to other simulators . finally , section [ sec - conclusion ] concludes the article with a summary and with the identification of future research directions . this section first reviews the scheduling scenarios that a cloud simulator might support . then an overview is presented on the most popular cloud simulation platforms . finally , the section concludes with a problem statement for the new simulator . there are seven common kinds of schedulers that could have an influence on the behaviour of a virtual infrastructure created on top of iaas cloud systems . in the following , a short overview is given of these kinds of schedulers with special attention on their requirements from a simulated environment . the list is presented from the schedulers that have the strongest user - side orientation to the most hidden schedulers in infrastructure systems . if a user has large enough resource demands , then its virtual infrastructure might include multiple virtual machines that could host a particular kind of task . in such cases , whenever a new task arrives , the user has to decide on which virtual machine it should actually run the task . the decision can be automated with a scheduler and a queuing system ( similar to local resource managers e.g. , @xcite ) . in order to support research on these kinds of schedulers , _ simulators should be able to provide past and present vm level performance metrics _ ( e.g. , temporal performance degradation of the vm s computing capabilities ) . when a user s resource demands are more dynamic and sometimes unpredictable , then he / she would frequently face heavy under- or over - utilisation of his / her virtual infrastructure . to better meet the demands of the newly arriving tasks , the virtual infrastructure should be able to automatically scale . this scaling is often achieved with a special scheduler ( e.g. , @xcite ) that decides when to instantiate / terminate a particular kind of vm . research on such schedulers need _ simulators that are capable of providing accurate vm management metrics _ ( e.g. , virtual machine instantiation time ) . some users have access to multiple cloud infrastructures . for such users , a new scheduler is needed ( e.g. , @xcite ) which can choose between various cloud providers and dispatch vms to them . the selection procedure is expected to take into account the availability , reliability and similar metrics of the various providers and also it should consider issues like placing processing close to big data . for such schedulers , _ simulators are required to offer infrastructure provider level metrics and data locality information . _ inside iaas systems , user requests are no longer represented as tasks but they are only seen as vms . as iaas systems are highly automated , decisions to place a particular vm on a physical machine must be also done by a scheduler . this kind of scheduler ( e.g. , @xcite ) could have two main tasks : @xmath0 for already existing virtual machines , a new vm to host mapping could be identified which would allow a vm arrangement that considers both the vms actual load and the providers current needs , and @xmath1 for newly requested virtual machines , the scheduler should determine the host where the vm could be run . as these schedulers have diverse tasks , _ simulators should have the capability to disseminate the load of currently running virtual machines and also the utilisation of physical machines_. energy conscious iaas systems aim at reducing their energy consumption in several ways . a simple way to do so is to consolidate vm load to the most energy efficient machines and switch the rest to a more energy efficient state . the automated decisions on which machines should be serving vms and which ones should be waiting in low power states ( e.g. , suspend , switch off ) are done by physical machine schedulers ( e.g. , @xcite ) . these schedulers should ensure that , because of their operations , virtual machine creation and quality of service do not degrade below certain levels . to support the development of physical machine ( pm ) schedulers , _ simulators must necessarily maintain the cost of pm power state changes _ ( e.g. , cold / suspend to ram boot - up procedures ) . schedulers are also present in virtual machine monitors ( like xen or kvm ) in order to allocate physical resources to virtual machines on a time - sharing basis ( e.g. , @xcite ) . although , these schedulers are not the main focus of research in cloud computing , they could have a direct impact on the quality of service if the above - mentioned vm placement strategies under - provision some virtual machines . for this reason , _ simulators should be able to correctly handle and report under - provisioning scenarios on physical machines_. the lowest levels of schedulers that may affect higher - level ( e.g. , task to vm assignment ) decisions are the process schedulers ( e.g. , @xcite ) of the operating system in the user s vms . in some cases the user could have an influence on the os scheduler , but in others users must use oss and schedulers that are prepared and accredited by the iaas providers . since these schedulers are generic os level schedulers , they are out of the scope of cloud computing research . but since higher - level schedulers might make decisions on how these process schedulers behave , simulators should give their users some information on their behaviour . for example , _ simulators should be capable of reporting if a particular vm is under - provisioned _ and tasks have no chance to access resources scheduled for them by the os level scheduler . cloudsim @xcite is amongst the most popular iaas cloud simulators . it was initially based on gridsim ( a widely used grid simulator developed by the same research institute @xcite ) but , after some performance and reliability issues , it was completely rewritten so it uses only some concepts ( e.g. , cloudlet gridlet analogy ) from its predecessor @xcite . cloudsim introduced the simulation of virtualised data centres mostly focusing on computational intensive tasks and data interchanges between data centres . later , they extended the simulation to better support internal network communications of a data centre with networkcloudsim @xcite . there are also extensions that simulate the energy consumption behaviour of the physical machines in the data centre based on specpower benchmarks and on dynamic voltage and frequency scaling @xcite . cloudsim also formed an ecosystem . several third parties offer extensions on top of cloudsim . some significantly change cloudsim behaviour ( e.g. , add performance improvements @xcite , add better support for inter - cloud operations @xcite , implement new energy consumption models @xcite , or introduce sla concepts into the simulation @xcite ) , while others wrap cloudsim and provide additional functionality ( like graphical user interfaces for teaching @xcite or for analytics @xcite ) . despite its wide use , cloudsim has several disadvantages : @xmath4 low performance for scheduling research where thousands of scheduling scenarios should be evaluated in a timely fashion , @xmath5 networking is simulated for tasks only ( e.g. , data centre operations that utilise the same network as user tasks like virtual machine image transfers are not simulated even though they could have significant effects on the user perceived network performance ) and @xmath6 using multiple extensions at once is frequently not possible ( e.g. , advanced networking and energy consumption modelling are not usable together since one would need to have virtual machines that inherit behaviour both from ` powervm ` and from ` networkvm ` classes ) . the simgrid framework @xcite is another widely used simulator for analysing the behaviour of distributed systems ( e.g. , grids , peer to peer systems ) . its resource sharing simulation is one of the most detailed ; for example , it contains one of the most accurate non - packet oriented network models @xcite . this simulator s focus was not particularly on clouds for a long time but recently its developers introduced extensions for virtualisation ( e.g. , hypervisors or live migration because of its distributed systems and grid background the simulator is inefficient in iaas cloud related situations . for example , this simulator stops at the virtual machine level , thus it would require significant effort to build a multi data centre / cloud simulation on top of it . while cloudsim and simgrid were heavily influenced by previous simulators for grids and distributed systems , for performance reasons they also make compromises on networking . to resolve such issues there are simulators like icancloud @xcite and greencloud @xcite that are built on network simulators ( e.g. , omnet++ or ns2 ) to more accurately simulate network communications in cloud systems . their efforts result in great accuracy if all iaas components and applications are modelled correctly network - wise ; otherwise , they just introduce serious performance penalties because of the packet level simulations without the expected accuracy . in addition to networking improvements , greencloud @xcite is also offering precise energy estimations for networking and computing components , while icancloud also offers a user oriented simulation which supports iaas utilisation decision - making @xcite on top of the regular iaas related simulation functionalities @xcite . next , groudsim a simulator developed at the university of innsbruck @xcite was analysed . this simulator aims at performance while it encompasses cloud concepts in a grid simulator environment . the simulator is also integrated with the askalon workflow system @xcite so it can be used to evaluate behavioural changes of real - life scientific workflows in the case of changes in the computing environment . although this simulator supports clouds , it does not provide implementation on the internals of iaas systems ( i.e. , it provides a black box implementation ) , thus it is not suitable for research studies that involve the internals of cloud infrastructures . and although groudsim supports both cpu and network resources , the networking implementation of groudsim is one of the least developed ones amongst the reviewed simulators . the above simulators focus more on the user related behaviour of data centres , but there is a class of cloud simulators which is more focused on supporting decisions related to data centre operations ( e.g. , @xcite ) . so even though these simulators could be used for examining user related behaviour , their detailed implementation of data centre behaviour reduces their usability in this context . on the other hand , these simulators offer some unique features that might be useful for research on iaas related schedulers . for example , speci @xcite is focused on offering a tool to analyse the scalability of iaas toolkits that will support future data centres . next , dcsim @xcite allows the analysis of new virtual machine management operations ( like relocation ) . finally , dcworms @xcite provides a unique view on data centre energy efficiency , including the heating , ventilation and air conditioning ( hvac ) system s airflow and high granularity resource ( e.g. , individual cpu , memory modules , network interfaces ) energy modelling . [ [ problem - statement ] ] problem statement + + + + + + + + + + + + + + + + + the analysis of the related work leads to the conclusion that existing simulators have many drawbacks for those who would like to investigate scheduling scenarios in iaas systems . to fulfil the needs of such scheduling scenarios , the rest of the article reveals a new infrastructure simulator that provides better insights on infrastructure behaviour for schedulers while maintaining the scalability of past simulators . ] figure [ fig - arch ] presents the overall architecture of the newly proposed simulator . the figure groups the major components with dashed lines into subsystems . each subsystem is implemented as independently from the others as possible . as a result , simulation developers do not need to understand the complexity of the entire simulator if they intend to work on one of its subsystems . there are five major subsystems ; they are listed in an order that follows their level of abstraction ( from the most abstract to the more specific to iaas systems ) : these components provide the time reference for simulations . this subsystem acts as a lightweight and extensible foundation to low - level computing resource sharing ( e.g. , cpu , i / o ) . with these components dissect - cf enables simulator developers to monitor and analyse energy usage patterns of each individually simulated resource ( e.g. , network links , disks ) . these components handle the behaviour of those iaas system parts ( e.g. , virtual machines ) that are the primary target of iaas related schedulers . this subsystem provides the user interface ( the vm management api ) and represents the high level functionalities ( e.g. , virtual machine schedulers ) of infrastructure clouds . in the following sections , these subsystems are individually discussed . the core of the dissect - cf simulator is a simple but high performance event generator ( reflected as ` timed ` in figure [ fig - arch ] ) . it is used to maintain the time within the simulated system and allow third parties to be notified if a particular time instance has been reached.the simulator is not aware of the applied time granularity ( i.e. , it is not known in the simulation if a single increase in the maintained time is equivalent to a single millisecond or a full hour ) . this enables flexibility in use , and allows simulation developers to have precision only when they assuredly need it ; otherwise , they can benefit from faster simulations . in later sections of the article , the smallest time granularity for the current simulation is denoted with @xmath7 and is expressed in seconds . thus any given time instance in the simulator can be specified as : @xmath8 , where @xmath9 and @xmath10 . here , @xmath11refers to the set of all possible time instances throughout a simulation . the simulator also assumes that notifications are recurring . thus , subscribing to events means specifying the frequency with which one would like to be notified . the simulator contains a construct ( called ` deferredevent ` ) for non - recurring events . creating a subclass of either the ` timed ` or the ` deferredevent ` classes allows simulation developers to receive custom time dependent notifications . finally , the ` timed ` class is also the control point for the simulation time . simulations have two distinct ways to influence simulated time : controls directly influence the timer . first , one can _ fire _ the events for the current time instance then advance the timer by one @xmath12 . second , it is possible to ask for a time jump that will progress the time with a given interval if within the interval there are no events expected . controls let the simulation flow for a given time interval without any intervention ( e.g. , one can simulate until all events from the queue are cleaned up ) . these controls also enable the progression of the timer while dropping irrelevant events that would occur in a given period of time . directly on top of basic time management lies the resource model of the simulator . the model is intended to capture low - level resource sharing behaviour ( e.g. , assigning tasks to virtual cpus of vms or virtual cpus to physical ones , or balancing network bandwidth utilisation ) . dissect - cf applies a provider - consumer scheme to resources where resource consumptions are intermediaries between consumers and providers . in the case of simulated cpus , consumptions represent instructions to be processed , thus cpu computing cycles of a physical machine are provided to virtual machines to consume . in a network analogy , consumptions represent data to be transferred between two network hosts ( where the sender acts as the provider and the receiver as the consumer ) . dissect - cf allows the definition of both providers and consumers with the help of the ` resourcespreader ` class ( see figure [ fig - arch ] ) . the set of all spreaders in a particular simulation will be referred as @xmath13 . the simulator uses the concept of resource consumptions as the intermediaries that represent the current processing demands of the actual consumers . resource consumptions are denoted with a triplet : @xmath14 , where @xmath15 represents the resource consumption , @xmath16 represents the processing that is currently under way , @xmath17 represents the remaining processing ( i.e. , processing that has not started yet ) and @xmath18 represents the limit for this processing in a single @xmath12 ( e.g. , simulation developers can specify that a resource consumption is single - threaded so it can use the processing power of a single processor of a consumable cpu resource only ) . @xmath19 represents all possible resource consumptions in a simulation : @xmath20 . at a given time instance , the function @xmath21 determines which provider offers the resources to be consumed . similarly , @xmath22 defines the consumer that utilises the offered resources . these functions are time dependent to allow the migration of resource consumptions amongst spreaders . at a given time , a particular resource consumption is processed in its provider by determining how much processing can be considered possible during a single @xmath12 . the possible processing has an upper bound of @xmath17 ( i.e. , if more processing could be possible than there is still remaining in @xmath15 , then the provider will have some non - utilised processing capabilities ) . also , the possible processing is limited by the provider s maximum processing capability and the processing limit @xmath18 of the consumption . @xmath23 where @xmath24 , @xmath25 and @xmath26 represent the processing under way , the remaining processing and the limit , respectively , for resource consumption @xmath15 at the time instance @xmath27 . it must be pointed out that @xmath28 is offering the provider side under processing value only . finally , @xmath29 reveals the processing power of a resource spreader ( in this current case the provider for resource consumption @xmath15 : @xmath30 ) at the time instance @xmath27 . dissect - cf simulates consumers with similar behaviour , so they remove utilised resources from @xmath16 . of course in this case the limit of utilisation is dependent on the consumer and the previously evaluated provider side possible processing value : @xmath31 thus , to determine the state of a particular resource consumption , dissect - cf first evaluates the provider side of resource consumptions , and then it processes the consumer side . after the simulator determined the @xmath16 value for a resource consumption , the remaining consumption @xmath32 can be determined as well by reducing with the increment of the @xmath16 value . this behaviour ensures that at the end of the consumer side processing both @xmath16 and @xmath17 will represent the resource consumption s state in the next simulated time instance ( @xmath33 ) . in order to determine how much processing can be done on resource consumptions at a particular time instance @xmath34 , resource spreaders apply the lowest level schedulers in dissect - cf based simulations . these schedulers share the processing capacities of the resources amongst those resource consumptions that the spreaders are currently dealing with ( @xmath35 , where the notation of @xmath36 is used to depict a power set ) . as simulation developers are expected to run simulations with thousands of resources and millions of resource consumptions , these low - level schedulers must be highly customisable and efficient . to enable simple customisability , dissect - cf provides efficient implementations for most common scheduling related tasks ( e.g. , resource consumption registration , de - registration , event generation for parties interested in resource consumption state ) , allowing providers of new schedulers to just focus on the scheduling logic that calculates the new @xmath34 values . + the scheduling logic is expected to deliver fair resource allocation for simultaneously occurring resource consumptions denoted as @xmath37 . to simplify the behaviour and complexity of schedulers , dissect - cf also introduces the concept of influence groups . these groups are formed from all resource spreaders that have a chance to influence each other s resource allocation schedules . with the help of influence groups even schedules for complex network structures can be simulated at close - to - real - life behaviour ( e.g. , the simulator can apply fair share algorithms over multiple related network links ) . these schedulers can utilise influence groups as their domain in which they have to guarantee a fair resource schedule for the spreader associated resource consumptions . to determine the membership of an influence group , the simulator uses the resource consumptions that link consumers and providers ( see figure [ fig - inf - gen ] ) . as a practical example , figure [ fig - inf - cpu ] shows how each simulated physical machine forms independent influence groups with the virtual machines it hosts via their respective cpu spreader implementations . formally , an influence group of a resource spreader at the particular time instance is defined as follows ( @xmath38 ) : @xmath39 where @xmath40 is a resource spreader , and the function @xmath41 defines the spreaders available at a particular time instance . the equation shows that @xmath42 includes all resource spreaders that are directly or transitively referred by the associated resource consumptions of the spreader @xmath43 @xmath44 . as a result , one could find as many influence groups as the number of resource spreaders existing at a given time instance in the simulation . on the other hand , these groups are frequently equivalent because determining the influence group of any member of a particular group will result in the original influence group : @xmath45 this last equation is derived from the definition of @xmath42 and reveals that after the proper calculation of @xmath42 there should not be any members of it ( e.g. , @xmath46 ) that would result in a different influence group than the original @xmath42 . although the definition of @xmath47 is quite straightforward , its evaluation in all necessary time instances for all relevant resource spreaders would result in significant simulation performance deterioration . therefore , dissect - cf provides an algorithm that significantly reduces the use of @xmath47 but still ensures that the influence groups are available for the scheduling logic in every time instance . to differentiate between the original function s results and the algorithm calculated values , the notation @xmath48 is used for the influence groups determined by the algorithm ( see algorithm [ alg - basopt ] ) . @xmath49 @xmath50 [ lin - ext - start ] @xmath51 [ lin - emptyext ] @xmath52 [ lin - identadd ] @xmath53 [ lin - extgadd ] @xmath54 [ lin - realext ] @xmath55 [ lin - igextcomplete ] [ lin - ext - stop ] [ lin - split - start ] @xmath56 @xmath57 [ lin - randsel ] @xmath58 [ lin - correct - group ] @xmath59 [ lin - split - upd - start ] @xmath60 [ lin - split - upd - stop ] [ lin - split - stop ] in the following few paragraphs , the internal behaviour of the new algorithm is discussed . it is built on the assumption that @xmath61 and it is composed of two distinct phases : influence group extension see lines [ lin - ext - start]-[lin - ext - stop ] and group dissolution see lines [ lin - split - start]-[lin - split - stop ] . let us first discuss the extension phase . during this phase , the algorithm first starts with an empty resource spreader set ( see line [ lin - emptyext ] ) that will later on hold the identified extensions of the input influence group @xmath49 . as a next step , line [ lin - identadd ] determines the resource consumptions that arrived to a particular resource spreader at the time instance @xmath33 . afterwards , the following line focuses on those newly added resource consumptions that introduce new resource spreader members into the input influence group . these non - member providers or consumers are added to the extension set in line [ lin - extgadd ] . this iteratively created extension set is used to actualise the input influence group in lines [ lin - realext]-[lin - igextcomplete ] . the extension phase completes only when there are no newly introduced resource spreader members in the input group . otherwise , the current phase is repeated to ensure finding even further group extensions via the resource consumptions associated with the introduced members ( this last step is shown in line [ lin - ext - stop ] ) . after there are no new extension possibilities found in the current influence group , the algorithm proceeds to its second phase in which it identifies all splits of the current influence group . for example , influence group # 5 in figure [ fig - inf - gen ] will have to be split when the resource consumption between provider @xmath62 and consumer @xmath63 finishes . to identify the need for splitting , the algorithm therefore first determines if there were some resource consumptions dropped from at least one member of the influence group ( see line [ lin - split - start ] ) . if there is a need for splitting , then the algorithm will maintain the not yet split parts of the original influence group in @xmath64 . in order to determine which parts have to be split from the not yet split parts , lines [ lin - randsel ] and [ lin - correct - group ] use the original @xmath65 on a randomly selected spreader from @xmath64 resulting in a new influence group called @xmath66 . in the next line , the not yet split parts are updated so that only those resource spreaders will be considered afterwards that are not in @xmath66 . finally , the algorithm updates its self - maintained influence group membership so all members of @xmath66 will be exactly the same ( see lines [ lin - split - upd - start]-[lin - split - upd - stop ] ) . to conclude , the newly introduced algorithm reduces the number of direct @xmath42 function calculations to those cases where there is a chance to have a group to be split . and even in that case , it ensures that the number of @xmath42 evaluations is limited by the number of influence groups created after a split . ] the last remaining part of the unified resource model is the simulation developer customisable low - level scheduling logic . to understand the customisation options dissect - cf offers , figure [ fig - rs ] presents the context of the scheduling logic . the figure reveals the role of the low - level scheduler through the illustration of the life of a single resource consumption that can be represented in three phases and denoted with different kinds of arrows : preparation regular lines ; resource consumption dashed lines ; and completion dotted lines . the next paragraphs provide a brief overview of these three phases . first , the preparation phase is initiated by the entity who is responsible for creating ( see _ step 1 _ in the figure ) a particular resource consumption @xmath67 . this entity could be an automated process ( e.g. , a workload generator ) or some higher - level entity of the simulator ( e.g. , the virtual machine representation ) . after creation , the registration can be initiated in any time instance @xmath27 after both the consumer @xmath68 and the provider @xmath69 spreaders are specified . the registration is accomplished in _ step 2 _ in the figure . the provider nudges the scheduler base in _ step 3 _ after both the consumer and the provider have registered the new resource consumption @xmath70 and @xmath71 . the _ scheduler base _ is implemented in the base class of all resource spreaders and is responsible for interfacing with the event system , the scheduler and the influence group management algorithm . before contacting the event system for subscription , the scheduler base first updates the influence groups with algorithm [ alg - basopt ] in _ step 4_. after the identification of all distinct influence groups , the scheduler base filters those groups that would need an updated schedule . such groups are identified via recently added or dropped resource consumptions to / from one of their member resource spreaders i.e. , @xmath72 . in _ step 5 _ , the scheduling logic is invoked for each of the filtered groups . during this step , it should assign the @xmath34 values for all resource consumptions that are currently taking place in a given influence group . with these assignments , the simulator calculates the earliest completion time of the currently managed resource consumptions . then , in _ step 6 _ , it subscribes to a notification from the event system in order to know when the next resource consumption will be removed from the filtered influence groups . in the following phase , the simulator handles the resource consumptions this phase is either done when the event system delivers the notification on a resource consumption completion ( see _ step 7 _ ) or alternatively upon the registration of a new resource consumption . in the second case , the resource consumption handling is automatically executed before influence groups are calculated ( i.e. , steps _ 8 - 12 _ could precede _ step 4 _ of the preparation phase if a resource consumption is registered at a resource spreader that has already had some prior resource consumptions ) . in practice , _ steps 8 - 9 _ evaluate eq . [ eq - prov - update ] and _ steps 10 - 11 _ evaluate eq . [ eq - cons - update ] for all simultaneously existing resource consumptions @xmath73 in the simulator . resource consumptions e.g. , @xmath74 are automatically marked for removal when they reach their completion i.e. , @xmath75 . as a final step for resource consumption handling , the scheduler base checks for resource consumptions marked for removal ( see _ step 12 _ ) and on all marked resource consumptions it executes the completion phase . in the final phase , the resource consumption s completion is simulated . consumptions can be considered complete in two cases : either they have no further processing to be done or they were cancelled by the entity using the resource sharing mechanism . the scheduler base notifies this entity in both cases see _ step 13_. then it checks if it has finished all the current resource consumptions if there are still further resource consumptions to process , then the scheduler base resumes operations from _ step 4 _ , otherwise it cancels further notifications from the event system . as can be seen , the simulator expects scheduling logic implementations to utilise a fairly narrow and well - defined interface with the scheduler base . through this interface the simulator ensures that whenever a new schedule is needed ( i.e. , new @xmath34 values ) , the simulation developer provided scheduling logic is always called . dissect - cf also provides two sample implementations for this scheduling logic : a max min fairness algorithm @xcite implementation with progressive filling , and a simple logic that does not deal with complex bottleneck situations but demonstrates the interfaces with the scheduler base . compared to other recently developed simulators , dissect - cf completely decouples energy modelling from resource simulation in order to allow accounting for such energy consumptions that are not in direct relation to the resource utilisation of data centres . with this approach , a more comprehensive energy and power modelling is achievable that enables the analysis of new sophisticated energy aware algorithms in the areas of virtual machine placement , task scheduling , etc . these algorithms previously were frequently limited because energy readings from heating , ventilation and air conditioning ( hvac ) units or higher - level iaas components ( like vm schedulers or iaas interfaces ) were scarcely available in past simulators . thus , this section presents how energy consumption related information is collected and accumulated so they can support future algorithms . later , section [ sec - hls ] discusses the foundations for physical machine schedulers that are expected to be the primary users of the models overviewed in this section . first , in order to enable the decoupling , dissect - cf offers _ resource utilisation counters _ both for producers and consumers . these counters allow an aggregated and time dependent view of the consumption of particular resources . counters are updated depending on the ` power state ` of a resource spreader ( see figure [ fig - arch ] ) . for example , a physical machine @xmath77 in suspend to ram ( str ) power state zeroes its processing power : @xmath78 the power states of the various entities in the simulation can be defined as needed ; the simulator only expects these states to define the basic power characteristics ( e.g. , minimum and maximum power draw ) and the resource processing behaviour of the given entity at the specific state . the simulator also provides a basic set of power states ( on , off , turning on , turning off ) for which the resource processing behaviour is already defined for the resources incorporated into a physical machine . based on this low - level power modelling functionality , the decoupling of energy models is accomplished through ` energy meter`s . these meters are organised around four functionalities : @xmath0 monitoring of energy consumption directly related to resource utilisation , @xmath1 indirect energy consumption estimation , @xmath2 aggregation of metering results from multiple meters and @xmath79 presenting up - to - date energy readings to their users . the remainder of this subsection reveals how these functions accomplish the decoupling and shows the ways customised , infrastructure specific metering can be achieved in dissect - cf . ] [ [ direct - resource - utilisation - related - energy - consumption - metering ] ] direct resource utilisation related energy consumption metering + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + based on the previously mentioned resource utilisation counters , the simulator can be requested to periodically evaluate the instantaneous utilisation percentage . then , energy ` consumption model`s use these percentages ( see figure [ fig - arch ] ) to estimate the instantaneous power draw of each monitored resource spreader ( later the resulting estimate and the metering period is used to calculate the direct meter s energy consumption estimate ) . consumption models are dependent on the actual power state a resource spreader is in , and the simulator developer can define them . as examples , the simulator provides two simple energy consumption model implementations : @xmath0 a linear interpolation between minimum and maximum power draw depending on current resource utilisation , to allow basic modelling of dynamic power behaviour or @xmath1 a constant minimum power draw , to allow the effortless modelling of _ off _ or _ str _ power states . figure [ fig - meters ] presents direct meters for each resource spreader in the shown physical machines . [ [ indirect - energy - consumption - estimation ] ] indirect energy consumption estimation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + to support more complex energy consumption estimates , the simulator also allows energy consumption to be derived from other properties of the simulated system . for example , these properties could include the virtual machine request rate of a particular data centre , or the utilisation of the data centre level storage subsystem ( e.g. , to estimate how many disk drives the currently stored data can occupy ) . these meters are expected to periodically evaluate the system state and accumulate their energy consumption estimates for those components of the simulated system that are not directly represented with resource spreaders . figure [ fig - meters ] reveals two indirect metering solutions to represent the internal actions of iaas systems and the energy consumption behaviour of data centre level hvac systems . [ [ meter - aggregators ] ] meter aggregators + + + + + + + + + + + + + + + + + in several cases , the energy consumption values from individual direct or indirect meters are not sufficient for higher - level energy aware decision makers ( e.g. , physical machine state schedulers see section [ sec - schedulers ] ) . for example , a physical machine in dissect - cf is represented with multiple resource spreaders ( e.g. , cpu , disk bandwidth ) , thus to have a complete view of a physical machine s energy consumption , one would need to monitor several direct energy meters . meter aggregators allow the automated collection and management of several meters in parallel and provide the higher - level view expected by decision makers . figure [ fig - meters ] shows a complex scenario for the use of aggregators . this scenario shows that a single aggregated meter can be constructed for a whole data centre , allowing even the inclusion of indirect metering results such as hvac . the figure also shows how a physical machine s resource spreader set can be metered as a single entity . as the estimation of energy consumption of the simulated entities might be time consuming , the simulator allows simulation developers to fragment and focus their measurements on those parts of the simulated systems that they are interested in . for instance , the simulation developers might be only interested in the energy consumption of a single virtual machine that is deployed on a simulated cloud with multiple data centres . in such cases , dissect - cf can limit the number of energy meters that are evaluated with two approaches : independent meters or adjusted aggregations . [ [ independent - meters ] ] independent meters + + + + + + + + + + + + + + + + + + this approach entails that metering results are only dependent on the metered component and the rest of the simulated system can not influence them . e.g. , the energy consumption reported for a cpu level resource spreader of a physical machine should not be dependent on the behaviour of the rest of the system . [ [ adjusted - aggregations ] ] adjusted aggregations + + + + + + + + + + + + + + + + + + + + + if there is a dependency between two metered components ( e.g. , a virtual machine that is hosted on a particular physical machine ) , then a special meter aggregator can be created . this meter aggregator will not just add the aggregated meters measurements . instead it allows simulation developers to define an aggregation function . dissect - cf uses the adjusted aggregation technique to handle derived energy consumptions such as vm level energy consumption . for example , when applying the linear interpolation based energy consumption model , virtual machine level power draw can be calculated using the power draw of the hosting physical machine , as follows : @xmath80 where @xmath81 is the derived instantaneous power draw of a particular vm while it is running . the equation s first part estimates the variable part of the power draw , while the second part provides an estimate for the idle part . the variable part is dependent on @xmath82 , which is the maximum variability of the physical machine s power draw . the variable part is proportional to @xmath82 , depending on the resource utilisation of the particular vm compared to the resource utilisation of all vms hosted on the same physical machine . the second part of the estimate is the idle part that is derived from the idle power draw of the physical machine @xmath83 . this part is proportional to the number of vms hosted by the physical machine at the given time instance ( this number is one less than the cardinality of the vm s influence group because the group contains also the resource spreader of its hosting physical machine ) . [ eq - powermodel ] reveals that the energy consumption model ( that estimates the instantaneous power draw ) of the virtual machine is not independent from the physical machine s behaviour . therefore , energy consumption can not be directly accounted to virtual machines . when such meters are requested , the simulator identifies them as dependent meters and instead of creating independent meters , it creates an adjusted meter aggregator including those meters that could provide the necessary information to calculate the energy consumption to be attributed to the originally requested meter . it must be noted that dependent meters consider energy consumptions multiple times ( e.g. , energy consumption is accounted to both physical and virtual machines ) . thus when meter aggregations are created they must only include meters that are not dependent on each other . to allow the simplified development of new vm placement algorithms and pm state schedulers , dissect - cf provides an implementation of relevant infrastructure components in iaas systems . these components are built on top of the previously discussed resource sharing and energy modelling techniques and provide abstractions for networked entities and for physical / virtual machines ( see figure [ fig - arch ] ) . network activities rarely play a role in scheduling decisions related to tasks or to physical / virtual machines . thus , to increase the performance of simulations , by default , dissect - cf offers a limited network model where two networked entities must be always directly connected ( therefore connection properties like bandwidth must be defined between all networked entities that should be able to communicate with each other in the simulation ) . this rudimentary behaviour could be sufficient even for some network aware schedulers , but to allow better representation of real networks , the simulator also allows the creation of intermediary network entities ( such as routers ) . the implementation of such entities should alter the processing limit ( @xmath84 ) of all resource consumptions that are directed through them . directly connected networked entities ( @xmath85 , where @xmath86 is the set of all possible networking entities ) are simulated with the ` networknode ` component . this component encapsulates an incoming and an outgoing network connection simulated with the unified resource sharing foundation ; thus connections are implemented as resource spreaders : @xmath87 , where @xmath88 . the processing power of these spreaders represents the network bandwidth ( either incoming or outgoing ) of the given network node . when a new network communication must take place , simulation developers are expected to request a resource consumption between the source network node s outgoing resource spreader and the target s incoming resource spreader . the component also introduces network latencies ( @xmath89 ) that can be defined between every networked entity for any given time instance . the latency values resulting from this function are used as delays preceding the registration of each resource consumption to the incoming or outgoing resource spreader of the node . for example , let us see what happens if a network communication ( represented as a resource consumption @xmath90 ) needs to be registered between a source and a target networked entity ( @xmath91 ) at the time instance @xmath92 : @xmath93 where @xmath94 is the network latency between the source and target nodes at the time of registration ( @xmath95 ) , and @xmath96 is the nonperforming spreader that never processes any resource consumption : @xmath97 . thus the equation shows , that the consumption is registered to the nonperforming spreader for the complete period of the latency , afterwards the simulator switches the consumption s registration to the originally designated spreaders ( which are the network output port @xmath98 of the source node @xmath99 and the input port @xmath100 of the target node @xmath101 ) . this last step allows the simulator to utilise its unified resource sharing mechanism after the latency period is over . in iaas systems , physical machines offer most of the user exploited simulated resources . thus , dissect - cf ` physical ` ` machine`s encapsulate a diverse set of resources : local disks ( via repositories see section [ sec - ui ] ) , network interfaces ( with the help of network nodes see section [ sec - nw ] ) , cpus ( using the unified resource sharing model of section [ sec - ursm ] ) and memory . besides the modelling of these resources , dissect - cf s physical machine behaviour also focuses on two additional functionalities : administering resource allocations and vm requests ; and modelling physical machine level power behaviour . as resource sharing and modelling has already been discussed in detail , this subsection mainly discusses the latter functionalities . [ [ resource - allocator - and - vm - request - handler ] ] resource allocator and vm request handler + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in order to maintain up - to - date information on the available and utilised resources of the physical machine , dissect - cf applies a resource allocator . the applied allocator can reserve resources ( e.g. , a given amount of memory or number of cpu cores ) from the physical machine . these reservations are represented as ` resource ` ` allocation ` instances that are used to maintain the free resource set of the physical machine . allocation instances are also passed as a token of resource availability towards virtual machine schedulers . other than the amount of resources associated with them , resource allocations also have the following properties : expiry time if unused or a link to the vm that uses it . when a resource allocation is created , the physical machine automatically initiates a ` deferredevent ` , which automatically cancels the allocation after the expiry time . to avoid this mechanism , the entity that requested the allocation must request a vm to use the reserved resources ( i.e. , establishing the link between the vm and the allocation ) . the automatic cancellation of the resource allocation is a self - defence mechanism of physical machines to avoid keeping resources out of use just because they received an unused allocation . for a single vm request , dissect - cf allows multiple ` resource ` ` allocation`s to be made across multiple physical machines . the multi - allocation technique can be used by schedulers to optimise for non - functional properties ( e.g. , past availability , expected energy consumption , environmental impact ) of those resources that a vm could bind to at a given moment . after a decision is made about the use of a particular allocation for the vm request , the rest of the allocations ( which were non optimal according to the non - functional requirements ) are expected to be cancelled by the schedulers . for complex vm instantiation scenarios , schedulers are also allowed to adjust the expiration time upon allocation request . with this mechanism , researchers can evaluate advanced reservation - like scenarios regarding vm instances . [ [ power - behaviour ] ] power behaviour + + + + + + + + + + + + + + + as dissect - cf aims at supporting the development of energy aware scheduling strategies in iaas clouds , the energy model of the physical machine is particularly important . in its default configuration , the simulator supports 4 power states ( and the transitions between them ) : off , switching on , running and switching off . although this power state set is fairly limited , the simulator already offers constructs that allow the modelling of more complex operations like : suspend to ram , suspend to disk , dynamic voltage & frequency scaling or core / cpu de- and reactivation . suspension related states can be modelled with the introduction of new power states , while the latter two are available because resource spreaders can alter their maximum processing capabilities . the modelling of the 4 supported states was done after real - life physical machine behaviour in the clouds of the university of innsbruck and mta sztaki . the real - life behaviour of the machines was observed through constantly monitoring their instantaneous power draw while they went through the following cycle : idling @xmath102 shutdown @xmath102 off @xmath102 switch on @xmath102 idling @xmath102 full cpu load . a sample measurement with a typical cloud node at innsbruck ( with 80 cpu cores , 128 gb ssd , 132 gb ram , and redundant power supplies ) can be seen in figure [ fig - pmbehaviour ] . as physical machine state schedulers can be highly influenced by the behaviour in non - running states ( e.g. , their decision on shutting off a machine could be dependent on the time it takes to boot the machine back and the expected power savings because of the completely off machine ) , dissect - cf offers a simplified and a more complex behaviour model . .simplified power state definition of a physical machine[tab - simpli ] [ cols="<,<,<,<,<",options="header " , ] during the performance evaluation , all parallel tasks of the trace started up in the first 10 seconds ( _ task spread _ ) and had a length variety between 10 - 90 seconds . the necessary amount of tasks for the 10-second - long simulation runtime is shown in table [ tab - taskdistrs ] . each evaluation was ran with different parallel task numbers between ( 1 - 100,000 ) to allow the investigation on how the increasing parallelism changed the performance of the simulators . the findings are shown in figure [ fig - pjruntime ] . as the figure shows , in the case of cloudsim , both its time - shared and space - shared vm scheduler was experimented on . the time - shared scheduler of cloudsim simulates parallelism , while the space - shared scheduler serialises the arrived tasks so at any given time there is only one task that can utilise the cpu , and the rest are queued until this task is finished . unfortunately , the time - shared scheduling mechanism of cloudsim is broken ; it lengthens task execution times significantly even in the case of a two - parallel - task setup ( e.g. , a simple setup like the one presented in figure [ fig - incpartask ] leads to completely different results than one would see in real - life ) on the other hand , the space - shared scheduler provides completely different task completion times ( because of its serialising behaviour ) . thus the cloudsim related details of the figure are only valid when considering their simulation runtime ; the simulated tasks do not finish at the expected times in either case . as depicted in the figure , cloudsim - time - shared and groudsim measurements abruptly finish at 1,000 and 50,000 tasks . for higher levels of parallelism , the runtime of the simulation took more than 8 hours and therefore was cancelled . in the case of cloudsim , the performance penalties mainly originate from its centralised design of data centres ( i.e. , most of the logic and event processing reside in the ` datacenter ` class , and the rest of cloudsim s classes are used for state representation only ) . groudsim , on the other hand , has a different design issue : it pre calculates all task completion times , puts them into the event queue and thus if a change is needed to them , the whole event queue has to be updated . finally , for all simulators , the task processing performance shown in the figure is expected to be faster than what one usually would see , because of the large amount of tasks that were executed for gathering data for the figure . the amount of repeated evaluations allowed the jvm to optimise the runtime behaviour of each simulator for the particular performance evaluation scenario . later on this effect is excluded from the evaluations . [ [ performance - influence - of - load - characteristics ] ] performance influence of load characteristics + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the rest of this sub - section analyses of how the performance of the resource sharing varies depending on changes in task arrival and length characteristics . for this analysis , a baseline measurement was collected with no parallel tasks but varying task lengths . than the assumption was made that from this baseline , the simulators would linearly degrade in performance ( i.e. , two times the parallelism would increase the simulation run time of a single task by two times ) . to compare how linearly a particular simulator behaves under particular load characteristics , the following scaling ratio function was designed : @xmath103 where @xmath104 is the scaling ratio , @xmath105 is the kind of simulator , @xmath106 is the chosen range of task length variety ( in the below detailed experiments it was either 10 - 90 s or 200 - 3600 s ) , @xmath107 is the number of parallel tasks , @xmath108 is the task spread ( in particular , 10 s or 200 s were used ) , and @xmath109 is the measurement function that evaluates the particular simulator with the given load characteristics . figure [ fig - jobscalingcompare ] compares dissect - cf with the two selected simulators via the scaling ratio function . the labels of the figure are presented in the following format : @xmath110 ( see eq . [ eq - scratio ] ) , where @xmath105 is one of ds / gs / cs , which translate to dissect - cf / groudsim / cloudsim , respectively . the x - axis of the figure shows the increase in @xmath107 . based on the distribution of the measured scaling values for the different load characteristics , the most tolerant to the change in workload is groudsim , while the least tolerant is cloudsim . degrading performance in scaling can be also observed for both cloudsim and groudsim . in the case of cloudsim , the degradation starts around 10 parallel tasks , and by the time the simulation reaches 50 parallel tasks the simulator becomes worse than linear ( @xmath111 ) . in the case of groudsim , the degradation starts around 100 parallel tasks , and becomes worse than linear around 20,000 parallel tasks . in both cases the early degradation is caused by the data structures and the indirect call structures used ( i.e. often these simulators experience rather deep call stacks during parallel event handling ) . finally , although it is not visible in this chart , the degradation starts for dissect - cf at around 10,000 parallel tasks , and based on estimates , its scaling becomes worse than linear around 1.5 million parallel tasks . fortunately , this huge amount of parallelism is unlikely for most simulations . but simulations of some highly under - provisioned systems might need levels of parallelism that could lead to degraded performance even in groudsim . the figure also reveals that the performance of dissect - cf is more dependent on the task spread , while cloudsim s scaling is limited more by the task length variety . in case of dissect - cf , the task spread dependency manifests because the simulator executes its resource sharing mechanism only once per @xmath12 , on new resource consumption arrivals and departures ( see figure [ fig - rs ] for details ) . however , the increased task spread decreases the likelihood that the resource sharing mechanism can be executed on multiple resource consumptions at once . therefore , the wider the spread the closer one gets to the worst possible resource sharing performance in dissect - cf . in fact , the 200 s long spread was used because this spread is already high enough to reveal close - to - worst - case performance . increasing the spread further did not introduce significant resource sharing performance drops , but its impact on cloudsim based simulations rendered the evaluation of the 1,000 parallel task experiment too time consuming . as synthetic loads often criticised because of their possible bias , dissect - cf was evaluated and compared using workloads that were collected from real - life computing infrastructures . thus , the evaluation required workload traces with the following characteristics : @xmath0 collected for extensive periods of time ( i.e. , at least a few months long ) to ensure the widest variety of observable load characteristics for the particular infrastructure ; @xmath1 tasks should be described in detail including their resource utilisation and submission , start and completion times ( so even if just task definitions are available , one can still translate them to the kinds of virtual machines the tasks would need for their execution in a cloud environment ) ; and @xmath2 if the traces contain virtual machine management logs , then task allocation details are also necessary to enable the analysis of new scheduling techniques that might aim at reallocating tasks or that would change vm management operations . based on these requirements , most of the traces ( e.g. , planetlab ) containing virtual machine management logs were not found suitable for the planned comparative study ; the rest of these log based traces are not collected for enough time to be used in large scale experiments . thus , although these virtual machine management log based traces would be the best candidates for analysing cloud characteristics , their immaturity necessitates to also search amongst traces collected from other large - scale infrastructures like grids . two appropriate sources were identified : the grid workloads archive ( gwa @xcite ) and the parallel workloads archive . because both at the university of innsbruck and in mta sztaki , there are earlier good experiences with the processing of gwa traces , this article presents a comparative study of the three selected simulators through the traces downloadable from gwa ( namely : das2 , grid5000 , nordugrid , auvergrid , sharcnet , lcg ) . [ [ trace - processing ] ] trace processing + + + + + + + + + + + + + + + + a trace loader was prepared for all three simulators so events were fired every time a task arrives . every time a new task arrival event is fired , the simulators are programmed to create a virtual machine that will host the task . once the vm was created the task was instantiated in it according to its definition in the trace file . when the task was completed according to the particular simulator , its hosting vm was also terminated . because of the known problems with the time - sharing mechanism in cloudsim , single vms did not receive parallel tasks ( e.g. , by requesting a vm that is sized to host multiple tasks first ) . instead , when parallel tasks were needed , multiple vms were created in the simulated cloud infrastructure . unfortunately , because of a few conceptual differences between dissect - cf and the other analysed simulators , the above - mentioned trace processing technique has several minor differences in their adaptations to the various simulators . first of all , dissect - cf has a queuing first - fit vm scheduler ( see section [ sec - hls ] ) that allows users to send vm requests right upon task arrival and wait for the vm scheduler s queuing mechanism to notify them about the vm s running state ( and readiness to receive the task ) . unfortunately , the rest of the simulators do not offer vm request queuing : they reject vm requests that can not be hosted according to the actual state of the simulated infrastructure . to have a similar behaviour to dissect - cf , either the vm request needs to be resent until it can be fulfilled ( which would introduce an unwanted busy waiting loop ) , or alternatively one must rely on user - side information . if the vm request is repeatedly resent , then the other simulators are significantly disadvantaged . thus , the second approach was used : previously non - servable vms were only re - requested once one of the previous vms have terminated ( i.e. , it was assumed that one can determine the state of all the running vms at a given time instance ) . with this technique the cpu share of the performance evaluation code ( as measured by java s embedded cpu sampler ) was kept around the same levels as it was for dissect - cf . tasks , which are utilising several cpu cores in parallel , are modelled in two ways . first , the simulation tries to request a vm with as many cpu cores as the task needs . unfortunately , this can not be achieved in cases when even the biggest possible vm is too small for the task s needs . for these huge tasks , the simulation requests multiple vms , enough to fulfil the parallelism in the task . in groudsim and dissect - cf it is possible to request multiple vm instances at once ( e.g. , similarly to how amazon ec2 behaves ) and they will be ensured to be available in parallel . on the other hand , cloudsim does not have such functionality ; therefore it was extended with a technique that only submits the tasks to their virtual machines once all necessary vms are available for the level of parallelism needed for the task . as groudsim is more focused on the user - side behaviour of cloud infrastructures , it has several deficiencies compared to the other two evaluated simulators : @xmath0 it can only handle tasks that occupy a single cpu core , @xmath1 its network sharing mechanism could result in network under utilisation if the communicating parties are using connections with different bandwidths and @xmath2 it does not provide data centre level simulation details e.g. , no vm / pm scheduling and pm level resource sharing is simulated . to overcome the first deficiency , multi - core tasks are simulated as several single core tasks in the same groudsim virtual machine . to avoid the problem with network sharing , the simulated data centres in all three simulators were constructed on such a way that every node was connected with the same bandwidth to the others . unfortunately , without significantly changing groudsim s code it was not possible to add the missing data centre level simulation details . thus , during the performance analysis one should keep in mind that these details are not simulated in groudsim . finally , for both groudsim and cloudsim , the instantiation time of a virtual machine is instantaneous , which is not realistic . as the transfer of the vm image is often the most time consuming operation in the vm instantiation procedure @xcite , this transfer was simulated with a download operation to the vm with both of the simulators . in groudsim , a network transfer was initiated right after the vm was created and this transfer delayed the creation of the task on the vm until the transfer s completion . in cloudsim , cloudlets ( computing tasks in cloudsim terminology ) could have input files defined for them . thus , tasks in cloudsim were specified so they must transfer an input file with the size of the vm image before they can start their processing . unfortunately , even with input files specified , the vms in cloudsim start immediately and execute their tasks right after the vm is created . this means that cloudsim based simulations can not be as accurate as the other two simulator s results . in order to reduce the impact of vm image transfer on task execution times and to allow all three simulations to have a similar simulation completion time , the size of the vm image was set to 100 mb ( which is the size of a rather small image nowadays ) . [ [ the - simulated - virtual - infrastructure ] ] the simulated virtual infrastructure + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in all three simulators , a single kind of physical machine acted as the foundation of the simulated infrastructure . this physical machine was modelled after a single node in mta sztaki s cloud and had the following properties : @xmath0 64 cpu cores , @xmath1 256 gb ram , @xmath2 5 tb local disk , and @xmath79 two 1 gb / s ethernet connection . in the experiments detailed below , it was possible to define how many of these machines should be in the data centre . the set up of the machines includes the creation of a network interconnecting all of them with a central switch . [ [ the - runtime - comparison - experiment ] ] the runtime comparison experiment + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for the first experiment the simulated infrastructure was set up so it was sufficient to host even the largest parallelism a single task of the traces could request ( i.e. , no task has had to be dropped because there were not enough physical machines to host its level of parallelism ) . in particular , this experiment required the simulation of 20 physical machines with the above - mentioned properties . on this infrastructure , the first @xmath112 tasks from a particular trace were submitted . then , a measurement was initiated for the real time passed between the submission of the first task and the completion of the last one . the infrastructure preparation , task submission and time measurement operations were repeated 10 times , each time starting with a new jvm . then , the trace specific average simulation duration for the measurement of @xmath113 tasks was calculated . afterwards , the whole measurement procedure was repeated for the remaining traces . finally , the _ aggregated simulation duration _ was calculated for @xmath113 tasks by averaging the trace specific averages from all traces . later , the aggregated simulation duration is referred as @xmath114 ( where @xmath105 is the measured simulator , @xmath107 is the number of tasks , @xmath115 is the number of physical machines to simulate and @xmath116 ) . figure [ fig - runtimecompare ] presents these aggregated simulation durations while the number of tasks @xmath107 were changed from a hundred to a million . as the measurements ran with a cold jvm , for the first 10,000 tasks one can see that the jvm s effects on the start - up dominate the runtimes . the first workload related differences of the simulators can be seen after the jvm s behaviour is no longer a significant contributor to the measurement . also , despite dissect - cf s focus on the internals of the infrastructure , its performance is always better than that of the other two simulators . this is especially apparent after around 200,000 tasks . in spite of groudsim doing the lightest - weight simulation ( see the deficiencies listed in the previous paragraphs ) , it is still significantly and consistently slower ( albeit in some cases like the lcg trace groudsim performs consistently better ) . the figure also shows that groudsim s network simulation code gives around 21% overhead for lower task counts , and for higher task counts the overhead could grow as high as 109% . this difference experienced during the network simulation is caused by groudsim s sub - par network modelling ( i.e. , for a single transfer to occur there are several events that needs to be handled in the event system , and in case of parallel transfers these events must be refreshed in the future event queue ) . [ [ the - scalability - experiment ] ] the scalability experiment + + + + + + + + + + + + + + + + + + + + + + + + + + to determine the scalability of the compared simulators , the previous experiment was repeated with varying sizes of infrastructure under them . thus , these experiments will not only be able to pinpoint how the various simulators deal with the increasing amounts of tasks and virtual machines , but also how the simulators cope with increasing size of infrastructures . infrastructures as small as a single physical machine were used but the experiments were reaching to the size of 500 machines . as even smaller infrastructures were evaluated than 20 machines ( which was the minimum for the level of parallelism found in some of the traces ) , there were tasks that could not fit to the infrastructures . these tasks were automatically filtered out by the trace processor and never reached the simulation . interestingly , even with the simulation with a single machine the filtered out tasks were never more than 6% of the total number of tasks in any of the traces . of course , the single machine still caused a significant serialising effect that lengthened the total execution time of the rest of the tasks . similarly to eq . [ eq - scratio ] , the scalability of the simulators was calculated compared to linear scaling . figure [ fig - scalingcompare ] presents how the various simulators scale under a particular number of physical machines . the figure is composed of twelve sub - charts , each delimited with a black line . between the black lines a large bidirectional arrow contains the title of a sub - chart . the title shows the number of tasks for which the final runtime measurement values were compared . for example : the title `` _ _ 100 - 200 tasks _ _ '' means the sub - chart shows how the final runtime measurement value of 200 tasks compares to the value of 100 tasks . the comparison is made according to the following equation : @xmath117 where @xmath118 , @xmath119 is the number of tasks for which the aggregated simulation duration is evaluated , @xmath105 is one of the evaluated simulators and @xmath115 is the physical machine count . thus , @xmath120 will become one if the particular simulator scales linearly in the task number range of @xmath118 and @xmath119 . if the scaling function above is lower than one then the particular simulator is worse than linear . according to the sub - charts in figure [ fig - scalingcompare ] all simulators scale significantly better than linear with fewer than 10,000 tasks ( again , this is mostly due to the jvm s class loading and start - up mechanism , since practically all measurements under this task count result in sub - second runtimes ) . afterwards , one can see that cloudsim is already scaling significantly worse than the other two investigated simulators , and by 50,000 tasks cloudsim already drops below linear scaling . it can be also observed that simulators handle the number of parallel machines completely differently . in the case of cloudsim , the bigger the machine number , the more likely that the scaling factor will become lower . in the case of groudsim , the opposite behaviour is observed ; while for dissect - cf , one can see a rather balanced case . in conclusion , groudsim behaves as expected : the missing simulation of physical machine behaviour and vm / pm scheduling techniques allows it to behave practically independently from infrastructure size . in this sense , cloudsim and dissect - cf should behave more closely to each other . unfortunately , the problematic implementation of cloudsim s time - shared vm scheduler ( see section [ sec - syntheval ] ) changes its apparent behaviour and reduces the similarities in the trends between dissect - cf and cloudsim . to conclude , dissect - cf scales comparably to other state - of - the - art simulators ( in fact it never drops below linear scaling , in contrast to the others ) while it offers significantly more detailed infrastructure level behaviour . this experiment was also used to _ validate the results of dissect - cf _ through a comparison with the other simulators . as the first step of validation , the simulator reported completion time of the last task in every trace specific measurement was collected ( from @xmath114 where @xmath107 was set between one and a million , and @xmath115 between one and five hundred ) . then , these task completion times were compared with each other . the median of the difference between the simulators was less than 0.001% ( the average difference from the median was 2.15% ) . the biggest differences occurred with simulated infrastructures containing the lowest number of physical machines : in such cases the median of difference between the simulators jumped to a little more than 0.29% ( with a sample standard deviation of 7.21% ) . the reason behind such a great deviation is the magnified effect of virtual machine instantiation simulation ( i.e. , as vms can only be instantiated on a single pm , they are mostly created and destroyed in a serial fashion ) . [ [ the - impact - of - energy - metering ] ] the impact of energy metering + + + + + + + + + + + + + + + + + + + + + + + + + + + + + so far , all presented measurements were executed without energy meters attached to the resource spreaders of dissect - cf . as energy metering is not available in groudsim , the other two simulators would have suffered a disadvantage because of their ongoing metering simulation . as dissect - cf was consequently a better performer than the other two simulators , its further evaluation was focused on how one could set up energy metering in order to get comparable performance to the other simulators meter - less operation . figure [ fig - energyperformancedrop ] analyses the performance drop of dissect - cf observed when energy metering was requested for the complete simulated infrastructure with 20 physical machines . just as before , the function @xmath121 was evaluated , but now instead of comparing dissect - cf to other simulators , the comparison was made between the newly received values to values from the previously presented simulations without energy metering . adding energy metering leads to significant performance differences between the various gwa traces . to show what is the possible performance degradation range in the various traces , the two extreme cases ( the best and worst performing ones ) are revealed in the figure . the presented cases show that selecting the energy measurement interval is crucial for well performing simulations . in general it is not recommended to use energy measurement intervals below one minute for long running simulations i.e. , where the trace s length is over a few thousand tasks . if higher precision is needed , then selective metering techniques should be used that evaluate energy meters for only the necessary parts of the infrastructure ; otherwise , one could experience a slowdown of over 50 times compared to meter - less runtimes . finally , in figure [ fig - energyequivalence ] , a comparison is shown about the performance drop caused by the metering to the meter - less setups of other simulators . the figure shows what the metering interval is despite the resulting performance drop that still provides equivalent performance to other simulators . as the figure shows , the performance drop also depends on the number of physical machines in the simulated infrastructure . overall , if energy metering is desired on the entire simulated infrastructure , the metering interval must be set to at least five - seconds to receive comparable performance to cloudsim . in contrast , groudsim s performance can be matched with a metering interval around 25 - 30 seconds . anything above these values will give a performance advantage to dissect - cf . on the other hand , metering intervals below these values could still cause a better performing dissect - cf but this is highly dependent on the particular trace or virtual machine request pattern used during simulations . this article outlined several iaas related schedulers that could be further improved with the use of cloud simulators . then , it has shown that current simulators barely meet the demands of the scheduling oriented researchers : @xmath0 they often limit the accessibility of information , @xmath1 they often hide the internal details of iaas systems , @xmath2 they frequently perform poorly in large - scale simulations , and @xmath79 they provide scarcely any options for introducing new energy management techniques inside iaas clouds . to overcome these issues , the article has proposed a new simulator called dissect - cf . the proposed simulator targets information accessibility issues with open apis and monitoring and performance related customisable events . the internal details of the cloud systems are also accessible , allowing simulation developers : @xmath0 to construct clouds in novel ways ( e.g. , introduction of new physical machine - virtual machine interaction techniques or cloud organisation topologies ) , and @xmath1 to experiment with new cloud side behaviour ( e.g. , new vm schedulers , power states ) . dissect - cf utilises a new unified resource sharing mechanism that allows centralised performance optimisations and ensures scaling independently of the size of the infrastructure and the amount of tasks processed by the simulated system . finally , the new simulator deeply integrates energy metering techniques ( ranging from resource usage counters and energy consumption models to meter aggregators ) . these techniques not only allow further extensions but they allow selective and composite power metering to ensure minimal performance drops during the metering sessions . the new simulator was evaluated by comparing it to small - scale but real - life environments . during this evaluation , it was presented how one should model cpu and memory intensive tasks , and it was also shown that the accuracy of the simulator s energy metering technique is also dependent on the new unified resource sharing mechanism . experiments shown that the relative error of the new resource sharing technique is around 1% in most cases . after the small - scale evaluation , dissect - cf was compared with two state - of - the - art simulators ( namely cloudsim and groudsim ) . the comparisons were focused on the scaling and performance characteristics of the simulators . the results revealed that these simulators often produce inaccurate results , while the new simulator not only provided more accurate outputs but also offered better performance . future research will consider several areas . first , there are plans to look how the simulator s resource sharing accuracy could be improved with stochastic sharing models and new low - level scheduling techniques . next , support for more complex network topologies and additional node types ( e.g. , routers and switches ) will also be prepared . as the current simulator merely provides raw data for user - side schedulers , an investigation of the necessary constructs and optimisations to better support the development and analysis of these schedulers also needs to be done . finally , memory behaviour is practically neglected in the current simulator , because applications and openly available traces do not offer any details on this . in the future , a task level complex memory model is expected to be delivered that not only considers memory bandwidth utilisation but also access patterns . with this model the simulator s target could include accurate live migration support . this article described the behaviour and features of dissect - cf version 0.9.5 . the source code of the simulator is open and available ( under the licensing terms of the gnu lesser general public license 3 ) at the following website : https://github.com/kecskemeti/dissect-cf . this work was partially supported by european union s horizon 2020 research and innovation programme under grant agreement no 644179 ( entice ) as well as by the cost program action ic1305 : network for sustainable ultrascale computing ( nesus ) , finally the research has also received partial funding from the austrian science fund project trp 237-n23 . a. ahmed and a.s . cloud computing simulators : a detailed survey and future direction . in _ advance computing conference ( iacc ) , 2014 ieee international _ , pages 866872 . ieee , february 2014 . doi:10.1109/iadcc.2014.6779436 . w. zhao , y. peng , f. xie , z. dai , modeling and simulation of cloud computing : a review , in : ieee asia pacific cloud computing congress ( apcloudcc ) , ieee , shenzhen , 2012 , pp . 2024 . doi:10.1109/apcloudcc.2012.6486505 . t. hirofuchi , a. lbre , adding virtual machine abstractions into simgrid : a first step toward the simulation of infrastructure - as - a - service concerns , in : third international conference on cloud and green computing ( cgc ) , ieee , karlsruhe , germany , 2013 , pp . 175180 . doi:10.1109/cgc.2013.33 . a. nez , j. l. vzquez - poletti , a. c. caminero , j. carretero , i. m. llorente , design of a new cloud computing simulation platform , in : computational science and its applications - iccsa 2011 , vol . 6784 of lecture notes in computer science , springer , santander , spain , 2011 , pp . 582593 . doi:10.1007/978 - 3 - 642 - 21931 - 3_45 . k. kurowski , a. oleksiak , w. pikatek , t. piontek , a. przybyszewski , j. wkeglarz , dcworms a tool for simulation of energy efficiency in distributed computing infrastructures , simulation modelling practice and theory 39 ( 2013 ) 135151 . doi:10.1016/j.simpat.2013.08.007 . d. kliazovich , p. bouvry , y. audzevich , s. khan , greencloud : a packet - level simulator of energy - aware cloud computing data centers , in : global telecommunications conference ( globecom 2010 ) , 2010 , pp . 15 . doi:10.1109/glocom.2010.5683561 . r. n. calheiros , r. ranjan , a. beloglazov , c. a. de rose , r. buyya , cloudsim : a toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms , software : practice and experience 41 ( 1 ) ( 2011 ) 2350 . doi:10.1002/spe.995 . s. ostermann , k. plankensteiner , r. prodan , t. fahringer , groudsim : an event - based simulation framework for computational grids and clouds , in : euro - par 2010 parallel processing workshops , vol . 6586 of lecture notes in computer science , springer , 2011 , pp . 305313 . doi:10.1007/978 - 3 - 642 - 21878 - 1_38 . velho , l. m. schnorr , h. casanova , a. legrand , on the validity of flow - level tcp network models for grid and cloud simulations , acm transactions on modeling and computer simulation ( tomacs ) 23 ( 4 ) ( 2013 ) 23 . doi:10.1145/2517448 . p. velho , a. legrand , accuracy study and improvement of network simulation in the simgrid framework , in : proceedings of the 2nd international conference on simulation tools and techniques , icst ( institute for computer sciences , social - informatics and telecommunications engineering ) , 2009 , p. 13 . doi:10.4108/icst.simutools2009.5592 . r. sobie , a. agarwal , i. gable , c. leavett - brown , m. paterson , r. taylor , a. charbonneau , r. impey , w. podiama , htc scientific computing in a distributed cloud environment , in : proceedings of the 4th acm workshop on scientific cloud computing , science cloud 13 , acm , new york , ny , usa , 2013 , pp . doi:10.1145/2465848.2465850 . h. song , j. li , x. liu , idlecached : an idle resource cached dynamic scheduling algorithm in cloud computing , in : 9th international conference on ubiquitous intelligence computing and 9th international conference on autonomic trusted computing ( uic / atc ) , 2012 , pp . 912917 . doi:10.1109/uic - atc.2012.24 . m. rodrguez , d. tapiador , j. fontn , e. huedo , r. montero , i. llorente , dynamic provisioning of virtual clusters for grid computing , in : e. csar , m. alexander , a. streit , j. trff , c. crin , a. knpfer , d. kranzlmller , s. jha ( eds . ) , euro - par 2008 workshops - parallel processing , vol . 5415 of lecture notes in computer science , springer berlin heidelberg , 2009 , pp . doi:10.1007/978 - 3 - 642 - 00955 - 6_4 . b. sotomayor , k. keahey , i. foster , combining batch execution and leasing using virtual machines , in : proceedings of the 17th international symposium on high performance distributed computing , hpdc 08 , acm , new york , ny , usa , 2008 , pp . doi:10.1145/1383422.1383434 . j. tordsson , r. s. montero , r. moreno - vozmediano , i. m. llorente , cloud brokering mechanisms for optimized placement of virtual machines across multiple providers , future generation computer systems 28 ( 2 ) ( 2012 ) 358 367 . doi:10.1016/j.future.2011.07.003 . e. elmroth , l. larsson , interfaces for placement , migration , and monitoring of virtual machines in federated clouds , in : eighth international conference on grid and cooperative computing . gcc 09 . , 2009 , doi:10.1109/gcc.2009.36 . k. tsakalozos , m. roussopoulos , a. delis , vm placement in non - homogeneous iaas - clouds , in : g. kappel , z. maamar , h. motahari - nezhad ( eds . ) , service - oriented computing , vol . 7084 of lecture notes in computer science , springer berlin heidelberg , 2011 , pp . 172187 . doi:10.1007/978 - 3 - 642 - 25535 - 9_12 . d. jayasinghe , c. pu , t. eilam , m. steinder , i. whally , e. snible , improving performance and availability of services hosted on iaas clouds with structural constraint - aware virtual machine placement , in : ieee international conference on services computing ( scc ) , 2011 , pp . doi:10.1109/scc.2011.28 . r. ghosh , v. naik , k. trivedi , power - performance trade - offs in iaas cloud : a scalable analytic approach , in : ieee / ifip 41st international conference on dependable systems and networks workshops ( dsn - w ) , 2011 , pp . doi:10.1109/dsnw.2011.5958802 . e. feller , c. rohr , d. margery , c. morin , energy management in iaas clouds : a holistic approach , in : ieee 5th international conference on cloud computing ( cloud ) , 2012 , pp . doi:10.1109/cloud.2012.50 . lizhe wang , samee u. khan , dan chen , joanna kolodziej , rajiv ranjan , cheng zhong xu , and albert zomaya . energy - aware parallel task scheduling in a cluster . , 29(7):1661 1670 , september 2013 . including special sections : cyber - enabled distributed computing for ubiquitous cloud and network services & amp ; cloud computing and scientific applications big data , scalable analytics , and beyond . doi:10.1016/j.future.2013.02.010 . d. ongaro , a. l. cox , s. rixner , scheduling i / o in virtual machine monitors , in : proceedings of the fourth acm sigplan / sigops international conference on virtual execution environments , vee 08 , acm , new york , ny , usa , 2008 , pp . doi:10.1145/1346256.1346258 . g. von laszewski , l. wang , a. younge , x. he , power - aware scheduling of virtual machines in dvfs - enabled clusters , in : ieee international conference on cluster computing and workshops . cluster 09 . , 2009 , pp . doi:10.1109/clustr.2009.5289182 . j. yang , x. zhou , m. chrobak , y. zhang , l. jin , dynamic thermal management through task scheduling , in : ieee international symposium on performance analysis of systems and software . ispass 2008 . , 2008 , pp . doi:10.1109/ispass.2008.4510751 . r. buyya , r. ranjan , r. n. calheiros , modeling and simulation of scalable cloud computing environments and the cloudsim toolkit : challenges and opportunities , in : international conference on high performance computing & simulation ( hpcs09 ) . , ieee , leipzig , 2009 , pp . doi:10.1109/hpcsim.2009.5192685 . r. buyya , m. murshed , gridsim : a toolkit for the modeling and simulation of distributed resource management and scheduling for grid computing , concurrency and computation : practice and experience 14 ( 13 - 15 ) ( 2002 ) 11751220 . doi:10.1002/cpe.710 . s. k. garg , r. buyya , networkcloudsim : modelling parallel applications in cloud simulations , in : fourth ieee international conference on utility and cloud computing ( ucc ) , ieee , victoria , nsw , 2011 , pp . doi:10.1109/ucc.2011.24 . a. beloglazov , r. buyya , optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in cloud data centers , concurrency and computation : practice and experience 24 ( 13 ) ( 2012 ) 13971420 . doi:10.1002/cpe.1867 . tom gurout , thierry monteil , georges da costa , rodrigo neves calheiros , rajkumar buyya , and mihai alexandru . energy - aware simulation with \{dvfs}. , 39:76 91 , december 2013 . energy efficiency in grids and clouds . doi:10.1016/j.simpat.2013.04.007 . x. li , x. jiang , p. huang , k. ye , dartcsim : an enhanced user - friendly cloud simulation system based on cloudsim with better performance , in : 2nd international conference on cloud computing and intelligent systems ( ccis ) , vol . 1 , ieee , hangzhou , 2012 , pp . 392396 . doi:10.1109/ccis.2012.6664434 . s. sotiriadis , n. bessis , n. antonopoulos , a. anjum , simic : designing a new inter - cloud simulation platform for integrating large - scale resource management , in : 27th international conference on advanced information networking and applications ( aina ) , 2013 , pp . doi:10.1109/aina.2013.123 . s. sotiriadis , n. bessis , n. antonopoulos , towards inter - cloud simulation performance analysis : exploring service - oriented benchmarks of clouds in simic , in : 27th international conference on advanced information networking and applications workshops ( waina ) , 2013 , pp . doi:10.1109/waina.2013.196 . y. shi , x. jiang , k. ye , an energy - efficient scheme for cloud resource provisioning based on cloudsim , in : ieee international conference on cluster computing ( cluster ) , ieee , austin , tx , 2011 , pp . doi:10.1109/cluster.2011.63 . alexandru - florian antonescu and torsten braun . sla - driven simulation of multi - tenant scalable cloud - distributed enterprise information systems . in florin pop and maria potop - butucaru , editors , _ adaptive resource management and scheduling for cloud computing _ , lecture notes in computer science , pages 91102 . springer international publishing , november 2014 . doi:10.1007/978 - 3 - 319 - 13464 - 2_7 . b. wickremasinghe , r. n. calheiros , r. buyya , cloudanalyst : a cloudsim - based visual modeller for analysing cloud computing environments and applications , in : 24th ieee international conference on advanced information networking and applications ( aina ) , ieee , 2010 , pp . doi:10.1109/aina.2010.32 . thiago teixeira s , rodrigo n. calheiros , and danielo g. gomes . . in zaigham mahmood , editor , _ cloud computing _ , computer communications and networks , pages 127142 . springer international publishing , october 2014 . doi:10.1007/978 - 3 - 319 - 10530 - 7_6 . h. casanova , simgrid : a toolkit for the simulation of application scheduling , in : first ieee / acm international symposium on cluster computing and the grid , ieee , brisbane , qld . , 2001 , pp . doi:10.1109/ccgrid.2001.923223 . t. hirofuchi , a. lbre , l. pouilloux , et al . , adding a live migration model into simgrid , one more step toward the simulation of infrastructure - as - a - service concerns , in : 5th ieee international conference on cloud computing technology and science ( ieee cloudcom ) , bristol , uk , 2013 , pp . 96103 . doi:10.1109/cloudcom.2013.20 . d. kliazovich , p. bouvry , s. u. khan , greencloud : a packet - level simulator of energy - aware cloud computing data centers , the journal of supercomputing 62 ( 3 ) ( 2012 ) 12631283 . doi:10.1007/s11227 - 010 - 0504 - 1 . a. nez , g. castane , j. vazquez - poletti , a. caminero , j. carretero , i. llorente , design of a flexible and scalable hypervisor module for simulating cloud computing environments , in : international symposium on performance evaluation of computer & telecommunication systems ( spects ) , ieee , 2011 , pp . 265270 . a. nez , j. l. vzquez - poletti , a. c. caminero , g. g. casta , j. carretero , i. m. llorente , icancloud : a flexible and scalable cloud infrastructure simulator , journal of grid computing 10 ( 1 ) ( 2012 ) 185209 . doi:10.1007/s10723 - 012 - 9208 - 5 . s. ostermann , k. plankensteiner , d. bodner , g. kraler , r. prodan , integration of an event - based simulation framework into a scientific workflow execution environment for grids and clouds , in : towards a service - based internet , vol . 6994 of lecture notes in computer science , springer , poznan , poland , 2011 , pp . 113 . doi:10.1007/978 - 3 - 642 - 24755 - 2_1 . i. sriram , speci , a simulation tool exploring cloud - scale data centres , in : cloud computing , vol . 5931 of lecture notes in computer science , springer , 2009 , pp . doi:10.1007/978 - 3 - 642 - 10665 - 1_35 . m. tighe , g. keller , m. bauer , h. lutfiyya , dcsim : a data centre simulation tool for evaluating dynamic virtualized resource management , in : 8th international conference on network and service management ( cnsm ) , ieee , 2012 , pp . 385392 . w. piatek , dcworms a tool for simulation of energy efficiency in data centers , in : energy efficiency in large scale distributed systems , vol . 8046 of lecture notes in computer science , springer , 2013 , pp . 118124 . doi:10.1007/978 - 3 - 642 - 40517 - 4_11 . m. tighe , g. keller , j. shamy , m. bauer , h. lutfiyya , towards an improved data centre simulation with dcsim , in : proceedings of the 9th international conference on network and service management , cnsm 2013 , ieee , zurich , switzerland , 2013 , pp . doi:10.1109/cnsm.2013.6727859 . g. kecskemeti , foundations of efficient virtual appliance based service deployments : new techniques for virtual appliance delivery and size optimization in infrastructure as a service clouds , isbn : 978 - 3 - 8484 - 0383 - 7 , lap lambert academic publishing , saarbrcken , 2012 .

cosmic reionization corresponds to the transition from a fully neutral to a highly ionized intergalactic medium ( igm ) , driven by ultraviolet ( uv ) radiation from the first stars and black holes . the transition is a key milestone in cosmic structure formation , marking the formation of the first luminous objects . reionization represents the last major epoch of cosmic evolution left to explore . study of the igm , galaxies , and quasars present during that time is a primary science driver for essentially all future large - area telescopes , at all wavelengths . recent observations of the gunn - peterson effect , i.e. , ly@xmath4 absorption by the neutral igm , toward the most distant quasars ( @xmath5 ) , and the large scale polarization of the cmb , corresponding to thompson scattering during reionization , have set the first constraints on the reionization process . these data suggest significant variance in both space and time , starting perhaps as far back as @xmath6 @xcite and extending to @xmath7 @xcite . current probes of the reionization are limited : present wmap - v data indicates the 5 @xmath8 detection of the e - mode of polarization which rules out any instantaneous reionization at @xmath9 at 3.5 @xmath8 level . for the gunn - peterson effect , the igm becomes optically thick to ly@xmath4 absorption for a neutral fraction as small as @xmath10 . it has been widely recognized that mapping the red - shifted hi 21 cm line has great potential for direct studies of the neutral igm during reionization @xcite . there are number of upcoming low - frequency arrays whose key science goal is to detect the hi 21 cm signal from the epoch of reionization ( eor ) . this includes the murchison widefield array [ mwa ] @xcite , precision array to probe epoch of reionization [ paper ] @xcite and low frequency array [ lofar ] @xcite . one of the major challenges for all of these upcoming arrays will be the removal of the continuum foreground sources in order to detect the signals from the eor . in this paper we discuss how the radio interferometric imaging techniques are going to affect the foreground source modeling and subsequent removal from the data - set in order to search for the eor signal . recently , there has been substantial research on foreground source modeling ( @xcite , @xcite , @xcite , etc ) at these low frequencies . similar effort has also been made in exploring different techniques to remove the foregrounds from the eor data - set by @xcite , @xcite,@xcite , @xcite , @xcite and @xcite . they primarily focus on the removal of the sources that are fainter than a certain @xmath11 ( @xmath12 jy ) level . most of these works do not consider the foreground sources brighter than @xmath11 and how accurately they need to be removed from the data - set by some real - time calibration or modeling technique @xcite in the uv - domain . in this paper we deal with the bright point sources above @xmath11 and the limitations that will be caused due to imperfect removal of such sources . one of the objectives of this paper is to demonstrate the effect of the frequency dependent side - lobes of the synthesized beam on the foreground source removal strategies . the effect of other frequency dependent terms like primary beam , ionosphere , etc will be addressed in future papers . in section-2 , we discuss briefly our choice of model for the eor signal , i.e. the largest expected cosmic stromgren sphere ( css ) which represents the only signature that can be detected in the image domain by the upcoming radio - telescopes . section-3 presents the detailed array parameters that have been used in the simulations . in section-4 , we discuss the foreground source model that has been used in every simulation performed for this paper . section-5 outlines the simulation methodology and a possible data reduction procedure that might be followed while processing the raw - data from the upcoming low - frequency radio telescopes in order to extract the eor signal . this procedure may not be exactly the same as what will be actually implemented for these upcoming telescopes . in section-6 , we discuss the propagation of different forms of error through the radio interferometric data reduction procedure . we consider two frequency independent errors : i ) position error in the global sky model ( gsm ) that is used to remove the bright sources above the @xmath11 level and ii ) residual calibration error . we also present the results showing how these errors propagate to the final residual ( foreground - removed ) spectral image - cube . finally , in the last section we discuss the implications of the results from our simulations and our recommendations for the upcoming low - frequency arrays in order to detect the signal of cosmic reionization . first generation low frequency arrays like mwa , paper , lofar , etc are meant to detect three potential signatures of eor : i ) 3d power spectrum , ii ) rare and large cosmic stromgren sphere ( css ) and iii ) hi 21 cm forest @xcite . all of the above signatures require the standard low - frequency radio - interferometric calibration and imaging like self calibration , frequency dependent calibration and source removal . for simplicity , we restrict ourselves to the large and rare cosmic stromgren sphere which are formed around the luminous quasars at the end of reionization . these css form the only potential eor signature that can be detected in the image domain by the upcoming radio telescopes . these css are rare and are expected to have a brightness temperature of 20 ( @xmath13 ) mk with a physical size of @xmath14 mpc physical size , where @xmath13 is the neutral fraction of the inter galactic medium ( igm ) . the physical size of these rare css has been derived from the ly-@xmath4 spectra @xcite . according to @xcite , we derive the angular size of the css ( @xmath15 ) from the co - moving radius of css ( @xmath16 ) : @xmath17 this gives the angular scale of the css to be @xmath18 arc - minutes along with a line - width of @xmath19 given by @xcite : @xmath20 the line - width of the css translates to 2.5 mhz . thus the total flux density of the css is about 0.24 @xmath13 mjy which gives a surface brightness of : @xmath21 where @xmath22 is the size of the synthesized beam for an array with maximum baseline 1.5 km at 158 mhz ( @xmath23 ) . for a late reionization model , @xcite predict that the @xmath24 field - of - view and 16 mhz bandwidth of mwa , will include atleast one of these large and rare hii regions ( @xmath25 mpc at @xmath26 ) . moreover , there is also the possibility of finding smaller hii regions ( @xmath27 mpc ) and up to @xmath28 fossil hii regions due to nonactive agn within the same field - of - view , depending on the duty cycle @xcite . table 1 outlines the basic array parameters that we have adopted . we note that most of these parameters reflect actual specifications for the upcoming mwa-512 array . figure 1 shows the array layout for the 512 element array with maximum baseline of 1.5 km . this might not be the final array design or specification for mwa . cc no . of tiles & 512 + central frequency & 158 mhz ( @xmath29 ) + field of view & @xmath30 15@xmath31 at 158 mhz . ( @xmath32 ) + synthesized beam & @xmath30 4.5 at 158 mhz . ( @xmath32 ) + effective area per tile & @xmath30 17 @xmath33 + maximum baseline & @xmath30 1.5 km . + total bandwidth & 32 mhz + @xmath34 & @xmath30 250 k + channel width & @xmath30 32 khz + thermal noise & @xmath30 15.4 @xmath35 jy / beam + & ( @xmath36 hours @xmath37 @xmath38 mhz ) + in order to detect the signal from cosmic reionization , arrays like mwa , paper , lofar , etc will have to overcome the thermal noise limitation @xcite : @xmath39 where @xmath40 denotes the final rms noise in the image from the channel width of @xmath41 after @xmath42 hours of integration . @xmath34 , @xmath43 and @xmath44 denotes the system temperature , effective collecting area of each element / tile and number of elements / tiles in the array . according to the above equation , the thermal noise is 15.4 @xmath35 jy / beam after @xmath45 hours of integration with @xmath46k at these low frequencies is dominated by the sky temperature ( @xmath47 ) @xcite . we have adopted the above noise value for 158 mhz . in practice , one should quote the noise figure for the lowest frequency edge which corresponds to the highest value for @xmath48 . ] and channel width of 2.5 mhz , which is also the spectral width of a css , as discussed in the previous section . most of the upcoming low frequency telescopes will be transit - instruments and will observe a field around its transit . hence we have used 6 hours of integrations for all the simulations , assuming that the telescopes will observe a field between @xmath49 3 hours in hour angle . here , we have assumed that the field - of - view of the instrument to be @xmath24 . in this paper , we aim to demonstrate the case for the 512-element array , as mentioned in table 1 . however , the simulations of 512-element are expensive . hence , we have considered a simpler geometry with 128-element array ( figure 1 ) , with a maximum baseline of 600 meters . this array design does not represent any of the upcoming telescopes but has been adopted in order to simplify the simulations . apart from the effective area and synthesized beam value the rest of the specifications for the 128-element array remains the same as in table 1 . in our test simulations , we have used identical residual calibration errors and created separate resultant spectral image cubes for the two different array specifications ( 128-element and 512-element ) . the rms noise level in identical regions of these two maps differed by a factor of @xmath50 . the synthesized beam or the psf ( point spread function ) image from these two array configurations shows that the rms noise level in psf side - lobes for the 512-element is a factor of @xmath50 lower than that in the 128-element array . this confirms that there is a scaling property between the 512-element and the 128-element array . the same scaling property is also evident in the results from our test simulations with the position errors . we have exploited this property and have used the 128-element array for all the simulations referred to in the later sections and have subsequently scaled down the rms noise values by the scaling factor , in order to represent the same for the 512-element array . we also note here that the thermal noise and the css surface brightness values that are quoted throughout the paper corresponds to the 512-element array configuration . since we are dealing with point sources in our foreground source model the change in the maximum baseline between the two configurations will not affect our conclusions . the only frequency dependent component in our simulations is the side - lobes of the point spread function ( psf ) . in this section we discuss the effects of different weighting schemes on the psf shape and its dependence on frequency . figures 2 and 3 show the effect of three different weighting functions on the psf side - lobes for the 512-element array . these figures show different plots from three well - known weighting schemes used in the synthesis imaging : natural , uniform and robust . @xcite showed that the robust weighting scheme smoothly varies between the uniform to natural weighting schemes . figure 3 shows how the near - by side - lobes are almost coherent for two different channels separated by 32 mhz while the far - away side - lobes become more stochastic in frequency . the effect of the different weighting scheme shows similar results for the 128-elements as well but with a higher psf side - lobe level . the rms value computed on the far away side - lobe level of the psf for the 128-element array is about @xmath50 times higher than that in the 512-element array . we recognize that the choice of optimal weighting scheme depends on a number of parameters based on the array configurations , telescope hardware specifications and also on precise data reduction algorithms used . hence all the upcoming eor experiments need an optimal weighting scheme for their individual data reduction procedures . however , exploring the optimal weighting scheme is beyond the scope of this paper . for our work , we have adopted natural weighting scheme . psf side - lobe level reduces with increase in hour angle coverage . in our simulations we have used six hours of observing time to produce each data - set . however , in the actual experiment the observing time for each data - set may be less than that . hence the psf side - lobe level will also degrade accordingly . our foreground sky model only includes point sources . no extended emission from galactic foreground is included as a part of the sky model . recently , there has been extensive research regarding the foreground source removal dealing with sources below @xmath51 jy level . most of these analysis implicitly assume that foreground sources above the @xmath11 level are being removed perfectly . but in reality any imperfect calibration will introduce artifacts in the residual data after source removal . hence in this paper our main aim is to explore the level of accuracy needed in these calibration procedures in order to ensure the residual errors from the strong foreground source removal do not obscure the detection of the signal from cosmic reionization . the choice of @xmath11 level is not totally arbitrary and will be discussed in the final section of this paper . the sky model is derived from the logn - logs distribution of sources and is termed as the global sky model ( gsm ) from now onwards . since our gsm only includes sources above 1 jy , we follow the source count from the 6c survey at mhz @xcite . the source count from the 6c survey is given as : @xmath52 for a field - of - view of @xmath24 the total number of sources ( @xmath53 jy ) equates to @xmath54 , following the above power - law distribution . the entire flux range between 1 - @xmath55 jy was divided into several bins . the source population in each of these bins has been predicted following the above logn - logs distribution . in order to assign fluxes to individual sources inside a bin , a gaussian random number generator has been used . the strongest source in our gsm is @xmath56 jy . the center of the field is chosen such that it coincides with one of the cold spots in the foreground galactic emission seen from the southern hemisphere . southern sky is chosen since both the upcoming arrays mwa and paper are being constructed in western australia . the exact field center used for the gsm is 4 hours in right ascension and -26 degree in declination . in order to assign a position to each of these sources within the field - of - view , another gaussian random number generator has been used which predicted the offset from the field center for respective sources . in the gsm all the foreground sources are flat spectrum , i.e. with zero spectral index ( @xmath57 ) . figure 4 shows the cleaned image of the gsm used for all of the simulations . the clean algorithm used in this process is a wide - field variant of the well - known clark - clean algorithm , which uses w - projection algorithm for 3-dimensional imaging . in practice , the upcoming eor telescopes should be using a similar global sky model ( gsm ) but created from the existing source catalogs available for the part of the sky and frequency ranges of their observations . these telescopes will use the gsm as the preliminary sky model in order to detect and subsequently remove the bright foreground sources in the observed data - set . all the simulations have been performed within the casa package . as mentioned before , simulations for both 512-elements as well as 128-elements have been conducted and compared in the later sections . the @xmath24 field - of - view will include @xmath54 bright sources ( @xmath58 jy ) . the individual flux densities of these foregrounds are @xmath59 time higher than the signal from cosmic reionization that these instruments are aiming to detect . so the challenge lies in calibration and subsequent removal of such bright sources from the raw data - sets . in order to add to the constraints , the data rates of most of these telescopes ( e.g. @xmath60 gb / s for mwa ; @xcite ) will not allow them to store the raw visibilities produced by the correlator . hence real - time calibration and imaging needs to be done in order to reduce the data volume and store the final product in the form of image cubes @xcite . the critical steps include removal of the bright sources above the @xmath11 level from the data - sets in these iterative rounds of real - time calibration and imaging procedure . as a result the residual image - cubes will not be dominated by these bright sources and the rest of the foregrounds can be removed in the image domain . however , the accuracy of the foreground source removal strategies are strongly dependent on the data reduction procedure . the likely data reduction procedure which will be followed by the upcoming telescopes can be outlined as : * the raw data - sets from the correlator will go through real - time calibration and subsequent removal of the bright sources based on some global sky model ( gsm ) , down to @xmath11 level , in the uv domain . * the residual data - sets will be imaged and stored as a cube for the future processing and removal of sources which are below @xmath11 . however , due to imperfect calibration during the process of real - time calibration and source removal if any systematic error is incorporated in the residual image - cubes they can not be minimized by simply averaging over a large number of such residual image - cubes . hence to understand the level of accuracy required in the calibration procedure to reduce the residual systematic errors in the data - set is critical . here , we outline the reduction pathway that have been followed throughout the paper : * the observed visibilities ( @xmath61 or @xmath62 ) are simulated for 6 hours ( @xmath63 hours in hour - angle ) from the gsm ( with no position or flux error ) and the array configuration from table : 1 . * model visibilities ( @xmath64 ) are also generated using the same gsm but they are now corrupted with either gsm position errors or with residual calibration errors . * model visibilities are subtracted from the observed visibilities to residual visibilities ( @xmath65 ) . this step will be referred to as uvsub @xcite . * we fit a third order polynomial in frequency to the dirty image - cube ( @xmath66 ) formed from @xmath67 . this step will be referred to as imlin @xcite . after the imlin step , we investigate the rms noise in the residual image cube . we have tried to use different orders of polynomial for the imlin step but our conclusions did not change significantly . this may be mainly due to the fact that the only frequency dependent term in these simulations is the psf . however , it should be mentioned in this context that on using higher order polynomials in the imlin step we will be taking out relevant structures at those scales in the frequency space @xcite . since the reionization signal is a spectral signature with a width @xmath68 mhz ( size of the largest possible css ) , using a very higher order polynomial in the imlin might remove the signature itself . we have also explored uvlin @xcite , which fits a polynomial in the uv domain to @xmath67 . but uvlin works perfectly only within a small field - of - view depending on the channel width in frequency @xcite . hence , in our proposed reduction pathway we are using uvsub followed by imlin only . in this section we describe the effects of different errors arising from radio interferometric data processing and how they propagate and show up as artifacts in the final residual image . here , we consider effects of the errors arising from the imperfect gsm and residual calibration errors on removal of strong foregrounds . the source positions in the global sky model ( gsm ) are accurate up to a level , e.g. the nvss catalog has an accuracy of @xmath69 arc - seconds in positions of the sources detected @xcite . the real - time calibration procedure for mwa like telescopes might implement some method to iteratively solve for the positions of bright sources taking the initial value from the gsm , e.g. asp - clean @xcite . an iterative scheme will also converge with some residual position and flux errors in the model . hence , any error in the final gsm position or flux will contribute to imperfect source removal . the resultant artifact will limit the dynamic range in the residual data - set and obscure the detection of the faint signal from reionization . the main aim of the simulations in this section is to determine the accuracy of the gsm ( position and flux ) required in order to reach the required rms noise level . the gsm position errors considered in the following simulations are actually systematic residual errors after the iterative position determination . here , we deal with residual gsm errors in position and will also mention briefly the effect of the gsm flux errors in the residual data - set . in order to introduce the position error in the gsm we have added an error term in the right ascension ( @xmath4 ) of each of the sources in the gsm . this error term is derived from a gaussian distribution:- @xmath70 where @xmath8 denotes the error level in the position , i.e. for nvss catalog @xmath71 arc - seconds . the steps in the simulation performed in order to explore the effect of the gsm error in position is as follows : @xmath61 are simulated . 2 . predict model visibilities from the gsm model with source - position error ( @xmath72 ) . 3 . subtract the model visibilities from the observed visibilities to get @xmath73 . 4 . make @xmath66 and apply imlin as mentioned befor.e in order to explain the above mentioned steps in detail we refer to figure 5 , which demonstrates the procedure of the reduction followed in the simulations involving gsm errors . figure 5a shows the dirty map ( peak value : @xmath74 jy / beam ) from the residual visibilities where the model visibilities where corrupted with a gsm position error of 0.01 arc - second . figure 5b ( peak value : @xmath75 jy / beam ) is generated after applying imlin to the residual image . the gray - scale color - bars represent the variation of the intensity in two images . here we discuss the implications of the results from the simulations with the gsm errors . figure 6a represents the variation in the rms noise level of three regions in : [ 1 ] true image of @xmath61 ( figure 4 ) , [ 2 ] image of @xmath76 after uvsub ( figure 5a ) and [ 3 ] image after imlin ( figure 5b ) . the gsm position errors used in this case is 0.01 arc - second . in figure 6b , the rms noise value in three different parts of the field - of - view have been plotted as a function of different residual gsm position errors . the rms values quoted here are from the images of residual visibilities after uvsub [ 2 ] . hence the solid line denotes the trend in the decrease of the rms noise in the uvsub image with the decrease in the gsm position error . the magnitude of the position error ( value of @xmath8 as in equation 6 ) has been varied from 6 arc - seconds ( as in nvss catalog ) down to an extremely low value of @xmath77 arc - seconds . the original simulations were performed with 128-element array and the resulting rms values are scaled down by factor of 5 in order to represent the same for the 512-element array . this scaling property has already been discussed in details in section-3 of this paper . according to the figure 6b it is evident that an accuracy of 0.1 arc - second in gsm position is needed to achieve the required rms noise in order to detect the reionization signal after removal of strong foreground point sources . the solid curve in figure 6b represents the mean curve of the rms noise level variation and has a mean slope of @xmath12 in the log - log space . this means that a decrease in the rms noise level by another order of magnitude is possible on achieving an extra order of magnitude accuracy in the gsm position estimates . @xcite have shown similar results in demonstrating the effect of the pixelation error on the dynamic range of the image . recently , @xcite have shown that a image made from a data - set using the vla ( 27-element ) obtains a dynamic range limitation of @xmath78 when the sources are shifted by 0.01 pixel in the image plane . this result is similar to the result quoted in figure 6b , where it is evident that gsm position error of 1 arc - second ( @xmath79 pixel ) results in an rms noise of @xmath80 jy / beam corresponding to a dynamic range ( dr ) of @xmath81 . the better psf side - lobe level in 512-element array over vla27-element array causes an extra order of magnitude gain in the dynamic range in the case of the former . here , we have considered only systematic residual gsm position errors which are the same from one day to the another . this will not reduce after averaging over residual image - cubes from successive days of observations . the real - time calibration system might implement an iterative scheme ( e.g. asp - clean , @xcite ) to determine the source positions . however , unless the systematic error determined by any such scheme is smaller than 0.1 arcsecond , averaging over long lengths of time will not mitigate the errors due to imperfect gsm . the gsm position error limit that we report in this paper is thus also applicable for the specifications of the accuracy that any such scheme delivers . in this section we discuss the results of introducing gsm flux error in our simulations . gsm flux errors have been derived from a gaussian random distribution ( similar to equation 6 ) with mean zero and @xmath82 of flux values of respective sources . similar to the steps of simulations discussed in section 6.1 , the gsm model with flux error is used to corrupt model visibilities . the residual image cube shows sources with flux values of @xmath83 of the original sources in the same position . since the gsm contains only sources between 1 jy and @xmath56 jy , the maximum residual flux density in the residual image is @xmath84 jy / beam and can be denoted as sources below the @xmath11 limit @xcite . the removal of such sources are beyond the scope of this paper . flux errors might be as high as @xmath85 , as quoted for most of the source catalogs such as nvss . the residuals will show up as sources with flux density as high as few jy / beam . in this case the residuals are well above the level of the @xmath11 and perfect removal of such sources in the subsequent imlin step still remains a challenge . this section deals with the effect of residual calibration error in the foreground source subtraction from the raw data - set coming out of the upcoming eor instruments . the antenna dependent complex gains can be defined as : @xmath86 where @xmath87 s are the antenna dependent complex gains , @xmath88 are the corrupted visibilities and @xmath89 are the true visibilities or @xmath62 in our case ; @xmath90 denotes time variation . the model for the complex gains used in the simulations is given by : @xmath91 where @xmath92 , @xmath93 , @xmath94 and @xmath95 are for amplitudes ( @xmath96 ) and phases ( @xmath97 ) . the gains in both amplitude and phase are derived from individual gaussian random distribution as in equation 6 . the process of simulation is as follows : 1 . @xmath61 are simulated . model visibilities are predicted from a perfect gsm and are corrupted with @xmath98 to compute @xmath99 3 . in uvsub we get : @xmath100 . 4 . make @xmath101 and apply imlin as mentioned before . the model for the complex antenna - dependent gain errors ( equation 7 ) includes two terms for amplitude ( @xmath92 ) and phase @xmath93 gains , which are constant for each antenna throughout each day of observing ( 6 hours ) . along with them there are also two small additive random offsets @xmath102 , which vary within a single day of observation . following @xcite , it has been determined through several simulations that the constant offsets ( @xmath103 ) are the dominant terms compared to the time dependent terms . moreover , since the time variation for @xmath104 are derived from a gaussian distribution similar to equation 6 , the effect of these terms decreases upon averaging over time . hence , in our simulations we have only included the constant gain errors @xmath105 and redefined equation 8 as : @xmath106 it is certainly evident that some variant of the self calibration algorithm will be implemented in the real - time calibration procedure for the upcoming mwa like telescopes in order to facilitate the bright foreground source removal . in our simulations , we have not implemented any form of calibration algorithm . instead , we have used the measurement equations with residual calibration errors in order to explore the accuracy with which any calibration procedure must work in order to achieve the desired rms noise limit and detect the signal from reionization . figure 7 demonstrates the three steps followed to simulate the effect of the residual calibration errors in the eor data - set . figure 7a shows the dirty image of the residual uvsub data and figure 7b shows the image after imlin . both these figures are from the simulation with @xmath107 amplitude error or @xmath108 degree phase error . figure 8a shows the variation in the rms noise level in the three regions of the images made using : [ 1 ] @xmath61 ( figure 4 ) , [ 2 ] @xmath76 in uvsub ( figure 7a ) and [ 3 ] after imlin ( figure 7b ) . it is clear from figure 8a that the rms noise level reached near the bright sources after the uvsub stage is @xmath10 jy / beam , corresponding to a dynamic range of @xmath109 . @xcite calculated the dynamic range limitation due to a time independent amplitude error in all the baselines that will be introduced to a point source data . if we denote the visibility function ( @xmath110 ) as : @xmath111 where @xmath112 are the amplitude and phase errors respectively , the dynamic range of the image is limited to : @xmath113 where n is the number of antennas . in our case , @xmath114 , @xmath115 and @xmath116 degree . these values yield a dynamic range of @xmath109 , which agrees with the dynamic range we obtained from our simulations . figure 8b denotes the rms noise value in different parts of the uvsubed image as a function of different residual calibration errors . the three different residual calibration errors considered here are @xmath117 , @xmath107 and @xmath118 in amplitude or 0.01 degree , 0.1 degree and 1 degree in phase respectively . the resultant curve in the log - scale has a slope of @xmath119 . this indicates that in order to achieve the desired rms noise level to detect the reionization signal , the residual calibration errors should be @xmath120 in amplitude or @xmath121 degree in phase . in practice , it is extremely challenging to achieve such accuracy with a real - time calibration procedure . here , we have only considered systematic residual calibration errors which will not vary from one day to the other . as before , these simulations were originally performed for the 128-element array and then the rms noise values were scaled down by a factor @xmath50 following the scaling property that has been discussed in initial sections . however , we have also repeated the simulations for the 512-element array with residual calibration error of @xmath122 in amplitude or 0.1 degree in phase and found the results consistent with our scaled values . in the previous section it is clearly shown that even the low systematic residual calibration errors restrict the dynamic range in the residual image to @xmath78 , whereas the desired dynamic range is @xmath123 in order to detect the signal from reionization . the obvious next step to follow is to add the uvsubed image cubes from successive days in order to check whether the residual calibration error reduces as @xmath124 , where @xmath125 denotes the number of days over which the uvsub images has been added and each day consists of 6 hours of observations . we can demonstrate the same from the expression of the residual visibilities ( @xmath100 ) . if the calibration procedure is able to remove all systematic gain errors from the corrupted data contributing to residual calibration errors that are purely random from day to day , then they will reduce upon averaging . figure 9 demonstrates the effect of the averaging over 20 days of uvsubed images with the residual calibration errors of @xmath122 in amplitude or @xmath108 degree in phase . figure 9a denotes the dirty image after the first day and figure 9b denotes the averaged image after 20 days . the figures demonstrate a significant reduction in the rms noise level after 20 days of averaging . figure 10a shows the reduction in the rms noise level in different parts of the image on averaging over @xmath126 days . the solid line denotes the theoretical noise curve that follows @xmath124 . hence , the reduction in rms noise on averaging over different days is consistent with the theoretical prediction ( @xmath124 ) . a uvsubed image from each day ( 6 hours ) is also limited by the thermal noise , which reduces down as @xmath124 . if the residual calibration errors are random between successive data - sets from different days of integration then it should also reduce down as @xmath124 . as a result , it should then be sufficient to achieve a resultant rms noise level in the images below the daily thermal noise limit . from table 1 we can estimate that the thermal noise limit from 1 day of integration ( 6 hours ) is @xmath127 jy / beam . figure 9b is the same as figure 8b except that we have plotted the new noise estimates from 6 hours of observation . this figure now explains that the residual calibration errors should be @xmath128 in amplitude or @xmath129 degree in phase in order to reach below the daily thermal noise limit . once the rms noise level in the uvsubed image is below the daily thermal noise limit , it can reduce down along with the thermal noise in order to detect the faint reionization signal within the time limit set by the telescope sensitivity ( i.e. @xmath45 hours ) . the most important caveat in the entire analysis is the fact that we assume that the residual calibration errors are not systematic from one day to the next . if there are any systematics left over from day to day ( e.g. due to antenna gain errors , antenna primary beam errors , pointing errors , etc ) then those will not go down as @xmath130 . moreover , even with these assumptions the much needed accuracy in calibration is alarmingly high . there are few important results and concerns that have come out from the simulations performed in this paper . the simulations have been performed for both 128-element array and 512-element array . necessary comparisons between both arrays have been discussed at essential stages of this paper . the comparison between the two configuration can be well explained by a simple scaling relation which is the ratio between the rms value of the psf side - lobe levels in two configurations . hence it was sufficient to perform the 512-element simulations only for the optimal value of the gsm position error and residual calibration error . the choice of @xmath51 jy in our gsm is not totally arbitrary keeping in mind the fact that the real - time calibration procedure needs to model and remove all the sources down to @xmath11 from the raw uv data - sets within a duration constrained by the instrumental specifications . since the errors we have discussed in this paper are dominated by the bright sources inside the field - of - view , our analysis will not change significantly by reducing the @xmath11 level and including more and more fainter sources . in the section dealing with the gsm position errors , we have only discussed about systematic errors which do not change from one day to the next . if the calibration procedure involves some iterative position calibration scheme that will evaluate the source position everyday then the errors from those procedure should vary from one day to the other . hence , there is a possibility of reaching the desired rms noise level even with lesser accuracy in the gsm position estimates . in practice , a better gsm can be obtained a priori from a deepsurvey of the specific fields to be observed with mwa , lofar , etc . using the present facilities like gmrt and wsrt @xcite at 150 mhz . mwa , lofar , etc can also improve the existing gsm after sufficient days of observations . the simulations with calibration errors assume that the image cubes are formed after each day of observations ( 6 hours ) . we assume that the calibration errors are incoherent beyond 6 hours and as a result the effect of calibration error will reduce with successive days of integration . if calibration error have shorter coherence time , then lesser accuracy in the calibration can be tolerated in order to reach below the thermal noise limit for the shorter integrations . however , the systematics need to be removed at those timescales to make sure that the residual errors are still random from one data - set to the next . it should be noted that the psf side - lobe level will also be higher due to shorter integration period , which in turn might make it harder to solve and remove the systematics at a shorter timescales . the data analysis pipeline adapted in this paper resembles mostly the upcoming mwa telescope . since the raw visibilities can not be stored , the calibration and removal of the brightest sources in the real - time calibration pipeline is one of the major step in foreground subtraction @xcite . the accuracy in gsm position and residual calibration ( discussed in this paper ) is critical for extraction of the eor signal . in other telescopes like lofar , where it is possible to store the raw visibilities , the calibration will be carried out through a combination of real - time processing and off - line reprocessing scheme . the ability to store the raw visibilities will allow more accurate calibration of the data in the off - line mode . however , the combination of real - time and off - line calibration pipeline for lofar still needs to achieve similar accuracy in calibration ( discussed in this paper ) in order to reach the desired dynamic range to extract the eor signal . in the present paper we have not discussed in great detail the step of polynomial fitting in the image domain . this is mainly because the lack of frequency dependence of the sky model as well as the instrumental gain model . our simulations do not include thermal noise . in practice , the imlin step should be applied to the residual image - cubes , averaged over total number of days required to beat thermal noise . hence , it should produce similar decrease in the rms noise level as obtained in this paper on `` noise - less '' data . the major feedback from all these simulations that we get is that the raw uv - data should be retained , even considering the huge data - rate issue , until the iterative real - time calibration procedures achieve the desired accuracy . upcoming telescopes like mwa , paper , lofar , etc will also be trying to detect the eor signal in the power spectral domain . hence , a relevant extension of our present work will be to extend these analysis into the power spectral domain . we are also planning to deal with frequency dependent and direction dependent calibration errors in subsequent publications . ad and cc are grateful for support from the max - planck society and the alexander von humboldt foundation through the max planck forshungspreise 2005 . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . , d. c. , parsons , a. , bradley , r. , parashare , c. , gugliucci , n. , mastrantonio , e. , herne , d. , lynch , m. , wright , m. , werhimer , d. , carilli , c. , datta , a. , & aguirre , j. 2007 , in bulletin of the american astronomical society , vol . 38 , bulletin of the american astronomical society , 967+ , g. , de bruyn , a. g. , brentjens , m. a. , ciardi , b. , harker , g. , jeli , v. , koopmans , l. v. e. , labropoulos , p. , offringa , a. , pandey , v. n. , schaye , j. , thomas , r. m. , yatawatta , s. , & zaroubi , s. 2009 , , 500 , 965 , d. s. , schwab , f. r. , & sramek , r. a. 1999 , in astronomical society of the pacific conference series , vol . 180 , synthesis imaging in radio astronomy ii , ed . g. b. taylor , c. l. carilli , & r. a. perley , 127+ , g. , zaroubi , s. , bernardi , g. , brentjens , m. a. , de bruyn , a. g. , ciardi , b. , jeli , v. , koopmans , l. v. e. , labropoulos , p. , mellema , g. , offringa , a. , pandey , v. n. , schaye , j. , thomas , r. m. , & yatawatta , s. 2009 , , v. , zaroubi , s. , labropoulos , p. , thomas , r. m. , bernardi , g. , brentjens , m. a. , de bruyn , a. g. , ciardi , b. , harker , g. , koopmans , l. v. e. , pandey , v. n. , schaye , j. , & yatawatta , s. 2008 , , 389 , 1319 , e. , dunkley , j. , nolta , m. r. , bennett , c. l. , gold , b. , hinshaw , g. , jarosik , n. , larson , d. , limon , m. , page , l. , spergel , d. n. , halpern , m. , hill , r. s. , kogut , a. , meyer , s. s. , tucker , g. s. , weiland , j. l. , wollack , e. , & wright , e. l. 2008 , arxiv e - prints , p. , koopmans , l. v. e. , jelic , v. , yatawatta , s. , thomas , r. m. , bernardi , g. , brentjens , m. , de bruyn , g. , ciardi , b. , harker , g. , offringa , a. , pandey , v. n. , schaye , j. , & zaroubi , s. 2009 , arxiv e - prints , c. j. , cappallo , r. j. , morales , m. f. , briggs , f. h. , benkevitch , l. , bowman , j. d. , bunton , j. d. , burns , s. , corey , b. e. , desouza , l. , doeleman , s. s. , derome , m. , deshpande , a. , gopalakrishna , m. r. , greenhill , l. j. , herne , d. , hewitt , j. n. , kamini , p. a. , kasper , j. c. , kincaid , b. b. , kocz , j. , kowald , e. , kratzenberg , e. , kumar , d. , lynch , m. j. , madhavi , s. , matejek , m. , mitchell , d. , morgan , e. , oberoi , d. , ord , s. , pathikulangara , j. , prabu , t. , rogers , a. e. e. , roshi , a. , salah , j. e. , sault , r. j. , udaya shankar , n. , srivani , k. s. , stevens , j. , tingay , s. , vaccarella , a. , waterson , m. , wayth , r. b. , webster , r. l. , whitney , a. r. , williams , a. , & williams , c. 2009 , arxiv e - prints , r. m. , zaroubi , s. , ciardi , b. , pawlik , a. h. , labropoulos , p. , jelic , v. , bernardi , g. , brentjens , m. a. , de bruyn , a. g. , harker , g. j. a. , koopmans , l. v. e. , mellema , g. , pandey , v. n. , schaye , j. , & yatawatta , s. 2008 , arxiv e - prints

one of the main industrial applications of bitumen is as a binder in asphalt pavement [ ] . bitumen links together mineral aggregates and filler particles to form a cohesive asphalt on the road surface . over the years , a chemical reaction takes place in bitumen increasing the number of heavy aromatic molecules [ ] . this process is called chemical aging . the aromatic molecules in bitumen , especially the asphaltene molecules , tend to align to form nanoaggregates . the aging reaction leads also to an increase in the nanoaggregate size [ ] correlated with an unwanted increase in bitumen viscosity and brittleness [ ] . the change in bitumen mechanical properties , which go from `` liquid - like '' to more `` solid - like '' during chemical aging finally results in cracks in the pavement and road deterioration . to prevent or reverse the effects of chemical aging , a first step is to gather more knowledge about the nanoaggregate structure and stability in bitumen . much progress has been made over the last 50 years in the experimental and numerical literature to determine the structure of the nanoaggregates and the conditions under which they are formed . one can cite in particular the design of the yen - mullins model [ , ] describing the nanoaggregate structure , the determination of the critical nanoaggregate concentration in different solvents [ , ] , the evaluation of the average size and polydispersity of a nanoaggregate [ ] on the experimental side . on the numerical side , different stable conformations of nanoaggregates were identified depending on the asphaltene structure using molecular mechanical calculations [ ] and molecular dynamics ( md ) simulations [ ] . md simulations were also used to determine the molecular orientation inside the nanoaggregates [ ] , and the effects of solvent and presence of other molecules on the nanoaggregate structure [ ] . however , analytical models for the thermodynamics stability and dynamics of the nanoaggregates in relation to their structure are still quite rare , to the notable exception of ref . [ ] . the purpose of this paper is precisely to suggest simple and generic models , which can reproduce md results on the nanoaggregate stability . we present molecular dynamics results concerning the nanoaggregate size in the stationary regime . in bitumen , nanoaggregates are composed of asphaltene molecules , the most heavy and aromatic fraction in bitumen , but also of lighter aromatic molecules like resin and resinous oil [ ] . we give results on the probability of having a nanoaggregate containing a given number of aromatic molecules and the probability of having a nanoaggregate containing a given number of asphaltene molecules . the two probabilities are shown to differ qualitatively . we first model the simpler case , where only asphaltene molecules are counted . the focus in this case is to establish a mechanism for the nanoaggregation of asphaltene molecules . from this aggregation mechanism , a master equation and a statistical mechanics model are derived and provide a simple theoretical framework for interpreting the simulation results on nanoaggregate stability . the results of the master equation approach are also compared to the aggregation dynamics observed in the md simulations . then , the more complicated case where all aromatic molecules are counted is discussed in terms of statistical mechanics arguments . the md simulations carried out in this work are based on the four - component united - atom - unit model developed in ref . [ ] in the framework of the cooee project [ ] . the model is shown to reproduce a generic bitumen reasonably well [ ] . the simulations were performed on graphic - processor - units ( gpu ) using the roskilde university molecular dynamics ( rumd.org ) package [ ] . the paper is organized as follows . in sec . [ sec : results ] , we provide simulation details , give a definition of a linear nanoaggregate , and present the md results about the probability of having a nanoaggregate of a given size . section [ sec : mastereq ] is devoted to model the results on the stability of linear aggregates where only asphaltene molecules are counted with a master equation approach . in sec . [ sec : statmech ] , a related statistical mechanics model is described and shown to reproduce the md results on asphaltene nanoaggregate stability . the more complicated case where all aromatic molecules are counted in a nanoaggregate is also discussed in sec . [ sec : statmech ] in terms of statistical mechanics arguments . [ sec : discussion ] includes a comparison of our results to existing experimental and numerical results and discuss the limit of our model . finally , sec . [ sec : conclu ] contains a summary and a conclusion . before presenting the molecular dynamics ( md ) results on nanoaggregate stability , we mention a few details about the simulations and define precisely a nanoaggregate . as mentioned in the introduction , the simulation method and molecular potentials are described in details in ref . only information necessary to understand the study on nanoaggregate stability carried out in the present paper is given here . the simulated system contains four types of molecule , chosen to resemble the sara classification [ ] : a saturated hydrocarbon , a resinous oil molecule , which is denoted aromatic in the sara scheme , a resin molecule and an asphaltene molecule . the molecular structures chosen are shown in fig . [ fig : molecule ] . the main system studied in this paper contains @xmath0 saturated hydrocarbons , @xmath1 resinous oil molecules , @xmath1 resin molecules and @xmath1 asphaltene molecules , which corresponds to @xmath2 united atom units . the methyl ( ch@xmath3 ) , methylene ( ch@xmath4 ) , and methine ( ch ) groups are represented by the same united atom unit of molar mass 13.3 g@xmath5mol@xmath6 and the sulfur atoms are represented by a united atom unit with a different molar mass 32 g@xmath5mol@xmath6 . the potential between the united atom units contains four terms : an intermolecular potential , corresponding to a lennard - jones potential with parameters @xmath7 @xmath8 and @xmath9 k , where @xmath10 is the boltzmann constant and three terms for the intramolecular potential . these three terms describe the bond length between two connected particles , the angle between three consecutive particles , and the dihedral angle between four consecutive particles . the parametrization of the intramolecular potential is described in details in refs [ ] . the simulations are performed in the canonical ensemble ( nvt ) at a constant temperature @xmath11 k and a constant density . the density @xmath12 kg@xmath5l@xmath6 is chosen to obtain an average pressure around the atmospheric pressure . a nos - hoover thermostat is used . the time step is @xmath13 fs and the duration of the simulations is @xmath14 ns . eight independent simulations are performed at the same state point . the molecular dynamics package rumd [ ] is used to perform the calculation . + it was shown in a previous work [ ] that the asphaltene , resin , and resinous oil molecules tend to align with respect to each other . they align at a distance of around @xmath15 , close to the minimum of the lennard - jones potential between the molecules . the lennard - jones potential used in the md simulations mimics the @xmath16-stacking interaction observed between aromatic molecules experimentally . the alignment of aromatic molecules in bitumen is the basis of the nanoaggregate formation . the definition of a nanoaggregate is described in detail in ref . it is based on the following rule : two aromatic molecules are nearest neighbors in the same nanoaggregate if they are `` well - aligned '' and `` close enough '' . a nanoaggregate is composed of all molecules connected by this rule . moreover , the asphaltene molecule chosen in this model has two parts , a flat head and a flat body which can rotate with respect to each other . for this reason , aromatic molecules can align in the direction of an asphaltene body or in the direction of an asphaltene head , thus creating branched nanoaggregates . these branches can link together purely linear nanoaggregates . we believe this is one mechanism explaining the formation of clusters of nanoaggregates , also observed experimentally [ ] . in this paper , we focus on purely linear nanoaggregates . it means that any aromatic molecules linked to an asphaltene head will not be considered as part of this asphaltene nanoaggregate . we define a linear aggregate as composed of asphaltene bodies , resin and resinous oil molecules , and not asphaltene heads because heads are the smallest parts and probably lead to the smallest interaction energy . figure [ fig : linearaggregate ] shows conformations of two molecules corresponding to limiting cases of the nanoaggregate definition used in this paper . figures [ fig : linearaggregate](a ) and ( b ) show two conformations of two asphaltene molecules where these molecules are considered as nearest neighbors . conversely , figs [ fig : linearaggregate](c ) to ( e ) show three conformations where the two molecules are not considered as nearest neighbors . a picture of a linear nanoaggregate obtained from the md simulations is shown in fig . [ fig : linearaggregate](f ) . note that this nanoaggregate is not typical as a resin and a resinous oil molecule are aligned on the same side of an asphaltene body . ( color online ) ( a ) to ( e ) scheme of limiting cases illustrating whether a molecule is or is not the nearest neighbor of the first molecule . the first molecule is an asphaltene molecule and is black . the second molecule is green if it is the neighbor of the first one and red otherwise . ( a ) head - to - head conformation . ( b ) head - to - tail conformation . ( c ) non - aligned molecules . ( d ) molecule aligned to the head of the asphaltene molecule . ( e ) molecules far from each other and aligned . modified with permission from c. a. lemarchand , t. b. schroder , j. c. dyre , and j. s. hansen , j. chem . 139 , 124506 ( 2013 ) . copyright 2013 american institute of physics . ( f ) snapshot of a linear nanoaggregate , obtained in molecular dynamics . asphaltene molecules are in blue , resin molecules in red and resinous oil molecules in green . , title="fig : " ] ( color online ) ( a ) to ( e ) scheme of limiting cases illustrating whether a molecule is or is not the nearest neighbor of the first molecule . the first molecule is an asphaltene molecule and is black . the second molecule is green if it is the neighbor of the first one and red otherwise . ( a ) head - to - head conformation . ( b ) head - to - tail conformation . ( c ) non - aligned molecules . ( d ) molecule aligned to the head of the asphaltene molecule . ( e ) molecules far from each other and aligned . modified with permission from c. a. lemarchand , t. b. schroder , j. c. dyre , and j. s. hansen , j. chem . 139 , 124506 ( 2013 ) . copyright 2013 american institute of physics . ( f ) snapshot of a linear nanoaggregate , obtained in molecular dynamics . asphaltene molecules are in blue , resin molecules in red and resinous oil molecules in green . , title="fig : " ] sections [ sec : simulationdetail ] and [ sec : defnanoaggregate ] gave necessary information about the simulations and the definition of linear nanoaggregates . the results of the molecular dynamics simulations concerning the nanoaggregate size and stability can now be presented . there are different possibilities to quantify the size of a linear nanoaggregate . for example , it can be quantified as the total number of aromatic molecules or as the number of asphaltene molecules which it contains . the first definition makes use of the fact that nanoaggregates are not only composed of asphaltene molecules , but also of smaller aromatic molecules . the second definition accounts for the fact that asphaltene molecules are the largest aromatic molecules and probably the most important in the nanoaggregate stability . this is why experimentalists often define the nanoaggregate size as the latter [ ] . the molecular dynamics simulations enable us to study the consequences of both definitions on the nanoaggregate stability . we quantify the stability as the probability of having a nanoaggregate of a given size in the stationary regime . the probability @xmath17 of having a nanoaggregate containing @xmath18 aromatic molecules is defined as : @xmath19 where @xmath20 is the number of linear aggregates containing @xmath18 aromatic molecules and @xmath21 is the total number of linear aggregates . in a similar way , the probability @xmath22 of having a nanoaggregate with @xmath18 molecules of type @xmath23 is defined as @xmath24 where @xmath25 is the number of linear aggregates containing @xmath18 molecules of type @xmath23 and @xmath26 is the total number of linear aggregates containing at least one molecule of type @xmath23 . the type @xmath23 can be @xmath27 for asphaltene , @xmath28 for resin , @xmath29 for resinous oil or @xmath30 for resin and resinous oil . the probabilities @xmath31 and @xmath32 are plotted in fig . [ fig : pdfall ] ( b ) and ( c ) , respectively . the probabilities @xmath33 and @xmath34 are plotted in fig . [ fig : pdfall ] ( d ) . two main points can be noticed in fig . [ fig : pdfall ] . the probability @xmath35 of having a nanoaggregate containing @xmath18 aromatic molecules seems to have two slopes in a log - lin scale , _ i. e. _ can be described by a biexponential . on the contrary , when only asphaltene molecules are counted , the probability @xmath36 has only one slope , _ i. e. _ can be characterized as a simple exponential . for the probability @xmath37 of having a nanoaggregate containing @xmath18 resin or resinous oil molecules , the biexponential shape seems to be prevailing . it is the same for the probability @xmath38 of having a nanoaggregate containing @xmath18 resin molecules . for the probability @xmath39 of having a nanoaggregate containing @xmath18 resinous oil molecules , the monoexponential shape seems to be recovered . the transition between the two slopes is not sharp for the probability @xmath38 . to highlight the existence of the two slopes in this case in fig . [ fig : pdfall ] ( d ) , a dashed line was drawn in continuation of the line corresponding to the first slope . it departs further and further away from the second slope . the rest of the paper is devoted to understand the molecular dynamics results presented in fig . [ fig : pdfall ] . more specifically , the paper is aimed at unraveling some of the dynamical and thermodynamical origins lying behind the monoexponential and biexponential behaviors . a dynamical approach will first be proposed for the monoexponential behavior . in fig . [ fig : pdfall ] ( b ) , the probability of having a nanoaggregate containing @xmath18 asphaltene molecules was shown to be monoexponential . the aim of this section is to establish a simple aggregation mechanism reproducing this behavior . the mechanism will be used to derive a master equation which provides a dynamical framework for understanding the monoexponential behavior . a simple mechanism which can be written in the asphaltene aggregation problem is @xmath40 where @xmath41 denotes an aggregate containing @xmath18 asphaltene molecules . the assumptions lying behind this mechanism are the following : only one asphaltene molecule at a time can attach to or detach from an existing aggregate , the medium is homogeneous and the aggregation dynamics is not limited by diffusion . the same mechanism was already suggested in ref . [ ] to model asphaltene aggregation from a thermodynamical point of view , but the possibility to obtain a master equation from this mechanism and consequently to model the dynamics of asphaltene aggregation was left aside . to obtain a master equation linked to this aggregation reaction , we make supplementary assumptions : the reaction rate constants do not depend on the size @xmath18 of the aggregate and the attachment rate does not depend on the probability of having free asphaltene molecules in the medium . under these assumptions , the master equation linked to the aggregation reaction eq . can be written as : @xmath42 where @xmath43 is the attachment rate constant , @xmath44 the detachment rate constant and @xmath45 the probability of having an aggregate containing @xmath18 asphaltene molecules at time @xmath46 , given that its initial size was @xmath47 . the probability denoted @xmath31 in sec . [ sec : mdresults ] corresponds to the stationary state to eqs . and and depends neither on time @xmath46 nor on the initial condition @xmath47 . the master equation eq . was solved analytically in the 1950 s [ , ] for different boundary conditions . this master equation and its generalized version in which the rate constants depend on the aggregate size @xmath18 have also been extensively used to model different stochastic processes such as alcohol clusters [ , ] , biological adhesion clusters [ ] or aggregation in freeway traffic [ ] , to cite only a few . the expression of this master equation deserves some further clarification in connection to the specific aggregation reaction used . following the law of chemical kinetics , the attachment rate linked to the aggregation mechanism should be @xmath48 , because the probability of having a collision between two small spherical molecules is usually proportional to the product of their concentrations . we can give empirical reasons explaining why it is not proportional to the probability @xmath49 of having a free asphaltene molecule here . we assume that the attachment rate is not due to the collision between two small spherical molecules , but limited by the steric hindrance around the aggregate and the orientation of the free asphaltene molecule @xmath50 . the attachment reaction is successful if the free asphaltene molecule is well oriented and placed at a reactive end of the aggregate . the number of free asphaltene molecules is higher than the number of any aggregate of a given size at any time in our md simulations ( not shown ) . as the number of free asphaltene molecules is high enough , there is always a free asphaltene molecule nearby an aggregate but it is not always well - oriented and prevent other molecules from approaching due to the density of the system . thus , the attachment rate is independent of the free asphaltene concentration and the attachment rate constant @xmath43 is averaged over possible orientations . the boundary condition eq . needs also some explanation . it corresponds to the fact that the system is closed . in other words , the total number of asphaltene molecules is kept constant . as the free asphaltene molecules are involved in each aggregation reaction , the derivative with respect to time of the probability @xmath51 to have a free asphaltene molecule depends on infinitely many terms . it leads to @xmath52 these series are then simplified using the fact that at every time @xmath46 : @xmath53 , to give the boundary condition eq . . to test the validity of this master equation for the present problem , two different quantities are evaluated : the stationary probability of having a nanoaggregate of a given size , as shown in sec . [ sec : mdresults ] and the aggregation dynamics , quantified as the time evolution of the fraction of aggregated asphaltene molecules . the stationary distribution of the master equation eq . with the boundary condition eq . is easy to derive . in the stationary state , the probability of having a nanoaggregate containing @xmath18 asphaltene molecules depends neither on time nor on the initial size of a nanoaggregate . for this reason , the stationary probability will be denoted @xmath31 . it satisfies the following equation : @xmath54 the solution of this equation is a geometrical law [ , ] . by induction and using the fact that @xmath55 , one can show that the solution of this equation is : @xmath56 in a log - lin scale , the stationary probability predicted by the master equation approach is a straight line of the form : @xmath57 it is a monoexponential distribution , as observed in fig . [ fig : pdfall ] ( b ) , obtained in md . the exponential distribution can be used to fit the molecular dynamics data , as shown in fig . [ fig : aonlybb ] . the value of the dynamical parameter @xmath58 is , in this case , @xmath59 . the master equation framework provides a dynamical interpretation of the parameter @xmath58 as the ratio between the attachment and detachment rate constants . figure [ fig : aonlybb ] also shows the probability of having a nanoaggregate containing @xmath18 asphaltene molecules obtained in md simulations for two different system sizes and the corresponding exponential fits . the results for a system containing @xmath1 asphaltene molecules and for a system containing @xmath60 times less molecules agree surprisingly well , giving the values @xmath59 and @xmath61 respectively . it shows that , while finite size effects are present , they are not very important when it comes to the number of asphaltene molecules in a nanoaggregate . more specifically , the relative probability of small nanoaggregates is accurately described even in the small system . the probability of having larger nanoaggregates is not so well described in the small system , as expected , but does not affect much the value of the parameter @xmath58 since these nanoaggregates are rare even in the large system . it is important to note , however , that physical properties due to large nanoaggregates or to nanoaggregates filling up the box in one direction , such as residual stresses , could be affected by finite size effects . as shown above , the stationary state predicted by the master equation agrees with the md results on the probability of having a nanoaggregate containing @xmath18 asphaltene molecules . if the master equation eq . correctly describes the aggregation process , it should also reproduce the aggregation dynamics . checking this fact is the purpose of this section . to compare the prediction of the master equation approach and the molecular dynamics results , we quantified the aggregation dynamics in the following way . we used the fraction @xmath62 of aggregated asphaltene molecules versus time . the fraction @xmath62 of aggregated asphaltene molecules is defined as the ratio of the number of asphaltene molecules in all aggregates containing at least two asphaltene molecules and the total number of asphaltene molecules in the system . the fraction @xmath62 verifies : @xmath63 where @xmath64 is the fraction of free asphaltene molecules . the fraction of free asphaltene molecules is defined as @xmath65 where @xmath66 is the number of asphaltene molecules in aggregate containing one asphaltene molecule and @xmath67 is the total number of asphaltene molecules in the system . the fraction @xmath64 can be expressed in terms of the probability @xmath68 of having an asphaltene aggregate of size @xmath69 at time @xmath46 and of the average size of an asphaltene aggregate @xmath70 , given that all aggregates had size 1 initially . indeed , we know that : @xmath71 where @xmath72 is the number of aggregates containing one asphaltene molecule and @xmath73 is the total number of aggregates containing at least one asphaltene molecule , as in eq . [ eq : padef ] . we also know that : @xmath74 where @xmath75 is again the total number of asphaltene molecules and @xmath73 is the total number of aggregates containing at least one asphaltene molecule . it is easy to show that @xmath76 and finally that @xmath77 to obtain the dynamics predicted by the master equation eq . with the specific boundary condition eq . , a numerical implementation of the scheme was carried out . in the numerical implementation the time step @xmath78 is a hundred times smaller than the inverse of the detachment rate constant @xmath79 . an aggregate can attach to a single molecule with the probability @xmath80 and release a single molecule with the probability @xmath81 . an aggregate of size @xmath69 can attach to a molecule but can not release one . the total number of molecules is kept constant in the numerical implementation of the master equation . the total number of molecules is chosen to be @xmath82 to reduce the statistical noise . initially , in the numerical implementation of the master equation , all the aggregates are of size @xmath69 . the fraction @xmath64 of free asphaltene molecules can be obtained through this numerical implementation . it depends _ a priori _ on the two rate constants @xmath43 and @xmath44 . however , the ratio @xmath83 is known from the stationary state result . it leaves us with one dynamical parameter , say @xmath43 , to fit . the initial state in the molecular dynamics simulations is not as well defined as in the master equation approach . to reach the desired density , a first md simulation where the system is compressed is performed . during the compression period , the asphaltene molecules begin to aggregate , but the data can not be recorded . the data are recorded just after the compression period , in a state where small asphaltene aggregates are already formed . the fraction of free asphaltene molecules in that state is @xmath84 . it is a second fitting parameter . the curve predicted by the numerical implementation of the master equation was shifted in time so that time @xmath85 corresponds to @xmath84 as in the md simulations . figure [ fig : dynamics ] shows the fraction of aggregated asphaltene molecules versus time in the molecular dynamics simulations and in the master equation approach . both results agree well , indicating that the master equation approach correctly describe the aggregation process of asphaltene molecules . the value of the dynamical parameter @xmath43 is found to be @xmath86 s@xmath6 . it is worth mentioning that the inverse rate constant @xmath87 s is much larger than the upper limit @xmath88 s , needed for an asphaltene molecule to diffuse of a distance equals to the average distance @xmath89 nm between the centers of mass of @xmath90 aromatic molecules in a homogeneous system of volume @xmath91 nm@xmath92 minus the intermolecular distance @xmath93 @xmath8 in an aggregate . this characteristic time is evaluated using the diffusion coefficient of a single asphaltene molecule in the docosane solvent : @xmath94 m@xmath95.s@xmath6 and the formula @xmath96 for three - dimensional diffusion . this time is an upper limit because only the distance between centers of mass is considered , whereas molecules are extended in space and can be close to each other even if the distance between their centers of mass is larger than @xmath97 . we can conclude from that fact that the nanoaggregation process is not limited by diffusion . in fig . [ fig : dynamics ] , there seems to be a discrepancy between the master equation approach and the md results at long times . this could indicate the existence of another aggregation process , taking place at a longer time scale . checking carefully the existence of this second process requires further investigation . the master equation approach related to the aggregation reaction eq . is able to reproduce both the stationary behavior and the dynamics of the asphaltene aggregation process at intermediate time scales . it indicates that the assumptions made to derive the master equation eq . are relevant . the master equation approach gives a dynamical framework to interpret the monoexponential distribution observed in md . even if the same master equation has already been used in the general field of clustering processes [ ] , it is the first time , to our knowledge , that it is applied to asphaltene aggregation in molecular dynamics simulations . section [ sec : mastereq ] provided a dynamical interpretation of the monoexponential behavior of the probability of having a nanoaggregate containing @xmath18 asphaltene molecules based on a master equation description on a simple aggregation reaction . the aim of the present section is to describe a statistical mechanics model based on the same aggregation scheme and consistent with the monoexponential behavior . in this section , we will also look into possible thermodynamical interpretations of the biexponential behavior , obtained when all aromatic molecules are counted . to formulate a simple statistical mechanics model linked to the aggregation reaction eq . , some assumptions are needed . we make the following common assumptions : 1 . the nanoaggregates are linear . 2 . the system is homogeneous and isotropic on average and the nanoaggregates are rigid . 3 . the aggregation process occurs at a characteristic time much larger than the equilibration of the pressure , temperature and solvent molecules . the system is dilute enough to consider no interaction between the nanoaggregates except through the aggregation reaction . in other words , the solution of nanoaggregates of different sizes in a solvent is ideal . the free energy of a nanoaggregate depends linearly on its size . the idea of considering a mixture of free molecules and dimers as an ideal mixture with an aggregation reaction occurring at a larger time scale dates back to the beginning of the 20@xmath98 century [ ] as is nicely explained in the recent review [ ] . the set of assumptions was then completed to include the case of linear aggregates of any size and widely used to describe rodlike micelles [ ] , linear polymer chains [ ] , and discotic liquid crystals [ ] . we propose in this section our own version of the derivation applied to the case of linear asphaltene aggregates and obtain an analytical formula for the probability of having an aggregate of a given size . the analogy between linear assembly in discotic liquid crystals and asphaltene aggregates was also made in ref [ ] , which derives very similar equations to the ones presented in this section , but do not present a successful comparison with their coarse - grained simulations of asphaltene - resin nanoaggregates at high density . the assumption of having an ideal mixture can be relaxed to take into account more complicated interactions between the aggregates such as excluded volume interactions [ ] , the gain in entropy when the chain breaks [ ] and interactions due to the flexibility of the aggregates [ ] , to cite only a few . for the sake of simplicity , these interactions are neglected in the case of linear asphaltene aggregates . the validity of the assumptions made in this section will be discussed in sec . [ sec : assumption ] . concurrently to these approaches expressing the probability of having an aggregate of a given size , another theory of aggregation based on statistical thermodynamics was developed by wertheim [ ] . this theory considers molecules of a reference liquid , typically a lennard - jones fluid , with a finite number of binding sites . the potential modeling the interaction between the binding sites is designed to take into account steric hindrance . the theory then counts the number of molecules with no bond on a given binding site instead of the number of aggregates of a given size . it is a clever way to count physically meaningful graphs . the outcome of the theory is an expression for the equilibrium pressure and the concentration of free molecules versus the composition of the system . these predictions were successfully checked numerically for strongly associating fluid [ ] . however , this theory does not provide an expression for the probability of having an aggregate of a given size . for this reason , it is not considered in further details in this section but is mentioned for the sake of completeness . in the framework of the common assumptions stated above , the free energy of the system can be written as [ ] : @xmath99 where @xmath100 is the number of nanoaggregates containing @xmath18 asphaltene molecules , @xmath101 the volume of the system , @xmath102 the effective free energy of an aggregate , @xmath10 the boltzmann constant and @xmath103 the temperature . the first term in eq . corresponds to the free energy of an ideal mixture of ideal gas and the second term corresponds to the energy of the aggregates . the effective free energy @xmath102 is to be understood as the energy of an asphaltene aggregate when the degrees of freedom due to solvent molecules have been integrated out . the expression eq . for the free energy of the system holds for each value of the number @xmath100 of aggregates of a given size , because a state of local equilibrium acting on entropy , pressure and solvent molecules is assumed for each step along the aggregation reactions . the total equilibrium of the system depends now only on the equilibrium of the aggregation reactions . the condition for chemical equilibrium of each aggregation reaction @xmath104 is @xmath105 where @xmath106 is the chemical potential of an asphaltene aggregate of size @xmath18 . using the definition of the chemical potential @xmath107 , we can obtain after some calculations the well - known mass action law : @xmath108 where @xmath109 is the equilibrium constant of the reaction @xmath104 . the equilibrium constant is given by : @xmath110 we now make use of assumption 5 , stating that the free energy of a nanoaggregate depends linearly on its size . this assumption amounts to write the effective free energy @xmath102 as : @xmath111 where @xmath112 is the free energy of a single asphaltene molecule and @xmath113 is the effective free energy between two asphaltene molecules . using this form in the expression of the equilibrium constant eq . gives : @xmath114 thus , the equilibrium constant does not depend on the size @xmath18 of the considered nanoaggregate under the set of assumptions considered . by induction , it is now easy to show from eq . that @xmath115 to compare the model to the simulations results it is more useful to obtain the probability @xmath31 of having a nanoaggregate of size @xmath18 , which is defined in eq.[eq : padef ] @xmath116 where @xmath117 is the total number of asphaltene nanoaggregates . to express @xmath31 , we make use of the conservation of the total number @xmath75 of asphaltene molecules : @xmath118 having this in mind , one can show ( see appendix [ ap : paramp ] ) that @xmath119 where @xmath120 and @xmath121 we have now recovered the exponential distribution observed in the md simulations and fig . [ fig : pdfall ] ( b ) . the md simulations provide a value of the parameter @xmath58 characterizing the exponential distribution : @xmath122 . according to the thermodynamical interpretation eq . , it leads to the value @xmath123 the simple statistical mechanics model provides an interpretation for the parameter @xmath58 in terms of an effective free energy @xmath113 between two asphaltene molecules . within the assumptions of this simple model , the effective free energy is the interaction energy between two asphaltene molecules when the degrees of freedom related to solvent molecules are integrated out . the physical meaning of free energy @xmath113 depends on its definition in the framework of the assumptions made here , but also on the validity of these assumptions . this will be discussed in the next section . the validity of each assumption and its consequences on the physical meaning of the free energy @xmath113 will now be listed . 1 . _ linear dependence of @xmath102 on the size @xmath18 of the nanoaggregate . _ + it is very easy to check that the potential energy , and not the free energy , of a nanoaggregate in vacuum depends linearly on its size . the potential energy of a linear nanoaggregate of size @xmath18 is plotted versus @xmath18 for nanoaggregates in vacuum in fig . [ fig : potenergya ] . the origin of the energy is set arbitrarily to zero in this figure . the nanoaggregate used to plot this figure is a linear nanoaggregate containing five asphaltene molecules found in one simulation . to find the potential energy of an aggregate of size @xmath124 , only the first @xmath18 asphaltene molecules in the aggregate were considered in vacuum . this figure clearly shows that the potential energy of a nanoaggregate in vacuum is linear and the slope @xmath125 corresponding to the interaction energy between two asphaltene molecules is equal to @xmath126 in practical terms , it means that when an asphaltene molecule is added to a nanoaggregate of size @xmath18 , this asphaltene molecule interacts with the energy @xmath127 @xmath128 with the molecule at the end of the nanoaggregate but do not interact with the other ones , which are further away . + the fact that the potential energy of a nanoaggregate in vacuum depends linearly on its size is a good indication that it might also be the case for the effective free energy @xmath102 of a nanoaggregate in a solvent . in that case , the effective free energy @xmath113 corresponding to the slope of @xmath102 versus the size @xmath18 of the nanoaggregate is well defined . + we remind the reader that the potential energy of a nanoaggregate in vacuum and the effective free energy @xmath102 are not identical and the slopes @xmath113 and @xmath125 are indeed quite different . within the framework of the statistical model presented in sec . [ sec : thermomodel ] , the two causes of this difference are : the interaction between the asphaltene molecules and the solvent and the entropic effects due to the effective integration of the degrees of freedom of the solvent . to quantify the effective free energy @xmath113 properly , a possibility is to set up umbrella sampling simulations controlling the distance between the center of mass of two asphaltene molecules in a bath of docosane molecules at the same temperature and pressure as the one used in the md simulations . the free energy of this system can be derived in terms of the distance between the two asphaltene molecules . the effective free energy @xmath113 would then be the difference between the free energy of such a system when the two asphaltene molecules are far away and the free energy of the system when the two asphaltene molecules are aligned and close . implementing umbrella sampling simulations is beyond the scope of this paper . moreover , the free energy @xmath113 found with this method might be slightly different from the value @xmath129 @xmath128 , found using the statistical mechanics model of sec . [ sec : thermomodel ] , because of the other assumptions made to derive this model . _ asphaltene nanoaggregates . _ + to derive the statistical mechanics model in sec . [ sec : thermomodel ] , we only considered the aggregation of asphaltene molecules , whereas resin and resinous oil molecules are also part of the nanoaggregates . the degrees of freedom related to the position of resin and resinous oil molecules should be integrated in an effective way to obtain the free energy @xmath113 , just as it was done for the degrees of freedom related to the solvent . the fact that the monoexponential behavior is valid for the probability of having a nanoaggregate containing @xmath18 asphaltene molecules means that this effective integration can be done . linear nanoaggregates . _ + the nanoaggregates considered in the simulations are only some linear portions of bigger branched nanoaggregates . the existence of the branches could also produce some degrees of freedom to be integrated to obtain the effective free energy @xmath113 . 4 . _ dilute limit . _ + the assumption stating that the solution of nanoaggregates is ideal neglects the interaction between the nanoaggregates . some of these interactions , for example between the nanoaggregates @xmath41 , @xmath130 and @xmath50 , are later taken into account through the aggregation reactions corresponding to the asphaltene aggregation . but in the md simulations , the asphaltene molecules interact in a more complicated way . they can not for example be too close to each other , due to short range repulsion . this could be taken into account through excluded volume interactions . it is known [ , ] , that taking into account excluded volume interactions preserves the monoexponential behavior . consequently , excluded volume interactions might play a role in the value of the effective free energy @xmath113 . _ time scale of the aggregation reactions . _ + to derive the statistical mechanics model , we assumed that the aggregation reactions occurred on a time scale much larger than the time scale associated with the equilibrium of pressure , entropy , and solvent molecules for a given number of each nanoaggregate . it is very difficult to predict the effect of the relaxation of this fundamental assumption . one way to check the assumption , however , is to compare once again the characteristic time of the aggregation reaction @xmath87 s and the upper limit @xmath88 s for the diffusion of an asphaltene molecule . a factor @xmath131 exists between the two characteristic times which should be enough to ensure the validity of the assumption . _ rigidity of the nanoaggregates and isotropy of the system . _ + [ sec : relaxrigidity ] the assumptions of rigid nanoaggregates and isotropic system can be considered together , because if they are both relaxed , they lead to a new term in the free energy of the system [ ] . this term depends on a persistence length @xmath132 and reads as [ , ] @xmath133^{1/2 } \nabla[\rho_n(\mathbf{u})]^{-1/2},\ ] ] where @xmath134 is the number density of asphaltene molecules part of a nanoaggregate of size @xmath18 oriented in the direction @xmath135 with respect to some reference direction and @xmath136 is the corresponding solid angle . this term is helpful to describe the nematic phase of liquid crystals , where long range order is seen . the addition of this term in the free energy expression leads to a biexponential behavior [ ] , which is not observed for the probability of having a linear nanoaggregates with @xmath18 asphaltene molecules . consequently , the assumption of rigid nanoaggregates and isotropic system probably holds , at least in an effective way , for the probability of having a linear nanoaggregate with @xmath18 asphaltene molecules and does not participate in the value of the effective free energy @xmath113 . the addition of this flexible term to the free energy is potentially of interest to explain the biexponential behavior of the probability of having a nanoaggregate containing @xmath18 aromatic molecules . to summarize , one can say that the statistical mechanics model developed in sec . [ sec : thermomodel ] provides a thermodynamical interpretation of the monoexponential behavior based on the effective free energy between two asphaltene molecules . this effective free energy should be understood as the interaction energy between two asphaltene molecules when all the degrees of freedom related to solvent molecules , resin and resinous oil molecules and branched nanoaggregates have been integrated out . the picture emerging from the study of the probability of having a linear nanoaggregate containing @xmath18 asphaltene molecules is that asphaltene bodies interact with an effective energy through the aggregation reaction . this leads to a monoexponential behavior . we know that the probability of having a linear nanoaggregate with @xmath18 aromatic molecules has a different behavior . it is biexponential as shown in fig . [ fig : pdfall ] ( a ) . this section is devoted to identifying possible statistical mechanics explanations for this different behavior . the first explanation that comes to mind is that when all molecules are counted , different interaction energies are involved . each interaction energy taken separately would lead to a specific monoexponential behavior and the combinations of several interaction energies could lead to a bi- or multiexponential behavior . to test this idea , we set up one - dimensional lattice monte - carlo simulations . a linear lattice of @xmath137 sites is created . each site can contain one asphaltene molecule , one resin molecule , one resinous oil molecule or nothing . one site can not contain two molecules . the total numbers of asphaltene molecules , resin molecules , and resinous oil molecules are constant . a monte - carlo move consists in exchanging the content of two sites providing that the content is different . this condition makes the molecules indiscernible . the system is initialised with the largest possible aggregate where all asphaltene molecules are next to each other , then comes resin molecules and then resinous oil molecules . ten million ( @xmath138 ) monte - carlo moves are realised using the metropolis algorithm . the algorithm converges quite quickly despite its elementary implementation . the potential energy of a nanoaggregate is calculated in the following way : when two asphaltene molecules are next to each other the interaction @xmath139 between two asphaltene molecules is added , when a resinous oil molecule is next to an asphaltene molecule or another resinous oil molecule the interaction energy @xmath140 is added , finally when a resin molecule is next to any other molecule the interaction energy @xmath141 is added . as two molecules can not be on the same site , effective excluded volume interactions are created . the assumptions underlying the establishment of the one - dimensional lattice monte - carlo simulations are similar to the ones made in sec . [ sec : thermomodel ] except for excluded volume interactions : 1 . the nanoaggregates can only be linear because the system is one - dimensional . 2 . the nanoaggregates are rigid . the system is one dimensional , so isotropy is not a criterion . 3 . the only energies involved are those related to the aggregation process . all potential energies related to interaction with and within solvent molecules are averaged out . no kinetic energy is involved . 4 . the system is not dilute and excluded volume interactions are taken into account . 5 . the energy of a pure asphaltene nanoaggregate depends linearly on its size . one of the main advantages of lattice monte - carlo simulations compared to the analytical approach is to take into account resin and resinous oil molecules and not only asphaltene molecules . we checked that in the dilute limit , when only asphaltene molecules are present in the monte - carlo simulations , the same monoexponential behavior with the same value for the parameter @xmath58 as the one predicted by the analytical approach is recovered . in the lattice monte - carlo simulations , it is possible to obtain a biexponential behavior for the probability of having @xmath18 molecules in a nanoaggregate as can be seen in fig . [ fig:1dlatmc]a . the biexponential behavior is characterized by the presence of two straight lines with different slopes in a log - lin scale . the biexponential behavior occurs in the lattice monte - carlo simulations when the interaction energies @xmath139 between asphaltene molecules on the one hand and @xmath140 and @xmath141 with resin and resinous oil molecules on the other hand are substantially different . for example , the choices @xmath142 @xmath128 , @xmath143 @xmath128 and @xmath144 @xmath145 at the same temperature and same volume as the md simulations give a biexponential behavior very close to the one obtained in md in fig . [ fig : pdfall ] ( a ) . it is shown in fig . [ fig:1dlatmc ] ( a ) . however , the biexponential behavior does not have the same causes as in the md simulations . in the monte - carlo simulations , the biexponential behavior is due to the fact that pure resin or resinous oil nanoaggregates and pure asphaltene nanoaggregates are formed . there are very few mixed nanoaggregates . thus , the first slope in the biexponential behavior is due to resin and resinous oil nanoaggregates and the second slope is due to asphaltene nanoaggregates . one consequence of this fact is that the probability of having a nanoaggregate with @xmath18 resin or resinous oil molecules has a monoexponential behavior with a slope very close to the first slope of the biexponential behavior . in the same way , the probability of having a nanoaggregate with @xmath18 asphaltene molecules has a monoexponential behavior with a slope very close to the second slope of the biexponential behavior . this can be seen in fig . [ fig:1dlatmc](a ) . on the contrary , in the md simulations , the two slopes of the biexponential behavior do not correspond to two different slopes in two different monoexponential behaviors . this can be seen in fig . [ fig:1dlatmc](b ) . the existence of mixed nanoaggregates can be checked directly in the md simulations . [ fig:1dlatmc](c ) displays the ratios @xmath146 , @xmath147 , and @xmath148 of asphaltene , resin , and resinous oil molecules respectively versus the size of the aggregate . for a given molecule type @xmath149 , the ratio @xmath150 is defined as : @xmath151 where @xmath152 is the number of molecules of type @xmath149 in the aggregate and @xmath18 the total number of molecules in the aggregate . [ fig:1dlatmc](c ) shows that there is indeed a change in the nanoaggregate composition with their size . the ratio of asphaltene molecules increases versus the size of the aggregates until it reaches an approximately constant value for aggregates of size @xmath153 . at the same time , the ratio of resin molecules decreases versus the size of the aggregates and reaches an approximately constant value for aggregates of size @xmath153 , while the ratio of resinous oil molecules is roughly constant . however , for any size the nanoaggregates contain all molecule types . we can conclude that , the biexponential behavior in md is probably not only due to the difference in effective interaction energies involved . another possible explanation of the biexponential behavior obtained for the probability of having a nanoaggregate containing @xmath18 aromatic molecules , is that it derives from the nanoaggregate flexibility . as mentioned in sec . [ sec : assumption ] , if the aggregates are flexible and the system is not isotropic , an extra term in the free energy of the system should be considered ( see eq . ) . this leads to a biexponential behavior . the existence of the biexponential behavior can be explained qualitatively in this way : bent nanoaggregates , less stable than straight ones , tend to be smaller and are responsible for the first slope of the biexponential behavior ; on the contrary straight nanoaggregates are larger and give rise to the second slope [ ] . this additional term is of course well - suited to explain the biexponential behavior observed for the probability of having a nanoaggregate with @xmath18 aromatic molecules . however , we are not convinced that it is the main reason explaining the biexponential behavior , but there are indications that it might play a role . first , some nanoaggregates are bent . a picture of a bent nanoaggregate can be seen in fig . [ fig : flexchain](a ) . to quantify the variation of the rigidity of an aggregate with the distance inside this aggregate , we computed the orientation correlation function @xmath154 , where @xmath155 is the unit vector normal to the molecule @xmath156 in a given linear aggregate , @xmath157 stands for the @xmath158 neighbor of molecule @xmath156 in the same linear aggregate , and @xmath159 corresponds to the average over different nanoaggregates and over time . [ fig : oriencorrel ] shows the variation of the orientation correlation function @xmath154 with @xmath160 . the error bars corresponding to the standard deviation over eight independent simulations are very large for this plot and are not shown for the sake of visibility . considering the large errors , only a qualitative discussion on the average trend is possible . we can see that there is , on average , an initial decrease of the orientation correlation with the number of neighbors . this trend shows that nanoaggregates are not perfectly rigid . for larger distances and consequently larger nanoaggregates , the orientation correlation seems to plateau around the value @xmath161 . this last trend matches the qualitative picture of small aggregates being more bent than larger aggregates . the change in the nanoaggregate composition observed in fig . [ fig:1dlatmc](c ) could explain the change in the nanoaggregate rigidity with the nanoaggregate size : aggregates containing many asphaltene molecules tend to be longer and more rigid . second , the system is not strictly isotropic . the isotropy was quantified using the nematic order parameter @xmath162 [ ] . to define the nematic order parameter , the following order tensor needs to be defined for each nanoaggregate : @xmath163 where @xmath18 is the number of molecules in the considered linear nanoaggregate , @xmath164 and @xmath165 cartesian coordinates , @xmath156 the index of a molecule inside the aggregate , @xmath155 the unit vector normal to molecule @xmath156 , and @xmath166 the kronecker delta . the nematic order parameter @xmath162 is the largest eigenvalue of the order tensor @xmath167 averaged over different nanoaggregates and time . the nematic order tensor is equal to @xmath69 in a system where all aggregates are perfectly aligned and to @xmath168 in a perfectly isotropic system . in the md simulations , the nematic order parameter is equal to @xmath169 where the error is the standard deviation corresponding to eight independent simulations . the value of the nematic order parameter indicates a system closed to being isotropic but not quite . it can be due to the fact that some linear nanoaggregates are branches of bigger aggregates . they can be connected through asphaltene heads and the angle between an asphaltene head and an asphaltene body is fixed by the dihedral potential and does not vary this much from one asphaltene molecule to another [ ] . it can also be due to steric hindrance : long nanoaggregates can not interpenetrate each other and consequently tend to align . the facts that nanoaggregates are flexible and that the system is not perfectly isotropic are consequently a plausible explanation for the biexponential behavior of the probability of having a nanoaggregate containing a given number of aromatic molecules . ( color online ) picture of a bent nanoaggregate , obtained in molecular dynamics . the color code is the same as in fig . [ fig : linearaggregate ] . ] the fact that the nanoaggregates are not strictly rigid and that the system is not strictly isotropic was also valid when we looked at the probability of having a nanoaggregate containing @xmath18 asphaltene molecules . however , the fact that this probability has a monoexponential behavior means that the additional term eq . can be neglected in this case . it could be because the effective persistence length is longer when only asphaltene molecules are considered . our bitumen model is quite simplified compared to a real bitumen . in this section , we will compare our results to other md results and to experimental results . we will also discuss the new idea that such a simplified model can bring to the field of asphaltene nanoaggregation and the specific aspects of asphaltene nanoaggregation which are neglected in this simplified model . our model relies on the fact that the @xmath16-stacking interaction is the origin of the nanoaggregate formation . in our md simulations , the @xmath16-stacking is modeled through the lennard - jones potential . recent molecular dynamics simulations of asphaltene molecules in toluene [ ] reported that the @xmath16-stacking interaction is indeed the most important one to explain the nanoaggregate formation , which justifies our model . in our md simulations , the average number of asphaltene molecules in a linear nanoaggregate is given by the exponential distribution . we can compare the average and standard deviation predicted by the exponential distribution to experimental results . to do that , we considered a renormalized distribution of the nanoaggregate size , where the aggregate of size @xmath69 are left out . in other words , free asphaltene molecules are not considered here as a nanoaggregate . the corresponding probability reads : @xmath170 in this case , the average number of asphaltene molecules in linear nanoaggregates is @xmath171 the average number of aromatic molecules in linear aggregates can be computed directly from the md results and is @xmath172 . these results are in agreement with the general consensus that nanoaggregates contain less than @xmath173 molecules [ ] . our simulation also offers a precise quantification of the polydispersity of the nanoaggregate size . the standard deviation of the distribution given in eq . is : @xmath174 again , the standard deviation on the number of aromatic molecules in linear nanoaggregates can be computed from the md results and is @xmath175 . these results are compatible with previous md results reporting for example that : for asphaltene nanoaggregates in toluene , the number of asphaltene molecules in the largest aggregate varies from 2 to 18 depending on the asphaltene structure [ ] ; for asphaltene nanoaggregates in vacuum , the number of asphaltene molecules in any aggregate varies from 1 to 5 and the precise distribution depends both on temperature and the asphaltene structure [ ] . on the experimental side , recent laser - based mass spectrometry experiments [ ] were able to obtain not only the average size of a nanoaggregate but also the polydispersity . the aggregation numbers are found to range roughly from @xmath176 to @xmath177 or from @xmath177 to @xmath178 depending on the bitumen chemical composition . it is also compatible with our results . moreover , our simulations bring the idea , already suggested in ref . [ ] without a successful match to simulation results at high density , that the simple monoexponential distribution is a typical distribution of the number of asphaltene molecules in linear aggregates . this distribution is consequently a good basis to model more complicated cases . one of these more complicated cases is considered here : when all aromatic molecules are counted in the nanoaggregates the probability of having a nanoaggregate of a given size becomes biexponential . the biexponential distribution can be seen as a modification of the exponential distribution when the flexibity of the aggregates and the anisotropy of the system are taken into account . many more specific aspects relevant for bitumen science can be considered and are not treated here . for example , the effect of having branched nanoaggregates on the distribution of the nanoaggregate size could be addressed . furthermore , it seems that the presence of long aliphatic chains in the asphaltene molecules modifies the typical structure of a nanoaggregate [ ] enhancing t - shaped geometry ( @xmath16-@xmath179 interaction ) and offset @xmath16-stacked geometry ( @xmath179-@xmath179 interaction ) compared to @xmath16-@xmath16 geometry [ ] . it would be very interesting to consider the effect of adding asphaltene and resin molecules with long aliphatic chains in our simulations on the shape of the nanoaggregate size distribution compared to the simple exponential distribution . in conclusion , we have shown that the aggregation reactions give a good description of the stability of linear asphaltene nanoaggregates as observed in molecular dynamics ( md ) simulations . from these aggregation reactions , we derived a master equation able to reproduce the monoexponential behavior of the stationary probability of having a nanoaggregate containing a given number of asphaltene molecules in md . the parameter of the monoexponential behavior is interpreted as the ratio between the attachment and detachment rate constants of a single asphaltene molecule to a nanoaggregate . the master equation approach is also able to reproduce the aggregation dynamics . from the same aggregation reactions , we also derived a statistical mechanics model giving a thermodynamics interpretation of the monoexponential behavior . the main parameter is then the effective free energy between two asphaltene molecules , when the degrees of freedom corresponding to solvent molecules , resin , and resinous oil molecules and branched nanoaggregates are integrated out . finally , a possible thermodynamic explanation for the biexponential behavior , observed for the stationary probability of having a nanoaggregate of @xmath18 aromatic molecules in md , is the flexibility of these nanoaggregates . to continue this work on bitumen nanoaggregate two directions are possible and equally interesting . a first direction is to consider a simpler system without resin and resinous oil and even without any possibility of branching . then , the integration of the degrees of freedom related to the solvent molecules could be done and the effective energy could be evaluated in an independent way . this direction would lead to a quantitative understanding of the effective free energy between two asphaltene molecules . the second and opposite direction is to add more molecule types to resemble a real bitumen . for example asphaltene molecules without a head and with long alkyl chains could be added and the consequences of this addition on the probability of having a nanoaggregate of a given size investigated . interesting md simulations have been performed recently using many molecules types [ ] and could serve as an inspiration . another interesting route is to quantify the evolution of the nanoaggregate size distribution with the composition of the bitumen mixture . this work is sponsored by the grants 1337 - 00073b and 1335 - 00762b of the danish council for independent research @xmath180 technology and production science . it is in continuation of the cooee project ( co@xmath4 emission reduction by exploitation of rolling resistance modeling of pavements ) , sponsored by the danish council for strategic research . the centre for viscous liquids dynamics `` glass and time '' is supported by the danish national research foundation s grant dnrf61 . the probability @xmath31 of having a nanoaggregate with @xmath18 asphaltene molecules is defined as : @xmath181 where @xmath182 is the number of aggregates with @xmath18 asphaltene molecules and @xmath117 the total number of asphaltene nanoaggregates . according to eq . , @xmath183 so that one can express the total number of asphaltene nanoaggregates : @xmath184 to obtain the last expression eq . , we assume that the number of asphaltene molecules is very large , so that the sum goes to infinity and that the ratio @xmath185 is smaller than 1 , _ i. e. _ the volume @xmath101 is big enough for the aggregates to develop given the total number of asphaltene molecules and the equilibrium constant . inserting the expression eq . of the total number of asphaltene aggregates back into eq . gives : @xmath186 it is a geometrical law of the form @xmath187 and the parameter @xmath58 can be identified as : @xmath188 an expression for @xmath72 , the number of asphaltene aggregates of size @xmath69 , is given by the conservation law : @xmath189 , where @xmath75 is the total number of asphaltene molecules . it leads to : @xmath190 the last expression eq . is a quadratic equation , whose solutions are : @xmath191 where @xmath192 the solution @xmath193 is the physical one , because in the limit where there is no interaction between the asphaltene molecules and even a large repulsion , all molecules should be in aggregates of size @xmath69 . in mathematical terms , it gives : @xmath194 only @xmath193 satisfies this last equation . it leads finally to : @xmath195 which is the same result as eq . . another useful expression is the one giving the equilibrium constant @xmath196 in terms of the parameter @xmath58 : @xmath197 finally , it is noteworthy that the minimization of the free energy of the system as given in eq . subject to the conservation condition eq . leads to the same result . the minimization can be done using a lagrange multiplier to guarantee the conservation of the total number of asphaltene molecules . in this calculation , the free energy @xmath102 of an aggregate should be expressed right away as @xmath198 , where the arbitrary origin of the energy @xmath199 should be equal to @xmath200 . in this case , @xmath113 represents the energetic penalty of having a free end . j. f. branthaver , j. c. petersen , r. e. robertson , j. j. duvall , s. s. kim , p. m. harnsberger , t. mill , e. k. ensley , f. a. barbour , and j. f. schabron , tech . shrp - a-368 , strategic highway research program ( 1993 ) . o. c. mullins , h. sabbah , j. eyssautier , a. e. pomerantz , l. barr , a. b. andrews , y. ruiz - morales , f. mostowfi , r. mcfarlane , l. goual , r. lepkowicz , t. cooper , j. orbulescu , r. m. leblanc , j. edwards , and r. n. zare , energy & fuels * 26 * , 3986 ( 2012 ) .

the use of cavity qed tools is now ubiquitous across different areas of physics ranging from quantum information @xcite to detection of dark matter @xcite . specifically , the atoms held in optical cavities play a vital role in studies of many - body physics @xcite . such systems are natural implementations of many - body jaynes - cummings ( jc ) or dicke hamiltonians @xcite . their high tunability and the possibility of achieving strong light matter coupling or probing the dynamics in real time make them very attractive experimental platforms . the prospect of probing phase transitions and the associated critical phenomena with these platforms have been put forward e.g. in @xcite and the non - equilibrium dynamics of the dicke model has been theoretically investigated in @xcite . specific many - body phenomena that can be studied with cavity qed include for example the physics of spin glasses @xcite or the self - organization of the atomic motion @xcite , to name a few . the self - organization has been subsequently observed in the experiments @xcite . so far we have mentioned only studies concerning a single cavity - generalizations to multiple cavity arrays implementing the hubbard physics have been reviewed in @xcite . although appealing in principle , the realization of many efficiently coupled cavities , each hosting a discrete - level quantum system is a challenging task . to make such experiment scalable requires miniaturization of the cavities . one possibility is the use of microcavities in photonic crystals @xcite . a further option is to use integrated optical circuits , where in principle arbitrary waveguide forms can be created with high precision by laser engraving in the silica substrate @xcite . they have been successfully used for the demonstration of a quantum gate operation @xcite , creation of classical and quantum correlations @xcite , multi - photon entangled state preparation @xcite , quantum random walk @xcite , discrete fourier transform @xcite or bloch oscillations @xcite . while it has been demonstrated that cavities can be fabricated by creating the bragg grating during the laser writing process @xcite , there is now an active experimental effort to combine the waveguides with atomic microtraps on a single device @xcite . motivated by these developments and the prospects of studying many - body physics using integrated optical circuits with trapped atoms , we theoretically analyze the non - equilibrium physics of such system , which we take to be a two - dimensional cavity array , where each cavity hosts a number of atoms . we first derive the effective hamiltonian describing the system in sec . [ sec : model ] . using this hamiltonian we find mean - field ( mf ) equations of motion whose solutions feature rich dynamics , including bistable behaviour , ising like phase transition or the occurrence of limit cycles , which we discuss in detail in sec . [ sec : mf ] . we then conclude in sec . [ sec : conclusion ] . is sketched by the large red circle . ( b ) atomic level scheme ( only a single @xmath0 mode is shown , see text for details).,width=453 ] we consider a two - dimensional array of identical three - level atoms each coupled to a horizontal cavity mode @xmath1 and vertical cavity mode @xmath2 with the same coupling constant @xmath3 , fig . [ fig : scheme]a . from here on we use latin and greek indices to denote horizontal and vertical degrees of freedom respectively . in order to simplify the notation , we omit the sum symbols and summation over occurrences of indices labeling the _ spatial _ position is understood ( i.e. @xmath4 ) . we will use indices with bar , e.g. @xmath5 , when we want to emphasize that the index @xmath5 is fixed and is not summed over . the atomic level structure is depicted in fig . [ fig : scheme]b . the @xmath6 transition is coupled to cavity mode @xmath7 ( @xmath8 ) with frequency @xmath9 ( @xmath10 ) . the @xmath11 transition is coupled using a strong transverse classical light , whose direction of propagation is perpendicular to the plane of the cavity array and which has rabi frequency @xmath12 , carrier frequency @xmath13 and detuning @xmath14 . the main reason for choosing a three level @xmath15 system is that it allows , in the limit of large @xmath14 , to adiabatically eliminate the excited state @xmath16 in order to avoid the losses due to the spontaneous emission and to obtain an effective hamiltonian in the ground state manifold subspace . this effective hamiltonian features a high tunability , cf . below . we now proceed with the derivation of the effective hamiltonian . we discuss the issue of cavity gain and loss later in sec . [ sec : mf ] . the system is described by the hamiltonian ( in the rotating wave approximation ) h & = & _ i a_i^a_i + _ b_^b_+ _ e _ i_i + _ s _ i_i + & & + ( _ i_i e^-i _ t t + _ i_i e^i _ t t ) + & & + g ( _ i_i ( a_i + b _ ) + h.c . ) . [ eq : h rwa 2d ] the excited state @xmath16 can be adiabatically eliminated in the standard way , see appendix [ app : ad.el ] for details . the resulting effective hamiltonian reads h = _ i a_i^a_i + _ b_^b_+ ^z_i + ( ^+_i ( a_i + b _ ) + h.c . ) + f_i , [ eq : h eff rwa 2d ] where @xmath17 are the usual pauli matrices in the @xmath18 basis , _ | & = & _ s - + & = & - + f_|i | & = & ( _ s - ) . [ eq : pars eff ] and @xmath19 , @xmath20 and @xmath21 . here , @xmath22 is an auxiliary frequency used in the adiabatic elimination which , in principle , can be chosen arbitrarily . its physical motivation and interpretation will be discussed momentarily in sec . [ sec : single cavity ] . the effective hamiltonian ( [ eq : h eff rwa 2d ] ) is the starting point in this section , where we analyze the effect of cavity loss and pump on the dynamics . also , it can be seen from the form of the hamiltonian and the expressions ( [ eq : pars eff ] ) , that the parameters are highly tunable through varying @xmath23 and @xmath24 . we first perform a mf analysis of a simpler system with only one cavity mode , which already contains rich physics as we show below . we then turn back to the multi - mode two - dimensional setup in sec . [ sec:2d ] . _ cavity without pump _ lets first consider the hamiltonian ( [ eq : h eff rwa 2d ] ) in the single cavity limit with the single mode @xmath0 , i.e. we drop the indices @xmath25 and @xmath26 . we rewrite the hamiltonian as h = ( + a^a ) ^z + _ ph a^a + ( ^+ a + a^^- ) , [ eq : h jc ] where @xmath27 are the global spin operators and _ at & = & _ s - + _ ph & = & - = - _ aux - + & = & - , with @xmath28 being the cavity frequency . note that the model ( [ eq : h jc ] ) without the @xmath29 term is the usual jc model @xcite . below we show , that the @xmath29 term is indeed at the origin of intriguing system dynamics ( see also @xcite for various other non - linear extensions of the jc model ) . one can now derive the equation of motion for the operator @xmath30 according to @xmath31 $ ] , where @xmath32 is given by ( [ eq : h jc ] ) . at the same time , any realistic cavity is subject to a decay of the electromagnetic field into the environment . the dissipation process is typically described by means of a master equation ( see e.g. @xcite ) , which corresponds to an extra term in the equation of motion for the cavity mode operator , @xmath33 , where @xmath34 denotes the cavity loss rate ( see fig . [ fig : scheme]a ) . introducing the expectation values of the operators @xmath35 , it is now straightforward to derive the the mf equations of motion , which read [ eq : eoms mf ] @xmath36 in the derivation we have used the mf decoupling @xmath37 and @xmath38 . we have also neglected the spin decay on the transition @xmath39 typically belong to some ground state manifold , where only magnetic dipole transitions are allowed between the states of that manifold . in turn , the spin decay is negligible compared to the cavity decay . ] first , we wish to find a steady state solution of the equations of motion ( [ eq : eoms mf ] ) . from ( [ eq : eoms mf a ] ) and ( [ eq : eoms mf w ] ) we have & = & - s c s [ eq : cond1 ] + s^ * & = & ^ * s. [ eq : cond2 ] substituting eq.([eq : cond1 ] ) to eq.([eq : cond2 ] ) yields c |s|^2 = c^ * |s|^2 . [ eq : restr ] when @xmath40 is complex , which occurs only for non - zero cavity decay @xmath41 , the condition eq.([eq : restr ] ) can be satisfied if @xmath42 ( and consequently @xmath43 ) , which results in a trivial solution with empty cavity and all spins down ( up ) in the steady state . this indicates that in order to obtain some non - trivial physics in the steady state limit and to counteract the cavity losses , one needs to input energy into the system@xmath16 transition are taken to be at optical frequencies . ] . in our case a natural choice is either through the pumping of the cavity mode or driving directly the @xmath39 transition . we focus here on the former case only . + _ cavity with pump _ the cavity pump can be described by the hamiltonian h_pump = a^^-i _ p t + ^ * a e^i _ p t. when adding @xmath44 to the system hamiltonian ( [ eq : h rwa 2d ] ) , the explicit time dependence in the total hamiltonian can be removed if the auxiliary frequency @xmath22 in ( [ eq : pars eff ] ) is chosen such that it is the frequency of the cavity pump , @xmath45 . the mf equations of motion ( [ eq : eoms mf ] ) then become _ r & = & - _ r + ( w + _ ph ) _ i - s_y + _ i + _ i & = & -(w + _ ph ) _ r - _ i - s_x - _ r + _ x & = & - 2 w _ i - ( _ at + 2 ||^2 ) s_y + _ y & = & - 2 w _ r + ( _ at + 2 ||^2 ) s_x + & = & 2 ( _ r s_y + _ i s_x ) , [ eq : eoms mf pump real ] where we have introduced real variables through @xmath46 , @xmath47 and @xmath48 . note that the equations ( [ eq : eoms mf pump real ] ) imply the conservation of the total spin w^2 + 4 |s|^2 = n^2 , [ eq : spin conservation ] where @xmath49 is the total number of the spins . this can be easily verified as _ t s^2 = _ t ( s_x^2 + s_y^2 + w^2 ) = _ t ( 4 s s^ * + w^2 ) = 0 , [ eq : diff spin conservation ] where we have parametrized the total spin as @xmath50 . ) for ( a ) @xmath51 and ( b ) @xmath52 . left hand side ( right hand side ) of ( [ eq : gon eq ] ) is represented by solid blue ( red ) lines . see text for details.,width=491 ] in order to investigate the steady state solutions , we first put @xmath53 to further simplify the problem ( see e.g. @xcite for related studies of the phase diagram of the jc and dicke models ) . we then turn back to the situation with @xmath54 in the next section . with these simplifications , the real equations for the steady state read [ eq : eoms mf pump real ss ] @xmath55 solving the set ( [ eq : eoms mf pump real ss ar],[eq : eoms mf pump real ss ai ] ) for @xmath56 and substituting to ( [ eq : eoms mf pump real ss sx]-[eq : eoms mf pump real ss w ] ) we obtain for the spin conservation ( [ eq : spin conservation ] ) w^2 - n^2 = , [ eq : poly w ] where @xmath57 . this is a 4th order polynomial for @xmath58 and its solutions in terms of radicals can in principle be found yielding rather complicated expressions , which are not of much practical use . instead , we will analyze the properties of ( [ eq : poly w ] ) as follows . since @xmath59 $ ] , the lhs of ( [ eq : poly w ] ) is non - positive , namely @xmath60 $ ] . at the same time the nominator of the rhs is clearly non - positive , so that the non - positivity of the rhs requires the quadratic polynomial in the denominator to be non - negative . we thus have g^4 w^2 + _ at^2 ^2 + _ ph _ at ( 2 g^2 w + _ ph _ at ) 0 . [ eq : cond poly ] the roots of this polynomial read w_= . it can be immediately seen that for @xmath41 , any @xmath61 would yield imaginary @xmath58 . this however does not mean that a solution of the original constraint ( [ eq : poly w ] ) is not possible for both @xmath62 . rather , it means that the point @xmath63 has some particular properties which we will investigate further . for @xmath63 , the steady state equations read [ eq : ss l0 d0 ] @xmath64 ( [ eq : ss l0 d0 sx],[eq : ss l0 d0 sy ] ) imply either @xmath65 or @xmath66 . if @xmath66 , the equations can be readily solved to yield s_x & = & - + s_y & = & + w & = & . [ eq : sol a=0 ] these solutions can be valid only for @xmath67 . in the case where @xmath65 , we can use the spin conservation to parametrize the spin as @xmath68 . ( [ eq : ss l0 d0 w ] ) then becomes = . [ eq : tan ] next we can express @xmath56 from ( [ eq : ss l0 d0 ar],[eq : ss l0 d0 ai ] ) as _ r & = & + _ i & = & - . [ eq : alpha tan ] substituting these expressions to ( [ eq : tan ] ) yields the condition for the angle @xmath69 which determines the solution and can be found numerically . in order to proceed further analytically , we put @xmath70 . the equation ( [ eq : tan ] ) can then be cast in the form z + = _ ph [ eq : gon eq ] or equivalently ( ^2 + _ ph^2 ) ^2 + 2 z + z^2 - _ ph^2 = 0 [ eq : cos eq ] with solutions = , [ eq : sol cos ] where @xmath71 . it is easy to understand the structure of solutions from the graphical representation of ( [ eq : gon eq ] ) , which is shown in fig . [ fig : gon ] . the left ( right ) hand side of ( [ eq : gon eq ] ) is represented by solid blue ( red ) line respectively . assuming @xmath72 , fig . [ fig : gon]a ( fig . [ fig : gon]b ) shows a situation for @xmath51 ( @xmath52 ) respectively . if @xmath51 , one of the solutions is negative , while both solutions are positive for @xmath52 . when tuning @xmath73 , the non negativity of the discriminant in ( [ eq : sol cos ] ) determines the maximal coupling @xmath74 up to which the solutions ( [ eq : sol cos ] ) can be found . it reads g^*_2 = . [ eq : g crit 2 ] note , that the region @xmath75 $ ] admits both solutions ( [ eq : sol a=0 ] ) and ( [ eq : alpha tan],[eq : sol cos ] ) indicating a bistable behaviour . in what follows we refer to the points @xmath76 as transition points . + _ stability study _ + the stability analysis of the steady state solutions ( [ eq : sol a=0 ] ) and ( [ eq : alpha tan],[eq : sol cos ] ) is performed in a standard way by linearizing ( [ eq : eoms mf pump real ] ) around the solutions , i.e. expressing the variables as @xmath77 , where @xmath78 denotes the steady state solution . formally this yields the linearized equations of motion = m v + b , [ eq : stability ] where @xmath79 . instability of the solutions is indicated by the positivity of the real part of the maximum eigenvalue of @xmath80 , see appendix [ app : general stability ] for details . it is noteworthy , that in any steady state solution we have either @xmath81 or @xmath82 and consequently the characteristic polynomial of the matrix @xmath80 becomes |m - y | = y p(y^4 ) , where @xmath83 is some polynomial which is 4th order in @xmath84 . we thus always have one eigenvalue @xmath85 , which is simply the consequence of the spin conservation law ( [ eq : spin conservation ] ) . examining numerically the negativity of real part of the roots of @xmath83 we identify stable and unstable solutions . these are depicted by solid and dashed lines respectively in steady state phase diagram fig . [ fig : stability ] , where we plot @xmath86 and @xmath58 as functions of the coupling @xmath73 . the region ii exhibits bistable behaviour , whereas regions i and iii admit only a single stable solution . we have also verified , by numerically solving the dynamical equations ( [ eq : eoms mf pump real ] ) , that they indeed evolve into the steady state solutions ( [ eq : sol a=0],[eq : alpha tan],[eq : sol cos ] ) ( not shown ) . in the next two subsections we investigate how the inclusion of the non - zero atomic detuning @xmath87 and non - zero coupling @xmath29 terms modifies the @xmath88 solution . , @xmath89 and @xmath90 respectively . stable ( unstable ) steady state solutions are indicated by solid ( dashed ) lines . ( a ) @xmath86 and ( b ) @xmath58 as functions of the coupling @xmath73 . region i exhibits two possible solutions corresponding to the two solutions of ( [ eq : alpha tan ] ) . region ii exhibits the coexistence of the solutions ( [ eq : alpha tan ] ) together with the solution ( [ eq : sol a=0 ] ) indicating bistability . the line colors are used as eye guide to help to identify corresponding @xmath86 and @xmath58 solutions . parameter values used are @xmath91.,width=491 ] regime : the shaded area represents the shrinking of the @xmath92 $ ] region ( denoted as region ii , see also fig . [ fig : stability ] ) with increasing @xmath29 . ( b ) @xmath93 regime : the shaded area corresponds to the region of @xmath73 exhibiting four solutions for a given @xmath73 ( denoted as region r , see also fig . [ fig : delta_grid]a , b ) as @xmath87 is increased . the critical value of @xmath87 at which the four simultaneous solutions cease to exist is represented by the red circle.,width=491 ] in order to shed light on the effect of the @xmath29 term independently of the @xmath87 term , we keep @xmath63 . from ( [ eq : eoms mf pump real ] ) the steady state equations read [ eq : ss lneq0 d0 ] @xmath94 finding the steady state solution encompasses solving a 6th order polynomial equation , which in general can only be done numerically . nevertheless , some information about the steady state solution can be obtained analytically as we now describe . substituting @xmath95 from ( [ eq : ss lneq0 d0 sx],[eq : ss lneq0 d0 sy ] ) to the spin conservation ( [ eq : spin conservation ] ) , we get the following condition for @xmath58 w^2 = n^2 . [ eq : cond w2 ] since @xmath96 , the relation ( [ eq : cond w2 ] ) indicates , that there is no further instability ( @xmath97 ) when changing @xmath29 , i.e. there are also two transition points @xmath76 for @xmath54 . next , one can find solutions for asymptotic values of the parameter @xmath29 . clearly , for @xmath98 , one should recover the solutions of sec . [ sec : linear ] . on the other hand , for @xmath99 , one can look for solution by substituting a perturbative expansion for all the variables of the form @xmath100 . we provide the details of this expansion in appendix [ app : lambda exp . ] . in order to simplify the analytic expressions we take the imaginary part of the pump to be zero , @xmath101 . in this case , to leading order in @xmath29 , the solutions read @xmath102 and [ eq : sx0 ] @xmath103 interestingly , since the spin conservation implies @xmath104 , the two solutions ( [ eq : sx0 1 ] ) , ( [ eq : sx0 2 ] ) yield the same transition point @xmath105 . this is in contrast to the @xmath53 case , where two distinct transition points @xmath106 were identified . here , the solution ( [ eq : sx0 1 ] ) is valid for @xmath107 , whereas ( [ eq : sx0 2 ] ) is valid for @xmath108 . this indicates that the transition point @xmath109 approaches @xmath110 as @xmath29 is increased until they become identical in the @xmath99 limit . in other words , the region ii of the phase diagram fig . [ fig : stability ] is shrinking to zero width as @xmath29 is increased . for the intermediate values of @xmath29 , the values of @xmath76 can be found numerically ( selecting only the physically meaningful solutions of the 6th order polynomial , corresponding to the real values of @xmath58 ) . we plot the shrinking of the @xmath92 $ ] region in fig . [ fig : g2 to g1]a . and the expectation value @xmath58 of the spin as functions of the coupling strength @xmath73 for various values of the nonlinearity @xmath29 : @xmath111 in ( a , b ) and @xmath112 in ( c , d ) . stable ( unstable ) solutions are shown as solid ( dashed ) lines . panels ( e , f ) show a magnification of the data shown in ( c , d ) in the vicinity of the critical point @xmath110 . the dashed dotted lines represent the analytic scaling solutions ( [ eq : a2 ] ) and ( [ eq : w0 ] ) respectively . in all plots the line colors are used as eye guide to help to identify the corresponding @xmath86 and @xmath58 solutions.,width=491 ] and @xmath113 ( point a in fig . [ fig : lambda_grid]b ) represented by the evolution of the spin on the bloch sphere . here the direction of time corresponds to the increasing opacity of the spin evolution line . ( b ) evolution of @xmath58 as a function of time . we take the initial ( final ) time to be the same in ( a ) and ( b).,width=491 ] in fig . [ fig : lambda_grid ] we show the phase diagrams for @xmath86 and @xmath58 for increasing values of @xmath29 . it can be seen from fig . [ fig : lambda_grid]a , b , that for some values of @xmath29 , there is a region with no apparent stable solution . in non - linear systems , this is typically a signature of the appearance of limit cycles which occur through a hopf bifurcation as the system leaves the stable fixed point by changing the coupling @xmath73 @xcite . we represent a limit cycle corresponding to the point a in fig . [ fig : lambda_grid]b as the time evolution of the global spin in fig . [ fig : limit_cycle ] . one can carry the analysis further in the large @xmath29 limit and look for asymptotic solutions in the vicinity of the unique transition point @xmath110 . the first non - trivial contribution to @xmath114 is of order @xmath115 and given by ( see appendix [ app : lambda exp . ] ) | |^2 = ( g_1^ * - ) ( + 13 ) + o((g_1^ * -)^ ) , [ eq : a2 ] which is valid for @xmath116 as the solution for @xmath52 is trivial ( @xmath43 ) . the scaling of the spin observables , say @xmath58 , is simply obtained from the expansion of ( [ eq : sx0 ] ) , which read [ eq : w0 ] @xmath117 where the sign @xmath118 depends on the branch of @xmath58 considered . note that the scaling @xmath119 is characteristic of the ising type model with @xmath120 symmetry and @xmath110 is the phase transition critical point . we compare the asymptotic solutions ( [ eq : a2]),([eq : w0 ] ) with the solutions obtained by numerically solving ( [ eq : ss lneq0 d0 ] ) in fig . [ fig : lambda_grid]e , f . and @xmath58 for @xmath121 ( a , b ) and @xmath122 ( c , d ) . r in ( a , b ) denotes a region exhibiting four distinct solutions . stable ( unstable ) solutions are shown as solid ( dashed ) lines . the line colors are used as eye guide to help to identify the corresponding @xmath86 and @xmath58 solutions . parameter values used are @xmath91.,width=491 ] our next aim is to explore the effect of the @xmath87 term on the solution in the absence of the non - linear @xmath29 term . in this case , the structure of the solutions is dictated by the 4th order @xmath58 polynomial ( [ eq : poly w ] ) and it is straightforward to obtain the solutions numerically . for small @xmath87 , there is a range of @xmath73-values , which admits four solutions ( corresponding to four real solutions of ( [ eq : poly w ] ) ) , fig . [ fig : delta_grid]a , b . as @xmath87 is increased , this region eventually disappears leaving us with only two solutions for all values of @xmath73 . the latter limit can be simply understood from the mf equations ( [ eq : eoms mf pump real ss ] ) . in the large @xmath87 limit , the leading contribution comes from ( [ eq : eoms mf pump real ss sx],[eq : eoms mf pump real ss sy ] ) with the trivial solutions @xmath123 implying @xmath124 . the evolution of the solutions towards this large @xmath87 limit can be seen rather clearly from fig . [ fig : delta_grid]d . next we have determined numerically the size of the region of @xmath73 admitting four solutions as a function of @xmath87 . this is shown in fig . [ fig : g2 to g1]b . clearly , there is some maximal @xmath87 after which there are only two possible solutions , as discussed . this limiting value is represented by the red circle in fig . [ fig : g2 to g1]b . note , that the inclusion of the @xmath87 term also lifts the transition point @xmath110 , i.e. that all solutions are smooth in the vicinity of @xmath110 . in this section we seek the generalization of the single cavity case to higher dimensional geometries . for concreteness , we consider a 2d square geometry depicted schematically in fig . [ fig : scheme]a . before diving in the details of the analysis , we motivate this section by asking whether a 2d square geometry offers mf solutions which are qualitatively different from the 1d case studied above . for example , it was shown in @xcite in the context of laser driven and interacting rydberg gases on a square lattice , that a homogeneous system admits a mf solution that breaks the lattice symmetry and exhibits antiferromagnetic ( af ) order . we start our discussion by deriving the 2d mf equations , which follow from the operator equations of motion ( [ eq : op eoms ] ) given by the hamiltonian ( [ eq : h eff rwa 2d ] ) . repeating the argument yielding ( [ eq : restr ] ) leads to the same conclusion , namely that in the absence of the cavity pump the system posses only a trivial solution , where the cavity modes are empty and all the spins are down ( up ) . we first focus on the situation without the non - linear term , @xmath93 . in the following , we consider the cavity pump @xmath125 , the cavity decay @xmath34 and the spin decay @xmath126 to be the same for all cavity modes and all spins respectively ( the motivation for adding the spin decay will become clear shortly ) . we obtain the set of mf equations [ eq : mf 2d ] @xmath127 - i \gamma \left ( w_{\bar{i } \bar{\nu } } + \mathds{1 } \right ) . \label{eq : mf 2d sz}\end{aligned}\ ] ] here , we allowed for the couplings @xmath73 and the photon detunings @xmath128 to be different for rows and columns while taking the remaining parameters to be the same for all cavities . clearly , the 2d system provides higher tunability by enlarging the parameter space . in the following , we study the problem considering two different perspectives , namely a homogeneous and cluster - mf ansatz . in order to further simplify the mf equations ( [ eq : mf 2d ] ) , it is reasonable to use the typical mf ansatz , namely that , due to translational symmetry in either direction of the 2d array ( along rows or columns ) , the corresponding cavity fields are the same ( @xmath129 , @xmath130 for all @xmath131 ) as well as the spins , @xmath132 . we can now proceed along similar lines as in sec . [ sec : linear ] in order to characterize the solutions . in analogy to sec . [ sec : linear ] we start with @xmath88 and we neglect the spin decay for the moment , @xmath133 . considering @xmath134 and @xmath135 for all @xmath131 to be the same along rows or columns , the steady state mf equations ( [ eq : mf 2d ] ) simplify to [ eq : mf 2d ass ] @xmath136 here @xmath137 ( @xmath138 ) is the number of rows ( columns ) respectively . since the fields @xmath139 and @xmath114 share the same spin @xmath140 , @xmath139 can be given directly in terms of the field @xmath114 . the situation is then essentially equivalent to the single cavity case analyzed in sec . [ sec : single cavity ] , though with some extra tunability provided by larger number of parameters . for example , one can find the transition point @xmath110 by combining ( [ eq : mf 2d ass a ] ) and ( [ eq : mf 2d ass b ] ) with @xmath141 , which is one possible solution of ( [ eq : mf 2d ass s ] ) . the transition point then corresponds to the situation where @xmath65 , i.e. |s|^2 = = @xmath142 for all other parameters fixed . as a consistency check , it is easy to verify that one recovers the single cavity expression @xmath143 by omitting all the `` @xmath144 '' variables and setting @xmath145 and @xmath146 , the number of spins . in the previous section , we have used the homogeneous mf ansatz and found that the solutions correspond effectively to a single cavity case with no intriguing spin configurations . however , it is known and was shown e.g. in @xcite , that a suitable mf ansatz can lead to a non - trivial configuration ( such as antiferromagnetic ordering ) even if the steady state mf equations are completely symmetric under exchange of the spin variables . here we will consider the simplest possible case of such ansatz where a cluster is formed by two adjacent inequivalent spins @xmath147 and @xmath148 . on the other hand , since the fields @xmath114 and @xmath139 couple to both @xmath149 and @xmath150 in the same way , we take @xmath151 to be the same along rows ( columns ) due to translational symmetry . in order to simplify the equations , we now take all the parameters to be the same along all rows and columns ( i.e. we set @xmath152 and @xmath153 for all @xmath154 ) . starting with the simplest case @xmath88 , the mf equations ( [ eq : mf 2d ] ) become [ eq : mf 2d symm ] @xmath155 - i \gamma \left ( w_{\bar{j } } + \mathds{1 } \right ) \label{eq : mf 2d symm sz},\end{aligned}\ ] ] where @xmath156 labels the different spins of the cluster and we assumed that each field @xmath151 couples to the same number of spins 1 and 2 , hence the factor @xmath157 in the second term of ( [ eq : mf 2d symm a]),([eq : mf 2d symm b ] ) . here @xmath49 denotes the total number of spins , i.e. there are @xmath158 spins along rows and along columns . until now we did not comment on the spin conservation in the 2d case . going back to the most general situation , where every individual spin and cavity mode is described in terms of the corresponding mf variables @xmath159 and @xmath160 respectively , the system evolves according to the full set of the 2d mf equations ( [ eq : mf 2d ] ) . one can verify by means of ( [ eq : diff spin conservation ] ) that ( [ eq : mf 2d ] ) actually imply both local w_|i |^2 + 4 | s_|i |^2 |^2 = 1 , @xmath133 . in ( [ eq : spin conservation global ] ) @xmath161 and @xmath162 are the global spin components . one should appreciate that ( [ eq : spin conservation global ] ) actually prevents any spin configuration incompatible with it , including the af order ( which corresponds to @xmath163 when using the here considered cluster - mf ) . in order to see e.g. the af order , such as in @xcite , one needs to break the global spin conservation . this is achieved by the inclusion of the spin decay @xmath126 , which breaks both the local and global spin conservations ( [ eq : spin conservation local]),([eq : spin conservation global ] ) . allowing for the spin decay , we can now ask , whether the solution of the cluster - mf equations ( [ eq : mf 2d symm ] ) features any non - trivial spin configuration ( @xmath164 ) . first , it follows from ( [ eq : mf 2d symm a ] ) and ( [ eq : mf 2d symm b ] ) that @xmath165 . next , expressing for @xmath166 from ( [ eq : mf 2d symm s ] ) and substituting to ( [ eq : mf 2d symm sz ] ) , it can be shown that = w_1 = w_2 and @xmath167 from ( [ eq : mf 2d symm s ] ) . once again we find that the structure of the equations ( [ eq : mf 2d symm ] ) reduces the problem effectively to the single cavity situation described by a simple set of variables @xmath168 and this even when allowing for inequivalent spin configurations and the spin decay . one might argue , that the equivalence to the single cavity case conjectured in sections [ sec : asym array ] and [ sec : symm array ] is an artefact of taking @xmath169 . indeed , when comparing ( [ eq : eoms mf ] ) and ( [ eq : mf 2d ] ) , one can note that there is a _ qualitative _ difference between the 1d and 2d situation . specifically , there is an extra coupling between the @xmath0 and @xmath144 modes , the last term in ( [ eq : mf 2d a ] ) and ( [ eq : mf 2d b ] ) . we have numerically verified that , in a general situation with variables @xmath170 , @xmath156 and @xmath171 , @xmath172 , @xmath173 , the solutions always yield @xmath174 . for completeness we have included in our numerical analysis also the special cases , where any possible combination of the following conditions can occur : @xmath175 , @xmath176 , @xmath177 , @xmath53 . in summary , the mf equations ( [ eq : mf 2d ] ) with the ansatz considered in sections [ sec : asym array ] and [ sec : symm array ] on a square lattice reduce to effectively one - dimensional description with no intriguing spin configurations . it would be desirable to perform a beyond mf study of the non - linear two - dimensional model in order to asses the true nature of the steady state and the corresponding spin and field configurations , including their mutual correlations . also we did not fully exploit the possibilities offered by the proposed implementation , such as taking different geometries of the array or allowing for disordered coupling strengths , which we leave for further investigations . motivated by the progress in integrated optical circuits , we have proposed a possible realization of a two - dimensional cavity array with trapped atoms , which is a promising scalable quantum architecture . we derived an effective description of the system in terms of jaynes - cummings like hamiltonian with highly tunable parameters and extra non - linear terms . we then analyzed the dynamics of the system using a mf approach . we have found a rich behaviour including bistable regions , ising like phase transition or occurrence of limit cycles through hopf bifurcations . in the present setup , we have not found conceptual differences between the one and two - dimensional cases at the level of the mf description and with the geometry considered . we hope that the present work lays down grounds for future studies of the cavity arrays realized with integrated optical circuits . the problems which might be addressed in the future are e.g. going beyond mf description , accounting for more exotic geometries or studying effective spin physics as a low energy limit of the presented cavity hamiltonian . we thank t. pohl , j. goette , d. jukic and all members of the quilmi consortium for fruitful discussions . j.m . would like to thank matteo marcuzzi for useful discussions . the research leading to these results has received funding from the european research council under the european union s seventh framework programme ( fp/2007 - 2013 ) / erc grant agreement no . 335266 ( escquma ) and the eu - fet grants hairs 612862 and quilmi 295293 . here we derive the effective hamiltonian resulting from the adiabatic elimination of the state @xmath16 and starting with the full system hamiltonian given in ( [ eq : h rwa 2d ] ) , which we rewrite in the hard - core boson representation with annihilation operators @xmath178 , @xmath179 for the atom at position @xmath180 h & = & _ i a_i^a_i + _ b_^b_+ _ e b^_e , i b_e , i + _ s b^_s , i b_s , i + & & + ( b^_e , i b_s , i e^-i _ t t + b^_s , i b_e , i e^i _ t t ) + & & + g ( b^_e , i b_g , i ( a_i + b _ ) + h.c . ) . [ eq : h rwa 2d hard core ] first , we determine the transformation to the rotating frame . we consider a general unitary transformation u = e^-it ( _ i a_i^a_i + _ b_^b_+ _ i b_e , i^b_e , i + _ i b_s , i^b_s , i ) , where @xmath181 are some arbitrary frequencies . the requirement of eliminating all explicit time dependencies in eq.([eq : h rwa 2d hard core ] ) leads to the conditions _ |i _ |i & = & _ |i | + _ | & = & _ |i | . [ eq : cond rwa 2d ] note that in general , it is not possible to bring the levels @xmath182 and @xmath183 to degeneracy for all atoms since @xmath184 can not be satisfied for all @xmath185 and @xmath186 at the same time . on the other hand , one has a freedom in the choice of frequencies @xmath187 , provided the conditions ( [ eq : cond rwa 2d ] ) are satisfied . we adopt the following choice _ |i & = & _ | = _ |i | = _ aux + _ |i | & = & _ aux - _ t , where @xmath22 is an arbitrary auxiliary frequency . the hamiltonian ( [ eq : h rwa 2d ] ) then becomes h & = & _ i a_i^a_i + _ b_^b_+ _ e b^_e , i b_e , i + _ s b^_s , i b_s , i + & & + ( b^_e , i b_s , i + b^_s , i b_e , i ) + g ( b^_e , i b_g , i ( a_i + b _ ) + h.c . ) , [ eq : h rwa 2d rot ] where @xmath19 , @xmath20 and @xmath21 . this leads to the equations of motion i _ e,|i| & = & _ e b_e,|i| + b_s,|i| + g b_g,|i|(a_|i + b_| ) + i _ g,|i| & = & g b_e,|i| ( a_|i^+ b_|^ ) + i _ s,|i| & = & _ s b_s,|i| + b_e,|i| + i _ |i & = & _ |i a_|i + g _ b^_g,|i b_e,|i + i _ | & = & _ | b_| + g _ i b^_g , i| b_e , i| , where the summation over @xmath26 and @xmath25 in the last two equations is emphasized . setting @xmath188 and substituting to the remaining equations yields i _ g,|i| & = & ( - ) ( a_|i^+ b_|^ ) ( a_|i + b_| ) b_g,|i| + ( - ) ( a_|i^+ b_|^ ) b_s,|i| + i _ s,|i| & = & _ s b_s,|i| + ( - ) b_s,|i| + ( - ) b_g,|i| ( a_|i + b_| ) + i _ |i & = & _ |i a_|i + _ ( - ) b_g,|i^b_g,|i ( a_|i + b _ ) + ( - ) b^_g,|i b_s,|i + i _ | & = & _ | b_| + _ i ( - ) b_g , i |^b_g , i | ( a_i + b_| ) + ( - ) b^_g , i | b_s , i | . switching back to the pauli matrices representation , now in the @xmath189 basis , the effective hamiltonian reads h = _ i a_i^a_i + _ b_^b_+ ^z_i + ( ^+_i ( a_i + b _ ) + h.c . ) + f_i , [ eq : h eff rwa 2d app ] where _ | & = & _ s - + & = & - + f_|i | & = & ( _ s - ) . the corresponding equations of motion for the operators read [ eq : op eoms ] @xmath190 \sigma^-_{\bar{i } \bar{\nu } } - \tilde{g } \left ( a_{\bar{i } } + b_{\bar{\nu } } \right ) \sigma^z_{\bar{i } \bar{\nu } } \label{eq : op eoms sm } \\ i \dot{\sigma}^z_{\bar{i } \bar{\nu } } & = 2\tilde{g } \left [ \sigma^+_{\bar{i } \bar{\nu } } \left ( a_{\bar{i } } + b_{\bar{\nu } } \right ) - { \rm h.c . } \right ] - i \gamma \left ( \sigma^z_{\bar{i } \bar{\nu } } + \mathds{1 } \right ) , \label{eq : op eoms sz}\end{aligned}\ ] ] where _ at & = & _ s - + _ ph^l & = & _ l - + & = & - and we have introduced the cavity and spin decays @xmath191 which we take to be the same for all cavity modes and all spins respectively . the matrix @xmath80 used in the stability study , equation ( [ eq : stability ] ) , can be simply obtained from the mf equations of motion ( [ eq : eoms mf pump real ] ) and reads m = - & _ ph + |w & 0 & - & |_i + -_ph - |w & - & - & 0 & -|_r + -4|_r |s_y & -2 |w -4|_i |s_y & 0 & - ( _ at + 2 |||^2 ) & -2 |_i + -2 |w + 4|_r |s_x & 4|_i |s_x & ( _ at + 2 |||^2 ) & 0 & -2 |_r + 2 |s_y & 2 |s_x & 2 |_i & 2 |_r & 0 , where @xmath192 and @xmath78 are the steady state solutions , @xmath193 . here we seek a perturbative solution of the algebraic steady state equations ( [ eq : spin conservation ] ) and ( [ eq : ss lneq0 d0 ] ) in the large @xmath29 limit . this can be done using a perturbative ansatz for the variables of the form v = _ n=0^^-n v^(n ) = v^(0 ) + ^-1 v^(1 ) + ^-2 v^(2 ) + ... the set of equations of order @xmath194 read _ i^(0 ) w^(0 ) & = & 0 + _ r^(0 ) w^(0 ) & = & 0 + ( . _ i^(0).^2 + ._i^(0).^2 ) s_y^(0 ) & = & 0 + ( . _ i^(0).^2 + ._i^(0).^2 ) s_x^(0 ) & = & 0 , [ eq : exp order -1 ] which yield the solution for @xmath195 . using this result , order @xmath196 equations simplify to in order to proceed , we realize that the equation ( [ eq : ss lneq0 d0 w ] ) at order @xmath115 reads _ r^(1 ) s_y^(0 ) = - _ i^(1 ) s_x^(0 ) . substituting for @xmath198 from ( [ eq : exp order 0 ar],[eq : exp order 0 ai ] ) , it can be cast to the form - = . in order to simplify the treatment further , we put @xmath70 which implies @xmath199 and by the sake of ( [ eq : exp order 0 ai ] ) @xmath200 . we will use these solutions in what follows . where @xmath207 and we have used the spin conservation @xmath208 . since @xmath209 , the solution ( [ eq : sol sx order 1a ] ) is valid for @xmath210 . next , we remark , that the function @xmath211 is monotonously decreasing with the limit _ s_x,+^(0 ) ( ) = n > n , i.e. is unphysical . we are thus left with the solution @xmath212 which is also monotonously decreasing function with the asymptotes _ 0 s_x,-^(0 ) ( ) & = & 0 + _ s_x,-^(0 ) ( ) & = & - n. it is thus clear that the solution @xmath213 is valid for @xmath214 $ ] , where @xmath215 is some critical value for which @xmath213 reaches the physically allowed maximum @xmath216 . it is easy to find , that @xmath217 . this completes the leading order solutions and yields the expressions ( [ eq : sx0 1]),([eq : sx0 2 ] ) . the equations ( [ eq : exp order -1]),([eq : exp order 0]),([eq : exp order 1 ] ) yield a closed set for spin variables up to the order @xmath115 and for @xmath114 up to @xmath218 . it is straightforward to find the solutions up to the respective order explicitly , giving however rather lengthy algebraic expressions . these can be simplified in specific situations . for example , as we discuss in the main text , when looking for solutions in the vicinity of the transition point @xmath219 , the leading order contribution to @xmath86 is of order @xmath218 , namely | |^2 = . _ r^(1 ) .^2 + o(^-3 ) , which gives the relation ( [ eq : a2 ] ) . similarly , the leading contribution to @xmath58 is of order @xmath196 yielding the expressions ( [ eq : w0 ] ) .

we investigate distributed supervisory control of concurrent discrete - event systems . supervisory control theory of discrete - event systems modeled as finite automata was introduced by ramadge and wonham @xcite and studied by many others . it aims to guarantee that the control specifications consisting of safety and of nonblockingness are satisfied in the controlled ( closed - loop ) system . safety means that the language of the closed - loop system is included in a prescribed specification , and nonblockingness means that all controlled behaviors can always be completed to a marked controlled behavior . supervisory control is realized by a supervisor that runs in parallel with the system and imposes the specification by disabling , at each state , some of the controllable events in a feedback manner . since only controllable specifications can be achieved , one of the key issues is the computation of the supremal controllable sublanguage of the specification , from which the supervisor can be constructed . supervisory control theory is well developed for monolithic systems , i.e. , systems where the plant is modeled as a single generator . however , most of the current complex engineering systems can be abstracted as a composition of many components . systems that model technological systems typically consist of small generators that communicate with each other in a synchronous @xcite or asynchronous @xcite way . such systems are often called modular ( or concurrent or distributed ) discrete - event systems . it is known that to compute the overall monolithic plant for such a system can be unrealistic because the number of states of a modular system grows exponentially with respect to the number of components . this limits the applicability of the monolithic supervisory control synthesis to relatively small systems . on the other hand , the purely decentralized control consisting of an independent construction of a supervisor for each subsystem is only applicable for a local ( decomposable ) specification . specifically , let @xmath0 and @xmath1 be two systems over the respective alphabets @xmath2 and @xmath3 modeled as finite generators forming the overall plant @xmath4 , the computation of which we want to avoid , and let @xmath5 denote a specification . there exists a monolithic supervisor for the monolithic plant @xmath4 if and only if the specification is controllable with respect to the plant . however , to avoid the computation of the monolithic plant @xmath4 , the naive approach to compute supervisors for each subsystem separately does not work in general . to demonstrate this , assume that we compute supervisors @xmath6 such that @xmath7 , where @xmath8 denotes the projection of the specification @xmath9 to the alphabet of the generator @xmath10 . then , it can still be the case that @xmath11 . the problem here is that the specification is not necessarily decomposable . recall that a languages @xmath12 is decomposable ( or separable ) with respect to alphabets @xmath2 and @xmath3 if @xmath13 , where @xmath14 is a natural projection ( see more details below ) . if a language is not decomposable , it would be natural to search for some `` maximal '' decomposable sublanguage . unfortunately , this is not algorithmically possible @xcite . to deal with this problem , we introduced the notion of _ conditional decomposability_. conditional decomposability of a language @xmath12 with respect to alphabets @xmath2 and @xmath3 requires to find another alphabet @xmath15 such that the language @xmath12 is decomposable with respect to @xmath16 and @xmath17 , i.e. , @xmath18 , see more details below . the difference between decomposability and conditional decomposability is that every language can be made conditionally decomposable , which overcomes the undecidable problem of finding a nonempty decomposable sublanguage discussed above . in other words , we can always find an alphabet @xmath15 such that the language becomes conditionally decomposable . indeed , one could take @xmath19 , but the aim is to find a reasonably small such alphabet . although to find a minimal alphabet is np - hard @xcite , a polynomial - time algorithm to find an acceptable alphabet is described in @xcite . the observation that conditional decomposability is always possible leads us to the formulation of the following problem that we investigate in this paper . [ problem1 ] let @xmath0 and @xmath1 be generators over the alphabets @xmath2 and @xmath3 , respectively . assume that a specification @xmath20 and its prefix - closure @xmath21 are conditionally decomposable with respect to @xmath2 , @xmath3 , and @xmath15 , for some alphabet @xmath15 such that @xmath22 . let @xmath23 be a coordinator that ensures the necessary communication between the systems @xmath0 and @xmath1 . the aim is to determine nonblocking supervisors @xmath24 and @xmath25 such that the supervised distributed system satisfies @xmath26 ) \parallel l_m(s_{2}/ [ g_2 \| g_k ] ) = k & & \text { and } & & l(s_{1}/ [ g_1 \| g_k ] ) \parallel l(s_{2}/ [ g_2 \| g_k ] ) = \overline{k}\ , , \end{aligned}\ ] ] i.e. , it fulfills the specification and the supervisors are nonconflicting . in other words , the question is whether it is possible to distribute the monolithic ( global ) supervisor with help of a coordinator in such a way that there exist nonblocking and nonconflicting supervisors @xmath27 and @xmath28 such that the supervised distributed closed - loop system satisfies safety and nonblockingness . note that we restrict ourselves to the case , where the coordinator @xmath29 does not affect the behavior of the whole system . the case where @xmath29 can influence the system , i.e. , the new plant is @xmath30 , is left for the future investigation . however , it should be pointed out that one has to be careful with this modification , since the coordinator could influence the system in such a way that it forbids some uncontrollable events , which may be physically impossible . a possible solution would be to consider only such coordinators @xmath29 that are controllable with respect to ( some projection of ) @xmath4 . in this paper , we investigate problem [ problem1 ] . we provide a necessary and sufficient condition for the existence of a solution , which results in a decision procedure and in an algorithm to compute the solution , if it exists . as usual , if the solution does not exist , we make use of the coordination control methods to construct an acceptable sub - specification for which a solution exists , measured with respect to the solution provided in the coordination control framework . moreover , we relax the coordination control framework so that the main results of the framework still hold , but the presentation is simplified and potentially larger solutions can be achieved . however , we do not claim that the use of the coordination control framework is the best here , we only show that it is possible to use it here . it is an interesting research topic whether other methods could be used to obtain a better solution . related work that could be useful to achieve this includes , but is not limited to , @xcite . the paper is organized as follows . in section [ sec : preliminaries ] , the necessary notation is introduced and the basic elements of supervisory control theory are recalled . in section [ sec : existenceofsolution ] , we provide necessary and sufficient conditions under which the solution of the problem exists . in section [ sec : relaxation ] , we recall the basic coordination control framework and introduce its relaxed variant . in section [ sec : comparison ] , we compare problem [ problem1 ] with the ( relaxed ) coordination control problem . in section [ sec : app ] , we make use of the relaxed coordination control to find a reasonable sub - specification for which a solution exists if a solution for the specification does not exist . an industrial example is provided in section [ sec : mri ] . we discuss only the full - observation case here . the approach can be extended to partial observations using either the notion of ( conditional ) normality of @xcite or the notion of ( conditional ) relative observability of @xcite . let @xmath31 denote a finite nonempty set of events ( an _ alphabet _ ) , and let @xmath32 denote the set of all finite words over @xmath31 . the empty word is denoted by @xmath33 . a _ language _ over @xmath31 is a subset of @xmath32 . the prefix closure of a language @xmath12 over @xmath31 is the set @xmath34 . a language @xmath12 is _ prefix - closed _ if @xmath35 . for more details , the reader is referred to @xcite . a _ generator _ is a structure @xmath36 , where @xmath37 is the finite set of states , @xmath31 is the finite nonempty set of events , @xmath38 is the partial transition function , @xmath39 is the initial state , and @xmath40 is the set of marked states . the transition function @xmath41 can be extended to the domain @xmath42 in the usual way . the behavior of @xmath43 is described in terms of languages . the language _ generated _ by @xmath43 is the set @xmath44 , and the language _ marked _ by @xmath43 is the set @xmath45 . obviously , @xmath46 . for @xmath47 , a _ ( natural ) projection _ is a mapping @xmath48 , which deletes from any word all letters that belong to @xmath49 . formally , it is a homomorphism defined by @xmath50 , for @xmath51 in @xmath49 , and @xmath52 , for @xmath51 in @xmath53 . it is extended ( as a homomorphism for concatenation ) from letters to words by induction . the _ inverse image _ of @xmath54 is denoted by @xmath55 . for three alphabets @xmath56 , @xmath57 , @xmath58 , subsets of @xmath31 , the notation @xmath59 denotes the projection from @xmath60 to @xmath61 . if @xmath62 , we simplify the notation to @xmath63 . similarly , the notation @xmath64 stands for the projection from @xmath32 to @xmath65 . the projection of a generator @xmath43 , denoted by @xmath66 , is a generator whose behavior satisfies @xmath67 and @xmath68 . it is defined using the standard subset construction , cf . @xcite . the synchronous product of languages @xmath69 over @xmath2 and @xmath70 over @xmath3 is the language @xmath71 , where @xmath72 is a projection , @xmath73 . a definition for generators can be found in @xcite . for generators @xmath0 and @xmath1 , @xmath74 and @xmath75 . languages @xmath9 and @xmath12 are _ synchronously nonconflicting _ if @xmath76 . we recall the basic elements of supervisory control theory . a _ controlled generator _ over an alphabet @xmath31 is a structure @xmath77 , where @xmath43 is a generator over the alphabet @xmath31 , @xmath78 is the set of _ controllable events _ , @xmath79 is the set of _ uncontrollable events _ , and @xmath80 is the _ set of control patterns_. a _ supervisor _ for the controlled generator @xmath77 is a map @xmath81 . the _ closed - loop system _ associated with the controlled generator @xmath77 and the supervisor @xmath82 is defined as the smallest language @xmath83 such that @xmath84 , and if @xmath85 , @xmath86 , and @xmath87 , then @xmath88 . the marked language of the closed - loop system is defined as @xmath89 , where @xmath90 is a specification . the intuition is that the supervisor disables some of the controllable transitions of @xmath43 , but never an uncontrollable transition , and marks in accordance with the specification . if @xmath91 , then the supervisor @xmath82 is called _ nonblocking_. in the automata framework , where a supervisor @xmath82 has a finite representation as a generator , basically it is the generator @xmath43 with some of the controllable events disabled , the closed - loop system is a synchronous product of the supervisor and the plant . thus , we can write the closed - loop system as @xmath92 . moreover , the supervisor keeps the information about the marked states , i.e. , we have that @xmath93 . the supervisor is then nonblocking if @xmath94 . control objectives of supervisory control are defined using a specification @xmath9 . the goal of supervisory control is to find a nonblocking supervisor @xmath82 such that @xmath95 . in the monolithic case , such a supervisor exists if and only if @xmath9 is _ controllable _ with respect to @xmath96 and @xmath97 ( i.e. , @xmath98 ) , cf . @xcite . if the specification fails to satisfy controllability , controllable sublanguages of the specification are considered instead . let @xmath99 denote the supremal controllable sublanguage of @xmath9 with respect to @xmath96 and @xmath97 , which always exists and is equal to the union of all controllable sublanguages of @xmath9 , cf . @xcite . the projection @xmath100 , with @xmath47 , is an _ @xmath12-observer _ for @xmath101 if , for all @xmath102 and @xmath103 , @xmath104 is a prefix of @xmath105 implies that there exists @xmath106 such that @xmath107 and @xmath108 , see @xcite . this property is well known and widely used in supervisory control of hierarchical and distributed discrete - event systems , and , as mentioned in @xcite , also in compositional verification @xcite and modular synthesis @xcite . if the projection does not satisfy the property , the co - domain of the projection can be extended so that it is satisfied . although the computation of such a minimal extension is np - hard , there exists a polynomial - time algorithm that finds an acceptable extension @xcite . more information can be found in @xcite . a language @xmath9 is _ conditionally decomposable _ with respect to alphabets @xmath2 , @xmath3 , and @xmath15 , where @xmath109 , if @xmath110 where @xmath111 is a projection , for @xmath73 . the definition is extended to more alphabets in a natural way . for simplicity , we formulated problem [ problem1 ] only for two subsystems . however , it is clear how to extend it to @xmath112 subsystems . it should perhaps be pointed out that for @xmath112 , the set @xmath15 contains all shared events @xmath113 . we define the coordinator @xmath29 as the parallel composition of projections . it is a folklore in automata theory that , in the worst case , the computation of a projected generator can be of exponential size with respect to the number of states of the original generator . however , it is known that if the projection satisfies the observer property , then the projected generator is smaller than the original generator @xcite . therefore , we suggest the following steps to compute the alphabet @xmath15 and the coordinator @xmath29 : 1 . let @xmath114 be the set of all shared events of the generators @xmath0 and @xmath1 . 2 . extend the alphabet @xmath15 so that @xmath9 and @xmath21 are conditionally decomposable with respect to @xmath2 , @xmath3 , and @xmath15 , cf . @xcite for a polynomial algorithm . 3 . if the computation of @xmath29 in the following step is not possible ( e.g. , because of the state explosion ) , extend the alphabet @xmath15 so that the projection @xmath115 is an @xmath116-observer , for @xmath73 , cf . 4 . define the coordinator @xmath29 as @xmath23 , where @xmath117 . we now provide necessary and sufficient conditions for the existence of a solution for problem [ problem1 ] . [ thm : char ] consider the setting of problem [ problem1 ] . there are nonblocking and nonconflicting supervisors @xmath24 and @xmath25 such that the distributed closed - loop system satisfies @xmath118 ) \parallel l_m(s_{2}/[g_2 \| g_k ] ) = k $ ] if and only if @xmath119 , @xmath120 , and @xmath27 and @xmath28 are nonblocking and nonconflicting supervisors with respect to @xmath121 and @xmath122 , respectively . the second part of the theorem gives us the following equations with variable languages @xmath123 and @xmath124 : @xmath125 let @xmath126 denote the infimal _ prefix - closed _ controllable superlanguage of @xmath9 with respect to @xmath12 , cf . . then we immediately have the following . [ lem1 ] consider the setting of problem [ problem1 ] . if the specification @xmath9 is prefix - closed , then there exists a solution of problem [ problem1 ] if and only if the languages @xmath127 and @xmath128 satisfy equations ( [ sol1 ] ) . the infimal controllable superlanguage is prefix - closed and it is not possible to compute the infimal non - prefix - closed controllable superlanguage , since it does not always exist . thus , the previous lemma can not be directly used for non - prefix - closed languages . however , we show that it is surprisingly sufficient to verify whether there exists a solution for the prefix - closure of the specification . [ thmmain ] consider the setting of problem [ problem1 ] . there exists a solution for specification @xmath9 if and only if there exists a solution for its prefix - closure @xmath21 . in other words , assuming that @xmath9 and @xmath21 are conditionally decomposable , it is sufficient to check whether there exists a solution for the prefix - closure @xmath21 of the specification @xmath9 using proposition [ lem1 ] . by theorem [ thmmain ] , this implies a solution for the specification @xmath9 , which can actually be derived from the solution for @xmath21 . if a solution of problem [ problem1 ] does not exist , we make use of the coordination control framework to obtain an acceptable sublanguage of the specification for which a solution exists . coordination control is based on the concept of conditional independence of probability theory , which allows for coordinated systems not only in discrete - event systems , but also in linear , stochastic , and hybrid systems , see @xcite . we first recall the original coordination control framework and then relax it so that it admits more and larger solutions , cf . example [ ex1 ] . we briefly recall the original coordination control framework developed in @xcite . the notation is that of @xcite . [ problem : origcontrolsynthesis ] consider generators @xmath0 and @xmath1 over the alphabets @xmath2 and @xmath3 , resp . , and a generator @xmath29 ( a _ coordinator _ ) over an alphabet @xmath15 , where @xmath129 . assume that a specification @xmath130 and its prefix - closure @xmath21 are conditionally decomposable with respect to @xmath2 , @xmath3 , and @xmath15 . the aim is to determine nonblocking supervisors @xmath27 , @xmath28 , and @xmath131 such that @xmath132 , @xmath133)\subseteq p_{i+k}(k)$ ] , for @xmath73 , and the closed - loop system with the coordinator satisfies @xmath134 ) \parallel l_m(s_2/ [ g_2 \| ( s_k / g_k ) ] ) = k$ ] . although the coordinator @xmath29 in the statement of problem [ problem : origcontrolsynthesis ] is a general automaton , we suggested to construct it in the same way as in problem [ problem1 ] . therefore , we may assume that @xmath23 . in what follows , we use the notation @xmath135 to denote the set of locally uncontrollable events of the alphabet @xmath56 . [ def : origconditionalcontrollability ] let @xmath0 and @xmath1 be generators over @xmath2 and @xmath3 , respectively , and let @xmath29 be a coordinator over @xmath15 . a language @xmath136 is _ conditionally controllable _ for generators @xmath0 , @xmath1 , @xmath29 and uncontrollable alphabets @xmath137 , @xmath138 , @xmath139 if ( i ) @xmath140 is controllable with respect to @xmath141 and @xmath139 and ( ii ) @xmath142 is controllable with respect to @xmath143 and @xmath144 , where @xmath145 , for @xmath73 . the main existential result follows . consider the setting of problem [ problem : origcontrolsynthesis ] . there exist nonblocking supervisors @xmath27 , @xmath28 , and @xmath131 such that @xmath146 ) \parallel l_m(s_2/[g_2 \| ( s_k / g_k ) ] ) = k$ ] if and only if the specification @xmath9 is conditionally controllable with respect to generators @xmath0 , @xmath1 , @xmath29 and uncontrollable alphabets @xmath137 , @xmath138 , @xmath139 . similarly as in the monolithic supervisory control , if the specification fails to be conditionally controllable , the supremal conditionally controllable sublanguage is computed . it always exists @xcite . consider the setting of problem [ problem : origcontrolsynthesis ] and define the languages @xmath147 for @xmath73 . if @xmath148 and @xmath149 are nonconflicting and @xmath150 is controllable with respect to @xmath141 and @xmath139 , then @xmath151 is the supremal conditionally controllable sublanguage of @xmath9 . otherwise , we compute a new supervisor @xmath152 denoted @xmath153 . this gives the following result , which is not stated in @xcite , but is in a more general form stated in @xcite . [ thm4 ] consider the setting of problem [ problem : origcontrolsynthesis ] and the languages defined above . if @xmath154 and @xmath155 are synchronously nonconflicting ( e.g. , prefix - closed ) for @xmath73 , then @xmath156 is the supremal conditionally controllable sublanguage of @xmath9 . in this section , we relax the coordination control problem . proofs of the results presented in this section can be found in the technical report @xcite . unlike the original approach , the supervisor for the coordinator is not computed and , thus , it is not included in the closed - loop system . we comment on this below the definition . [ problem : relaxed ] consider generators @xmath0 and @xmath1 over the alphabets @xmath2 and @xmath3 , respectively . let @xmath15 be an alphabet such that @xmath22 . a generator @xmath29 over the alphabet @xmath15 is called a coordinator . assume that a specification @xmath130 and its prefix - closure @xmath21 are conditionally decomposable with respect to @xmath2 , @xmath3 , and @xmath15 . the aim is to determine nonblocking supervisors @xmath27 and @xmath28 such that @xmath157)\subseteq p_{i+k}(k)$ ] , for @xmath73 , and the closed - loop system satisfies @xmath158 ) \parallel l_m(s_2/ [ g_2 \| g_k ] ) = k$ ] . the construction of a coordinator remains unchanged , compared to the original framework . the difference between the original and the relaxed problem is that the supervisor @xmath131 and , hence , the closed - loop system @xmath159 for the coordinator part of the specification is not computed . the original motivation to include the supervisor @xmath131 was the anti - monotonicity of the basic supervisory control operator ( the supremal controllable sublanguage ) . indeed , if the specification is fixed , then making the plant language smaller results in increasing permissiveness of the supervisors given by the supremal controllable sublanguages . however , it turns out that decreasing the plant @xmath160 for the local supervisors by replacing @xmath29 with , in general , a smaller closed - loop system @xmath159 does not change the permissiveness of the local closed - loop systems @xmath161 $ ] . this is because of the transitivity of controllability , since the closed - loop @xmath159 is always controllable with respect to @xmath29 . this means that a language @xmath162 is controllable with respect to @xmath163 if and only if it is controllable with respect to @xmath164 . for this reason , the supervisor for the coordinator does not help and is not used in the relaxed framework . [ def : conditionalcontrollability ] let @xmath0 and @xmath1 be generators over the alphabets @xmath2 and @xmath3 , respectively , and let @xmath29 be a coordinator over the alphabet @xmath15 . a language @xmath136 is _ relaxed conditionally controllable _ for generators @xmath0 , @xmath1 , @xmath29 if the projected specification @xmath142 is controllable with respect to @xmath165 and @xmath144 , where @xmath166 , for @xmath73 . by definition , any solution of the coordination control problem is also a solution of the relaxed coordination control problem . the opposite does not hold as demonstrated in the following simple example . [ ex1 ] let @xmath167 and @xmath168 , where the set of controllable events is @xmath169 . let the specification @xmath9 be the composition of @xmath170 and @xmath171 . then @xmath9 it is conditionally decomposable with respect to @xmath172 , @xmath173 , and @xmath174 . notice that @xmath175 and @xmath176 . therefore , @xmath140 is _ not _ controllable with respect to @xmath141 , thus it is _ not _ conditionally controllable and , hence , not a solution of the original coordination control problem . moreover , there does not exist a nonempty controllable sublanguage of @xmath140 that is controllable with respect to @xmath141 . on the other hand , @xmath177 and @xmath178 , hence @xmath142 is controllable with respect to @xmath179 , for @xmath73 . since nothing is required for @xmath140 in the relaxed framework , @xmath9 is a solution of the relaxed coordination control problem . the advantage of the relaxed framework is thus not only a simplification of the notation , but also the fact that there exist solutions in the relaxed framework that can not be achieved in the original coordination control framework . however , we should mention that the relaxed framework also restricts the set of potential solutions , i.e. , there exist solutions of problem [ problem1 ] that can not be achieved in the relaxed framework . this will be clarified in section [ sec : comparison ] . therefore , further relaxations of this framework would be of interest . we now show that the main results of the original framework hold in the relaxed framework , too . [ th : controlsynthesissafety ] consider the setting of problem [ problem : relaxed ] . there exist nonblocking supervisors @xmath27 and @xmath28 such that the closed - loop system satisfies @xmath180 ) \parallel l_m(s_2/[g_2 \| g_k ] ) = k$ ] if and only if the specification @xmath9 is relaxed conditionally controllable for generators @xmath0 , @xmath1 , @xmath29 . if the specification can not be achieved as the resulting behavior of the coordinated system according to theorem [ th : controlsynthesissafety ] , we describe a procedure to compute the maximal sublanguage of the specification that satisfies both these conditions . we use the notation @xmath181 to denote the supremal relaxed conditionally controllable sublanguage of @xmath9 with respect to plant @xmath182 and the sets of uncontrollable events @xmath137 , @xmath138 , @xmath139 , which exists and equals to the union of all relaxed conditionally controllable sublanguages of the language @xmath9 . furthermore , we define the supremal controllable languages for the local plants combined with the coordinator . consider the setting of problem [ problem : relaxed ] and define the languages @xmath183 where @xmath184 denotes the supremal controllable sublanguage of @xmath9 with respect to @xmath12 and @xmath97 . we now generalize the sufficient conditions for a distributed computation of the supremal conditionally controllable sublanguage from the original to the relaxed framework . the sufficient conditions are formulated in terms of controllability of the composition of local supervisors projected to the coordinator alphabet . the first main constructive result of the relaxed coordination control framework to guarantee relaxed conditional controllability in a maximally permissive way is stated below . [ thm2b ] consider the setting of problem [ problem : relaxed ] and the languages defined in ( [ eqcnr ] ) . if @xmath185 and @xmath186 are synchronously nonconflicting and @xmath187 is controllable with respect to @xmath141 and @xmath139 , then @xmath188 , for @xmath182 . the question remains what happens if the intersection of the projections is not controllable as required in theorem [ thm2b ] . we believed in the past that a solution to the coordination control problem ( in terms of a conditionally controllable sublanguage ) can only be computed as a product of languages in some special cases , where sufficient conditions such as those presented in @xcite hold . however , due to the presented relaxation , it becomes clear that such a solution can always be computed in a distributed way for prefix - closed specifications . namely , it suffices to make the resulting language @xmath189 controllable with respect to @xmath141 , as required in theorem [ thm2b ] , by computing a supervisor for it as the following result suggests . [ prop2 ] consider the setting of problem [ problem : relaxed ] and the languages defined in ( [ eqcnr ] ) . let @xmath190 denote the language @xmath191 . if the languages @xmath192 and @xmath155 are synchronously nonconflicting ( e.g. , prefix - closed ) for @xmath73 , then @xmath193 is a relaxed conditionally controllable sublanguage of @xmath9 . interestingly , we rediscover this way the role of a supervisor for the coordinator that is postponed to the end of the coordination control synthesis and it is used only when needed , i.e. , when controllability of @xmath189 with respect to @xmath141 does not hold . moreover , the supervisor @xmath153 can be computed in a distributed way , i.e. , we can compute @xmath194 . their composition ( here intersection ) is then never computed and they operate locally in conjunction with local supervisors @xmath192 , @xmath73 . compared to theorem [ thm4 ] , proposition [ prop2 ] only states that the result is a relaxed conditionally controllable sublanguage of the specification . we do not know whether the supremality also holds here . nevertheless , we should point out that the languages @xmath192 , @xmath73 , actually form a solution of problem [ problem1 ] as discuss in section [ sec : comparison ] below , even though their parallel composition is not necessarily relaxed conditionally controllable . from this point of view , to decrease the language by the composition with an additional coordinator , @xmath153 , is not relevant for us . this demonstrates the already mentioned drawback of the ( relaxed ) coordination control framework the restriction of the set of possible solutions , compared to the set of possible solutions for problem [ problem1 ] ; see more details in section [ sec : comparison ] below . to compare the solutions of problem [ problem1 ] and those of ( relaxed ) coordination control problem , we need the following . [ thm_inc_opt_sol ] consider the setting of problem [ problem : relaxed ] and the languages defined in ( [ supccn ] ) , ( [ eqcnr ] ) . then the language @xmath195 , i.e. , the supremal relaxed conditionally controllable sublanguage is included in our distributed solution . if @xmath185 and @xmath186 are synchronously nonconflicting ( e.g. , prefix - closed ) , then the parallel composition is controllable , i.e. , @xmath196 . we now discuss conditions that ensures optimality . the proofs are similar to those in @xcite . to this end , we use the notions of output control consistency ( occ ) , cf . @xcite , or local control consistency ( lcc ) , cf . @xcite . [ thm_opt_cond ] consider the setting of problem [ problem : relaxed ] and the languages defined in ( [ eqcnr ] ) . if the projection @xmath64 is an @xmath197-observer and occ ( lcc ) for @xmath197 , @xmath73 , then the parallel composition of the supervisors contains the optimal solution , i.e. , @xmath198 . note that we assume that @xmath64 is an @xmath199-observer and occ ( lcc ) for @xmath199 . however , we do not want to compute the language @xmath199 . instead , we can assume that ( i ) @xmath200 is an @xmath201-observer and ( ii ) @xmath64 is occ ( lcc ) for the language @xmath202 . these can be ensured by an appropriate extension of the alphabet @xmath15 , cf . @xcite and @xcite , respectively . here we highlight the difference between problem [ problem1 ] and the relaxed coordination control problem . on one hand , we show the drawback of relaxed coordination control that prevents it from fully solving problem [ problem1 ] . on the other hand , we appreciate it since it can be used to find a solution . we now point out the main difference between the two problems . let @xmath203 be a plant , and let @xmath204 denote a specification as in problem [ problem1 ] . consider the computation of supervisors @xmath185 and @xmath186 as defined in ( [ eqcnr ] ) . for simplicity , we assume that the languages under consideration are prefix - closed . assume that the parallel composition @xmath205 is not relaxed conditionally controllable . this is not an unrealistic assumption . then these two languages do not form a solution in the relaxed coordination control framework . however , note that ( a generator for ) @xmath192 is a supervisor for the plant @xmath160 , since @xmath192 is controllable with respect to @xmath160 by definition . let @xmath206 . then @xmath207 is conditionally decomposable and controllable with respect to the original plant , and @xmath208 which means that @xmath192 , @xmath73 , is a solution of problem [ problem1 ] that can not be achieved in the relaxed coordination control framework , since @xmath207 is not relaxed conditionally controllable by definition . ( here we slightly abuse the notation and use @xmath192 to denote both the generator and the language of that generator , depending on the context . ) the inclusion @xmath209 is required in the definition of the relaxed coordination control problem . but it does not hold in general , therefore we do not assume the inclusion to be satisfied in problem [ problem1 ] . this is the main difference between problem [ problem1 ] and the ( relaxed ) coordination control problem . the following example shows that there exist solutions that can not be achieved in the relaxed framework . let @xmath0 and @xmath1 be plants as shown in fig . [ figexl1plants ] , and let @xmath9 denote the specification with the generator depicted in fig . [ figexl1spec ] . and @xmath1,title="fig : " ] and @xmath1,title="fig : " ] let @xmath210 . notice that @xmath9 is conditionally decomposable . the coordinator @xmath29 is depicted in fig . [ figexl1coord ] . ] then @xmath211 and @xmath212 and it is not hard to see that the language @xmath213 is not controllable with respect to @xmath214 , hence @xmath9 is not relaxed conditionally controllable . if we use the relaxed coordination control framework to compute supervisors , we obtain @xmath148 and @xmath149 depicted in fig . [ figexl1ccsups ] . and @xmath149 computed in the relaxed coordination control framework , title="fig : " ] and @xmath149 computed in the relaxed coordination control framework , title="fig : " ] this is not the optimal solution , i.e. , their parallel composition is not equal to @xmath9 . on the other hand , we can obtain the specification if we consider infimal superlanguages of @xmath142 , namely for @xmath215 . then we obtain the supervisors depicted in fig . [ figexl1infopt ] that form the optimal solution . and @xmath28 for the problem obtained by considering infimal superlanguages , title="fig : " ] and @xmath28 for the problem obtained by considering infimal superlanguages , title="fig : " ] however , if the solution of problem [ problem1 ] does not exist , the relaxed coordination control framework allows us to compute a sublanguage of the specification for which a solution exists . the other advantage of this is that we can compare the obtained solution with the supremal relaxed conditionally controllable sublanguage of the specification ; see below . consider the setting of problem [ problem1 ] . theorem [ th : controlsynthesissafety ] says that if @xmath9 is relaxed conditionally controllable , then there exist nonblocking and nonconflicting supervisors @xmath24 and @xmath25 such that @xmath118 ) \parallel l_m(s_{2}/[g_2 \| g_k ] ) = k\ , . $ ] recall the languages @xmath185 and @xmath186 defined above in ( [ eqcnr ] ) . by definition , we immediately have that @xmath216 i.e. , it is potentially a solution of problem [ problem1 ] . we now show that it is actually a solution and compare it with the optimal solution of the relaxed framework . the following is basically theorem [ thm_inc_opt_sol ] . [ optimality ] consider the setting of problem [ problem1 ] and the languages defined in ( [ supccn ] ) , ( [ eqcnr ] ) . if the supervisors @xmath185 and @xmath186 are synchronously nonconflicting ( e.g. , prefix - closed ) , then their parallel composition is controllable . if the languages ( supervisors ) in the previous theorem are conflicting , we still have a solution for the global plant @xmath4 , namely , the language @xmath217 , where @xmath218 denotes the nonblocking generator of @xmath219 . then @xmath220 is a supervisor such that @xmath221 and @xmath222 , so it can be considered as a solution of the monolithic case with the supervisor @xmath220 . however , our aim is not to compute the parallel composition of supervisors @xmath27 and @xmath28 to obtain a single huge supervisor , but rather to distribute the supervision to local plants . taking a look at the previous theorem , we can notice that @xmath223 . however , if the supervisors @xmath27 and @xmath28 are conflicting , we only have that @xmath224 , i.e. , the overall supervised closed - loop system is blocking . to solve nonblockingness here , we can use the language @xmath225 where the projection @xmath226 is a @xmath154-observer , @xmath73 , which serves as a coordinator for nonconflictingness . the following can be proved similarly as in @xcite , based on the results of @xcite . [ thm22 ] consider the notation of problem [ problem1 ] and the languages defined in ( [ eqcnr ] ) and ( [ eq2 ] ) . then the language @xmath227 is controllable with respect to the plant @xmath228 . we can now summarize the method as an algorithm . [ alg ] consider the above . 1 . check whether problem [ problem1 ] has a solution using proposition [ lem1 ] . if so , stop ; otherwise , continue . 2 . compute @xmath185 and @xmath186 as defined in ( [ eqcnr ] ) . 3 . let @xmath229 and @xmath230 . 4 . extend the alphabet @xmath53 so that the projection @xmath226 is both a @xmath185- and a @xmath186-observer . 5 . define the coordinator @xmath231 as the minimal nonblocking generator such that @xmath232 from ( [ eq2 ] ) . see more comments on this algorithm in @xcite . to demonstrate our approach on an industrial example , we consider the model and specification of an mri scanner presented by theunissen @xcite . the plant consists of four parts @xmath233 where each part is again a composition of several smaller parts . however , we do not go into these details and consider these parts as the four subsystems that form the plant . the specification consists of the parts @xmath234 which do not exactly correspond the the four parts of the plant . for all computations , we used the c++ library ` libfaudes ` @xcite . we now observe the following . \(1 ) vreq is a specification that concerns only the plant vaxis . this is a simple case with one plant , hence a monolithic approach was used . the specification has 12 states and 44 transitions , the plant has 15 states and 50 transitions , and the computed supervisor has 15 states and 36 transitions . \(2 ) similarly , hreq is a specification that concerns only the plant haxis . again , the monolithic supervisor was computed . the specification has 112 states and 736 transitions , the plant has 128 states and 1002 transitions , and the computed supervisor consists of 80 states and 320 transitions . \(3 ) the specification hvreq concerns the plant @xmath235 . the specification consists of 7 states and 35 transitions , the plant hvnormal consists of 1 state and 1 transition ( the other two parts of the plant are mentioned above ) . we computed a coordinator consisting of 160 states and 1287 transitions and three supervisors with 516 , 1132 and 283 states and 3395 , 10298 and 1692 transitions , respectively . it was verified that their parallel composition is nonblocking and corresponds to the supremal controllable sublanguage of the specification hvreq with respect to the plant @xmath236 . \(4 ) the specification uireq concerns the whole plant @xmath237 . therefore it is the most interesting part to demonstrate our approach . let us mention that uireq consists of 256 states and 2336 transitions , and ui consists of 2 states and 15 transitions ( the rest is described above ) . we computed a coordinator with 4 states and 30 transitions , and four supervisors with 432 , 768 , 12 , and 96 states and 3488 , 6652 , 74 and 808 transitions , respectively . the overall minimal monolithic supervisor for the whole system would consist of 68672 states and 616000 transitions . we have verified that our solution is an optimal solution to problem [ problem1 ] . we have verified that it forms a solution in both coordination control architectures discussed in the paper . due to the modeling skills of the modeler and the nature of the model , there were no problems with blockingness at any step of the computation . all the computed supervisors are nonblocking and nonconflicting , thus no coordinator for nonblockingness was needed . the results are formulated for general , non - prefix - closed languages , for which synchronous nonconflictingness is required . one could find it an issue , but these assumptions are trivially satisfied for , e.g. , prefix - closed languages . to verify whether a synchronous product ( of an unspecified number ) of generators is synchronously nonconflicting is pspace - complete @xcite . it is only the worst case . some optimization techniques exist , e.g. @xcite , or a maximal nonconflicting sublanguage can be computed @xcite . the good news of pspace - completeness is that it is computable in polynomial space . all concepts and results can be extended from the generic case @xmath238 in a straightforward manner . however , with an increasing number of components it is likely that the coordinator will , in some situations , grow , e.g. , many events will have to be included into the coordinator alphabet to make the global specification conditionally decomposable . therefore we have recently proposed a multilevel coordination control architecture @xcite . the approach of this paper can easily be implemented in such a multilevel structure . a challenging and important problem is to find a convenient alphabet @xmath15 , from which the coordinator is computed . a natural step is to take a minimal such alphabet . there are two issues with this choice . first , to compute the minimal @xmath15 is np - hard , second , there exist examples showing evidence that the use of a minimal alphabet results in no solutions ( in empty supervisors ) . this means that we need to find a larger alphabet for which a solution can be obtained . this is always possible , since @xmath15 can be taken as the global alphabet . however , the aim is to characterize a reasonable choice of the alphabet that would provide nonempty solutions . in the future , we plan to take into account communication delays and losses and to consider more complicated forms of communications among local controllers . the research was supported by rvo 67985840 , by mmt in project music ( grant lh13012 ) , by gar in project ga15 - 02532s and by the dfg in project diamond ( emmy noether grant kr 4381/1 - 1 ) . lemma [ thm : char ] consider the setting of problem [ problem1 ] . there exist nonblocking and nonconflicting supervisors @xmath24 and @xmath25 such that the distributed closed - loop system satisfies @xmath118 ) \| l_m(s_{2}/[g_2 \| g_k ] ) = k $ ] if and only if @xmath239 , @xmath240 , and @xmath27 and @xmath28 are nonblocking and nonconflicting supervisors with respect to @xmath121 and @xmath122 , respectively . proposition [ lem1 ] consider the setting of problem [ problem1 ] . if the specification @xmath9 is prefix - closed , then there exists a solution of problem [ problem1 ] if and only if the languages @xmath127 and @xmath128 satisfy equations ( [ sol1 ] ) . ( if ) assume that @xmath251 and @xmath252 . since @xmath253 and @xmath254 are prefix - closed , hence nonblocking and nonconflicting , and controllable with respect to the corresponding plants , that is , they are supervisors for those plants , the implication follows from lemma [ thm : char ] . ( only if ) assume that there is a solution of problem [ problem1 ] . by lemma [ thm : char ] , there exist supervisors @xmath27 and @xmath28 satisfying equations ( [ sol1 ] ) . since @xmath6 is a supervisor , its language ( for simplicity also denoted by @xmath6 ) is controllable with respect to @xmath160 . moreover , @xmath6 contains @xmath142 , hence @xmath255 , for @xmath73 . then , @xmath256 thus , @xmath253 and @xmath254 satisfy equations ( [ sol1 ] ) . let @xmath27 and @xmath28 denote the solution for specification @xmath9 . if @xmath239 , then @xmath257 by nonconflictingness and nonblockingness of the supervisors . similarly for the other equation thus , @xmath258 and @xmath259 form a solution for @xmath21 . let @xmath260 and @xmath261 denote the generators of the solution for the prefix - closed language @xmath21 , that is , @xmath262 for @xmath73 , we now define the marking on @xmath260 and @xmath261 as follows : @xmath263 note that @xmath264 , since @xmath265 . we show that @xmath266 is nonblocking , that is , @xmath267 . to this end , let @xmath268 . if @xmath269 , then @xmath270 by definition . if @xmath271 , then there exists a word @xmath272 such that @xmath273 . hence @xmath274 . thus , @xmath275 . moreover , @xmath276 is controllable with respect to the plant @xmath164 , because , by assumption , @xmath277 is . thus , @xmath266 is a marking nonblocking supervisor for @xmath160 . furthermore , we have that @xmath278

in the last few years , great progress has been made towards the understanding of the out of equilibrium dynamics of _ classical _ systems with slow dynamics and aging effects . experiments in several glassy materials such as polymer glasses,@xcite spin - glasses,@xcite orientational glasses,@xcite simple liquids like glycerol@xcite and gels , @xcite have shown that all these systems share an extremely slow dynamics at low temperatures . although some aspects of the dynamic evolution as the precise scaling laws or the dependence on cooling rates vary from system to system , all the examples mentioned above are characterized by the existence of a non - equilibrium low temperature phase with _ aging effects_. this means that measurements show a strong dependence upon the time elapsed since its preparation by a thermal quench or annealing from a high temperature phase . needless to say , the standard equilibrium approach is not suited to describe such a phenomenology . in order to describe the spin - glass case , a number of theoretical methods have been exploited , ranging from scaling arguments based on coarsening ideas;@xcite phenomenological models founded on particular assumptions on the structure of phase space;@xcite analytical solutions to models with infinite - range interactions;@xcite and numerical simulations of finite and infinite dimensional models.@xcite all of these approaches have succeeded in capturing some of the effects observed experimentally . importantly enough , the solution of simplified models with infinite range interactions has been fruitful for several reasons : ( i ) it has shown that indeed a large family of such models do capture aging effects that resemble , at least qualitatively , the experimental observations . ( ii ) it has provided a general framework to analyze the non - equilibrium dynamics with new predictions , like scaling laws for the time - delayed correlation function and asymptotic violations of the fluctuation - dissipation theorem , that are not restricted to the infinite dimensional case . some of these have been confirmed by numerical studies of more realistic models@xcite and will be probed experimentally soon.@xcite ( iii ) it has allowed to extend the _ connection@xcite between mean - field theories of disordered systems and the mode - coupling theory of super - cooled liquids@xcite to the low temperature glassy phase.@xcite this is most welcome since many of the similarities observed in the behavior of real glasses of very different origin can now be understood theoretically . ( iv ) problems of a more general nature such as the dynamics of manifolds in a random quenched environment , relevant to the description of e.g. the pinning of the flux lattice by disorder in dirty high @xmath1 superconductors,@xcite can be studied in a similar way.@xcite the dependence on the internal dimension of the manifold can be included in the formalism and the analysis yields , among other results , dynamic scaling laws for the displacements with aging effects . the studies we have mentioned so far are mostly concerned with the analysis of _ classical _ systems . this treatment is in general justified since , in many cases , the critical temperature at which the transition to the glassy phase " occurs is sufficiently high as to make quantum effects irrelevant . nevertheless , in many cases of practical interest , the critical temperature can be lowered by tuning an external parameter . experimental realizations of this situation are given by insulating magnetic materials@xcite such as liho@xmath2y@xmath3f@xmath4 and randomly mixed hydrogen - bonded ferro - antiferroelectric crystals of the rb@xmath3 ( nh@xmath4)@xmath5h@xmath6po@xmath4 type ( known as radp).@xcite theoretical studies addressing the effects of quantum fluctuations in disordered media have been considered over the last twenty years . a variety of techniques including replica theory,@xcite renormalization group@xcite and monte carlo simulations@xcite have been employed in order to understand the _ static _ properties and critical behavior of quantum systems such as quantum rotors@xcite , transverse field ising models in finite dimensions@xcite and mean - field systems like the sherrington - kirkpatrick model in a transverse field.@xcite from the study of _ classical _ glassy systems described in the first paragraph , one knows that at low temperatures their dynamics becomes so slow that the systems are not able to reach thermal equilibrium at any experimentally accessible time . the same occurs regarding numerical simulations of large systems . it is then plausible that even at very low temperatures , say @xmath7 , with only quantum fluctuations driving the dynamics , the materials and/or models will need an unaffordable long time to reach equilibrium and that the relevant evolution will be characterized by non - equilibrium effects . indeed , it is well - known that certain glasses in the limit of very low temperatures have aging effects@xcite and hence the approaches based on the assumption of equilibration have a restricted domain of application . instead , one should necessarily start from a quantum _ dynamic _ description in order to obtain sensible information about the systems . to the best of our knowledge this problem has not been theoretically tackled , from a microscopic point of view , yet . the question then arises as to how do quantum fluctuations modify the _ real time _ dynamics of glassy systems . in this paper , we treat the case of a disordered model in contact with a thermal quantum bath . our first aim is to present a general framework that allows us to study the real time dynamics of a system with quantum _ and _ thermal fluctuations starting from an arbitrary initial condition . in the case of classical models this is most easily done within the langevin approach . one generally starts by postulating that the dynamics of the microscopic variables is given by a stochastic langevin equation . the noise and friction terms account for the bath - system interaction . a path - integral formulation of the generating functional , known as the martin - siggia - rose formalism@xcite ( msr ) , yields an elegant and simple formulation which is particularly well suited for problems with disorder . however , the generalization of this procedure to the quantum case is far from being straightforward . the langevin equation does not seem to be a good starting point since , being phenomenological , it can not be quantized in an obvious way.@xcite in other words , it is not clear how to make the langevin approach compatible with the intrinsic microscopic dynamics of a quantum system . the approach we use here results from the combination of different techniques which have been intensively discussed in the literature and successfully applied to a wide range of phenomena . essentially , it consists of the schwinger and keldysh@xcite closed - time path - integral formalism applied to a system in contact with a thermal bath , combined with the integration of the bath variables leading to the feynman and vernon influence functional.@xcite a subsequent integration over the disorder gives the quantum analogue of the msr generating functional for disordered systems . though the formalism allows us to consider general cooling procedures or / and various temperature variations , we shall not study involved thermal histories here but concentrate instead on a quench from a configuration associated to the disordered phase . for the sake of definiteness we focus on a system whose classical counterpart , the so - called `` @xmath0-spin spherical spin - glass model '' , has been studied in great detail . this model is described by a potential energy @xmath8 = - \sum_{i_1 < \dots < i_p}^n j_{i_i\dots i_p } \ , \sigma_{i_1 } \dots \sigma_{i_p}\ ] ] with @xmath9 quenched independent gaussian variables and @xmath10 a parameter . let us attempt a short review of the behavior of the _ classical _ @xmath0-spin models . in its ising version it was first introduced by derrida in relation with the random energy model.@xcite several studies of its static properties , in particular using replicas , followed . @xcite kirkpatrick and thirumalai@xcite studied the equilibrium dynamics , la sompolinsky,@xcite and remarked that the dynamic equations , in the high temperature phase , are those appearing as @xmath11 models in the mode - coupling theory of super - cooled liquids developed to describe the glass transition.@xcite the spherical version , where a time - dependent lagrange multiplier is introduced to enforce the spherical constraint @xmath12 , was first studied by jones , kosterlitz and thouless@xcite for @xmath13 . its statics is fully solvable either by using replicas a replica symmetric ansatz is exact or through a direct calculation . its dynamics@xcite is also fully solvable and it renders explicit the fact that the model is formally related@xcite to the @xmath14 model in the large @xmath15 limit and dimension @xmath16 . in the classical limit , it is then a toy model for ferromagnetic domain - growth . in the quantum case one feels tempted to relate it to coarsening in quantum systems . the analysis of the spherical model for @xmath17 is due to crisanti and sommers,@xcite who also studied the equilibrium dynamics la sompolinsky in collaboration with horner.@xcite in the classical limit and at the static level , this model is the prototype spin - glass solved by a parisi one - step replica symmetric ansatz . still at the classical level and in the large @xmath15 limit , one derives exact evolution equations that exhibit a first order discontinuous dynamic transition at @xmath18 . above @xmath18 , as in the ising case , one recovers the mode - coupling equation for model @xmath11.@xcite below @xmath18 it is well - known that the dynamic approach used in ref.[@xcite ] is incorrect.@xcite the dynamics in the low - temperature phase was solved in ref . [ ] showing that the dynamics of mean - field spin - glasses strikingly resembles that of real systems . in the spin - glass phase the evolution is intrinsically out of equilibrium and the systems age in similar way to the one observed in experiments . in order to define the model at the quantum level , we need to specify the dynamics of the variables @xmath19 . for simplicity , we consider a kinetic term of the form @xmath20 where the dot denotes a time - derivative . calling @xmath21 the lagrange multiplier enforcing the spherical constraint , the model we here consider is then described by the action @xmath22 = \int dt \ , \left [ \frac{m}{2 } { \dot { \bbox \sigma}}^2 - \frac{z}{2 } \left ( { \bbox \sigma}^2 -n \right ) + \sum_{i_1 < \dots < i_p}^n j_{i_i\dots i_p } \ , \sigma_{i_1 } \dots \sigma_{i_p } \right ] \ ; .\ ] ] our main goals are to detect a dynamic transition from a disordered phase to an ordered glassy phase and to characterize the dynamics in these phases . with this purpose , we pay particular attention to the departure of correlation and response functions from stationarity and , we search for one of the hallmarks of a glassy phase , namely , violations of the fluctuation - dissipation theorem . we also intend to discuss our results in view of previous and possible experimental tests . the paper is organized as follows . in section [ ctp ] we discuss the closed - time path - integral formalism of keldysh and schwinger and the coupling of the quantum system to a quantum heat bath . we obtain a generating functional with an effective action that takes into account the effect of the coupling to the bath . in three appendices we give details of the keldysh schwinger formalism ( appendix [ appkeldysh ] ) , we recall the quantum fdt ( appendix [ appfdt ] ) and we discuss the classical limit ( appendix [ classicallimit ] ) . section [ meanfieldequations ] is devoted to the derivation of the dynamic equations through a saddle - point approximation of the generating functional . the method is similar to the one used in the classical case when manipulating the martin - siggia - rose generating functional . the general scenario for the real - time dynamics of this system is discussed in section [ scenario ] . it is inspired by the behavior of the classical counterpart but some generalizations are necessary . in section [ equilibriumdynamics ] the equations and their solution in the paramagnetic phase are studied with special emphasis set on the analysis of the out of phase susceptibility close to the transition line . in section [ nonequilibrium ] the dynamic equations in the glassy phase are studied . the non - equilibrium dynamics is shown to survive quantum fluctuations in a finite region of the phase diagram . further details of the approach to solve the equations are given in appendix [ integrals ] . the modification of fdt needed to construct an ansatz to solve these equations and its relation to effective temperatures are discussed in section [ generalfdt ] . finally , in section [ conclusions ] we present a summary of our results , our conclusions and some ideas as to how to continue investigating along the lines here proposed . a short account of some of these results has appeared in ref . the closed - time path - integral ( ctp ) formalism of keldysh and schwinger@xcite for treating the non - equilibrium dynamics of a quantum system has been reviewed several times in the literature . @xcite in order to set the notation we briefly present it in section [ generalframework ] . we then consider the situation in which the system under study is coupled to a quantum thermal bath and derive the effective action following feynman and vernon @xcite ( section [ couplingtoaheatbath ] ) . in section [ ctpfordisorderedsystems ] we finally address the case of a disordered system and we obtain the effective action from which the dynamic equations are derived in section [ meanfieldequations ] . the basic ingredient of this formulation is the `` in - in '' generating functional from which , after derivation with respect to the external sources @xmath23 , @xmath24 , the different green functions can be obtained . let us consider a model with degrees of freedom described by a field @xmath25 in the heisenberg picture . denoting @xmath26 the density matrix at the initial time @xmath27 ( which from now on we set to zero ) , the generating functional is defined as @xmath28 & = & \mbox{tr } \left[t^ * \exp\left(- \frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^-(t ) { \bbox \phi}(t)\right ) t \exp\left ( \frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^+(t ) { \bbox \phi}(t ) \right ) \hat\rho(0 ) \right ] \ ; \label{generating}\end{aligned}\ ] ] where @xmath29 and @xmath30 are the @xmath15-vector external sources and a scalar product between @xmath15 vectors is assumed . the symbols @xmath31 and @xmath32 are the time and anti - time ordering operators : @xmath33 defining ensemble averages in the usual way @xmath34 one obtains @xmath35 which is directly related to the feynman causal propagator . in the same way , @xmath36 { \ln } { \cal z } \right|_{{\bbox \xi}=0 } \ ; \end{aligned}\ ] ] where @xmath37 is the _ symmetrized correlation function _ ( @xmath38 ) . the response function @xmath39 is defined as the variation of the averaged field @xmath40 with respect to the `` strength '' @xmath41 of a perturbation that modifies the potential energy according to @xmath42 : @xmath43 in linear - response theory , it can be expressed in terms of the averaged commutator : @xmath44 \rangle \\ & = & \left . \frac{\hbar}{i } \left [ \frac{\delta^2}{\delta \xi_i^+(t ) \delta \xi_j^+(t ' ) } + \frac{\delta^2}{\delta \xi_j^+(t ) \delta \xi_i^-(t ' ) } \right ] { \ln } { \cal z } \right|_{{\bbox \xi}=0 } \ ; .\end{aligned}\ ] ] one of the main advantages of this formalism is that the generating functional ( [ generating ] ) admits a path - integral representation which is a slight modification of the usual `` in - out '' feynman path integral . this will allow us to make formal manipulations similar to the ones performed in the study of the dynamics of classical systems via the martin - siggia - rose@xcite ( msr ) formalism . in a short - hand notation , the generating functional reads @xmath28 & = & \int { \cal d}{\bbox \phi}^+ { \cal d}{\bbox \phi}^- \exp\left [ \frac{i}{\hbar } \ , \left ( s[{\bbox \phi^+ } ] -s [ { \bbox \phi^- } ] + \int dt \ , { \bbox \xi}^+(t ) { \bbox \phi}^+(t ) - \int dt \ , { \bbox \xi}^-(t ) { \bbox \phi}^-(t ) \right ) \right ] \nonumber\\ & & \;\;\;\;\;\;\;\;\;\ ; \times \langle { \bbox \phi}^+ | \hat \rho(0 ) | { \bbox \phi}^- \rangle \label{aaab}\end{aligned}\ ] ] where @xmath45 $ ] is the action of the system and @xmath46 stands for the matrix element of the density matrix which has support only at @xmath47 ( see appendix [ appkeldysh ] for a proof of this equation ) . hereafter , we omit the limits of the time - integrals that , unless otherwise stated , go from @xmath48 to @xmath49 . in the following it will be useful to write the correlation and response in terms of the fields @xmath50 , @xmath51 : @xmath52 notice that as a result of having two external sources ( one related to the evolution with time ordering and the other with anti - time ordering ) a doubling of degree of freedom is necessary when writing the path - integral . as we explicitly show in appendix [ classicallimit ] , the doubling of degrees of freedom characteristic of this description is intimately linked to the introduction of lagrange multipliers in the msr description of the classical model . the time - integration in eq . ( [ aaab ] ) can be interpreted as being closed , going forward from @xmath53 to @xmath54 and then backwards from @xmath54 to @xmath53 . this motivates the name of the method . up to now , the discussion has been completely general and it applies to systems with an arbitrary number of degrees of freedom . we now consider the coupling of the system of interest to a thermal quantum bath , assumed to be in equilibrium . a convenient way to deal with the interaction between the system and the bath in the path - integral formalism is due to feynman and vernon . @xcite in this approach one starts by describing the system plus bath by @xmath55 with @xmath56 denoting the variables of the system and @xmath57 , @xmath58 , denoting the variables of the bath . the total action is @xmath59= s_s[{\bbox \sigma } ] + s_b[{\bbox v}^a ] + s_{sb}[{\bbox{\sigma , v}^a}]\ ] ] where @xmath60 is the action characterizing the system ( and eventually depending upon disorder ) , @xmath61 is the action for the bath , and @xmath62 contains the system - bath interaction terms . for definiteness we assume that our system interacts linearly with the bath which we model as a set of independent harmonic oscillators , @xmath63 = s _ { s}[{\bbox \sigma } ] + \frac12 \sum_{a=1}^{n_b } \ , m_a \ , \left ( \dot { \bbox v}_a^2 - \omega_a^2 \ , { \bbox v}_a^2 \right ) - \sum_{a=1}^{n_b } c_a { \bbox v}_a { \bbox \sigma } \ ; .\ ] ] @xmath64 are their masses , @xmath65 their frequencies and @xmath66 the coupling constants between the oscillators and the system . of course , more general couplings between bath and system are possible but this is the simplest choice that allows us to go a bit farther analytically . if we further assume that the initial density matrix @xmath67 is factorizable as @xmath68 where @xmath69 is the boltzman distribution for the bath variables at equilibrium at a temperature @xmath70 , one can integrate out the bath variables , obtaining in this way an effective thermal action @xmath71 $ ] , that enters the _ feynman - vernon influence functional _ : @xmath72 & = & - \int dt \int dt ' \ ; \left ( { \bbox \sigma}^+(t ) - { \bbox \sigma}^-(t ) \right ) \ , \eta(t - t ' ) \ , \left ( { \bbox \sigma}^+(t ' ) + { \bbox \sigma}^-(t ' ) \right ) \nonumber\\ & & + i\int dt \int dt ' \ ; \left ( { \bbox \sigma}^+(t ) - { \bbox \sigma}^-(t ) \right ) \ , \nu(t - t ' ) \ , \left ( { \bbox \sigma}^+(t ' ) - { \bbox \sigma}^-(t ' ) \right ) \ ; . \label{sthermal}\end{aligned}\ ] ] the noise and dissipative kernels @xmath73 and @xmath74 are given by @xcite @xmath75 in these equations , @xmath76 is the spectral density of the bath : @xmath77 the system - bath interaction manifests via the appearance of two quadratic and non - local terms ( [ sthermal ] ) in the effective action . as it can be most easily seen in the classical limit presented in appendix [ classicallimit ] they can be related to the noise and dissipative terms of the associated langevin formulation of the classical dynamics . in addition , it is easy to see that the kernels @xmath73 and @xmath74 are related to the correlation and response functions of the bath . in fact @xmath78 is the response of the bath while @xmath79 is the correlation of the bath . they are related in a way dictated by the fluctuation - dissipation theorem ( fdt ) : @xmath80 ( see appendix [ appfdt ] for the derivation and properties of the quantum fdt . ) this equation is the quantum counterpart of the fluctuation - dissipation relation of classical relaxational dynamics , that fixes the dissipative coefficient in terms of the strength of the noise correlation of the langevin equation . it expresses the fact that the bath is assumed to be in equilibrium at the initial time @xmath48 and that it stays in equilibrium at all subsequent times @xmath81 . it is important to notice that the assumption of thermal equilibrium of the bath _ does not imply _ that the system that is in contact with it is in equilibrium at the initial time nor that it will reach equilibrium at a given equilibration time @xmath82 . it is known that `` mean - field '' classical disordered models with relaxational dynamics the @xmath0 spin - glass model is an example do not reach equilibrium at any time if the limit of @xmath15 going to infinity is taken at the outset of the calculation.@xcite it is one of the main aims of this paper to show that this result carries through to the quantum case . as a consequence of the out of equilibrium behavior , we shall find that the fdt _ does not hold _ between response and correlation of the system in a sub - critical region of the phase diagram . the integrals in the @xmath74 and @xmath73 definitions are , in general , ill - defined ( divergent ) and might need to be regularized . a way to regularize them@xcite is to observe that one expects @xmath76 to decrease to zero for large @xmath83 and introduce an explicit cut - off : @xmath84 by modifying @xmath76 in this way , we regularize the kernels @xmath74 and @xmath73 at the same time preserving the fdt relation for the bath variables ; in other words , we do not break the assumption of equilibrium of the bath . different environments are characterized by different behaviors of @xmath76 such that for @xmath85 : @xmath86 a typical example is @xcite @xmath87 where @xmath88 is a constant that plays the rle of a friction coefficient . throughout this paper , we concentrate in the ohmic case . the kernel @xmath74 is @xmath89 in the limit where the cut - off tends to infinity it becomes @xmath90 of particular importance is the zero temperature limit of the kernel @xmath73 : @xmath91 one sees that the quantum fluctuations yield a non - trivial kernel even in the absence of thermal fluctuations . in the zero temperature limit the characteristic time goes to infinity and the long - time decay of @xmath73 becomes power - law : @xmath92 in the classical limit , for any temperature , @xmath73 becomes @xmath93 note that for a finite cut - off @xmath94 , the classical limit of the kernel @xmath73 also has a rather slow , power - law decay at large time differences . if , next , the cut - off is sent to infinity one recovers a delta function @xmath95 characteristic of e.g. a white thermal noise . the classical limit of these terms is further discussed in appendix [ classicallimit ] where we show that the dynamic equations of the corresponding classical model correspond to a friction coefficient @xmath96 . the coupling to the quantum thermal bath introduces three time - scales into the problem . first , the inverse cut - off @xmath97 characterizes the memory of the bath . second , the coupling strength @xmath98 yields the inverse relaxational characteristic time . third , @xmath99 accounts for the relative importance of quantum to thermal effects . the ctp formalism results to be well suited for problems with disorder.@xcite let us suppose that the system under consideration is described by the action , @xmath100 = s_{\sc o}[\bbox{\sigma } ] - \int dt \ , v[\bbox{\sigma } , j]\ ] ] where @xmath101 is the disordered independent part of the action and @xmath102 $ ] is the potential energy that depends upon random couplings collectively denoted by @xmath103 . as usual in disordered systems , we concentrate on quantities that are averaged over the disorder distribution @xmath104 $ ] . we wish to consider the real - time dynamics of the system starting from a _ random initial condition _ at time @xmath48 when it is set in contact with the environment ( that is itself in equilibrium at a constant temperature @xmath31 ) . given that the initial condition is chosen to be random , it is not correlated with the disorder . therefore , the density operator @xmath67 does not depend upon disorder and the generating functional without sources @xmath105 = \mbox{tr } \ , [ \ , \hat\rho(0 ) \ , ] \ ] ] _ is also independent of disorder_. this property is equivalent to the independence of the classical martin - siggia - rose ( msr ) generating functional @xcite without sources upon disorder.@xcite as in the classical case , it allows us to write dynamic equations for random initial conditions without having to compute the average over disorder of @xmath106 $ ] and hence without resorting to the use of replicas . we are then interested in the averaged generating functional : @xmath107 } = \int d j p[j ] \ ; { \cal z } [ \bbox{\xi}^+ , \bbox{\xi}^- , j ] \ ; , \ ] ] from which any averaged operator can be computed as @xmath108 } { \delta { \bbox \xi}^+(t ) } } \right| _ { { \bbox \xi } = 0 } = \left . \frac{1}{{\cal z}[0,0,j ] } \ ; \frac{\delta \overline{{\cal z}[{\bbox \xi}^+,{\bbox \xi}^-,j ] } } { \delta { \bbox \xi}^+(t ) } \right| _ { { \bbox \xi } = 0 } \ ; , \end{aligned}\ ] ] with all sources set to zero and @xmath109 . here and in what follows , the overline represents an average over the disorder . in many cases of interest , the integration over the disorder can be performed explicitly , @xmath110 \ ; \exp \left[-\frac{i}{h } \int dt \left ( \ , v[{\bbox \sigma}^+ , j ] - v[\bbox{\sigma } ^- , j ] \ , \right ) \right]= \exp \left[- \frac{i}{h } v_{\sc d } [ \bbox { \sigma}^+ , \bbox{\sigma}^- ] \right ] \;\ ] ] and it introduces a non - linear interaction term that is , in general , non - local both in time and in the `` space '' indices @xmath111 . a rather general way of introducing randomness is to consider a model with a gaussian random potential energy term @xmath112 $ ] depending upon the variables @xmath56 and correlated as @xmath113 v[{\bbox \sigma } ' , j ] } = - \frac{n}{2 } { \cal v } \left ( \frac { ( { \bbox \sigma } - { \bbox \sigma}')^2}{n } \right ) \label{randompotential } \ ; .\ ] ] the variables @xmath19 may represent spins leading to a spin - glass model or a position in an @xmath15-dimensional space becoming the problem of the dynamics and/or diffusion of a particle in a random potential . since we are interested in following the real time dynamics of such systems , @xmath19 will be a time - dependent @xmath15-vector . for the sake of definiteness , we shall solve a particular realization of the random potential that in the classical case defines the so - called @xmath0 spin - glass : @xcite @xmath114 = - \sum_{i_1 < \dots < i_p}^n j_{i_i\dots i_p } \ , \sigma_{i_1 } \dots \sigma_{i_p } \;.\ ] ] @xmath0 is a parameter , @xmath115 . the interaction strengths are quenched independent random variables with a gaussian distribution @xmath116 = \sqrt { \frac{n^{p-1}}{{\tilde j}^2\pi p ! } } \ ; \exp \left ( - \frac{n^{p-1}}{{\tilde j}^2 p ! } \sum_{i_1 \neq \dots \neq i_p } ( j_{i_i\dots i_p})^2 \right ) \;\;\;\;\;\;\;\ ; \rightarrow \;\;\;\;\;\;\;\ ; \overline { ( j_{i_i\dots i_p})^2 } = \frac{{\tilde j}^2 p!}{2 n^{p-1 } } \ ; .\ ] ] this corresponds to a random potential correlation @xmath117 and it leads to an effective , non - local in time , interaction @xmath118 & = & \frac{{\tilde j}^2 ni}{4 \hbar } \int dt \int dt ' \left [ \ ; \left ( \frac1n { \bbox \sigma}^+(t ) { \bbox \sigma}^+(t ' ) \right)^p + \left ( \frac1n { \bbox \sigma}^-(t ) { \bbox \sigma}^-(t ' ) \right)^p \right . \nonumber\\ & & \left . \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; - \left ( \frac1n { \bbox \sigma}^+(t ) { \bbox \sigma}^-(t ' ) \right)^p - \left ( \frac1n { \bbox \sigma}^-(t ) { \bbox \sigma}^+(t ' ) \right)^p \;\ ; \right ] \ ; . \label{actionv}\end{aligned}\ ] ] in its spherical version,@xcite @xmath119 are continuous dynamic variables @xmath120 , @xmath121 , that satisfy the global spherical constraint @xmath122 at each instant . we enforce this constraint by introducing a time - dependent lagrange multiplier @xmath123 . the disordered - independent part of the action is @xmath124 = \int dt \ , \left [ \frac{m}{2 } { \dot { \bbox \sigma}}^2 - \frac{z}{2 } \left ( { \bbox \sigma}^2 -n \right ) \right ] \label{actiono } \ ; .\ ] ] as explained in the introduction , for simplicity we have chosen a kinetic term that dictates a microscopic dynamics with second derivatives in time . other choices , for example leading to first time - derivatives , are also possible . the derivation of the dynamic equations that we explain in section [ meanfieldequations ] applies with only minor modifications . we now have all the elements to derive the equations which describe the dynamic evolution of a disordered quantum system . according to the discussion in section [ ctp ] , the system is described by the generating functional @xmath125 } = \int { \cal d}{\bbox \sigma}^+ { \cal d}{\bbox \sigma}^- \exp\left [ \frac{i}{\hbar } \ , \left ( s_{\sc eff}[{\bbox \sigma^+ } , { \bbox \sigma^- } ] + \int dt \ , { \bbox \xi}^+(t ) { \bbox \sigma^+}(t ) - \int dt \ , { \bbox \xi}^-(t ) { \bbox \sigma^-(t ) } \right ) \right]\ ] ] with @xmath126 = s_{\sc o}[{\bbox \sigma}^+ ] - s_{\sc o}[{\bbox \sigma}^- ] + s_{\sc t}[{\bbox \sigma}^+ , { \bbox \sigma}^- ] - v_{\sc d}[{\bbox \sigma}^+ , { \bbox \sigma}^- ] \ ; . \label{action1}\ ] ] we remind that @xmath101 is the disorder independent part of the action given by eq . ( [ actiono ] ) , @xmath127 accounts for the the system - bath interaction , eq . ( [ sthermal ] ) , and @xmath128 is the effective potential which arises as a result of the integration over the disorder and is given by eq . ( [ actionv ] ) . the quantum dynamic equations of motion follow from similar steps to those usually used @xcite to obtain the classical equations of motion in the martin - siggia - rose formalism . we introduce macroscopic order parameters and derive , through a saddle - point point approximation of the ks generating functional ( that becomes exact when @xmath129 ) , the dynamic equations of motion . since we are interested in considering the dynamics of the system for _ large but finite _ times with respect to @xmath15 , out of equilibrium effects are expected . we shall therefore make no assumption about the time - dependence of the order - parameters . the quadratic terms in the action can be condensed into one term by introducing the operator @xmath130 in such a way that @xmath126 = -\frac12 \int dt \int dt ' { \bbox \sigma}^\alpha(t ) \ , \mbox{op}^{\alpha\beta}(t , t ' ) { \bbox \sigma}^\beta(t ' ) - v_{\sc d}[{\bbox \sigma}^+ , { \bbox \sigma}^-]\ ] ] where greek indices label @xmath131 and the sum convention is assumed . introducing the identity @xmath132 into the generating functional , the full action can be rewritten as @xmath133 & = & - \frac{1}{2}\int dt \int dt ' \ ; { \bbox \sigma^\alpha}(t ) \ , \left ( \ , \mbox{op}^{\alpha\beta}(t , t ' ) + \lambda^{\alpha\beta}(t , t ' ) \right ) { \bbox \sigma}^\beta(t ' ) \nonumber\\ & & + \frac{n}{2 } \int dt \int dt ' \lambda^{\alpha\beta}(t , t ' ) \ , q^{\alpha\beta}(t , t ' ) + \frac{n}{2 } \int dt \left ( z^+(t ) - z^-(t ) \right ) \nonumber\\ & & + \frac{i { \tilde j}^2 n}{4 \hbar } \int dt \int dt ' \left [ \ , \left(q^{++}(t , t ' ) \right)^p + \left ( q^{--}(t , t ' ) \right)^p \right . \nonumber\\ & & \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; - \left . \left ( q^{+-}(t , t ' ) \right)^p - \left ( q^{-+}(t , t ' ) \right)^p \ , \right ] \ ; .\end{aligned}\ ] ] the stationary - point values of the order parameters @xmath134 with @xmath135 are related to the `` physical '' correlations and responses defined in eqs . ( [ defcorr ] ) and ( [ defresp ] ) as follows @xmath136 with @xmath137 it is easy to check that these functions satisfy the identity for all pairs of times @xmath138 . in the classical case this is a requirement to be imposed upon the saddle - point two - point functions in order to preserve causality . in the quantum case , this condition emerges as a property of the green functions . ] @xmath139 the functional integration over @xmath140 and @xmath141 is now quadratic and can be performed . this amounts to replace the quadratic term in @xmath142 by @xmath143 at this stage , all terms in the action depend upon the `` macroscopic '' quantities @xmath144 and @xmath145 and are proportional to @xmath15 . since it is easier to write the equations in matrix notation , we encode @xmath146 and @xmath147 in two matrices @xmath148 and we define @xmath149(t , t ' ) & \equiv & \left ( \begin{array}{rl } ( q^{++}(t , t'))^{p-1 } \ ; & \ ; -(q^{+-}(t , t'))^{p-1 } \\ -(q^{-+}(t , t'))^{p-1 } \ ; & \ ; ( q^{--}(t , t'))^{p-1 } \end{array } \right ) \label{saddlef } \ ; .\end{aligned}\ ] ] we denote with a cross the standard operational product @xmath150 where a sum over @xmath151 is assumed . the saddle - point with respect to @xmath152 yields @xmath153 the matrix and time - operator inverse of @xmath154 is denoted @xmath155 . the saddle - point equation with respect to @xmath134 yields @xmath156(t , t ' ) \ ; . \label{saddleq}\ ] ] equations ( [ saddlelambda ] ) and ( [ saddleq ] ) imply , in a compact matrix and time - operator notation , @xmath157 \otimes { \cal q } \ ; , \label{saddleq}\ ] ] where @xmath158 is the identity : @xmath159 . the saddle - point with respect to @xmath145 yields @xmath160 and these equations lead , as expected , to the spherical constraint . in the limit @xmath129 one could also proceed as in refs . [ ] and write the full action @xmath161 in terms of a _ single _ variable . this is at the expense of modifying the thermal kernel and the interaction term in a self - consistent way , through the introduction of terms arising from the non - linear interactions ( the vertex @xmath162 and the self - energy @xmath163 , respectively ) . this procedure is not of particular usefulness for the analysis of the model we treat in this paper since the single variable effective action is gaussian . it does however become useful for dealing with quantum models whose single - spin effective action has higher order interaction terms . an example is the quantum sherrington - kirkpatrick model . one could then envisage to derive an effective quantum langevin equation for the single variable@xcite and study this equation with an adequate numerical algorithm as the one developed by eissfeller and opper.@xcite the dynamic equations for the auto - correlation and response follow from the set of equations ( [ saddlef])-([saddleq ] ) and the definitions of the dynamic order parameters given in eqs . ( [ q++])-([q ] ) . more precisely , the equation of motion for the response function follows from the subtraction of the @xmath164 and @xmath165 components of eq . ( [ saddleq ] ) : @xmath166 \ , r(t'',t ' ) \ ; , \label{eqr1}\end{aligned}\ ] ] and the equation of motion for the auto - correlation function follows from the addition of the @xmath165 and @xmath167 components of eq . ( [ saddleq ] ) : @xmath168 \label{eqc1 } \ ; .\end{aligned}\ ] ] written in this way , eq . ( [ eqc1 ] ) is complex . its imaginary part yields @xmath169 moreover , since the response is causal , products of advanced @xmath170 and retarded @xmath171 responses vanish identically for all @xmath172 : @xmath173 and one can show that for any integer @xmath174 and any constants @xmath175 , @xmath176 @xmath177 ^k= \left [ c(t , t ' ) + c_1 r(t , t ' ) \right ] ^k + \left [ c(t , t ' ) + c_2 r(t',t ) \right ] ^k - c^k(t , t ' ) \ ; .\ ] ] using this property one has @xmath178^{p-1}\ ] ] and @xmath179 = \nonumber\\ & & 2 c(t'',t ' ) \mbox{im } \left [ c(t , t'')-\frac{i \hbar}{2 } r(t , t '' ) \right]^{p-1 } -\hbar r(t',t '' ) \mbox{re}\left [ c(t , t'')-\frac{i\hbar}{2 } ( r(t , t'')+ r(t'',t ) ) \right]^{p-1 } \ ; . \nonumber\\\end{aligned}\ ] ] we can identify the self - energy @xmath163 and the vertex @xmath162 as @xmath180^{p-1 } \ ; , \label{sigmatilde } \\ d(t , t')- 2 \hbar \nu(t - t ' ) & \equiv & \tilde{d}(t , t ' ) \equiv \frac{p \tilde{j}^2}{2 } \mbox{re}\left [ c(t , t')-\frac{i\hbar}{2 } ( r(t , t')+ r(t',t ) ) \right]^{p-1 } \ ; . \label{dtilde}\end{aligned}\ ] ] for @xmath181 , they can be encoded in a single complex equation : @xmath182 note that the total self - energy @xmath183 and vertex @xmath184 are real and have two contributions of different origin : one arises from the interaction of the system and the bath ( @xmath74 and @xmath73 ) and one is caused by the non - linearities stemming from the average over disorder ( that we called @xmath163 and @xmath162 in eqs . ( [ sigmatilde ] ) and ( [ dtilde ] ) ) . if @xmath13 , @xmath163 and @xmath162 are identical to the classical ones . instead , if @xmath185 , the non - linear terms acquire an explicit dependence upon @xmath186 . the dynamic equations can then be written in a compact form @xmath187 that we call later the @xmath188-eq . and @xmath189-eq . , respectively . it is important to realize that the self - energy @xmath190 is proportional to the response function @xmath191 , which in turns implies @xmath192 this means that the upper limit of integration in eqs . ( [ schwingerr ] ) and ( [ schwingerc ] ) is @xmath193 , which renders the equations explicitly causal . there are no more independent equations for @xmath188 and @xmath189 . the other two equations that can be obtained from eq . ( [ saddleq ] ) are the equation for @xmath194 , that is equivalent to the @xmath188-eq above , and one equation that identically cancels by virtue of the identity ( [ identityqo ] ) . in their integrated form as schwinger - dyson equations the dynamic equations read : @xmath195 with the propagator given by @xmath196 real and imaginary parts of eqs . ( [ z+ ] ) and ( [ z- ] ) combined with eq . ( [ saddlelambda ] ) imply the equal - times conditions @xmath197 in addition , from eq . ( [ eqr1 ] ) one obtains that the first derivative of the response function is discontinuous at equal times : @xmath198 while from eq . ( [ eqc1 ] ) one obtains that the correlation is continuous : @xmath199 the equation for @xmath200 can be derived from the schwinger - dyson equation ( [ intschwingerc ] ) , by imposing the spherical constraint through the evaluation at @xmath201 . multiplying operationally by @xmath202 one obtains @xmath203 \nonumber\\ & & + m \int_0^t dt '' \int_0^t dt ' '' \ , ( \partial_t r(t , t '' ) ) \,d(t'',t''')\ , ( \partial_t r(t , t ' '' ) ) \ ; .\end{aligned}\ ] ] the last term is a consequence of having a kinetic term with second derivatives . had we chosen a first derivative term in the kinetic energy the last term would have not appeared . it can be easily identified with minus the second - derivative of the correlation at equal times by taking the limit @xmath204 in eq . ( [ schwingerc ] ) . thus @xmath205 - m \left . \frac{\partial^2 } { \partial t^2 } c(t , t ' ) \right|_{t'\to t^- } \ ; . \label{zeq}\ ] ] in conclusion , eqs . ( [ schwingerr ] ) , ( [ schwingerc ] ) and ( [ zeq ] ) are the complete set of equations that determines the dynamics of the system . in the following we make a time reparametrization @xmath206 that implies @xmath207 and @xmath208 the system of units we use is @xmath209 & = & [ \hat r ] = [ \hat \hbar ] = 1 \ ; , \nonumber\\ \left[\hat m \right ] & = & [ \hat \beta ] = [ 1/\hat{\tilde j } ] = [ 1/\hat{\lambda } ] = [ \hat{t } ] \ ; .\end{aligned}\ ] ] hereafter we write the equation in reparametrized variables and drop all hats . as already mentioned , the set of integro - differential equations ( [ schwingerr ] ) , ( [ schwingerc ] ) and ( [ zeq ] ) is causal as in the classical case and one can attempt a numerical solution by a simple algorithm . the correlation @xmath210 and response @xmath191 are _ two time quantities _ , that is , they depend on @xmath193 ( which physically corresponds to the time of observation ) and @xmath211 ( which corresponds to the age of the system ) . in a normal situation , one expects that after an equilibration time @xmath82 ( in general model dependent ) all two - time functions will depend only on the time difference @xmath212 . in other words , all two - time functions become time - translation invariant ( tti ) . under these circumstances , to which we refer as equilibrium evolution , response and correlation are not independent quantities but are linked via the fluctuation - dissipation theorem ( fdt ) . nevertheless , it is also known that certain systems never reach ( at least for experimentally accessible time - scales ) an equilibrium dynamics . for arbitrary large @xmath211 and in a certain region of the parameter space , although one - time quantities , such as the energy - density , tend to an asymptotic limit , two - time quantities , such as the correlation and response , depend both on @xmath193 and @xmath211 in a non - trivial way . this is indeed the typical situation for a glassy system.@xcite the question we would like to explore is whether eqs . ( [ schwingerr ] ) , ( [ schwingerc ] ) and ( [ zeq ] ) encode a non - equilibrium evolution as described above . we know that indeed this is the case for classical disordered systems . as shown in ref . [ ] , this kind of equations describe , at least partially , the non equilibrium dynamics of glassy systems , notably _ aging _ effects of spin - glasses . the salient feature of these solutions is that , below a critical temperature @xmath213 , and for large @xmath211 , there exist at least two time scales : for @xmath212 small with respect to a characteristic time @xmath214 , that depends upon the age of the system , the dynamics is similar to an equilibrium evolution ( stationary regime ) but for @xmath212 large with respect to @xmath214 the dynamics becomes extremely slow with aging effects . in this regime , correlation and response are related in a manner that violates fdt . in this paper we show that this also happens in the quantum problem in a region of parameter space . in what follows we concentrate on the case @xmath17 . as in the classical case , the @xmath13 model is connected to @xmath14 models and has a different behavior . in fig . [ phasediagram ] we present a schematic phase diagram @xmath215 for @xmath216 . as we show below , the dynamics in the disordered ( paramagnetic ) and ordered ( spin - glass ) phases are characterized by equilibrium and `` out - of - equilibrium '' effects , respectively . the dynamics for the quantum system in the glassy phase has a similar pattern to the classical one , in the sense that there are at least two time - sectors in which the evolution is of a different nature . inspired by the classical problem,@xcite we propose that also in the quantum case the weak - ergodicity breaking @xcite and the weak long - term memory @xcite properties hold . these two assumptions are based in part on the insight coming from the numerical solution of the full equations . the _ weak - ergodicity breaking scenario _ states that , for @xmath181 , the correlation function decays in such a way that @xmath217 in these equations @xmath218 is the edwards - anderson order parameter that characterizes the spin - glass phase . this means that for large @xmath181 such that @xmath212 is small with respect to @xmath214 , the correlation function first decays from 1 to @xmath218 in a time - translational invariant manner . @xmath214 is a growing function of @xmath211 whose precise form depends on the model . the correlation goes further below @xmath218 to eventually reach @xmath219 in a manner that depends both upon @xmath193 and @xmath211 ( the aging effect ) . this behavior suggests the presence of at least two time - sectors in which the dynamics is stationary and non - stationary , respectively . we then write the correlation as the sum of a stationary and an aging contribution : @xmath220 the matching conditions at equal times between @xmath221 and @xmath222 are @xmath223 with @xmath224 together with eq . ( [ limitqq ] ) they ensure that in the two - time sector in which @xmath221 decays from @xmath225 to @xmath219 , @xmath222 is just a constant @xmath218 . instead , in the time - sector in which @xmath222 decays from @xmath218 to @xmath219 , @xmath221 vanishes identically . the name weak - ergodicity breaking @xcite reflects the fact that for short time - differences the system behaves as if it would be trapped in some region of phase space of `` size '' @xmath218 suggesting ergodicity breaking . however , it is always able to escape this region in a time - scale @xmath214 that depends upon its age @xmath211 . hence , trapping is gradual and ergodicity breaking is _ weak_. we recall that in the classical and purely relaxational case ( langevin dynamics ) the correlation functions are monotonic with respect to both times @xmath193 and @xmath211 . however , in the presence of inertial terms there can be oscillations and the decay can be non - monotonic . this will depend upon the relative value of the mass @xmath226 with respect to the other parameters in the problem . for the parameters we study in this paper , the oscillations appear only in the stationary regime , the aging dynamics having a monotonic decay towards zero . this is relevant since it will allow us to use the general properties of monotonic correlation functions proven in ref . [ ] to find the two - time scaling of @xmath227 . regarding the response function , we propose @xmath228 with @xmath229 when a system is in equilibrium , the response is simply related to the correlation via fdt . here we need to extend this relation to a non - equilibrium quantum situation . as in the classical case we assume that the dynamics in the stationary regime is like an equilibrium dynamics and satisfies fdt . we then assume that @xmath230 and @xmath231 are related by @xmath232 ( see appendix [ appfdt ] ) . in contrast , in the time - sector in which the dynamics is non - stationary and manifestly non - equilibrium , fdt will be modified . in section [ generalfdt ] we suggest an extension of fdt to quantum non - equilibrium systems with slow dynamics , that we later check in the particular model under study . the _ weak long - term memory scenario _ states@xcite that the system keeps a weak memory of what happened in the past . more precisely , the response function tends to zero when times get far apart , @xmath233 and its integral over a _ finite _ time - interval also vanishes @xmath234 but , its integral over an interval that grows with time gives a finite contribution : @xmath235 note that the property ( [ shortr ] ) implies @xmath236 an important quantity that reflects the decay of the response function is the two - time susceptibility defined as the integral of the response over the time - interval @xmath237 $ ] : @xmath238 ( @xmath239 is the strength of the applied perturbation , see eq . ( [ respdef ] ) ) . in the classical case , a useful description of the departure from fdt is obtained by plotting @xmath240 vs @xmath241 for several @xmath242 using @xmath243 as a parameter . since the classical fdt implies @xmath244 , violations of the theorem manifest as departures from a straight line plot . in the quantum case , the relation between @xmath188 and @xmath189 is more involved and _ a priori _ one does not expect these plots to be useful . as we show in section [ generalfdt ] , it turns out that even for a quantum system these parametric plots are relevant . in the paramagnetic phase ( see fig . [ phasediagram ] ) one expects that , after a short non - equilibrium transient , a time - translational invariant ( tti ) solution establishes . we show that a tti ansatz implies that fdt holds between @xmath188 and c and between @xmath163 and @xmath162 . by solving numerically the full set of equations ( [ schwingerr ] ) , ( [ schwingerc ] ) and ( [ zeq ] ) we show the correctness of this ansatz and later discuss some properties of the solutions . after the short initial transient , one - time quantities such as @xmath123 should reach a limit @xmath245 . we evaluate it as @xmath246 - m \left . \partial^2_{t } c(t - t ' ) \right|_{t'\to t^{-}\to\infty } \ ; .\ ] ] the fourier - transform of eqs . ( [ schwingerr ] ) and ( [ schwingerc ] ) yields _ the quantum mode - coupling equations _ : @xmath247 thus , @xmath248 if we now assume that @xmath163 and @xmath162 are related by fdt , it is easy to check that eqs . ( [ 111 ] ) , ( [ 222 ] ) and ( [ imromega ] ) imply that @xmath188 and @xmath189 are also related by fdt . the proof is completed by showing that fdt between @xmath188 and @xmath189 implies fdt between @xmath163 and @xmath162 . this is easier to do in the time - domain , by verifying that eq . ( [ fdttemporal ] ) holds between @xmath163 and @xmath162 if it does between @xmath188 and @xmath189 . a solution of this type exists only in a restricted region of the phase diagram . in order to demonstrate the existence of a dynamic transition , we solve the _ full _ dynamic equations numerically and find that at a critical line , both the tti and fdt assumptions break down . for the purpose of solving eqs . ( [ schwingerr ] ) , ( [ schwingerc ] ) and ( [ zeq ] ) numerically , we proceed as in the purely relaxational classical case.@xcite notice though that the presence of non local kernels @xmath74 and @xmath73 , that resemble the delta - function and appear convoluted with the correlation and response , renders the numerical solution harder . the larger the cut - off @xmath94 , the smaller the iteration step @xmath249 we need to compute these integrals with a good precision . we found an acceptable data collapse for @xmath250 if @xmath251 . note that the values of @xmath249 typically used in the purely relaxational case are at least one order of magnitude larger . this is the reason why the time - window we explore here is rather narrow though it suffices to show the trend of the solution . even after rescaling time and all parameters in the problem , there are still many free parameters left : @xmath252 we here concentrate on the dependence upon only a small fraction of this set . we fix hereafter @xmath216 , @xmath253 , @xmath254 and @xmath251 and study the behavior upon the remaining parameters @xmath255 . we discuss the data for the correlation , response and integrated response for three points in the paramagnetic phase : * @xmath256 . this corresponds to the classical problem with colored noise and inertia . the classical critical temperature for the purely relaxational case is @xmath257 . in fig . [ fight2 ] we show the correlation @xmath258 vs. the time - difference @xmath259 for @xmath260 . in fig . [ fight3 ] we plot the response @xmath261 vs @xmath259 for the same waiting times . in both cases the decay is very fast , typical of upper critical dynamics , and tti quickly sets in ( cfr . the curves for @xmath262 and @xmath263 that are indistinguishable in both plots ) . in the inset , we plot the integrated response @xmath264 vs. @xmath258 for @xmath265 using @xmath259 as a parameter . the @xmath266 vs. @xmath189 plot very soon becomes a straight line of slope @xmath267 showing that the classical fdt is satisfied . * @xmath268 . this corresponds to a strong quantum regime without thermal noise . the data for correlations and response in figs . [ fighhbar3 - 1 ] and [ fighhbar3 ] are qualitatively similar to the ones for the classical problem at high temperatures in the sense that an equilibrium regime is quickly attained . the decay is fast and tti holds . in the inset of fig . [ fighhbar3 ] we display the @xmath266 vs. @xmath189 plot . in this strong quantum case at @xmath7 , the curve severely deviates from the classical expectation of a straight line of infinite slope . the quantum fluctuation - dissipation theorem is verified in fig . [ rfoul5hbar6t0 ] by comparing the response function obtained numerically with the convolution of the kernel @xmath269 with the numerical correlation function , see eq . ( [ fdtfourier ] ) . these two curves should be equal in equilibrium , as imposed by fdt , and we here check that this is indeed the case in the paramagnetic phase . * this is the quantum problem with both thermal and quantum fluctuations in the paramagnetic phase . in fig . [ mixedhb1t2 ] we check fdt for these parameters and observe , in the @xmath266 vs. @xmath189 curve displayed in the inset , that the fdt becomes the classical . we continue the study of the paramagnetic phase by analyzing the variation with @xmath31 and @xmath186 , close to the transition line , of the out of phase susceptibility @xmath271 this is the quantity that is most commonly measured experimentally . for instance , in the context of the study of quantum phase transitions , the dynamics of the randomly diluted , dipolar coupled , ising magnet liho@xmath2y@xmath3f@xmath4 was studied by measuring,@xcite among other quantities , @xmath272 . we follow two routes to the transition : * @xmath273 , @xmath274 . one approaches the classical model by decreasing the strength of quantum fluctuation at fixed temperature @xmath275 . in fig . [ chit13hbar ] we plot @xmath272 vs. @xmath83 . in the curves for small @xmath186 we observe a local minimum that is associated to the appearance of a plateau in the correlation function , at the value @xmath218 , when approaching @xmath18 . this is typical of `` discontinuous transitions '' as the one in the classical @xmath0-spin model@xcite ( and in super - cooled - liquids@xcite ) . we here note that this feature survives the presence of small quantum fluctuations and progressively disappears when @xmath186 increases . for @xmath276 , the minimum has been substituted by a single peak . * @xmath277 , @xmath278 . one approaches the critical line @xmath279 from above by modifying the strength of the quantum fluctuations while keeping the temperature fixed @xmath280 . figure [ chit01hbar ] displays @xmath272 vs. @xmath83 for this series of @xmath186 . ( we can not get closer to the transition line because the dynamics gets so slow that the stationary limit can not be reached . ) the curves have a distinct peak that moves towards decreasing frequencies for decreasing @xmath186 . the spectral width of the bell - shaped curves also increases when approaching the critical line . the low frequency part of the curves is importantly suppressed with increasing @xmath186 . all these features show that quantum fluctuations radically affect the relaxation . + contrary to fig . [ chit13hbar ] , there is no local minimum in any of the curves . this is in accordance with a decreasing and eventually vanishing plateau value @xmath218 when the parameters approach the quantum critical point . we shall further discuss the behavior of the model close to the transition line in section [ criticalline ] . these results demonstrate the existence of a cross - over in the disordered phase , separating a region where the scaling laws are controlled by the quantum critical point @xmath281 from another region with different scaling laws controlled by the classical critical point @xmath282 . before concluding our analysis of the paramagnetic phase , let us compare our results with the situation encountered for the system@xcite liho@xmath2y@xmath3f@xmath4 , which is believed to be an experimental realization of the transverse ising model . naturally , we do not expect to find the same behavior in both models as we know that in the classical limit they are different:@xcite the classical transition being first order for the @xmath0-spin model and second order for the ising case . our analysis indicates that the dynamic transition of the @xmath0-spin model remains first order along the critical line but becomes second order at the quantum critical point . this is the mirror situation to what has been observed in the dipolar glass . in ref . [ ] , it has been suggested that the quantum spin - glass transition may be first order . figure [ chit13hbar ] displays the effect of quantum fluctuations close to the classical critical point and looks rather similar to fig . 2 in ref . [ ] which instead shows the dynamics close to the quantum critical point . figure [ chit01hbar ] looks similar to fig . 4 in ref . [ ] , where the dynamics close to the classical critical point is presented . this comparison indicates that this model belongs to a different class from the system liho@xmath2y@xmath3f@xmath4 and probably from the transverse field ising model.@xcite in this section we focus on the non - equilibrium dynamics observed in the glassy phase . in order to justify the scenario discussed in section [ scenario ] , we first present the results from the numerical solution of the full equations . next , we analyze these equations in the stationary and aging regimes and obtain a self - consistent relation that determines the edwards - anderson parameter . the numerical solution shows a very different behavior below the critical line @xmath283 . in order to illustrate the slow dynamics , the breakdown of tti and the weak - memory of the system we choose the following parameters : * @xmath284 this is the classical , glassy model studied in ref . [ ] but with the addition of inertia . the effect of inertia amounts to the presence of oscillations around @xmath218 but it does not change the qualitative behavior of the model . for example , as we shown analytically below , the value of @xmath218 is not modified by inertia if @xmath226 is small . * @xmath285 , @xmath286 . this corresponds to pure quantum fluctuations . figure [ ct0hb1 ] shows the correlation @xmath258 vs. @xmath259 for the waiting times @xmath287 . the curves have a stationary and an aging regime separated by the edwards - anderson parameter @xmath288 , see eq . ( [ limitqq ] ) . the correlations decay to zero at far apart times , as in eq . ( [ limit0 ] ) . this demonstrates the weak ergodicity breaking scenario . + in fig . [ rt0hb1 ] we plot the response @xmath261 vs. @xmath259 for the same waiting times . one observes a small time - difference regime , @xmath289 , where tti rapidly establishes and a large time - difference regime , @xmath290 , where the response shows a decaying queue with a @xmath242 dependence . the integration of this queue over a time - interval scaling with @xmath242 will tell us about the violations of fdt in the quantum case ( see fig . [ ml5hb01t0 ] below ) . it also shows that the system has a weak long - term memory as described in section [ scenario ] . * @xmath291 , @xmath286 . the qualitative behavior is similar to that obtained for @xmath7 and @xmath292 apart from the fact that thermal fluctuations slightly decrease the value of the edwards - anderson parameter and accelerate significantly the decay in the aging regime , as demonstrated in fig . [ ct05hb1 ] . the lagrange multiplier @xmath123 is a one - time quantity , i.e. it depends only upon the total time @xmath193 . in this formalism we assume that all one - time quantities reach a well - defined limit asymptotically . in appendix [ integrals ] we describe how we obtain the following equation for the asymptotic limit of @xmath123 : @xmath294 - m \left . \partial^2_\tau c_{\sc st}(\tau)\right|_{\tau\to 0 } \ ; .\end{aligned}\ ] ] let us discuss each term in this expression . we called @xmath295 the aging contribution : @xmath296 \ ; . \end{aligned}\ ] ] the kernels @xmath73 and @xmath74 , that are related to the bath , do not contribute to @xmath295 . this is so because we have assumed that for @xmath297 the kernels have already decayed to zero in such a way that they do not contribute to these integrals . more precisely , we are neglecting terms of the form @xmath298 where @xmath299 is either @xmath73 or @xmath74 and @xmath300 is either @xmath222 or @xmath301 . the second and third terms come from the constant ( non - zero ) limit of the first decay of the correlation @xmath302 and equivalently of the vertex @xmath303 in the model under study @xmath304 if we use @xmath305 , a property of the weak long - term memory scenario . if @xmath306 are such that @xmath307 , the discussion in section [ scenario ] implies @xmath308 and @xmath309 . the schwinger - dyson equation for @xmath188 in this time sector reads @xmath310 and it keeps the same form as in the high - temperature regime , apart from the fact that the constant @xmath245 has contributions from the aging regime . the schwinger - dyson equation for @xmath189 reads @xmath311 \ ; . \label{cst}\end{aligned}\ ] ] one can now fourier - transform both equations @xmath312 the formal solution to the equation for @xmath221 is @xmath313 the first term on the right - hand - side has an imaginary and a real part . the imaginary part vanishes identically since , due to fdt , both @xmath314 and @xmath315 are proportional to @xmath316 which is zero at @xmath317 . concerning the real part of this first term , as we have assumed that @xmath231 goes to zero for @xmath318 , we need to impose the self - consistent condition @xmath319 this is the condition that fixes the edwards - anderson parameter . we shall find it again in the next section as the matching condition between the stationary and aging regimes . the final equation for @xmath320 is @xmath321 one can check that these calculations are consistent with the results from @xmath245 . actually , the integrals in @xmath21-eq . involving the stationary parts can be evaluated with the help of the equations for @xmath322 and @xmath221 , eqs . ( [ rstat ] ) and ( [ cstat ] ) , and yield once again eq . ( [ cond2 ] ) . similarly to the high - temperature case one can now show that fdt for @xmath323 and @xmath324 implies fdt for @xmath322 and @xmath221 . the remainder of the proof , i.e. to show that fdt between @xmath322 and @xmath221 implies fdt between @xmath323 and @xmath324 depends only upon the from of @xmath323 and @xmath324 as functions of @xmath322 and @xmath221 and is not modified from the one discussed in section [ equilibriumdynamics ] . if we now choose the times @xmath138 to be well - separated so as to have @xmath325 and @xmath326 , the weak - ergodicity breaking and weak long - term memory hypotheses allow us to throw the second time derivatives on the left - hand - side . we assume that their contribution is much weaker than the one of each of the integral terms on the right - hand - side . this is an assumption that we have to verify at the end of the calculation , once the solution for @xmath222 and @xmath301 is known . it corresponds to the over - damped limit : @xmath327 using the approximation described in appendix [ integrals ] , the @xmath188-eq in the aging regime becomes @xmath328 and we call it the @xmath301-eq . similarly , the @xmath189-eq becomes @xmath329 and we call it the @xmath222-eq . in all integrals we approximated @xmath330 and @xmath331 since @xmath332 and @xmath333 are very rapidly decreasing functions . the edwards - anderson parameter @xmath218 is determined self - consistently from the matching of @xmath334 . taking the limit @xmath204 in the @xmath301-eq and @xmath222-eq one obtains @xmath335 the first equation admits the solution @xmath336 since @xmath337 is proportional to @xmath338 see eq . ( [ sigmatilde ] ) . this corresponds to the high - temperature solution where there is no aging regime . here we concentrate on the other possibility , that is to say when @xmath339 in actual fact , we can further approximate this equation by using one of the assumptions already used all over this section : that the response becomes smaller and smaller as time passes though its integral over an infinite interval gives a finite contribution . if we neglect all terms that are proportional to @xmath338 with respect to terms that are proportional to @xmath218 , only the first term in the power expansions of @xmath163 and @xmath162 survive : @xmath340 that in our model become @xmath341 in accord with the large @xmath259 limit of the stationary values ( [ dqstat ] ) . equations ( [ hola ] ) and ( [ cq ] ) become @xmath342 the second equation is the same as the one arising from the end of the stationary regime , eq . ( [ cond2 ] ) . we now study these equations with the help of eqs . ( [ rstat ] ) and ( [ cstat ] ) for the stationary functions . using @xmath343 one has @xmath344 we remind that the factor @xmath345 can be written in terms of the stationary correlation function using fdt ; therefore this is a closed equation for the correlation that gives the edwards - anderson parameter @xmath218 . in the case of the @xmath0-spin model it reads @xmath346 in the classical case , the integral can be readily computed and the final equation for @xmath218 is @xmath347 that coincides with the result for the purely relaxational dynamics.@xcite for @xmath17 fixed , @xmath218 is a function of temperature ; it equals one at @xmath7 and tends to @xmath348 at the dynamic critical temperature @xmath213 . if @xmath349 , the edwards - anderson parameter @xmath218 depends upon @xmath186 and @xmath31 , @xmath350 . it decreases when thermal and/or quantum fluctuations increase . one observes @xmath351 . at the critical line @xmath352 , @xmath353 as long as @xmath354 . at the _ quantum critical point _ the nature of the transition changes and @xmath355 . we shall explain this result after introducing the modification of the quantum fdt needed to complete the solution . figure [ plotq ] shows the tendency of @xmath356 for @xmath357 . one of the important characteristics of the dynamic evolution in equilibrium is the fluctuation - dissipation theorem theorem ( fdt ) that establishes a model independent relationship ( in the sense that only involves @xmath31 and @xmath186 but no other parameter in the model ) between correlation and response functions . as recalled in appendix [ appfdt ] , this result can be easily obtained , for the quantum and classical cases , under the hypothesis of equilibrium . in general , such a relation is not expected to hold in a non equilibrium situation and we can then say that fdt is violated " ( of course what is violated is not the theorem but the hypothesis under which the result is derived ) . it is nevertheless surprising that in the glassy phase of certain classical models , the violation of fdt @xcite takes such a simple form that it is possible to formulate a generalized version of this relation through the introduction of an effective temperature @xmath358 . @xcite in this way the fdt takes the standard equilibrium form but with a model dependent effective temperature which is different from the bath temperature @xmath31 ( in general @xmath358 is higher or equal than @xmath31 and it can take different values in different time - sectors ) . we would like to explore in this section how these ideas carry over to the quantum case . the standard quantum fluctuation - dissipation theorem states @xmath359 with @xmath360 guided by the classical case , we introduce the effective temperature@xcite @xmath361 and we propose , as a generalization of fdt to the out of equilibrium case , the following relation @xmath362 with @xmath363 and @xmath364 a function of both times @xmath193 and @xmath211 . evidently , eqs . ( [ xfdt1 ] ) and ( [ xfdt2 ] ) reduce to eqs . ( [ fdt1 ] ) and ( [ fdt2 ] ) when @xmath365 if the evolution is tti and the factor @xmath366 is set to one . let us assume , for reasons that are intimately related to the time - separation discussed in section [ scenario ] , that @xmath364 is characterized by @xmath367 for times in the stationary regime @xmath368 this implies : @xmath369 with @xmath370 if we suppose that ( even at @xmath7 ) @xmath371 is finite in the glassy phase , it is then clear that in the aging sector the fourier integral is dominated by @xmath372 . after using @xmath373 we obtain @xmath374 for a standard quantum system in equilibrium at finite temperature , there is a characteristic time , that depends upon temperature and the strength of quantum fluctuations , beyond which the evolution becomes classical . concerning fdt , this means that the classical relation establishes with @xmath375 . this is clear , for instance , for the model under consideration at finite temperature in the paramagnetic phase ( see the inset in fig . [ fight3 ] ) . if our assumptions are correct , the situation in the glassy phase is different . for small times compared with @xmath376 , now a _ waiting - time dependent _ characteristic time , correlation and response are related by the standard ( quantum ) fdt . instead , for time differences larger than @xmath376 , they are related in the same way as in the classical case _ but _ with a @xmath31 and @xmath186 dependent effective temperature that is different from the one of the environment . in general , the proposal given in eq . ( [ fdt6 ] ) will allow us to reduce the @xmath222-eq . and @xmath301-eq equations to single equation for @xmath222 . it is simple to check that , at least for the model under consideration , this equation has the same structure as the one for the classical problem . all dependence on @xmath186 enters through @xmath218 , @xmath377 and @xmath322 . hence , once the explicit form of @xmath377 is known , the aging equations are solved by the same ansatz as in the classical case : @xmath378 note that the general arguments introduced in ref . [ ] to characterize the behavior of _ monotonic _ correlation functions apply to the correlations in the aging regime and justify the solution ( [ agingc])-([agingr ] ) . even if the inertia might make the correlation oscillate , it does only in the stationary regime since it oscillates around @xmath218 , and it becomes monotonic in the aging in eq . ( [ xtwoscales ] ) , we assumed that @xmath364 can take only two values in view of the model we study . however , in the classical case , a more general situation arises in models with multiple time - scales , for which the factor measuring the violation of fdt is a non - constant function of the correlation in the aging regime.@xcite we expect the same structure to carry through to quantum problems whose classical counterparts are of the multi - scale type . we check , both analytically and numerically , the viability of our assumptions for the @xmath0-spin model . the factor @xmath377 measuring the violation of fdt is determined by eqs . ( [ firstq ] ) and ( [ secondqq ] ) @xmath380 using eq . ( [ fdt6 ] ) and the equivalent relation between @xmath381 and @xmath382 , we obtain @xmath383 and @xmath384 in the classical limit @xmath385 and the result in ref . [ ] is recovered . note that both in the classical and quantum case , @xmath386 if @xmath13 . since the case @xmath13 is formally connected to ferromagnetic domain growth ( in the mean - field approximation ) there is still no - memory in the quantum domain growth . the classical critical point @xmath387 is characterized by@xcite @xmath388 following the critical line towards the quantum critical point @xmath389 , one obtains @xmath390 and a finite zero - frequency stationary susceptibility @xmath391 . at the quantum critical point , @xmath218 vanishes and the zero - frequency stationary susceptibility diverges : @xmath392 ( it is interesting to notice that the case @xmath13 is clearly different . ) on the other hand , if approaching the quantum critical point from below along the axis @xmath393 @xmath394 then @xmath395 [ ml5hb01t0 ] we next check numerically the relation between response and correlation in the glassy phase , @xmath396 corresponding to pure quantum fluctuations . first , we use fourier analysis to check the quantum fdt for short time differences . in fig . [ fdthbarfourier ] we compare two different ways of obtaining @xmath397 for total time @xmath193 fixed and equal to @xmath398 . the thin curve is the result from the direct numerical solution of the dynamic equations and shows a fast evolution for short time differences and a tail that gives rise to the weak memory of the system . the bold curve is obtained from eqs . ( [ fdt3 ] ) and ( [ fdt4 ] ) using @xmath399 with @xmath400 obtained numerically . we have estimated the edwards - anderson parameter as @xmath401 . one observes that for small time - differences the two curves coincide showing that fdt holds in this time - sector . second , we check the generalized fdt in the aging regime . for this purpose , it is convenient to use the @xmath266 vs. @xmath189 plots defined in section [ scenario ] . in fig . [ chic1 ] we observe , as predicted by eqs . ( [ fdt3 ] ) , ( [ fdt4 ] ) and ( [ fdt6 ] ) , two behaviors according to the relative value of @xmath189 and @xmath218 . the portion of the graph that is significant for the aging regime corresponds to @xmath402 . as claimed in eq . ( [ fdt6 ] ) the @xmath266 vs @xmath189 curve approaches a time - independent asymptotic limit that is a straight line of slope @xmath403 . in this case , @xmath404 and this shows that even a @xmath7 , @xmath405 . the effective temperature depends on @xmath31 and @xmath186 in a non trivial way . in figure [ chic2 ] we show the @xmath266 vs. @xmath189 plot for @xmath186 = 1 and @xmath7 . in this case @xmath406 . in fig . [ teffmixed ] we work at fixed @xmath407 and study the variation of @xmath358 with t. from the @xmath266 vs @xmath189 plots we determine how @xmath358 varies with @xmath31 by measuring the slopes of the different curves and by using eq . ( [ xfinal ] ) . the six curves above the dotted straight line correspond to temperatures below @xmath1 . from top to bottom @xmath408 . one notices that @xmath409 in all these curves . the curve below the dotted straight line corresponds to @xmath410 . it has a slope @xmath411 ( indicated by the dots ) and @xmath412 . in the inset we plot the temperature dependence of the effective temperature . @xmath358 is linear in @xmath31 when the bath temperature is above @xmath1 but severely deviates from the linear behavior for temperatures below the critical line . two curious features are that , in the glassy phase , @xmath358 increases when either @xmath31 or @xmath186 decrease . the surprising increase of @xmath358 as @xmath31 decreases is also obtained in the classical limit.@xcite in this paper we presented a formalism that allows us to analytically study the behavior of quantum glassy models within the typical situation encountered in experiments , i.e. out of equilibrium . faced to the question as to how relevant quantum fluctuations are for the asymptotic real - time dynamics , one would have naively answered that they are irrelevant since quantum mechanics is not expected to play a relevant role in the large time sector , meaning large @xmath242 and for all @xmath413 . we show in this paper that this would have been a wrong conclusion and that the dynamic behavior is much richer : quantum effects , together with temperature , determine _ where _ in parameter space a glassy phase exists . we show that sufficiently strong quantum fluctuations destroy the glassy phase at arbitrary low temperatures driving the system towards a paramagnetic phase . we predict then the existence of a dynamic quantum phase transition . this has no analogue in the classical model where glassiness is always present at @xmath7 . quantum fluctuations also dictate _ when _ , in the glassy phase , non - equilibrium effects manifest . for any waiting - time @xmath242 after preparation , there exists a characteristic time @xmath376 that determines the end of a stationary regime and the entrance into the aging regime . this means that for time differences @xmath414 the dynamics is stationary while for @xmath415 two - point functions have explicit waiting - time dependences . the characteristic time @xmath376 depends on temperature , the strength of quantum fluctuations and the waiting - time @xmath242 . the dynamics in the paramagnetic phase satisfies the fluctuation - dissipation theorem . in the glassy phase , we found that the quantum fdt is satisfied in the stationary regime and that a generalization is needed in the aging regime . this generalization defines a two - time function @xmath364 that is consistent with the definition of an effective temperature @xmath416 and that we later characterized , based on the numerical evidence for this model . when @xmath358 is different from zero , the modification of the quantum fdt takes a similar form as the classical one but with a @xmath31 and @xmath186 dependent effective temperature . the characteristic time @xmath376 is reminiscent of a decoherence time in the sense that it separates the stationary regime in which the influence of quantum effects is very strong and explicit from the aging regime in which the influence of quantum effects is implicit , appearing through , e.g. , the @xmath186 dependence of @xmath358 . one of the virtues of the approach we have followed is that it is close in spirit to the dynamic approach to classical systems . this encourages us to attempt to extend many of the recent advances in the understanding of classical glassy dynamics to the quantum problem . we mention below some of the questions that merit further attention . the model here studied has a modification of the quantum fdt given by a piecewise function @xmath417 if @xmath418 ( @xmath419 ) and a constant @xmath420 if @xmath421 ( @xmath422 ) . it is known that in classical systems a more general situation can arise . another family of models exists with the factor measuring the fdt violation given by @xmath417 if @xmath418 and a non - constant function of the auto - correlation @xmath423 if @xmath421 . we expect the arguments put forward in section [ generalfdt ] to apply to this kind of system with minor modifications . it would be of particular interest to see how this structure appears in a concrete model as , for instance , the problem of a particle in a long - range correlated random potential . in the classical case , an intriguing connection between the factor that measures the fdt violation@xcite in the non - equilibrium approach @xmath424 and the parisi function @xmath425 , @xmath426 of the replica analysis in equilibrium , has been noticed and checked in several occasions . the structure of both functions is the same and , on top of this , their explicit values coincide if an argument of `` marginality '' is used to determine @xmath425 in the equilibrium case . in ref . [ ] , several quantum models related , but not equal , to the one considered here were studied with a gaussian variational method and replica theory in the imaginary time matsubara representation . in order to obtain a physical value for the conductivity of the model , giamarchi and le doussal used the marginality condition to fix the replica breaking point parameter @xmath427 , @xmath428 . a careful comparison of the dynamic @xmath377 with the static @xmath429 is now in order . nevertheless , this can not be readily done by comparing the result in ref . [ ] with ours since the way in which both systems are coupled to the environment are not necessarily equivalent . thence , if a formal relation between fdt - violations and the replica @xmath425 exists at the quantum level is an interesting question that deserves further study . the last remark we wish to make about the fdt violations is that in the classical case it has been related to the production of entropy during evolution.@xcite it would be desirable to see whether such a connection also holds at the quantum level . concerning the formalism here developed , it can be extended in several directions . on the one hand , one would like to study models with a spatial structure such as the one associated to the motion of a _ manifold _ of internal dimension @xmath430 in a random quenched potential . this quantum problem has many interesting applications and , in the case of an embedding space with a large dimensionality can be treated , for example , within a dynamic gaussian variational approach.@xcite on the other hand , it would be extremely interesting to study the opposite limiting case of low dimensions and analyze , for example , the real - time dynamics of the ising chain in a transverse field . in addition , one could investigate the effect induced on the dynamics by different environments like considering sub or super ohmic baths , non - linear couplings and even baths of a completely different nature as those generated by spin variables.@xcite the martin - siggia rose generating functional for classical stochastic processes takes a particularly simple form if the time - space is enlarged to a superspace by introducing a couple of grassmann coordinates , and the variables of the problem are encoded in a superfield.@xcite this can also be done for systems with disorder.@xcite the dynamic equations are then written in terms of a single super - correlator that encodes the auto - correlation and response . the equilibrium theorems , viz . causality , invariance under time - translations and the fluctuation - dissipation theorem , follow from the symmetries of the msr effective action written in this way.@xcite the supersymmetric formulation of the classical dynamics has been helpful to find a solution of mean - field like models with an internal dimension like the problem of the manifold in a random media but , most importantly , it has highlighted the formal connection between the out of equilibrium formalism and the replica approach to the statics . it would be very interesting to identify a supersymmetric structure in the quantum problem , probably at the level of the keldysh - schwinger generating functional . this could provide a means to establish a relation between the quantum dynamic formalism here presented and the quantum replica approach . we believe that an analytical study of aging effects along the lines presented in this paper will be relevant in the understanding of realistic glassy systems . indeed , even at very low temperatures , where quantum effects are important , a glass is a non - equilibrium system that evolves over many orders of magnitude in time.@xcite aging effects , demonstrating this non - equilibrium evolution , have been observed with , for example , burning hole experiments in organic glasses ( see ref . [ ] and references therein ) . in the context of magnetic disordered systems with explicit quenched disorder like liho@xmath2y@xmath3f@xmath4 , a great effort has been devoted to the study of the quantum phase transition and the dynamics close to the quantum critical point . it would be interesting to perform dynamic measurements in the glassy phase of these compounds to search for aging effects and check if a scenario as the one here discussed with , for example , a two step relaxation of correlation functions and fdt violations is also present . another area of research where the relevance of quantum effects in the very low temperature dynamics is currently under study are systems of magnetic nano - particles such as @xmath431fe@xmath6o@xmath432 particles ( maghemite ) embedded in a silica matrix . the search for a non - stationary evolution in samples at small concentrations ( @xmath433 ) has given a negative result . @xcite however , in experiments performed at higher temperatures ( where quantum effects are not expected to play an important role ) with samples with a larger concentration ( @xmath434 ) much of the phenomenology of spin - glasses has been recovered.@xcite it is then licit to ask whether there is a region in the temperature - concentration parameter - space where the systems exhibit aging driven by quantum fluctuations . in the area of magnetic nano - particles , a notion of an effective temperature greater than the temperature of the environment was advocated in ref . [ ] and it would be interesting to explore the connection with effective temperature defined in this paper . we have chosen to comment on three kind of materials where effects like the ones obtained in this article might be observed . surely enough , many other glassy systems with or without explicit disorder might exhibit a similar phenomenology . * acknowledgments * we want to especially thank j. kurchan for very useful discussions and constant encouragement during the preparation of this work . we also wish to thank t. evans , a. georges , t. giamarchi , d. grempel , p. le doussal , r. revers , m. rozenberg and e. vincent for useful discussions . at the early stages of this work we have also benefited from discussions with l. bettencourt and h. de vega . the authors want to thank the international center for theoretical physics , trieste , italy for hospitality during the `` summer college in condensed matter on statistical physics of frustrated systems '' . l.f.c . wishes to thank the institute for theoretical physics , university of california at santa barbara for hospitality during the work - shop `` jamming and rheology '' partially supported by the grant no phy94 - 07194 . g.l is financially supported by the ec grant n erbfmbict 961226 . [ appkeldysh ] the generating functional ( [ generating ] ) is more conveniently expressed with the help of path - integrals : @xmath28 & = & \int d{\bbox \phi } \ ; \langle { \bbox \phi } | t^ * \exp\left ( -\frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^-(t ) { \bbox \phi}(t ) \right ) t \exp\left ( \frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^+(t ) { \bbox \phi}(t ) \right ) | \hat \rho(0 ) | { \bbox \phi } \rangle \nonumber\\ & = & \int d{\bbox \phi}d{\bbox \phi}_1 d{\bbox \phi}_2 \ ; \langle { \bbox \phi } | t^ * \exp\left ( -\frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^-(t ) { \bbox \phi}(t ) \right ) \nonumber\\ & & \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\ ; \langle { \bbox \phi}_1 | t \exp\left ( \frac{i}{\hbar } \int_{0}^\infty dt \ , { \bbox \xi}^+(t ) { \bbox \phi}(t ) \right ) \langle { \bbox \phi}_2 | \hat \rho(0 ) | { \bbox \phi } \rangle \ ; .\end{aligned}\ ] ] the two matrix elements are now expressed as usual feynman path - integrals @xmath28 & = & \int d{\bbox \phi } d{\bbox \phi}_1 d{\bbox \phi}_2 \int_{{\bbox \phi}^+(0)={\bbox \phi}_2}^{{\bbox \phi}^+(\infty)= { \bbox \phi}_1 } { \cal d}{\bbox \phi}^+ \int_{{\bbox \phi}^-(0)={\bbox \phi}}^{{\bbox \phi}^-(\infty)= { \bbox \phi}_1 } { \cal d}{\bbox \phi}^- \nonumber\\ & & \exp\left [ \frac{i}{\hbar } \ , \left ( s[{\bbox \phi^+ ] - s [ \bbox \phi^- } ] + \int dt \ , { \bbox \xi}^+ { \bbox \phi}^+ - \int dt \ , { \bbox \xi}^- { \bbox \phi}^- \right ) \right ] \;\ ; \langle { \bbox \phi}_2 | \hat \rho(0 ) | { \bbox \phi } \rangle \ ; .\end{aligned}\ ] ] the first integrals @xmath435 , etc . are standard while we denote @xmath436 the path - integrals . the doubling of degrees of freedom @xmath437 is a consequence of having two matrix elements , one for each source @xmath438 . formally this expression is rewritten as @xmath28 & = & \int { \cal d}{\bbox \phi}^+ \int { \cal d}{\bbox \phi}^- \exp\left [ \frac{i}{\hbar } \ , \left ( s[{\bbox \phi^+ ] - s [ \bbox \phi^- } ] + \int dt \ , { \bbox \xi}^+ { \bbox \phi}^+ - \int dt \ , { \bbox \xi}^- { \bbox \phi}^- \right ) \right ] \nonumber\\ & & \;\;\;\;\;\;\;\;\;\;\ ; \times \langle { \bbox \phi}^+ | \hat \rho(0 ) | { \bbox \phi}^- \rangle \ ; .\end{aligned}\ ] ] [ appfdt ] in this appendix we recall the quantum fluctuation - dissipation theorem . proofs and descriptions of this theorem can be found in several textbooks.@xcite we express it in the time - domain in a form that we use in section [ equilibriumdynamics ] to show that the dynamic equations are compatible with a tti - fdt ansatz . we also write it in a mixed time - fourier notation that gives us insight as to how extend it to the case of glassy non - equilibrium dynamics . if at time @xmath211 the system is characterized by a density functional @xmath439 , the two - time correlation functions read @xmath440 \label{corr}\ ] ] where the time - dependent operators , in the heisenberg representation , are defined as @xmath441 the trace is defined in the usual way , @xmath442 \equiv \sum_\alpha \langle \psi_\alpha | \bullet | \psi_\alpha \rangle$ ] , with @xmath443 an orthonormal basis in fock space . the normalization is given by @xmath444 $ ] . since operators do not commute , the quantum two - time auto - correlation functions are not symmetric in times @xmath445 one thus defines the symmetrized and antisymmetrized correlation functions : @xmath446}(t , t ' ) & = & \frac12 \ , \langle a(t ) b(t ' ) - b(t ' ) a(t ) \rangle \ ; , \end{aligned}\ ] ] respectively . in linear response theory @xmath447 ( see eq . ( [ respdef ] ) for the definition ) and the correlation @xmath448}(t , t')$ ] are related by the kubo formula @xmath449 \ , \rangle = \frac{2 i}{\hbar } \ ; \theta(t - t ' ) \ , c_{[a , b]}(t , t ' ) \label{fdtasym } \ ; .\ ] ] if the system has reached equilibrium with a heat - bath at temperature @xmath31 at time @xmath211 , the density functional @xmath439 is just the boltzmann factor @xmath450 . it is then immediate to show that , in equilibrium , time - translation invariance holds @xmath451 in addition , eq . ( [ corr ] ) with @xmath452 imply the kms condition @xmath453 using now the kms properties and assuming , for definiteness , that @xmath454 it is easy to verify the following equation @xmath455 where @xmath456 . this is a way to express fdt through an analytic continuation to complex times that we use in section [ equilibriumdynamics ] to show that a tti and fdt solution holds in the paramagnetic phase . in terms of the fourier transformed @xmath457 defined by @xmath458 the kms relations read @xmath459 and lead to the following relation between fourier transforms of the correlation functions : @xmath460}(\omega ) & = & \left(1 - \exp(-\beta \hbar \omega ) \right ) \ , c_{ab}(\omega ) \ ; , \nonumber\\ 2 c_{\{a , b \}}(\omega ) & = & \left(1 + \exp(-\beta \hbar \omega ) \right ) \ , c_{ab}(\omega ) \ ; , \nonumber\\ c_{[a , b]}(\omega ) & = & \tanh\left ( \frac{\beta \hbar \omega}{2 } \right ) \ , c_{\{a , b \}}(\omega ) \ ; .\end{aligned}\ ] ] back in eq . ( [ fdtasym ] ) this implies fdt @xmath461 using @xmath462 one has @xmath463 from which we obtain the real and imaginary relations @xmath464 if @xmath465 , @xmath466 and eq . ( [ fdtfourier ] ) becomes the classical fdt : @xmath467 with @xmath468 . in the zero temperature limit , and for @xmath469 , fdt reads @xmath470 a finite integrated response @xmath471 implies @xmath472 . in the glassy model we study in the paper , below @xmath283 , the response and correlation turn out to be the sum of two contributions . one is a stationary part that satisfies fdt . since one expects it to yield a finite contribution to the susceptibility @xmath264 , one must have : @xmath473 thus , the stationary part of the correlation function oscillates around zero or , in other words , the full correlation oscillates around @xmath218 at zero temperature . this is observed in the numerical solution to the dynamic equations . [ classicallimit ] the classical limit can be checked at every stage of the calculation . we shall here show how one can recover ( i ) the martin - siggia - rose ( msr ) action from the keldysh - shwinger ( ks ) action ( [ action1 ] ) ; ( ii ) the equations of motion , for the classical correlation and response from the equations of motion for their quantum partners . this appendix is intended for readers already familiar with the msr formalism . we refer to ref . [ ] for a detailed description of the method . let us identify , in the limit @xmath474 , the quantum linear combinations on the left with the classical variables on the right @xmath475 in the classical limit @xmath476 , and when the cut - off tends to infinity , the kernels @xmath74 and @xmath73 become @xmath477 with these identifications , we establish a connection between the ks and the msr actions . the friction and thermal terms become @xmath478 by calling @xmath96 we recover two terms of the msr action . the kinetic and constraint terms reduce to @xmath479 by calling @xmath480 and @xmath481 the last two terms impose the spherical constraint and are usually included in the path - integral measure , while the second term is the one left in the msr action : @xmath482 finally , the terms that depend upon disorder become @xmath483 - v[{\bbox \sigma^- } , j ] = - \int_0^\infty dt \ ; i\hat{\bbox s } \ , \frac{\delta v[{\bbox s } , j ] } { \delta { \bbox s } } \ ; .\ ] ] putting everything together , in the large friction limit @xmath484 such that one neglects the second derivative terms in the kinetic part of the action , one recovers the usual msr action for the langevin dynamics of the classical model in a thermal bath at temperature @xmath31 : @xmath485 } { \delta { \bbox s}(t ) } i \hat{\bbox s}(t ) \right ] \ ; .\ ] ] in order to study the classical limit of the dynamic equations ( [ schwingerr ] ) and ( [ schwingerc ] ) we recognize the classical limit of the correlation and response as follows : @xmath486 in a similar way , it is now easy to verify that the classical dynamic equation for the @xmath0-spin model follow from the @xmath487 limit of eqs . ( [ schwingerr ] ) and ( [ schwingerc ] ) . [ integrals ] integrals over time windows spanning the interval @xmath488 $ ] appear in the dynamic equations ( [ schwingerr ] ) and ( [ schwingerc ] ) . in the large times limit we approximate them in the way we here describe . integrals of the form @xmath489 appear , for example , in @xmath123 . the idea is to separate the integration time - interval as @xmath490 = [ 0,\delta ] \;\ ; u \;\ ; [ \delta , t^- ] \;\ ; u \;\ ; [ t^-,t ] \ ; .\ ] ] if @xmath249 is chosen to be a finite time , all functions can be approximated by @xmath491 that vanishes when @xmath492 . since the integration interval is finite , this term can be neglected . in the second interval the functions vary in the aging regime and in the third interval they vary in the stationary regime . thus @xmath493 \ ; . \nonumber\\\end{aligned}\ ] ] we assume that this separation is sharp and that we can neglect the corrections associated to mixing of the three regimes . in the third term we replaced @xmath299 and @xmath300 in terms of @xmath494 , @xmath495 . since in all cases either @xmath299 or @xmath300 is proportional to the response , we can neglect the first term using the property ( [ finitetimes ] ) of the weak long - term memory scenario . we can then replace the lower limit of the second integral by @xmath219 and the upper limit of the second integral by @xmath193 . in addition , assuming that @xmath300 is proportional to the response , @xmath496 this yields @xmath497 another type of integrals is : is @xmath498 in particular , if @xmath499 , @xmath500 and @xmath501 , this integrals yields the susceptibility . let us assume that @xmath193 and @xmath211 are far apart ; we start as for the first type of integrals , by dividing the time interval in three subintervals @xmath502 = [ t',t'^+ ] \;\ ; u \;\ ; [ t'^+,t^- ] \;\ ; u \;\;[t^-,t]\ ] ] and by approximating the functions by their functional form inside each of the intervals : @xmath503 in the first term @xmath504 can be replaced by @xmath505 . the same applies to @xmath299 in the last term . all functions vary fast in the stationary regime but very slowly in the aging regime . the next assumption is that functions in the aging regime that are convoluted with functions in the stationary regime , can be considered to be constant and taken out of the integral . that is to say @xmath506 where we used @xmath507 and @xmath508 . typically , @xmath509 and @xmath510 are proportional to the response function . using fdt the integrals in the first and third term can then be performed in the classical limit or they can be expressed as functions of the correlation in the quantum case . the second term instead depends exclusively on the aging dynamic sector . e - mail address : leticia@physique.ens.fr e - mail address : lozano@ipno.in2p3.fr unit propre du cnrs , associe lecole normale suprieure et luniversit de paris sud . unit de recherche des universits paris 11 et 6 associe au cnrs . l. lundgren , p. svedlindh , p. nordblad and o. beckmann , phys . lett . * 51 * , 911 ( 1983 ) . e. vincent , j. hammann , m. ocio , j.p . bouchaud and l. f. cugliandolo , _ slow dynamics and aging _ in sitges conference on glassy systems , m. rub ed . ( springer - verlag , 1997 ) , cond - mat/9607224 . f. alberici , p. doussineau and a. levelut , j. phys . i france * 7 * , 329 ( 1997 ) ; europhys . lett . * 39 * 329 ( 1997 ) . f. alberici - kious , j - p bouchaud , l. f. cugliandolo , p. doussineau and a. levelut , _ aging in _ ta@xmath432 : _ a domain growth interpretation _ , cond - mat/9805208 . a. j. bray and m. a. moore , j. phys . * c17 * , l463 ( 1984 ) and in _ heidelberg colloquium on glassy dynamics _ , lecture notes in physics * 275 * , ed . j. l. van hemmen and i. morgenstern ( springer - verlag , berlin ) . g.j . koper and h.j . hilhorst , j. phys . ( france ) * 49 * , 429 ( 1988 ) . d.s . fisher and d.a . huse , phys . lett . * 56 * , 1601 ( 1986 ) ; phys . rev . * b38 * , 373 ( 1988 ) . v. dotsenko , j. phys . * c18 * , 6023 ( 1985 ) . v. dotsenko , m. feigelman and l. b. ioffe , _ spin - glasses and related problems _ , soviet scientific reviews * 15 * ( harwood 1990 ) . m. v. feigelman and v. vinokur , j. phys . i ( france ) * 49 * , 1731 ( 1988 ) . g. parisi , phys . lett . * 79 * , 3660 ( 1997 ) . w. kob and j - l barrat , phys . * 78 * , 4581 ( 1997 ) . a. barrat , phys . rev . * e57 * , 3629 ( 1998 ) . h. yoshino , _ aging dynamics of an elastic string diffusing in a disordered media _ , cond - mat/9802283 . j - l barrat and w. kob , _ fluctuation dissipation ratio in an aging lennard - jones glass _ , cond - mat/9806027 . t. d. kirkpatrick and d. thirumalai , phys . * b36 * , 5388 ( 1987 ) . t. r. kirkpatrick and p. wolynes , phys . rev . * a35 * , 3072 ( 1987 ) ; phys * b36 * , 8552 ( 1987 ) . t. r. kirkpatrick , d. thirumalai , p. g. wolynes ; phys . rev . * a40 * , 1045 ( 1989 ) . j - p bouchaud , l. f. cugliandolo , j. kurchan and m. mzard , _ out of equilibrium dynamics in spin - glasses and other glassy systems _ , in spin - glasses and random fields , a. p. young ed . ( world scientific , singapore , 1997 ) , cond - mat/9702070 . w. wu , b. ellmann , t. f. rosenbaum , g. aeppli and d. h. reich , phys . * 67 * , 2076 ( 1991 ) . w. wu , d. bitko , t. f. rosenbaum and g. aeppli , phys . lett . * 71 * , 1919 ( 1993 ) . t. f. rosenbaum , j. phys . * c8 * , 9759 ( 1996 ) . e. courtens , j. phys.lett . ( paris ) * 43 * , l199 ( 1982 ) ; phys . lett . * 52 * , 69 ( 1984 ) . e. matsushita , t. matsubara , prog . . phys . * 71 * , 235 ( 1984 ) . r. pirc , b. tadic and r. blinc , z. phys . * b61 * , 69 ( 1985 ) ; phys . rev . * b36 * , 8607 ( 1987 ) . p. shukla and s. singh , phys . rev . * b23 * , 4661 ( 1981 ) . a. j. bray and m. a. moore , * 13 * l665 ( 1985 ) . k. d. usadel , solid state commun . * 58 * , 629 ( 1986 ) . t. yamamoto and h. ishii , j. phys . * c20 * , 6053 ( 1987 ) . y. y. godschmidt and p. y. lai , phys . 64 * , 2467 ( 1990 ) . y. y. godschmidt , phys . rev . * b41 * , 4858 . v dobrosavljevic and d thirumalai , j. phys . * a23 * , l767 ( 1990 ) . l. de cesare , k. lubierska - walasek , i. rabuffo and k. walasek , physica * a214 * , 499 ( 1995 ) ; j. phys . * a29 * , 1605 ( 1996 ) . t. k. kope , phys . rev . * b54 * , 3367 ( 1996 ) . h. rieger and a. p. young , _ quantum spin - glasses _ ( springer - verlag , berlin , 1996 ) , cond - mat/9607005 . r. n. bhatt , _ quantum spin - glasses _ , in spin - glasses and random fields , a. p. young ed . ( world scientific , singapore , 1997 ) . c. p. martin , e. siggia and h. a. rose , phys . * a8 * , 423 ( 1973 ) . h. k. janssen , z. phys . * b23 * , 377 ( 1976 ) ; _ dynamics of critical phenomena and related topics _ , lecture notes in physics * 104 * , c. p. enz ed . , ( springer - verlag , berlin , 1979 ) . the relevance of the quantum langevin equation has been extensively treated in the literature . some references are g. w. ford , m. kac and m. mazur , j. math . phys . * 6 * , 504 ( 1965 ) . g. w. ford and m. kac , j. stat . phys * 46 * , 803 ( 1987 ) . t. jonsson , j. mattsson , c. djurberg , f. a. kahn , p. nordblad and p. svedlindh , phys . lett . * 75 * , 4138 ( 1995 ) . t. jonsson , j. mattsson , p. nordblad , p. svedlindh , j. magn . magn . mat . * 168 * , 269 ( 1997 )

in galactic astronomy , the most intensely studied of the problems of pattern formation is the problem of the formation of spiral structure , which has brought forth numerous theoretical and numerical approaches . even though at the present time there exists , as yet , no satisfactory theory of the origin and conservation of the spiral structure so prominent in the milky way and many other giant flattened galaxies , the majority of experts in the field has yielded to the opinion that the study of the stability of the small - amplitude cooperative oscillations in disk - shaped galaxies of stars is the first step towards an understanding of the phenomena . this is because in the milky way and many other disk galaxies , the bulk of the luminous mass , probably @xmath1 , is in the stars . therefore , it seems likely that the spiral structures are intimately connected with the stellar constituent of a galaxy , and that stellar dynamical phenomena play a basic role . because of the long - range nature of the gravitational force between stars , a galaxy exhibits collective modes of motions modes in which the stars in large regions move coherently or in unison . a galaxy is characterized also by a high temperature " ( a dispersion of random velocities of stars ) and hence a high thermal " energy ; this thermal energy is much larger than the interaction potential between pairs of stars . because of this , binary encounters produce only small - scale scattering of the motions of the stars . over long enough times , these gradually build up to produce large deflections that constitute collisions . in many cases these collisional effects are so weak that we consider the galaxy as collisionless on the hubble time @xmath2 yr ( chandrasekhar , 1960 ; binney and tremaine , 1987 ) . thus , the galaxy is dominated by the collective motions and the free streaming of the stars ( kinetic effects ) . as usual it is very difficult determine a priori whether a particular linearization of nonlinear equations made in analytical studies constitutes a valid approximation . this can be determined , however , by direct computer simulations of the nonlinear theory that mimic the behaviour of stars in galaxies . such modeling gives more detailed information than can be obtained analytically , so that the important physical phenomena can be determined . in this paper , the results of the approximate theoretical analysis described in the appendix are compared with our own many - body ( @xmath3-body ) simulations . one of the goals of the simulation is to test the validities of a modified local criterion for stability of jeans - type gravity perturbations in a self - gravitating , infinitesimally thin and collisionless disk . the fact that the nonaxisymmetric perturbations in the differentially rotating system are more unstable than the axisymmetric ones is taken into account in this modified toomre - like ( toomre , 1964 , 1977 ) criterion . in addition to the stability criterion we are interested in how model particles diffuse in velocity . as for the present study , such random velocity diffusion is caused by stellar disk turbulence . in particular , we verify the analytical prediction that as the jeans - unstable perturbations grow the mean - square velocity increases linearly with time . the observation of this behavior is a sensitive test of the model as well as the kinetic theory of stellar disks . in fact , morozov ( 1981 ) and griv _ et al . _ ( 1999a ) already attempted to confirm the modified criterion numerically . however , because of the very small number of model stars , @xmath4 , morozov s results are subject to considerable uncertainties , and additional simulations are clearly required to settle the issue . furthermore , for that number of model particles , the two - body relaxation time scale is comparable to the crossing time , even with morozov s modest softening parameter , raising some question about the applicability of his simulations to actual almost collisionless galaxies . increasing the number of model stars is definitely a more reliable procedure . in turn , griv _ et al . _ used a different numerical approach of the so - called local @xmath3-body simulations . the @xmath3-body experiments in a local or hill s approximation has been pioneered by toomre ( 1990 ) , toomre and kalnajs ( 1991 ) and salo ( 1995 ) . in these simulations dynamics of particles in small regions of the disk are assumed to be statistically independent of dynamics of particles in other regions . the local numerical model thus simulates only a small part of the system and more distant parts are represented as copies of the simulated region . unlike morozov ( 1981 ) and griv _ ( 1999a ) , we study some aspects of dynamical behavior of stellar systems by global simulations using a sufficiently large number of particles . different methods are currently employed to simulate global evolution of collisionless point - mass systems of galaxies by @xmath3-body experiments . one algorithm for a simulation code that has found wide use was developed by hohl and hockney ( 1969 ) , hohl ( 1971 , 1972 ) , miller ( 1971 ) , hockney and brownrigg ( 1974 ) , quirk ( 1972 ) , athanassoula and sellwood ( 1986 ) , combes _ ( 1990 ) , pfenniger and friedli ( 1993 ) , merritt and sellwood ( 1994 ) , donner and thomasson ( 1994 ) , and others . the method samples the density field with a usually uniform grid . the poisson equation is then solved on this grid using one of a number of the fast solvers that are available , usually fast fourier transforms . this is an analog of plasma particle - in - cell ( pic ) codes . it is believed that simulating many billions of stars in actual galaxies by using only several ten or hundred thousands representative stars in pic experiments will be enough to capture the essential physics , which includes the ordinary star star binary relaxation and wave - like collective motion . in other fields , such as the simulation of spiral structures , pic codes may be used with moderate success . this is because these fine - scale @xmath5 kpc structures can basically be governed by collisionless processes . by increasing the number of cells to reproduce the microstructures , one reduces the average number of particles per cell , and thus increases the undesirable effect of particle gravitational encounters . in addition , if one uses pic codes , then one can not resolve density fluctuations with wavelengths smaller than the size of a cell @xmath6 . this puts a lower limit on the wavelengths that it makes any sense to keep ; this lower limit is given by @xmath7 . other disadvantages of pic codes are possibilities of late - time loss of energy conservation , finite - grid and aliasing instabilities , and other numerical artifacts . pic method works well only for regions where there are many particles per cell . the problem is serious because there are regions in the phase space which are important to the problem but in which the distribution function is small . in general , since we will at most have a hundred particles in one cell we will have fluctuation of the order of @xmath8 or more which for many problems is unacceptable . in another type of @xmath3-body simulation , ostriker and peebles ( 1973 ) , morozov ( 1981 ) , grivnev ( 1985 ) , griv _ et al . _ ( 1994 ) , griv and zhytnikov ( 1995 ) , griv and chiueh ( 1998 ) investigated a stellar disk by direct integration of newtonian equations of motion . following them , we simulate here the evolution of a model for an isolated thin disk of a galaxy by direct integration of the equations of motion of stars : the force on a star @xmath9 due to all other stars is given by @xmath10^{3/2 } } , \ ] ] where @xmath11 is the constant of gravitation , @xmath12 , @xmath13 , @xmath14 and @xmath15 are the mass , position and velocity of the @xmath9-th and of the @xmath16-th particle , respectively , and @xmath17 is the so - called cutoff radius . this softening " of the gravitational potential is a device often used in @xmath3-body simulations to avoid numerical difficulties at very close but rare encounters ( miller , 1971 ; romeo , 1998 ) . the constant parameter @xmath18 reduces the interaction at short ranges and puts a lower limit on the size of the particles , i.e. the stars in the system can no longer be considered as point - masses they are in fact plummer spheres with a scale size @xmath18 . in addition , the softening parameter @xmath18 leads to a more collisionless " situation . as is known , the latter @xmath3-body models suffer from graininess due to the finite number of particles . the graininess manifests itself as a numerical noise " in the calculation . the use of many model particles @xmath19 decreases drastically the noise level . although the direct integration method sounds simple and straighforward , practical computational limitations require the use of only few ten thousands representative stars . this is because we evaluate the sum on the right - hand side of eq . ( 1 ) for each star or @xmath3 times for @xmath3 stars . the sum itself contains @xmath3 terms ; each term requires a number of arithmetic operations for the sake of estimating , let us say , ten . the total number of arithmetic operations required to evaluate the force will be of the order of @xmath20 . for the standard runge kutta calculation involving @xmath21 stars the total number of operations would be about @xmath22 . assuming that we can do an operation in @xmath23 sec , simply evaluating the forces would require about @xmath24 min . a typical calculation requires several thousand time steps so that @xmath25 h would be required . calculations of this magnitude are totally impractical for using such models to explore the physics of galaxies . the problem is solved by using the nowadays concurrent supercomputers : with the multiple processors in supercomputers it is possible to reduce the running time by processing more than one group of particles at a time . the use of , say , @xmath26 parallel computers would make such calculations practical . in the present study , we explore the inter university @xmath27 processors sgi origin @xmath28 supercomputer in israel . note that pic codes only formally utilized a larger number of particles than direct integration codes . this is because in pic experiments , the disk plane was covered with a difference grid . the stars within grid cells as well as within a whole model did not interact gravitationally ; the computation of the mass density in cells followed a fast fourier transform to obtain the potential only at a restricted set of points at the cell centers by interpolation rules and thus to find the very approximate forces at the positions of the particles . this avoids the need to compute forces between particle pairs , and makes the amount of computation necessary to obtain the forces independent of @xmath3 . clearly , nothing similar exists in nature , and therefore it follows that one can not compare the number of particles used in these different kinds of simulations in any way at all . modern parallel computing technology on concurrent processors had led to a dramatic increase in the computing power . the latter allows us to use direct integration methods with a sufficiently large number of model stars @xmath0 . in general , the numerical procedure is first to seek stationary solutions to the boltzmann kinetic equation in the self - consistent field approximation for the density distribution , the angular velocity of regular rotation , the dispersion of random velocities of the stars , and then to calculate the stability of those solutions to small - amplitude gravity perturbations . at the start of the @xmath3-body integration , our simulation initializes the particles on a set of @xmath26 circular rings with a circular velocity @xmath29 of galactic rotation in the @xmath30plane ; the system is isolated in vacuum . consider an uniformly rotating model disk of stars with a surface mass density variation given by @xmath31 where @xmath32 is the central surface density and @xmath33 is the radius of the initial disk . as a solution of a time - independent collisionless boltzmann equation , to ensure initial equilibrium the uniform angular velocity to balance the zero - velocity - dispersion disk , @xmath34 was adopted ( hohl , 1971 , 1972 ) . then the position of each particle was slightly perturbed by applying a pseudorandom number generator . for the uniformly rotating disk , the maxwellian - distributed random velocities with radial @xmath35 and azimuthal @xmath36 dispersions in the plane @xmath37 according to the familiar toomre s criterion ( toomre , 1964 , 1977 ) , @xmath38 may be added ( @xmath39 ) to the initial circular velocities @xmath40 , where @xmath41^{1/2}$ ] is the ordinary epicyclic frequency ( hohl , 1971 , 1972 ) . it is crucial to realize that in this case , according to lau and bertin ( 1978 ) , lin and lau ( 1979 ) , bertin ( 1980 ) , lin and bertin ( 1984 ) , morozov ( 1980 ) , polyachenko ( 1989 ) , griv and peter ( 1996 ) , polyachenko and polyachenko ( 1997 ) , initially the disk is jeans - stable against the small - scale axisymmetric ( radial ) perturbations but unstable against the relatively large - scale nonaxisymmetric ( spiral ) perturbations . the initial vertical velocity dispersion was chosen equal to @xmath42 . finally , the angular velocity @xmath43 was replaced by ( hohl , 1971 , 1972 ) @xmath44 \right\}^{1/2 } .\ ] ] the sense of disk rotation was taken to be counterclockwise , and units are such that @xmath45 and the mass of a star @xmath46 . the cutoff radius was chosen @xmath47 . within the framework of our model , dynamically young stars form the very thin disk @xmath48 . therefore , at the start of simulations the disk thickness was chosen @xmath49 . a time @xmath50 is taken to correspond to a single revolution of the initial disk . in all experiments the simulation was performed up to a time @xmath51 between @xmath52 and @xmath53 . it should be noted here that after about three rotations the picture is always stabilized and practically no significant changes in gross properties of the model over this time are observed . any difference between the results of simulations with and without applying the so - called quiet starts procedure to select the very regular initial coordinates of particles was not found . ( basically , by applying the method of quiet starts , one uses no random numbers in the initial conditions to suppress the noise level in a system . such techniques have proven useful in obtaining realistic noise levels without the use of a large number of particles ; byers and grewal , 1970 . ) moreover , tests indicated that the results were insensitive to changes in other parameters , e.g. the number of stars in the range @xmath54 , the cutoff parameter in the range @xmath55 and the initial disk thickness in the range @xmath56 . spiral amplitudes detected in our simulations are greatly in excess of an expectation from the level of particle noise and do not scale with @xmath3 . this behavior is clearly inconsistent with the hypothesis that the nonaxisymmetric structure in the models can be attributed to collisional effects , inappropriate values of @xmath18 or swing - amplified particle noise as advocated by toomre ( 1990 ) . finally , slight corrections have been applied to the resultant velocities and coordinates of the model stars so as to ensure the equilibrium between the centrifugal and gravitational forces , to preserve the position of the disk center of gravity at the origin and to include the weak effect of the finite thickness of the disk to the gravitational potential . thus , the initial model is very near the dynamical equilibrium for all radii . the system of equations of motion ( 1 ) for @xmath3 particles was integrated by the standard runge kutta method of the fourth order . all the particles move with the same constant runge kutta time step @xmath57 , where @xmath58 is the orbital period . in order to be compared with an actual galaxy of stars , the numerical model of the disk should properly simulate a collisionless system . otherwise , as it has been shown , e.g. by morozov _ et al . _ ( 1985 ) and griv _ et al . _ ( 2000a ) , a collisional disk is unstable with respect to a secular dissipative - type instability . the secular instability of spontaneous gravity perturbations might arise merely from the dissipation of the energy of regular rotation into ever larger amounts of heat when the system has moderate dissipation ; it results from perturbations which have negative energy in the dissipative medium ( morozov _ et al . _ , 1985 ; fridman and polyachenko , 1984 , vol . 2 , p. 243 ) . the secular instability produces structures completely unrelated to the effects we would like to model . in the spirit of the simple molecular - kinetic " theory by chandrasekhar ( 1960 ) , the collisional relaxation time for a three - dimensional system from the point of view of deflections suffered by a test star in encounters with a given distribution of field objects is @xmath59 where @xmath60 is the averaged velocity dispersion , @xmath61 is the field " particle mass and @xmath62 its three - dimensional number density . also @xmath63 is the so - called newton ( or coulomb in plasmas ) logarithm , by means of which the long - range nature of the gravitational force is taken into account ; to the order of magnitude @xmath64 , where @xmath65 is the total number of field particles . in actual galaxies @xmath66 . according to eq . ( 5 ) , the chosen quantities @xmath60 , @xmath3 , etc . guarantee that the model may be considered as a collisionless one to a good approximation at least during @xmath67 rotations . so we expect , in our numerical calculations because @xmath3 is large , relaxation from two - body gravitational interactions is unimportant . we suggest that the structures observed in the @xmath3-body simulations originate from the collective modes of practically collisionless galactic models gravitational jeans - type modes and bending firehose - type modes ( see below section 4.2 ) . consider a system of mutually gravitating particles . the local distribution function @xmath68 must satisfy the boltzmann kinetic equation @xmath69 where @xmath70 is the total gravitational potential determined self - consistently from the poisson equation , @xmath71 @xmath72 is the so - called collisional integral which defines the change of @xmath73 arising from ordinary interparticle collisions , @xmath74 is the collision frequency and @xmath75 is the quasi - steady state distribution function ( binney and tremaine , 1987 , p. 506 ; griv _ et al . _ , 2000a ) . in plasma physics , lifshitz and pitaevskii ( 1981 , p. 115 ) have discussed phenomena in which interparticle collisions are unimportant , and such a plasma is said to be collisionless ( and in the lowest - order approximation of the theory one can neglect the collision integral in the kinetic equation ) . it was shown that a necessary condition is that @xmath76 , where @xmath77 is the typical frequency of collective oscillations : then the collision operator in the kinetic equation is small in comparison with @xmath78 . in the appendix , we have shown that the frequency of jeans oscillations in a stellar disk @xmath79 . therefore , in the gravitation case in the lowest - order approximation of the theory we can neglect the effects of collisions between particles on a time scale of many rotations if @xmath80 . lifshitz and pitaevskii have pointed out that collisions may be neglected also if the particle mean free path is large compared with the wavelength of collective oscillations . then the collision integral in eq . ( 6 ) is small in comparison with the term @xmath81 . in this subsection , we test numerically if the models used in our @xmath3-body simulations are being correctly modeled as collisionless boltzmann ( vlasov ) systems . the direct method of checking if the system is being modeled as a collisionless system is to repeat a calculation using a mass spectrum ( rybicki , 1971 ) . it is obvious that as a result of gravitational collisions there is a tendency towards energy equipartition between the various masses . hohl ( 1973 ) has determined the experimental relaxation time and compared it with a theoretical prediction for the collisional relaxation time of a two - dimensional disk . here we do such a comparison by using the three - dimensional system . let us consider the three - dimensional computer model consisting of @xmath82 stars of mass @xmath83 , @xmath84 stars of mass @xmath85 and @xmath86 stars of mass @xmath87 . the total number of stars , which are distributed in the system in accord with the law ( 2 ) is @xmath88 . initially , the different mass groups of stars are distributed with the same velocity dispersion ( with different temperatures ) . in the appendix of the present paper we demonstrate that the disk becomes almost stable gravitationally for @xmath89 . in such a jeans - stable system collective effects associated with the classical gravitational instability will not affect the random velocity dispersion of particles : then the change of velocity dispersion can be explained only by usual two - body encounters . for this reason the initial condition was chosen to be a quasi - stable uniformly rotating disk with @xmath90 . following chandrasekhar ( 1960 ) and hohl ( 1973 ) , let us define the relaxation time @xmath91 as the time required for the mean change of the kinetic energy per unit mass of the test star to equal the initial kinetic energy . in figure 1 we show the change of the ratio of the mean particle ( kinetic ) energy , @xmath92 , @xmath93 and @xmath94 ( in units of the total kinetic energy of the system ) , for the different mass groups , where @xmath95 is the total velocity of a given mass group . as is expected , the two groups of heavy stars lose energy while the group of lightest stars gains an approximately corresponding amount of kinetic energy . also as is expected , one can see the decrease in the change of the kinetic energy with time . this is because the collisional frequency @xmath96 is inversely proportional to the velocity dispersion ( eq . [ 5 ] ) , and thus the encounters only weakly affect the stars with high random velocities . as one can see , the mean slope of the curves shown in figure 1 will result in energy equipartition after about @xmath97 rotation periods . it is crucial to realize that these relaxation times even for this relatively small number of model stars are much longer that the time of a single disk revolution . we conclude that the computer models used in the present study may be considered as collisionless ones to a good approximation at least during the first @xmath53 rotations which are of especial interest in spiral - galaxy simulation . therefore , we argue that the collective effects studied in the paper were apparent before the collisional time scale associated with the binary interactions of stars was reached . to study the development of spiral structure and to confirm the theoretical predictions made in the appendix , warm disks with initial radial velocity dispersion of identical stars equal to toomre s ( 1964 ) stabilizing velocity dispersion @xmath98 were modeled . figure 2 shows an example of the development of spiral structures . the structure grows rapidly on a dynamical time scale @xmath99 . at a time @xmath100 a multi - armed filamentary spiral structure forms with several short , tightly wound trailing arms . soon after , at @xmath101 a bar structure appears with two almost symmetrical open spiral arms . however , sometimes , e.g. at @xmath102 or at @xmath103 , one can see a one - armed " galaxy consisting of a prominent arm and a short , diffuse second one . after @xmath104 rotations , the spiral arms almost disappear , and a bar forms in the central parts of the system . the @xmath105 instability shifts the point with highest density from the center of mass . it is interesting to note that in many disk - shaped galaxies , e.g. in the spiral galaxies m 101 and ngc 1300 , there appears to be a deviation from rotational symmetry . in principle , such a deviation may be due to this one arm instability . note that the latter has already been suggested by b. lindblad ( see toomre , 1996 ) . in fact , asymmetries in the distribution of light in spiral galaxies have been known for a long time . et al . _ ( 1980 ) presented a number of spiral galaxies in which the distribution of neutral gas is lopsided , i.e. the gas extends further out on one side of the galaxy than on the other . in many systems asymmetries are also found to be well developed in the stellar disks as indeed confirmed recently by near - infrared observations ( zaritsky and rix , 1997 ) . rudnick and rix ( 1998 ) quantified the mean asymmetry of 54 early - type disk galaxies using the amplitude of the @xmath105 azimuthal fourier component of the @xmath33-band surface brightness . it was found that @xmath106 of all disk galaxies have azimuthal asymmetries in their stellar disk disribution , confirming lopsidedness as a dynamical phenomenon . based on the @xmath3-body simulations , we can offer a tentative hypothesis which accounts lopsided asymmetries in spiral galaxies by the @xmath105 jeans - unstable spiral mode . following athanassoula and sellwood ( 1986 ) , in figure 3 the time evolution of the fourier spectrum of the spiral pattern shown in figure 2 is plotted . only the azimuthal @xmath107 components are shown . the complex fourier coefficients , @xmath108 , are determined from the following summation @xmath109 \ } , \ ] ] where @xmath3 is the number of particles and @xmath110 are the polar coordinates of the @xmath16-th particle . the pitch angle of an @xmath111-armed logarithmic spiral @xmath112 is given by @xmath113 , and positive @xmath114 corresponds to trailing spirals and negative @xmath114 to leading ones . this logarithmic spiral fourier analysis shows that the peaks of the signals move to positive @xmath114 with increasing time , reach maxima at @xmath115 , and eventually decay . it follows , then , that after a time @xmath116 the model becomes practically stable with respect to gravitational jeans - type modes . we can observe that the @xmath117 modes are the most unstable ones . an interesting feature is that all these modes have roughly the same pitch angle . the amplitudes of the signals grow aperiodically , supporting the assumption that we have deal with the aperiodic jeans - type instability of gravity disturbances . in actual galaxies , discrete spiral modes were already found by rix and zaritsky ( 1995 ) , block and puerari ( 1999 ) and block _ et al . _ it was stated that in the near - infrared , the morphology of older star - dominated disk indicates a simple classification scheme : the dominant fourier @xmath111-mode in the dust penetrated regime , and the associated pitch angle . a ubiquity of low @xmath105 and @xmath118 modes was confirmed . ( 1998 ) detected @xmath119 spiral modes in relatively young stellar population of the nearly face - on galaxy ngc 3631 from observations in the h@xmath120 line . from the edge - on view pictured in figure 4 , one can see that from an initial very thin model , a fully three - dimensional disk develops immediately at @xmath121 with a mean height @xmath122 above the plane , corresponding to the force balance between the gravitational attraction in the plane and the pressure " due to the velocity dispersion ( i.e. temperature " ) in the vertical , @xmath123-direction . a straightforward estimate shows that @xmath124 where @xmath125 is the three - dimensional mass density in the mid - plane . after a time @xmath121 there is no change in the edge - on structure until at @xmath126 ( when the jeans instability is almost switched off by increasing the planar velocity dispersion ) . at somewhat later times , @xmath127 , a box - shaped " or peanut - shaped " bar structure is developed . the development of the box / peanut bar structure is explained in the terms of the bending firehose - type instability ( raha _ et al . _ , 1991 ; merritt and sellwood , 1994 ; griv and chiueh , 1998 ) . one sees that at times @xmath128 , the bending instability fiercely develops in the central , almost rigidly rotating parts of the system and is switched off at @xmath101 . at later times , no dramatic evolution is observed in our simulations . to emphasize , this type of bending instabilities develops in the central , rigidly rotating parts of the system . the three - dimensional simulations show that the bending instabilities mentioned above destroy the strong bar formation ( griv and chiueh , 1998 ) . note that the bending modes have been found experimentally by merritt and hernquist ( 1991 ) and merritt and sellwood ( 1994 ) , but in the nonrotating models . ( 1990 ) and raha _ et al . _ ( 1991 ) found the bending instability in the central regions of a fast - rotating @xmath3-body disk . it was stated that this instability may play a part in the formation of triaxial bulges in flat galaxies and explanation was given why many bulge stars are less than 10 gyr old ( raha _ et al . _ , 1991 ) . bureau and freeman ( 1997 ) have argued that new observations constitute a strong case in favor of the bending ( or bar - buckling ) mechanism for the formation of boxy / peanut - shaped bulges in spiral galaxies . observations suggest very strongly that many of ordinary spiral galaxies , including the milky way , have similar small triaxial bars in their central regions ( dwek _ et al . _ , 1995 ) . apparently , this instability of a sufficiently thin stellar disk has been predicted by toomre ( 1966 ) by using the simplified theory based on moment equations . toomre considered the collisionless analog of the kelvin helmholtz instability in an infinite , two - dimensional , nonrotating sheet of stars . this is the usual way to discuss the conditions of the firehose instability in plasma physics ( krall and trivelpiece , 1986 ) . it was demonstrated by toomre that the bending instability is driven by the stellar pressure " anisotropy : the source of free energy in the instability is the intrinsic anisotropy of a velocity dispersion ( temperature " ) . the bending firehose - type instability was also discovered independently by kulsrud _ et al . _ ( 1971 ) and mark ( 1971 ) with a more accurate kinetic theory . et al . _ have described the mechanism of the instability in terms of the balance of centrifugal force acting on the stars of a bending layer and the gravitational attraction toward the plane of the practically nonrotating , pressure - supported disk . fridman and polyachenko ( 1984 ) have discussed the role of the instability in explaining the existence of maximum oblateness in almost nonrotating elliptical galaxies and the formation of the bulges of disk - shaped galaxies of stars . as is seen in figures 2 and 4 , the basic form of warm disk evolution is the expansion of the outermost parts together with the contraction of a large part of the mass towards the center ; in the final quasi - steady state the surface density decreases exponentially outwards ( fig . 5 ) . in a future publication we intend to explain the exponential surface density distribution of the final quasi - steady state model by applying the quasi - linear theory of a stellar disk s stability . note only that the study of surface photometry have been shown that most spiral and s0 galaxies have an exponential disk with radial surface - brightness distribution @xmath129 ( freeman , 1970 ) . at first , the surface density and rotation velocity change rapidly ( figs . 5 and 6 ) , but after a time @xmath130 there is little further change during the following two rotations , except in the central portions of the disk . the model generates a nonflat rotation curve ( fig . 6 ) ; in addition growing density waves ( spiral arms ) transfer the angular momentum of the circular motion outwards ( see the difference in the rotation curves at @xmath131 and @xmath132 in figure 6 ) . as mentioned above , the model practically shows no the time evolution of the surface density and the velocity curve after @xmath133 . the latter seem to indicate that at that time the system has nearly reached a quasi - steady state ( cf . hohl , 1971 ) . the time evolution of the velocity dispersions @xmath35 , @xmath36 and @xmath134 is shown in figure 7 . initially , during the first rotation the velocity dispersions along each axis are changed rapidly . during the following two rotations the velocity dispersions are changed with time slightly , confirming that between @xmath135 and @xmath136 the system has nearly reached a quasi - steady state . however , in the central portions of the disk the velocity dispersion @xmath134 increases rapidly even after the time @xmath137 , when the bending instability begins to operate . as was stated , one of the goals of the present research is to compare toomre s ( 1964 ) critical dispersion ( 4 ) and the modified dispersion given by eq . ( 25 ) with the experimental radial velocity dispersion of the quasi - steady state model of a stellar disk . in what follows , we are doing such a comparison . as is seen in figure 8 , the modified critical dispersion , in contrast to toomre s ( 1964 ) critical dispersion , is in fair agreement with the experimental data , especially for the outer , differentially rotating parts of the disk . the agreement is quite reasonable considering the rough nature of the theory . in the central parts of the model the agreement with the theory is not good . this is to be expected since the condition of nearly circular orbits adopted in the appendix does not fulfill in the central almost nonrotating regions of model disks ( and in spiral galaxies , including both barred and normal galaxies , and in disks of elliptical galaxies ) . therefore in the central parts of the disk ( @xmath138 ) the condition for the local criteria ( both the toomre and the modified criterion ) to be applicable is violated . one of the important problems of stellar disks is the determination of the random velocity diffusion . such velocity diffusion can be caused by stellar disk turbulence . velocity diffusion produced by turbulence tells us something about the turbulence process . to compute velocity diffusion we calculate the mean - square spread in the planar random ( residual ) velocity @xmath139 as a function of time of model particles for different radii @xmath140 . in general @xmath141 has a time dependence like that shown in figure 9 . as is seen , in the differentially rotating ( @xmath142 ) , jeans - unstable ( at times @xmath143 ) parts of the disk approximately @xmath144 . this behavior has been predicted analytically ( see the appendix ) . interestingly , observations indicate a similar increase of @xmath141 with time in the solar neighborhood ( binney and tremaine , 1987 , p. 470 ; gilmore _ et al . _ , 1990 ) . the time evolution of the hot disk was investigated , in which the initial dispersion of random radial velocities of model stars was chosen to be greater than toomre s critical dispersion , namely , @xmath145 ( see eq . [ 24 ] ) . for such calculations we obtain plots like those shown in figure 10 . in agreement with the theory , the system becomes gravitationally ( jeans- ) stable . the linear lin shu density wave theory is considered now as well established ( lin and shu , 1966 ; lin _ et al . _ , 1969 ; shu , 1970 ; bertin , 1980 ; binney and tremaine , 1987 , p. 347 ; griv and peter , 1996 ; griv _ et al . _ , 2000b ) . however , as soon as the amplitude of the gravity perturbations becomes appreciable a new generation of problems , namely , nonlinear problems , become important . below we present a quantitative weakly nonlinear theory of the classical jeans instability of small - amplitude gravity disturbances ( e.g. those produced by a barlike structure , a spontaneous perturbation and/or a companion galaxy ) in self - gravitating , rapidly rotating stellar disks of flat galaxies . in the rotating frame of a disk galaxy , the collisionless motion of an ensemble of identical stars in the plane of the system can be described by the boltzmann equation for the distribution function @xmath68 without the integral of collisions ( lin _ et al . _ , 1969 ) : @xmath146 where @xmath147 are the galactocentric cylindrical coordinates , the total azimuthal velocity of the stars was represented as a sum of the random @xmath148 and the basic rotation velocity @xmath149 , and @xmath150 is the random velocity in the radial direction . the quantity @xmath151 denotes the angular velocity of galactic rotation at the distance @xmath140 from the center , and the epicyclic frequency @xmath152 varies from @xmath153 for a rigid rotation to @xmath154 for a keplerian one . random velocities are small compared with @xmath155 . collisions are neglected here because the collision frequency is much smaller than the cyclic frequency . in the kinetic equation ( 9 ) , @xmath70 is the total gravitational potential determined self - consistently from the poisson equation ( 7 ) . the equilibrium state is assumed , for simplicity , to be an axisymmetric and spatially homogeneous stellar disk . secondly , in our simplified model , the perturbation is propagating in the plane of the disk . this approximation of an infinitesimally thin disk is a valid approximation if one considers perturbations with a radial wavelength that is greater than the typical disk thickness . we assume here that the stars move in the disk plane so that @xmath156 . this allows us to use the two - dimensional distribution function @xmath157 such that @xmath158 , where @xmath159 is the surface density . we expect that the waves propagating in the disk plane have the greatest influence on the development of structures in the disk . the latter suggestion is strongly supported by numerical simulations ( hohl , 1978 ) . the disk in the equlibrium is described by the following equation : @xmath160 where @xmath161 and the angular velocity of rotation @xmath162 is such that the necessary centrifugal acceleration is exactly provided by the central gravitational force @xmath163 , where @xmath164 is the axisymmetric equilibrium potential . equation ( 10 ) does not determine the equilibrium distribution @xmath165 uniquely . for the present analysis we choose @xmath165 in the form of the anisotropic maxwellian ( schwartzschild ) distribution @xmath166 the schwarzschild distribution function is a function of the two epicyclic constants of motion @xmath167 and @xmath168 , where @xmath169 . these constants of motion are related to the unperturbed star orbits : @xmath170 ; \quad v_r = v_\perp \cos \left ( \phi_0 - \kappa t \right ) ; \nonumber \\ & & \varphi = \frac{2\omega}{\kappa } \frac{v_\perp}{r_0 \kappa } \left [ \cos ( \phi_0 - \kappa t ) - \cos \phi_0 \right ] ; \nonumber \\ & & v_\varphi \approx r_0 \frac{d\varphi}{dt } + r_0 \frac{v_\perp}{\kappa}\frac{d\omega } { dr } \sin \left ( \phi_0 - \kappa t \right ) \approx \frac{\kappa}{2\omega } v_\perp \sin \left ( \phi_0 - \kappa t \right ) , \label{eq : orbits } \end{aligned}\ ] ] where @xmath171 , @xmath172 are constants of integration , @xmath173 , @xmath174 is the mean epicycle radius , and we follow lin _ et al . _ ( 1969 ) , shu ( 1970 ) and lynden - bell and kalnajs ( 1972 ) , making use of expressions for the unperturbed epicyclic trajectories of stars ( 12 ) in the equilibrium central field @xmath175 . in eqs . ( 12 ) , @xmath176 is the radius of the circular orbit , which is chosen so that the constant of areas for this circular orbit @xmath177 is equal to the angular momentum imtegral @xmath178 , and @xmath179 . also , @xmath180 is the position angle on the circular orbit , @xmath181 . the quantities @xmath154 , @xmath152 and @xmath35 are evaluated at @xmath176 . in eq . ( 11 ) the fact is used that as follows from eqs . ( 12 ) in a rotating frame the radial velocity dispersion @xmath35 and the azimuthal velocity dispersion @xmath36 are connected through @xmath182 . in the solar vicinity , @xmath183 . the distribution function @xmath165 has been normalized according to @xmath184 , where @xmath185 is the equilibrium surface density . such a distribution function for the unperturbed system is particularly important because it provides a fit to observations ( lin _ et al . _ , 1969 ; shu , 1970 ) . we proceed by applying the standard procedure of the quasi - linear ( weakly nonlinear ) approach ( lifshitz and pitaevskii , 1981 ; alexandrov _ et al . _ , 1984 ; krall and trivelpiece , 1986 ) and decompose the time dependent distribution function @xmath186 and the gravitational potential @xmath187 with @xmath188 and @xmath189 for all @xmath190 and @xmath51 . the functions @xmath191 and @xmath192 are oscillating rapidly in space and time , while the functions @xmath75 and @xmath193 describe the slowly developing background " against which small perturbations develop ; @xmath194 and @xmath195 . the distribution @xmath75 continues to distort as long as the distribution is unstable . linearizing eq . ( 9 ) and separating fast and slow varying variables one obtains : @xmath196 @xmath197 where @xmath198 means the total derivative along the star orbit ( 12 ) and @xmath199 denotes the time average over the fast oscillations . to emphasize it again , we are concerned with the growth or decay of small perturbations from an equilibrium state . in the epicyclic approximation , the partial derivatives in eqs . ( 13 ) and ( 14 ) transform as follows ( lin _ et al . _ , 1969 ; shu , 1970 ; morozov , 1980 ; griv and peter , 1996 ) : @xmath200 to determine oscillation spectra , let us consider the stability problem in the lowest wkb approximation ; this is accurate for short wave perturbations only , but qualitatively correctly even for perturbations with a longer wavelength , of the order of the disk radius @xmath33 . in this local wkb approximation in eqs . ( 13 ) and ( 14 ) , assuming the weakly inhomogeneous disk , the perturbation is selected in the form of a plane wave ( in the rotating frame ) : @xmath201 where @xmath202 , @xmath203 are amplitudes that are constant in space and time , @xmath111 is the nonnegative azimuthal mode number (= number of spiral arms ) , @xmath204 is the doppler - shifted wavefrequency and @xmath205 ( lin _ et al . _ , 1969 ; shu , 1970 ; griv and peter , 1996 ) . the solution in such a form represents a spiral wave with @xmath111 arms whose shape in the plane is determined by the relation @xmath206 with @xmath207 increasing in the rotation direction , we have @xmath208 for trailing spiral patters , which are the most frequently observed among spiral galaxies . a change of the sign of @xmath209 corresponds to changing the sense of winding of the spirals , i.e. leading ones . with @xmath210 , we have the density waves in the form of concentric rings that propagate away from the center when @xmath211 or toward the center when @xmath212 . in eq . ( 13 ) using the transformation of the derivatives @xmath213 and @xmath214 given by eqs . ( 15 ) , one obtains the solution @xmath215 where @xmath216 . in this equation making use of the time dependence of perturbations in the form of eq . ( 16 ) and the unperturbed trajectories of stars ( 12 ) in the exponential factor occurring in the formula ( 18 ) , it is straightforward to show that @xmath217 where @xmath218 is the bessel function of the first kind of order @xmath219 , @xmath220 , @xmath221 \sin^2 \psi \}^{1/2}$ ] is the effective wavenumber , @xmath112 is the pitch angle of perturbations , @xmath222 , and we used the expansion @xmath223 and the usual bessel function recursion relation @xmath224 . in eq . ( 19 ) , the denominator vanishes when @xmath225 . this occurs near corotation ( @xmath226 ) and other resonances ( @xmath227 ) . the lindblad resonances occur at radii where the field @xmath228 resonates with the harmonics @xmath229 ( inner resonance ) and @xmath230 ( outer resonance ) of the epicyclic ( radial ) frequency of equilibrium oscillations of stars @xmath152 . clearly , the location of these resonances depends on the rotation curve and the spiral pattern speed ; the higher the @xmath111 value , the closer in radius the resonances are located ( lin _ et al . _ , 1969 ; shu , 1970 ) . in this paper , only the main part of the galactic disk is studied which lies sufficiently far from the resonances : below in all equations @xmath231 . in the wkb approximation , the linearized poisson equation ( 7 ) is readily solved to give the improved lin shu expression for the perturbed surface density of the two - dimensional disk ( lau and bertin , 1978 ; lin and lau , 1979 ) : @xmath232 on the other hand , integrating eq . ( 19 ) over velocity space @xmath233 , and equating the result to the perturbed density given by the improved lin shu solution ( eq . [ 20 ] ) , the dispersion relation @xmath234 is obtained : @xmath235 where @xmath236 , @xmath237 is a bessel function of imaginary argument with its argument @xmath238 , this dispersion relation generalizes the lin shu kalnajs one for nonaxisymmetric perturbations ( @xmath239 ) propagating in a homogeneous disk excluding the resonance zones ( @xmath240 ) . analyzing eq . ( 21 ) , it is useful to distinguish between two limiting cases : the cases of of epicyclic radius that is small compared with wavelength , @xmath241 , and of epicyclic radius that is large compared with wavelength , @xmath242 . we study the problem when the frequency of disk oscillations is smaller than the epicyclic frequency . in the opposite case of the high perturbation frequencies , @xmath243 , the effect of the disk rotation is negligible and therefore not relevant to us . this is because in this rotationless " case the star motion is approximately rectilinear on the time and length scales of interest which are the wave growth / damping periods and wavelengths , respectively ( cf . _ , 1984 , p. 113 ) . thus , the terms in series ( 19 ) and ( 21 ) for which @xmath244 may be neglected , and consideration will be limited to the transparency region between the turning points in a disk , i.e. between the inner and outer lindblad resonances . the @xmath226 harmonic that defines the corotation resonance dominates in series . thus , @xmath245 less than the epicyclic frequency of any disk star . in this case , in eq . ( 21 ) function @xmath246 starts from @xmath247 , reaches a maximum @xmath248 at @xmath249 , and then decreases . hence , the growth rate of the instability has a maximum at @xmath250 ( griv _ et al . _ , 1999b , fig . 2 in their paper ) . in addition , we consider only low-@xmath111 perturbations ( which are important only for the problem of spiral structure ) : from now on in all equations @xmath251 . using the well - known asymptotic expansions of bessel functions , the simplified @xmath252 dispersion relation reads @xmath253 where @xmath254 is the squared jeans frequency and @xmath255 is the so - called reduction factor , which takes into account the fact that the wave field only weakly affects the stars with high random velocities . the latter means that the spiral structure is seen more distinctly in objects with the lowest velocity dispersion . in the limit @xmath256 , @xmath257 $ ] ( but , of course , in order to be appropriate for a wkb wave we consider the perturbations with @xmath258 ) . in the opposite limit @xmath242 , @xmath259 $ ] . different forms of @xmath260 are given by athanassoula ( 1984 ) . the aperiodic jeans instability ( gravitational collapse ) is possible when @xmath261 . that is , the real part of the frequency of this unstable motion vanishes in a rotating frame . the jeans instability is driven by a strong nonresonant interaction of the gravity fluctuations with the bulk of the particle population , and the dynamics of jeans perturbations can be characterized as a fluid - like interaction . therefore , such an instability , which implies the displacement of macroscopic portions of a gravitating system , may be also analyzed through the use of relatively simple hydrodynamic equations ( lau and bertin , 1978 ; lin and lau , 1979 ; lin and bertin , 1984 ) . in general , the growth rate of the jeans instability is relatively high @xmath262 ; perturbations with wavelength @xmath263 have the fastest growth rate ( griv _ et al . _ , 1999a , b ; griv _ et al . _ , 2000a , b ) . in the solar vicinity of the galaxy , @xmath264 kpc . equation ( 22 ) has two roots which describe two branches of oscillations : long - wavelength , @xmath256 , and short - wavelength , @xmath242 , jeans branches . the dispersion law of low-@xmath111 jeans oscillations is @xmath265 here @xmath266 for jeans - stable ( @xmath267 ) perturbations and @xmath268 for jeans - unstable ( @xmath269 ) perturbations . the jeans perturbations can be stabilized by the random velocity spread . indeed , if one recalls that such unstable perturbations are possible only when @xmath270 , then by using the condition @xmath271 for all possible @xmath272 , from eq . ( 22 ) a local stability criterion to suppress the instability of arbitrary jeans - type gravity perturbations can be easily obtained ( griv _ et al . _ , 1999a , b ; griv _ et al . _ , 2000b ) : @xmath273 \sin^2 \psi \}^{1/2 } , \label{eq : criterion}\ ] ] where @xmath35 is the radial - velocity dispersion , @xmath274 is the critical radial velocity dispersion and usually in the main part of a disk - shaped galaxy the parameter @xmath275 . it is clear from this local criterion that stability of nonaxisymmetric jeans perturbations @xmath276 in a differentially rotating disk @xmath277 requires a larger velocity dispersion than toomre s critical value @xmath98 . it is only for rigidly rotating disks ( @xmath278 ) and/or for axisymmetric perturbations ( @xmath279 ) that the critical value of @xmath35 is @xmath98 . hence , the ordinary toomre s velocity dispersion @xmath98 should stabilize only small - scale axisymmetric ( radial ) perturbations of the jeans type . the differentially rotating disk however is still unstable against relatively large - scale nonaxisymmetric ( spiral ) modes with @xmath280 , in particular very open modes with @xmath281 . according to polyachenko ( 1989 ) and polyachenko and polyachenko ( 1997 ) the marginal stability condition for jeans perturbations of an arbitrary degree of axial asymmetry has been available since 1965 ( goldreich and lynden - bell , 1965 ) , though in a slightly masked form . a relationship exists between eq . ( 24 ) and what toomre ( toomre , 1981 ; binney and tremaine , 1987 , p. 375 ) called swing amplification . " summarizing , for jeans - stable differentially rotating stellar disks , the critical value of the radial velocity dispersion must be greater than ( although of the order of ) @xmath98 . in accordance with eq . ( 24 ) , @xmath282 might be the new " toomre s velocity dispersion to suppress all jeans perturbations in a stellar disk , including the most dangerous open ones . note that the destabilizing effect of tangential ( @xmath283 ) gravitational forces ( pitch angle dependent effect ) was first considered by lau and bertin ( 1978 ) and lin and lau ( 1979 ) using a much simpler gaseous approach , and a stability criterion that is similar to eq . ( 25 ) was also derived . clearly , the criterion ( 25 ) is only approximate one . in the present study , we test the validities of this modified criterion for stability of jeans perturbations in a self - gravitating , infinitesimally thin and almost collisionless disk of stars . correspondingly , for jeans - stable differentially rotating stellar disks , the critical value of the widely used toomre s instability parameter @xmath284 must be greater than @xmath285 . ( toomre s @xmath286-value is a measure of the ratio of thermal and rotational stabilization to self - gravitation . ) observations already demonstated that stellar ( and gaseous ) disks of a large number of galaxies , including milky way s disk , are near the boundary of the gravitational stability with @xmath286 between 2 and @xmath287 over a large range of galactic disks ( fridman _ et al . _ , 1991 ; bottema , 1993 ) . a general impression of how the spectrum of nonaxisymmetric jeans perturbations behaves in a homogeneous nonuniformly rotating disk can be gained from figure 11 , which shows the dispersion curves in the cases of jeans - unstable systems ( ( a ) and ( b ) ) , a marginally jeans - stable system ( c ) and a jeans - stable one ( d ) for values of @xmath288 ( as determined on a computer from eq . [ 21 ] ) . in this figure , the ordinate is the effective wavenumber @xmath289 measured in terms of the inverse epicyclic radius @xmath174 and the abscissa is @xmath290 , i.e. the dimensionless angular frequency at which the stars meet with the pattern , measured in terms of the epicyclic frequency @xmath152 . in general , for fixed wavefrequency @xmath291 there are two solutions in @xmath292 , comprising a long - wavelength wave , @xmath293 , and a short - wavelength wave , @xmath294 . a property of the solution ( 21 ) is that in a homogeneous system the jeans - stable modes those with @xmath295 are separated from each other by frequency intervals where there is no wave propagation : gaps occur between each harmonic ( cf . the bernstein modes in a magnetized plasma ) . the growth rate of the jeans instability of a homogeneous stellar disk @xmath296 as determined on a computer from the expression @xmath297 is shown in figure 12 . as one can see visually in this figure , the growth rates have maxima with respect to the critical mode number @xmath298 . it means that of all harmonics of initial perturbation , one perturbation with the maximum of the growth rate and with @xmath299 will be formed asymptotically in time . the low-@xmath111 spiral modes ( @xmath300 and @xmath301 ) are more unstable than the radial ones ( @xmath302 ) and the high-@xmath111 ones ( @xmath303 ) . these low-@xmath111 spiral modes are only important in the problem of galactic spiral structure because in contrast to the high-@xmath111 modes , they do extend essentially over a large range of the galactic disk ( lin _ et al . _ , 1969 ; shu , 1970 ; griv and peter , 1996 ) . the shape and the number of spiral arms depend on the equilibrium parameters of a galaxy . for the galaxy the most probable pattern is that of @xmath304 arms , the radial distance between the arms being @xmath305 kpc . such a pattern corresponds to a spiral wave mode with maximum of growth rate . next we substitute the solution ( 19 ) into eq . taking into account that the terms @xmath306 vanish for axially symmetric functions @xmath75 , after averaging over @xmath172 we obtain the equation for the slow part of the distribution function : @xmath307 as usual in the weakly nonlinear theory ( or weak turbulence theory ) , in order to close the system one must engage an equation for @xmath308 . averaging over the fast oscillations , we have @xmath309 where suffixes @xmath310 denote the @xmath310th fourier component . equations ( 26 ) and ( 27 ) form the closed system of weakly nonlinear equations for jeans oscillations of the rotating homogeneous disk of stars , and describe a diffusion in velocity space . the spectrum of oscillations and their growth rate are given by eq . ( 21 ) and @xmath311 , respectively . thus after a time @xmath312 the jeans - unstable disk will assume the form described by eq . ( 17 ) . in the solar vicinity , @xmath313 yr . a very important feature of the instability under consideration is the fact that it is aperiodic ( the real part of the wavefrequency vanishes in a rotating frame we are using ) . as an application of the theory we investigate the relaxation of low frequency and jeans - unstable , @xmath314 and @xmath315 , respectively , oscillations in the homogeneous galactic disk . indeed , already in the 1940s it was observed that in the solar neighborhood the random velocity distribution function of stars with an age @xmath316 yr is close to a schwarzschild distribution a set of gaussian distributions along each coordinate in velocity space , i.e. close to equilibrium along each coordinate . in addition , older stellar populations have a higher velocity dispersion than younger ones . on the other hand , a simple calculation of the relaxation time of the local disk of the milky way due to pairwise star star encounters brings the standard value @xmath317 yr ( chandrasekhar , 1960 ; binney and tremaine , 1987 , p. 489 ) , which considerably exceeds the lifetime of the universe . according to our approach , collisionless collective - type relaxation does play a determining role in the evolution of stellar populations of the galactic disk . evidently , the unstable fluid - like jeans oscillations must influence the distribution function of the main , nonresonant part of stars in such a way as to hinder the wave excitation , i.e. to increase the velocity dispersion . this is because the jeans instability , being essentially a gravitational one , tends to be stabilized by random motions ( griv and peter , 1996 ) . therefore , along with the growth of the oscillation amplitude , random velocities must increase at the expense of circular motion , and finally in the disk there can be established a quasi - stationary distribution so that the jeans - unstable perturbations are completely vanishing and only undamped jeans - stable waves remain : jeans instabilities are thought to heat disks up to velocity dispersion values of @xmath318 . greater than it follows from approximate criterion ( 25 ) . ] in the following , we restrict ourselves to the most dangerous , " in the sense of the loss of gravitational stability , long - wavelength perturbations , @xmath319 and @xmath320 ( see the explanation after eq . [ 21 ] ) . then in eqs . ( 21 ) and ( 26 ) one can use the expansions @xmath321 and @xmath322 . equation ( 26 ) takes the simple form @xmath323 where @xmath324 are functions of @xmath51 . as is seen , the velocity diffusion coefficient for nonresonant stars @xmath325 is independent of @xmath171 ( to lowest order ) . this is a qualitative result of the nonresonant character of the star s interaction with collective aggregates . by introducing @xmath326 and @xmath327 ( alexandrov _ et al . _ , 1984 ; krall and trivelpiece , 1986 ) , eqs . ( 27 ) and ( 28 ) are rewritten as follows : @xmath328 which has the solution @xmath329 ( we have taken into account the observations that the distribution of newly born stars is close to the @xmath330-function distribution , @xmath331 ; grivnev and fridman , 1990 . ) two general physical conclusions can be deduced from the solution ( 30 ) . first , during the development of the jeans instability , @xmath332 , the schwarzschild distribution of random velocities ( a gaussian spread along @xmath333 coordinates in velocity space ) is established . it sharp contrast to a normal gas with frequent interparticle collisions , it is the collisionless collective - type interactions between the stars which transform an arbitrary initial distribution into the maxwellian - like form . secondly , the energy of the oscillation field latexmath:[$\propto \sum_{\bf k } in the nonresonant - particle distribution . as the perturbation energy increases , the initially monoenergetic distribution spreads ( @xmath75 becomes less peaked ) , and the effective temperature ( or @xmath141 ) grows with time ( a gaussian spread increases ) : @xmath335 . that is , during the development of the jeans instability the planar mean - square velocity increases linearly with time . in section 4.2 of the present paper , we verify the @xmath336 dependence by @xmath3-body simulations . from the above , this nonlinear wave star interaction increases the velocity dispersion of stars in milky way s disk after they are born ( and of particles in jeans - unstable @xmath3-body models ) . subsequently , sufficient velocity dispersion prevents jeans instability from occuring . the diffusion " of nonresonant stars takes place because they gain mechanical ( oscillatory ) energy as the instability develops . the velocity diffusion , however , presumably tapers off as jeans stability is approached : the radial velocity dispersion @xmath35 becomes greater than the critical one @xmath337 . this is large enough to turn off the jeans instability in a stellar disk . thus , the true time scale for relaxation in the milky way may be much shorter than its standard value @xmath317 yr for the classical chandrasekhar ogorodnikov collisional relaxation ; it may be of the order @xmath338 yr , i.e. comparable with @xmath53 periods of the milky way rotation in the solar vicinity . the above relaxation time is in fair agreement with both observations and @xmath3-body simulations of milky way s disk ( binney and tremaine , 1987 , p. 478 ; grivnev and fridman , 1990 ; hohl , 1971 ) . the results show that a velocity dispersion given by toomre s ( 1964 ) criterion will stabilize a disk only against axisymmetric ring - like gravity perturbations . however , such disks are unstable against nonaxisymmetric spiral - like perturbations . jeans - unstable spiral perturbations heat " the system ; the mean - square random velocity increases linearly with time . the results are in agreement with analytical predictions . we were able to generate an axisymmetric , gravitationally stable disk . the initial condition for the axisymmetric stable disk was obtained analytically and confirmed experimentally : the radial dispersion of random velocities of stars should be equal ( or greater than ) to the modified dispersion at each radii . this work was performed in part under the auspices of the israeli ministry of immigrant absorption , the israel u.s . binational science foundation , the israel science foundation and the academia sinica in taiwan . the authors thank tzi - hong chiueh , arthur d. chernin , alexei m. fridman and shlomi pistinner for their interest in the work and valuable suggestions . in part , this work was done while one of the authors ( e. g. ) was a senior postdoctoral fellow at the asiaa during 19961997 . e. g. would like to thank yi - nan chin , minho choi , yi - jehng kuan , jeremy lim , kwok - yung lo and jun - hui zhao for many useful discussions during his two - year stay , and all the staff of asiaa for their hospitality .

hamilton cycles occupy a position of central importance in graph theory , and are the subject of countless results . the most famous is of course dirac s theorem @xcite , which states that a hamilton cycle can always be found in any @xmath1-vertex graph with all degrees at least @xmath7 . much more work has been done to determine conditions for hamiltonicity in graphs , digraphs , hypergraphs , and random and pseudo - random instances of these objects . see , e.g. , any of @xcite . there has also been a long history of research concerning conditions for the existence of multiple edge - disjoint hamilton cycles . indeed , nash - williams discovered that the dirac condition already guarantees not just one , but at least @xmath8 edge - disjoint hamilton cycles . his questions in @xcite started a line of investigation , leading to recent work by christofides , khn , and osthus @xcite , who answered one of his conjectures asymptotically by proving that minimum degree @xmath9 is already enough to guarantee @xmath10 edge - disjoint hamilton cycles . for random graphs , these `` packings '' with hamilton cycles are even more complete . bollobs and frieze @xcite showed that for every fixed @xmath11 , one can typically find @xmath11 edge - disjoint hamilton cycles in the random graph process as soon as the minimum degree reaches @xmath12 . kim and wormald @xcite established a similar result for random @xmath11-regular graphs , proving that such graphs typically contain @xmath13 edge - disjoint hamilton cycles . the previous statements are of course best possible , but invite the natural question of what happens when @xmath11 is allowed to grow . along these lines , frieze and krivelevich showed in @xcite that one can pack @xmath14 hamilton cycles in @xmath15 , up to @xmath16 , where @xmath17 is the minimum degree of the graph . for large @xmath6 , they discovered in @xcite that one can pack almost all edges into hamilton cycles . this was later improved to essentially the full range of @xmath6 by knox , khn , and osthus @xcite . in the hypergraph setting , the study of this hamilton cycle packing problem was initiated by frieze and krivelevich in @xcite . although the notion of a hamilton cycle in an ordinary graph is clear , there are several ways to generalize the notion to hypergraphs . indeed , for any @xmath18 , we may define a @xmath19-uniform hypergraph @xmath5 to be a _ hamilton cycle of type @xmath20 _ if there is a cyclic ordering of the vertices of @xmath5 so that every edge consists of @xmath19 consecutive vertices , and every pair of consecutive edges @xmath21 in @xmath5 ( according to the natural ordering of the edges ) has @xmath22 . the extreme cases @xmath23 and @xmath24 are the most obvious generalizations of graph hamiltonicity , and cycles of those types are often called _ tight _ and _ loose _ , respectively . in @xcite , the first two authors studied the problem of covering almost all the edges of a given @xmath19-uniform hypergraph with disjoint hamilton cycles of a fixed type @xmath20 . they considered @xmath20 on the looser end of the spectrum , determining sufficient conditions for the cases @xmath25 . however , their methods did not extend to the regime @xmath26 , which seems more difficult . in this paper , we introduce several new techniques which enable us to prove the first results for packing _ tight _ hamilton cycles , i.e. , with @xmath27 . to give the clearest presentation of the new methods involved , we concentrate on the concrete case of 3-uniform hypergraphs , which we refer to as _ 3-graphs _ from now on . our approach can be naturally extended to the general @xmath19-uniform case , but the explanations necessarily become more involved . we focus on 3-graphs with certain pseudo - random properties . consequently , our results will yield corollaries in the standard random 3-graph model , denoted @xmath28 , in which each of the @xmath29 possible edges independently appears with probability @xmath6 . recall that there are several essentially equivalent notions of pseudo - randomness in graphs , most notably the ones involving uniformity of edge distribution , the second eigenvalue , or the global count of 4-cycles . ( see , e.g. , the survey @xcite . ) however , even in ordinary graphs , these global conditions are insufficient for hamilton cycle packing , because hamilton cycles visit every vertex . therefore , even a single non - conforming vertex can doom the entire packing process . this situation can be prevented by localizing the specification of the pseudo - random criteria . importantly , the resulting stronger conditions are still satisfied by the random objects in question . for example , in the graph case , this can be done by controlling the degrees and codegrees ( see , e.g. , @xcite ) . note that this is a special case of the more general notion of controlling the numbers of _ extensions _ to certain specific subgraphs , uniformly over all base sites , instead of just the global number of copies of a certain graph ( @xmath30 , say ) . the following definition generalizes that concept to 3-graphs . [ def:3graph - ext ] let @xmath31 be distinct vertices of a 3-graph @xmath0 , and let @xmath32 be a graph on vertex set @xmath33 $ ] . then , we define @xmath34 to be the number of vertices @xmath35 such that @xmath36 for every edge @xmath37 . * example 1 . * if @xmath38 and @xmath32 is a single edge , then @xmath39 is the number of edges which contain the pair @xmath40 . this can be interpreted as a generalized notion of degree . * example 2 . * in general , if @xmath32 is the complete graph on @xmath41 vertices , then @xmath42 is the number of @xmath43 which simultaneously form edges with all pairs @xmath44 . this can be interpreted as a generalized notion of codegree . we can now state our localized version of pseudo - randomness , which we will later show is satisfied by the random 3-graph @xmath28 * whp * is said to occur _ with high probability _ ( abbreviated * whp * ) , if @xmath45 . ] when @xmath6 is not too small . since we need to control many quantities within certain ranges , we will employ the notation @xmath46 as a shorthand for the pair of inequalities @xmath47 . [ def:3graph - props ] we say that an @xmath1-vertex 3-graph @xmath0 is * @xmath48-uniform * if for every auxiliary graph @xmath32 on @xmath49 vertices and @xmath50 edges , and every choice of distinct vertices @xmath51 , we have @xmath52 * remark . * we do not actually need the full strength of this definition , as will be clear in the proof of lemma [ lem : d ] in section [ sec:3graph ] . indeed , there are only 5 specific graphs @xmath32 with respect to which we require control ( depicted in figure [ fig : gamma ] of section [ sec:3graph ] ) . however , we feel that the description is more succinctly captured in the above statement . the main result of this paper establishes the first known packing result for tight hamilton cycles in pseudo - random hypergraphs . here , and in the rest of this paper , we write @xmath53 if @xmath54 as @xmath55 . [ thm:3graph ] suppose that @xmath1 is a sufficiently large multiple of four , and @xmath56 . then every @xmath57-uniform 3-graph with @xmath1 vertices can have all but at most @xmath58-fraction of its edges covered by a disjoint union of tight hamilton cycles . note that in @xmath28 , for each graph @xmath32 with @xmath50 edges on a set of @xmath59 vertices , the value of @xmath60 has distribution @xmath61}$ ] . so , the chernoff bound in section [ sec : tools ] ( fact [ fact : chernoff ] ) shows that as long as @xmath62 , we have @xmath63 } = o(n^7 ) \cdot \sum_{s=1}^6 { \mathbb{p}\left[{\text{\rm bin}\left[n , p^s\right ] } \neq ( 1 \pm \e ) np^s\right ] } = o(1),\ ] ] giving the following immediate corollary . [ cor:3graph ] suppose that @xmath64 satisfy @xmath65 . then whenever @xmath1 is a multiple of four , @xmath28 can have all but at most @xmath58-fraction of its edges covered by a disjoint union of tight hamilton cycles * whp*. * remark . * although both results are stated for @xmath1 divisible by 4 , we expect that they are true in general . note , however , that a divisibility condition is unavoidable in the general case of packing hamilton cycles of type @xmath20 in @xmath19-uniform hypergraphs , since @xmath20 must divide @xmath1 . along the way , we also prove a new result about packing hamilton cycles in pseudo - random digraphs . the result differs from that in @xcite as our definition of pseudo - randomness is local instead of global , and therefore may be easier to apply in some situations . indeed , the previous result required a minimum degree condition , together with bounds on the edge distributions across almost all cuts . we can now replace the latter conditions with two more local statements the specific conditions that we impose below have been tailored for the task of producing hamilton cycles in digraphs . [ def : digraph - props ] we say that an @xmath1-vertex digraph is * @xmath48-uniform * if it satisfies the following properties : ( i ) : : every vertex @xmath66 has out - degree @xmath67 and in - degree @xmath68 . ( ii ) : : for every pair of distinct vertices @xmath69 , all three of the following quantities are @xmath70 : the number of common out - neighbors @xmath71 , the number of common in - neighbors @xmath72 , and the number of out - neighbors of @xmath66 which are also in - neighbors of @xmath73 . ( iii ) : : given any four vertices @xmath74 , which are all distinct except for the possibility @xmath75 , there are @xmath76 vertices @xmath43 such that @xmath77 , @xmath78 , @xmath79 , @xmath80 are all directed edges . * remark . * it is not clear that this is the minimal set of pseudo - random conditions which enable hamilton cycle packing in directed graphs . we choose the above statements because they naturally arise from our analysis , and are therefore the most convenient for our purposes . importantly , one can easily see that they are satisfied by random digraphs * whp * , as long as the edge probability is not too small . under these easily - verifiable conditions , we are able to prove the following packing result for digraphs , which has the obvious corollary for random digraphs which are not too sparse . [ thm : digraph ] suppose that @xmath81 , and @xmath1 is a sufficiently large even integer . then every @xmath82-uniform digraph can have its edges partitioned into a disjoint union of directed hamilton cycles , except for a set of at most @xmath83-fraction of its edges . although we originally developed this result only as a building block for our 3-graph analysis in section [ sec:3graph ] , we feel it is worth bringing attention to , as it may be of independent interest . in particular , it is easier to apply than its counterpart in @xcite , because our pseudo - randomness conditions are easier to verify . the key insight in the proof of theorem [ thm:3graph ] is the following connection between tight hamilton cycles in @xmath0 and hamilton cycles in an associated digraph . for a random permutation @xmath84 of the vertices of @xmath0 , define an @xmath85-vertex digraph @xmath86 with vertex set @xmath87 . note that each vertex of @xmath86 corresponds to an ordered pair of vertices of @xmath0 , so @xmath86 will have an even number of vertices , since the number of vertices of @xmath0 is a multiple of 4 . place a directed edge from @xmath88 to @xmath89 if and only if both hyperedges @xmath90 and @xmath91 are present in @xmath0 . in this construction , hamilton cycles in @xmath86 give rise to tight hamilton cycles in @xmath0 . to extract edge disjoint hamilton cycles from a digraph @xmath86 with an even number of vertices , we use an approach similar to that taken in @xcite . let @xmath92 be a random permutation of the vertices of @xmath86 with @xmath93 , and define @xmath94 and @xmath95 . define a bipartite graph @xmath96 with bipartition @xmath97 , and place an edge between @xmath98 and @xmath99 whenever @xmath100 and @xmath101 are both edges of @xmath86 . now perfect matchings in @xmath32 give rise to hamilton cycles in @xmath86 , and previous approaches in @xcite show how to pack perfect matchings in pseudo - random bipartite graphs . however , not all hamilton cycles in @xmath86 arise from perfect matchings in one particular @xmath32 . similarly , not all hamilton cycles in @xmath0 arise from hamilton cycles in a single @xmath86 . we overcome both obstacles with the same iterative approach , which we illustrate for the hypergraph packing . roughly speaking , instead of stopping after generating a single @xmath86 , we sequentially generate digraphs @xmath102 in the above manner , extracting a large set of edge disjoint directed hamilton cycles from each , and deleting the corresponding edge - disjoint hamilton cycles from @xmath0 . at each step , we verify that the pseudo - random properties are maintained . we repeat the process until we have packed the required number of cycles . since the digraph packing and hypergraph packing proofs are essentially independent ( though similar ) , we separate them into sections [ sec : digraph ] and [ sec:3graph ] , respectively . all required concentration inequalities are collected in section [ sec : tools ] for the reader s convenience . the final section contains some concluding remarks and open problems . we will implicitly assume throughout that @xmath103 are small , e.g. , less than @xmath104 . our results have them tending to zero . the following ( standard ) asymptotic notation will be utilized extensively . for two functions @xmath105 and @xmath106 , we write @xmath107 or @xmath108 if @xmath109 , and @xmath110 or @xmath111 if there exists a constant @xmath112 such that @xmath113 for all sufficiently large @xmath1 . we also write @xmath114 if both @xmath115 and @xmath116 are satisfied . all logarithms will be in base @xmath117 . we recall the chernoff bound for exponential concentration of the binomial distribution . the following formulation appears in , e.g. , @xcite . [ fact : chernoff ] for any @xmath118 , there exists @xmath119 such that any binomial random variable @xmath120 with mean @xmath121 satisfies @xmath122 } < e^{-c_\epsilon \mu},\ ] ] where @xmath123 is a constant determined by @xmath124 . when @xmath125 , we may take @xmath126 . a binomial random variable is the sum of independent indicator variables . we also need concentration in settings which are still product spaces , but are somewhat more complicated than simple binomials . a random variable @xmath127 defined over an @xmath1-dimensional product space @xmath128 is called @xmath5-lipschitz if changing @xmath129 in any single coordinate affects the value of @xmath127 by at most @xmath5 . the hoeffding - azuma inequality ( see , e.g. , @xcite ) provides concentration for these distributions . [ fact : azuma ] let @xmath120 be a @xmath5-lipschitz random variable on an @xmath1-dimensional product space . then for any @xmath130 , @xmath131 } | > t \right ] } \leq 2 \exp\left\ { -\frac{t^2}{2 c^2 n } \right\}.\ ] ] we also need concentration in settings where the probability space is not a simple @xmath1-dimensional product space , but rather the set of permutations on @xmath1 elements . the following concentration inequality can be found in , e.g. , section 3.2 of @xcite or lemma 11 of @xcite . [ fact : conc - perm ] let @xmath120 be a random variable on the uniformly distributed probability space of permutations on @xmath1 elements , and let @xmath5 be a real number . suppose that whenever @xmath132 differ by a single transposition , @xmath133 . then , @xmath134 } \right| \geq t\right ] } \leq 2\exp\left\ { -\frac{2t^2}{c^2 n } \right\}.\ ] ] in this section , we show how to complete the outline in section [ sec : overview ] for hamilton cycle packing in digraphs . recall that the main idea in this part is to reduce the problem to packing perfect matchings in bipartite graphs . we begin by showing how to achieve the final step . here , we will show that we can efficiently decompose a suitably uniform bipartite graph into perfect matchings . this is contained in the following proposition , whose statement is very similar to lemma 6 of @xcite . for completeness , we provide a slightly different proof here . [ prop : graph ] let @xmath135 be a bipartite graph with parts @xmath136 and @xmath137 , each of size @xmath138 . suppose that @xmath139 are given such that every vertex has degree @xmath140 , and every pair of distinct vertices has codegree at most @xmath141 . also assume that @xmath142 . then the edges of @xmath135 can be partitioned into the disjoint union @xmath143 , where @xmath144 and each @xmath145 with @xmath146 is a perfect matching . we mentioned in the introduction that although pseudo - randomness can be deduced from the global 4-cycle count , local conditions are required to pack spanning objects such as hamilton cycles or perfect matchings . for bipartite graphs , the above proposition s conditions on codegrees and degrees provide this localized property . indeed , the concentrated codegree condition implies the global bound on the number of 4-cycles , simply by summing over all vertex pairs . the following lemma converts this into a suitable bound on the uniformity of edge distribution another pseudo - random property , which we actually need to complete this proof . [ lem : flow - prelim ] let @xmath135 be a bipartite graph with parts @xmath136 and @xmath137 , each of size @xmath138 . let @xmath120 and @xmath147 be subsets of @xmath136 and @xmath137 , respectively , with @xmath148 and @xmath149 . suppose that @xmath103 are given such that every vertex in @xmath120 has degree @xmath140 , and every pair of distinct vertices in @xmath120 has codegree at most @xmath141 . then @xmath150 . * * let @xmath151 and @xmath152 . for each vertex @xmath153 , let @xmath154 denote the number of neighbors @xmath155 has in @xmath120 . define the random variable @xmath156 by sampling a uniformly random vertex @xmath153 , and setting @xmath157 . since every vertex in @xmath120 has degree at least @xmath158 , we have @xmath159 , and so @xmath160 } \geq ( 1-\epsilon ) xp$ ] . from the codegree condition , the number of labeled paths of length 2 from @xmath120 to @xmath120 is at most @xmath161 . this quantity is also equal to @xmath162 so @xmath163 . therefore , @xmath164 } \leq ( 1+\epsilon ) ( x^2 p^2 + xp)$ ] , which is at most @xmath165 since we assumed @xmath166 . this implies that @xmath167 } = { \mathbb{e}\left[z^2\right ] } - { \mathbb{e}\left[z\right]}^2 \leq 5\epsilon(xp)^2 $ ] , and this low variance will allow us to conclude that vertices in @xmath147 can not have an average @xmath168 which is too low . formally , if we assume for contradiction that @xmath169 , then jensen s inequality gives ( here , @xmath170 is a shorthand for @xmath160}$ ] ) : @xmath171 } & \geq & \left ( { \mathbb{e}\left[z-\overline{z } \mid v \in y\right ] } \right)^2 \\ & = & \left ( \frac{e(x , y)}{|y| } - \frac{e(x , b)}{|b| } \right)^2 \\ & \geq & \left ( ( 3\epsilon^{1/3 } - \epsilon)(xp ) \right)^2 \\ & > & 5 \epsilon^{2/3 } ( xp)^2.\end{aligned}\ ] ] yet a uniformly random vertex in @xmath137 lies in @xmath147 with probability at least @xmath172 , so @xmath173 } \geq { \mathbb{e}\left[(z-\overline{z})^2 \mid v \in y\right ] } \cdot { \mathbb{p}\left[v \in y\right ] } > 5\epsilon ( xp)^2,\ ] ] contradiction . @xmath174 now that we know the edges are distributed fairly uniformly , we can prove the packing result using the maxflow - mincut theorem . * proof of proposition [ prop : graph ] . * first , observe that if we can pack @xmath175 edge - disjoint perfect matchings into @xmath135 , then the proportion of unused edges is at most @xmath176 we will use the maxflow - mincut theorem to show that we can indeed pack @xmath19 matchings . observe that one can pack @xmath19 edge - disjoint perfect matchings in @xmath135 if and only if the following network has an integer flow of size exactly @xmath177 . give each edge in @xmath135 a capacity of 1 , and direct it from @xmath136 to @xmath137 . add a source vertex @xmath178 , with an edge of capacity @xmath19 to each vertex in @xmath136 , and add a sink vertex @xmath179 , with an edge of capacity @xmath19 from each vertex in @xmath137 . clearly , an integer flow of size @xmath177 is achieved if and only if the subset of used edges between @xmath136 and @xmath137 forms a @xmath19-regular graph . since every @xmath19-regular bipartite graph can be decomposed into @xmath19 edge - disjoint perfect matchings , this completes the argument . all capacities are integers , so there is an integer flow which achieves the maximum size . therefore , by the maxflow - mincut theorem , it remains to show that every cut has size at least @xmath177 . consider an arbitrary cut . let @xmath180 be those vertices of @xmath136 which are on the same side as @xmath178 , and let @xmath181 be those vertices of @xmath137 which are on the same side as @xmath179 . let @xmath151 and @xmath182 . therefore , it suffices to establish the following inequality for all choices of @xmath180 and @xmath181 : @xmath183 this is purely an inequality about the original graph @xmath135 . since it is symmetric in @xmath120 and @xmath147 , assume without loss of generality that @xmath184 . we will use lemma [ lem : flow - prelim ] to estimate @xmath185 , but first , we will need to dispose of the boundary cases @xmath186 and @xmath187 , in which it does not apply . in the first case , observe that we automatically have @xmath188 as well , so @xmath189 is already at least @xmath177 . ( we may assume @xmath190 . ) in the second case , note that if @xmath191 , then @xmath189 is already at least @xmath177 . so , we may assume that @xmath192 . yet every vertex in @xmath120 has at least @xmath158 neighbors in @xmath137 , and only at most @xmath193 of them can be outside @xmath147 . therefore , @xmath194 \geq x \cdot \left [ ( 1-\epsilon)mp - \frac{1}{\epsilon p } \right ] \geq x \cdot ( 1 - 2\epsilon^{1/3})mp,\ ] ] since we assumed that @xmath195 . yet the final quantity exceeds @xmath196 , so we also have inequality in this case . in all remaining cases , we see that lemma [ lem : flow - prelim ] applies , so @xmath197 the right - hand side is bilinear in @xmath43 and @xmath198 , so it is minimized when @xmath199 . if either of @xmath43 or @xmath198 are 0 , then one of the first two terms is already @xmath177 . on the other hand , if @xmath200 , then the expression is precisely @xmath201 as well . therefore , all cuts in the network have size at least @xmath177 , so by the maxflow - mincut theorem , there is a flow of size @xmath177 , completing the proof . @xmath174 now that proposition [ prop : graph ] allows us to efficiently pack perfect matchings , we can use it as the base to which we reduce the problem of packing hamilton cycles in directed graphs . throughout this section , all directed graphs have no loops , and no repeated edges in the same direction . however , they may have edges in both directions between pairs of vertices . the main objective of this section is to prove theorem [ thm : digraph ] , which allows us to efficiently pack hamilton cycles in digraphs that have the pseudo - random properties specified in definition [ def : digraph - props ] . as mentioned in the proof overview in section [ sec : overview ] , we extract hamilton cycles from digraphs by connecting them to perfect matchings in ordinary graphs . consider the following random procedure , which constructs an undirected bipartite graph @xmath32 from a digraph @xmath86 with an even number of vertices . * procedure 1 . * this takes as input a digraph @xmath86 with an even number of vertices . 1 . generate a random permutation @xmath202 of the vertices of @xmath86 . consider this sequence as two consecutive segments of length @xmath85 , and let @xmath203 and @xmath204 . the graph @xmath32 will be a bipartite graph with parts @xmath136 and @xmath137 . 2 . define _ the successor of @xmath205 _ to be @xmath206 , unless @xmath207 ( in which case the successor is @xmath208 ) or @xmath209 ( in which case it is @xmath210 ) . similarly , define _ the predecessor of @xmath205 _ to be @xmath211 , unless @xmath212 ( in which case it is @xmath213 ) , or @xmath214 ( in which case it is @xmath215 ) . 3 . for @xmath216 and @xmath217 , place the edge @xmath218 in @xmath32 if and only if @xmath86 contains both directed edges @xmath219 and @xmath220 , where @xmath221 is the successor of @xmath205 as defined in the previous step . since an edge in @xmath32 corresponds to two edges in @xmath86 , we account for this by defining @xmath222 to be the digraph containing all @xmath223 , for each @xmath224 with @xmath216 . the key observation is that perfect matchings in @xmath32 cleanly correspond to hamilton cycles in @xmath225 . specifically , given a perfect matching @xmath112 in @xmath32 , one can recover a hamilton cycle in @xmath225 . indeed , for each @xmath216 , @xmath112 matches @xmath205 to a distinct @xmath226 . so , @xmath227 is a directed hamilton cycle in @xmath225 ; call this the hamilton cycle _ associated with _ @xmath112 . [ lem : hamilton - matching ] let @xmath228 be edge - disjoint perfect matchings in @xmath32 . then their associated hamilton cycles @xmath229 are also edge - disjoint in @xmath225 . * proof . * since @xmath230 and @xmath231 are disjoint , it is clear that @xmath232 and @xmath233 can not overlap on any edges directed from @xmath136 to @xmath137 . on the other hand , if both @xmath234 contain the same edge @xmath235 , where @xmath217 and @xmath236 , then they also both contain @xmath237 , by definition of @xmath234 . then both @xmath238 contain the edge @xmath239 , contradiction . @xmath174 the next step , as we mentioned in the proof overview in the introduction , is to show that the randomly constructed bipartite graph inherits the pseudo - random properties of the initial digraph . [ lem : gamma ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath240 . let @xmath86 be an @xmath82-uniform digraph on @xmath1 vertices ( @xmath1 even ) , and randomly construct the undirected bipartite graph @xmath32 according to procedure 1 . then , with probability @xmath241 , @xmath32 satisfies all of the following properties : ( i ) : : all degrees are @xmath242 . ( ii ) : : all codegrees between pairs of vertices on the same side of the bipartition are @xmath243 . * proof . * by construction , @xmath32 is a bipartite graph with parts @xmath244 and @xmath245 . there are essentially four claims to prove , as the parts @xmath136 and @xmath137 are not constructed symmetrically . * claim 1 : degrees in @xmath246 are correct . * to show this , for an arbitrary vertex @xmath66 , define a random variable @xmath247 as follows . let @xmath73 be the successor of @xmath66 , as defined in step 2 of procedure 1 . @xmath247 is of no interest when @xmath248 but allowing @xmath248 makes the proof a trifle simpler . by definition , @xmath66 and @xmath73 are either both in @xmath136 or both in @xmath137 ; let @xmath247 be the number of vertices @xmath43 in the other part such that both @xmath77 and @xmath78 are edges of @xmath86 . note that @xmath247 is actually the degree of @xmath66 in @xmath32 if @xmath66 happens to end up in @xmath136 . therefore , it suffices to show that @xmath249 with probability @xmath250 . note that the random variable @xmath247 is completely determined by the permutation @xmath178 . expose @xmath178 in stages . first , expose which index @xmath251 has @xmath252 . next , expose the identity of the vertex @xmath73 which is the successor of @xmath66 . this determines @xmath205 and @xmath221 , where @xmath253 is either @xmath254 , @xmath255 , or @xmath256 . conditioned on these , the associations to the remaining vertices @xmath257 are a uniform permutation over the remaining @xmath258 vertices . yet the last part of property ( ii ) for @xmath82-uniformity shows that in @xmath86 , the number of vertices @xmath43 with @xmath77 and @xmath78 in @xmath86 is @xmath259 . as each @xmath43 will land in the part opposite @xmath260 with probability exactly @xmath261 , we immediately have @xmath262 } = ( 1 \pm 2\epsilon)\frac{np^2}{2}$ ] . for concentration , we use fact [ fact : conc - perm ] . we already conditioned on @xmath66 and @xmath73 , so the remaining randomness is from a uniformly random permutation of @xmath258 elements . transposing two of those elements can only change @xmath247 by at most 1 , so by fact [ fact : conc - perm ] , @xmath247 deviates from its expectation by @xmath263 with probability at most @xmath264 therefore , @xmath247 is indeed @xmath265 with probability @xmath266 , as desired . * claim 2 : degrees in @xmath267 are correct . * consider an arbitrary vertex @xmath73 . let @xmath268 be the set of ordered pairs of distinct vertices @xmath269 with @xmath270 . by property ( i ) of @xmath271-uniformity , @xmath73 has @xmath272 in - neighbors , and @xmath273 out - neighbors . since @xmath274 , this implies that @xmath275 . let @xmath276 be the number of these pairs that , after the permutation @xmath178 , have the additional properties that * ( i ) * both @xmath43 and @xmath198 are on the opposite side of the bipartition to @xmath73 , and * ( ii ) * @xmath198 is the successor of @xmath43 . it suffices to show that @xmath276 is @xmath277 with probability @xmath266 , because @xmath276 is precisely the degree of @xmath73 whenever @xmath73 happens to end up in @xmath137 . for this , we employ the same strategy as used in the previous claim . first expose the index @xmath251 for which @xmath278 . then , the locations of the remaining vertices are uniformly permuted amongst the remaining @xmath279 positions . for a given ordered pair @xmath280 , the probability that @xmath43 lands on the opposite side of @xmath73 is precisely @xmath281 . then , conditioned on this , the probability that @xmath198 is the successor of @xmath43 is precisely @xmath282 . therefore , @xmath283 } = |s_b| \cdot \frac{n/2}{n-1 } \cdot \frac{1}{n-2 } = ( 1 \pm 4\epsilon ) \frac{np^2}{2}.\ ] ] for concentration , we again expose @xmath73 first , and then consider the resulting @xmath284-permutation . consider two such permutations @xmath178 and @xmath285 differing only on a single transposition . transposing two of those elements can only change @xmath276 by at most 4 , so by fact [ fact : conc - perm ] , the probability that @xmath276 deviates by over @xmath286 from its expectation is at most @xmath287 as before . hence @xmath276 is indeed @xmath242 with the desired probability @xmath266 . * claim 3 : co - degrees in @xmath246 are correct . * this is similar to claim 1 . fix any two distinct vertices @xmath66 and @xmath73 , and let @xmath288 and @xmath289 be their respective successors as defined in step 1 of procedure 1 . note that we may have @xmath75 or @xmath290 , but not both . let @xmath291 be the number of vertices @xmath43 such that @xmath43 is on the opposite side of @xmath66 , and @xmath292 . expose the indices @xmath293 for which @xmath252 and @xmath294 , and expose their successors @xmath288 and @xmath289 . the remaining vertices are uniformly permuted over the remaining @xmath295 positions ( or @xmath296 positions if @xmath297 or @xmath298 , although the argument will be the same ) . by property ( iii ) of @xmath82-uniformity , there are @xmath299 candidates for @xmath43 . as each @xmath43 will land in the part opposite @xmath66 with probability either @xmath300 ( if @xmath73 is on the same side as @xmath66 ) or @xmath301 ( if @xmath73 is opposite @xmath66 ) , it follows that @xmath302 } = ( 1 \pm 2\epsilon ) \frac{np^4}{2}$ ] . a transposition in the @xmath303-permutation can only affect @xmath291 by at most 1 , so the probability that @xmath291 deviates from its expectation by over @xmath304 is at most @xmath305 taking a union bound over all pairs @xmath306 yields the desired result . * claim 4 : co - degrees in @xmath267 are correct . * this is similar to claim 2 . fix any two distinct vertices @xmath66 and @xmath73 , and let @xmath307 be the set of ordered pairs of distinct vertices @xmath269 with @xmath308 . by property ( ii ) of @xmath82-uniformity , @xmath66 and @xmath73 have @xmath70 common in - neighbors , and @xmath309 common out - neighbors . since @xmath310 , this implies that @xmath311 . let @xmath291 be the number of these pairs that , after the permutation @xmath178 , have the additional properties that * ( i ) * both @xmath43 and @xmath198 are on the opposite side of the bipartition from @xmath66 , and * ( ii ) * @xmath198 is the successor of @xmath43 . expose the indices @xmath293 for which @xmath252 and @xmath294 ; the remainder is then an @xmath312-permutation over the leftover positions . a similar calculation to claim 2 shows that @xmath302 } = ( 1 \pm 4\epsilon ) \frac{np^4}{2}$ ] , and a similar concentration argument shows that @xmath291 is within @xmath304 of its mean with probability @xmath313 . therefore , a union bound over all @xmath314 completes the proof of this final claim , and the proof of lemma [ lem : gamma ] @xmath174 at this point , we could immediately apply the results of section [ sec : graph ] to pack @xmath32 with perfect matchings , which then correspond to hamilton cycles in @xmath86 . however , it is unfortunate that this would miss most of the edges of @xmath86 , since not all edges of @xmath86 are in correspondence with edges of @xmath32 . the solution is to iterate procedure 1 several times . some care must be taken because the edges of @xmath86 will be covered many times by different @xmath32 . we will specify how to deal with this in procedure 2 below , but first we collect two intermediate results which control this multiplicity over several iterations of procedure 1 . [ lem : gamma - cover ] suppose @xmath1 and @xmath124 satisfy @xmath315 . let @xmath86 be an @xmath82-uniform digraph on @xmath1 vertices ( @xmath1 even ) , and randomly and independently construct @xmath316 graphs @xmath317 according to procedure 1 . let @xmath318 be their corresponding digraphs . then , with probability @xmath319 , every edge of @xmath86 is covered @xmath320 times by the @xmath318 . * fix an arbitrary edge @xmath321 , and let @xmath322 be the probability that it appears in @xmath225 if @xmath32 is constructed according to procedure 1 . since the @xmath318 are independent , the number of times @xmath321 is covered will then be @xmath323}$ ] , and we will use the chernoff bounds to prove concentration . so , let @xmath324 be the permutation which produces @xmath32 , and let @xmath136 and @xmath137 be the two sides of the bipartition as defined in procedure 1 . there are two ways that @xmath321 can appear in @xmath225 : either @xmath325 and @xmath153 and @xmath326 for the successor @xmath327 of @xmath328 , or @xmath329 and @xmath330 and @xmath331 for the predecessor @xmath332 of @xmath155 . these two cases are clearly disjoint , so @xmath333 , where @xmath334 and @xmath335 are the respective probabilities in these two cases . to calculate @xmath334 , the probability that @xmath336 is precisely @xmath337 . conditioned on this , the probability that @xmath153 is precisely @xmath281 . by property ( i ) of @xmath271-uniformity , @xmath338 , so the probability that some out - neighbor of @xmath155 becomes the successor of @xmath328 is @xmath339 . therefore , @xmath340 since @xmath341 . an analogous calculation shows that @xmath342 as well . hence @xmath343 . now , the number of times that @xmath321 is covered by the @xmath318 is precisely @xmath323}$ ] . since @xmath344 , the chernoff bound implies that the probability that the binomial deviates from its mean by more than a factor of @xmath345 is at most @xmath346 taking a union bound over all @xmath347 edges @xmath321 , we obtain the desired result . [ lem : gamma - rare - consec ] suppose @xmath348 . construct @xmath11 independent @xmath349 according to procedure 1 . then with probability @xmath319 , every pair of distinct vertices @xmath350 has the property that @xmath73 is the successor of @xmath66 in at most 5 of the permutations for @xmath349 . * for a fixed pair @xmath306 , the probability that @xmath73 is the successor of @xmath66 in a single run of procedure 1 is exactly @xmath351 . therefore , the probability of this occurring more than 5 times in @xmath11 independent runs is at most @xmath352 . taking a union bound over all pairs @xmath306 , we obtain the result . @xmath174 since a single run of procedure 1 packs very few edges into hamilton cycles , we must repeat the procedure multiple times , deleting the packed edges from @xmath86 after each round . however , it is important to maintain the pseudo - random properties through the iterations , and the set of removed edges after a single run of procedure 1 would be too sparse to control the changes in the pseudo - random counts . the following extension provides one way to achieve this , by repeating procedure 1 enough times to uniformly involve all edges of @xmath86 . * procedure 2 . * this takes as input a digraph @xmath86 with an even number of vertices , and an integer parameter @xmath11 . 1 . independently generate the random undirected bipartite graphs @xmath353 , , @xmath354 according to procedure 1 , and let @xmath355 , @xmath318 , @xmath356 , and @xmath357 be their corresponding permutations , digraphs , and bipartitions , respectively . 2 . for each edge @xmath358 , let @xmath359 . if @xmath360 , then independently select a uniformly random index in @xmath361 to label @xmath321 with . 3 . for each @xmath349 , define a subgraph @xmath362 by keeping each edge @xmath363 ( @xmath364 and @xmath365 ) if and only if both @xmath366 and @xmath367 are labeled by @xmath251 , where @xmath368 is the successor of @xmath66 according to @xmath355 . 4 . for each @xmath251 , let @xmath369 be the digraph containing all @xmath370 , for each @xmath371 , with @xmath372 , where @xmath368 is the successor of @xmath66 according to @xmath355 . observe that the final step ensures that the @xmath369 are all disjoint . so , lemma [ lem : hamilton - matching ] shows that we can work independently on each @xmath369 , packing hamilton cycles by packing perfect matchings in @xmath362 with proposition [ prop : graph ] . after this , we will remove all @xmath369 from @xmath86 , and show that their distribution is sufficiently uniform for us to maintain the necessary pseudo - random properties . the following lemma shows that the @xmath362 themselves are sufficiently pseudo - random for us to apply proposition [ prop : graph ] . [ lem : gamma - prime ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath373 . let @xmath86 be an @xmath82-uniform digraph on @xmath1 vertices ( @xmath1 even ) , and conduct procedure 2 with @xmath316 . let @xmath374 . then , with probability @xmath319 , every @xmath362 satisfies the following properties : ( i ) : : all degrees are @xmath375 . ( ii ) : : all codegrees between pairs of vertices on the same side of the bipartition are @xmath376 . * proof . * our restrictions on @xmath64 allow us to apply lemmas [ lem : gamma ] and [ lem : gamma - cover ] , so we have that with probability @xmath319 , after step 1 of procedure 2 : ( a ) : : every @xmath349 has all degrees @xmath377 and all same - side codegrees @xmath378 . ( b ) : : every edge in @xmath86 is covered @xmath379 times by the @xmath318 . condition on the above outcome of step 1 , and consider an arbitrary @xmath362 , which is derived from @xmath349 with bipartition @xmath380 . it suffices to show that in the randomness of step 2 , with probability @xmath266 each individual @xmath362 has the desired properties , since @xmath381 by the given restrictions on @xmath64 . there are four cases to consider : degrees in @xmath356 , degrees in @xmath357 , codegrees in @xmath356 , and codegrees in @xmath357 . fortunately , they will all follow by essentially the same argument . we begin with the degree of an arbitrary vertex @xmath382 . by ( a ) above , the degree of @xmath328 in @xmath349 is @xmath383 . note that this actually corresponds to exactly @xmath384 pairs of directed edges in @xmath86 , of the form @xmath385 , and all @xmath386 directed edges involved are distinct . step 2 of procedure 2 assigns labels to all directed edges , and the degree of @xmath328 in @xmath362 is precisely the number of the above pairs of directed edges for which both edges are labeled @xmath251 . since every directed edge is covered @xmath379 times by ( b ) and the edges are labeled independently , the probability that both of a given pair above receive label @xmath251 is @xmath387^{-2}$ ] . therefore , the expected degree of @xmath328 in @xmath362 is @xmath388 . furthermore , since all directed edges involved are distinct , the chernoff bound shows that the probability of the degree deviating from its mean by more than a factor of @xmath389 is at most @xmath390 because @xmath391 . therefore , with probability @xmath313 , the degree of @xmath328 in @xmath362 is @xmath392 . taking a union bound over all @xmath382 establishes part ( i ) for those degrees . the arguments for the other three parts of the lemma are similar . for a vertex @xmath393 , its incident edges in @xmath349 correspond to @xmath394 disjoint pairs of directed edges , so the exact same argument as above produces the bound for these degrees . the codegree of a given pair of vertices @xmath395 corresponds to disjoint quadruples of distinct directed edges . from ( a ) , there are @xmath243 such quadruples . the probability that a given quadruple is completely labeled by @xmath251 is @xmath396^{-4}$ ] by ( b ) , so the expected codegree in @xmath362 is @xmath397 . by the chernoff bound , the probability that the codegree deviates from its expectation by a factor of more than @xmath398 is at most @xmath399 since @xmath400 . taking a union bound over all pairs of vertices @xmath395 produces the desired bound for their codegrees . the argument for codegrees in @xmath357 is similar . @xmath174 the next few lemmas build up to a result which controls how @xmath86 is affected by the deletion of all edges in the @xmath369 . the first one controls the `` first - order '' effect of the deletion process . [ lem : d - delete - out - nbrs ] condition on the first step of procedure 2 covering every edge of @xmath86 @xmath379 times by the @xmath318 . fix any vertex @xmath401 and any set @xmath402 of edges incident to @xmath66 ( all oriented in the same way with respect to @xmath66 ) . suppose that @xmath403 . then with probability @xmath404 , the number of edges of @xmath402 which are covered by the @xmath369 is @xmath405 , * proof . * we start with the case when all edges in @xmath402 are directed out of @xmath66 . let the random variable @xmath406 be the number of these edges which are covered by the @xmath369 . since we conditioned on the first step of procedure 2 , the only remaining randomness is in the independent assignments of the edge labels . we expose these labels in three stages . first , expose the labels of @xmath402 . now each edge @xmath407 has its label @xmath408 , so we can identify its partner edge in @xmath409 which must also receive @xmath408 in order for both to remain in @xmath410 ; let @xmath411 be the set of all partner edges found in this way . next , expose all labels outside @xmath412 . finally , expose the labels of @xmath411 . it is clear that the second stage does not affect @xmath406 at all . thus , after conditioning on the result of the second stage , we are at the following situation : every edge @xmath413 has a label @xmath408 , and it will only still be covered by the @xmath369 if its partner in @xmath409 also receives the label @xmath408 . since its partner is in @xmath379 different @xmath318 , the probability that @xmath414 is covered by @xmath369 is the inverse of this multiplicity ; linearity of expectation then gives @xmath415 } = ( 1 \pm 1.04\epsilon ) \frac{|e|}{\kappa}$ ] . we will use the hoeffding - azuma inequality to show the concentration of @xmath406 . indeed , the third stage exposure is a product space of dimension @xmath416 . consider the effect of changing the label of a single edge @xmath417 . by the definition of @xmath411 , the edge @xmath418 is either some @xmath419 directed into @xmath66 , or some @xmath78 not incident to @xmath66 with @xmath420 . in the latter case , @xmath77 is the only edge of @xmath402 which can be affected by the label of @xmath418 , so @xmath406 can change by at most 1 . for the remaining case @xmath421 , suppose that the label of @xmath418 was changed from @xmath422 to @xmath19 . by construction , only the ( single ) partner edge of @xmath418 in @xmath423 could suffer from changing @xmath418 s label away from @xmath422 ; this could decrease @xmath406 by at most 1 . similarly , only the partner edge of @xmath418 in @xmath424 could benefit from changing @xmath418 s label to @xmath19 , and this would only increase @xmath406 by at most 1 . we conclude that @xmath406 is 1-lipschitz , so the hoeffding - azuma inequality implies that the probability @xmath406 deviates from its expectation by over @xmath425 is at most @xmath426 therefore , with probability @xmath404 , we have @xmath427 , in the case when all edges of @xmath402 are directed out of @xmath66 . the case when all edges are directed into @xmath66 follows by essentially the same argument . @xmath174 the next lemma controls the `` second - order '' effect of the deletion process . [ lem : d - delete - co - out ] fix any distinct vertices @xmath350 and any set @xmath120 of vertices such that either ( i ) : : for all @xmath428 , @xmath429 are edges of @xmath86 ; or ( ii ) : : for all @xmath428 , @xmath430 are edges of @xmath86 ; or ( iii ) : : for all @xmath428 , @xmath431 are edges of @xmath86 . condition on the first step of procedure 2 covering every edge of @xmath86 at least @xmath432 times by the @xmath318 , and on the fact that @xmath73 is the successor of @xmath66 in at most 5 of the permutations . suppose that @xmath433 . then with probability @xmath434 , the number @xmath406 of vertices in @xmath120 which still have both of their designated edges above still covered by the @xmath369 is at most @xmath435 . * proof in situations ( i ) and ( ii ) . * these two cases follow by very similar arguments ; for concreteness , let us begin with ( i ) . let @xmath402 be the set of @xmath436 edges specified in ( i ) . since we conditioned on the first step of procedure 2 , the only remaining randomness is in the independent assignments of the edge labels . we expose these labels in three stages . first , expose the labels of @xmath402 . now each edge @xmath407 has its label @xmath408 , so we can identify its partner edge in @xmath409 ; denote the partner by @xmath437 , and let @xmath411 be the set of all partner edges found in this way . next , expose all labels outside @xmath412 . finally , expose the labels of @xmath411 . it is clear that the second stage does not affect @xmath406 at all . thus , after conditioning on the result of the second stage , we are at the following situation : every edge @xmath407 has a label @xmath408 , and @xmath414 will only still be covered by the @xmath369 if its partner @xmath437 also receives the label @xmath408 . the random variable @xmath406 counts the number of vertices @xmath43 for which both @xmath77 and @xmath438 are still covered by the @xmath369 . note that their partner edges @xmath439 and @xmath440 are distinct unless they both coincide as some @xmath441 . if they are distinct , then clearly the probability of having both receive their correct labels is at most @xmath442 . on the other hand , if they coincide , then we must have that in both permutations @xmath443 and @xmath444 , the vertex @xmath288 is the successor of @xmath66 and @xmath73 . yet @xmath288 has a unique predecessor , so @xmath445 . therefore , it is actually impossible for the ( coincident ) partner edge to receive a label which suits both @xmath77 and @xmath438 , so the probability is 0 . in all cases , we have an upper bound of @xmath442 , so @xmath415 } \leq \frac{|x|}{\kappa^2}$ ] . we use the hoeffding - azuma inequality to probabilistically bound @xmath406 , since the third exposure stage is a product space . consider the effect of changing the label of a single edge @xmath417 . by the definition of @xmath411 , the edge @xmath418 is either some @xmath419 or @xmath78 , or some @xmath441 with @xmath77 and @xmath438 both in @xmath86 . in the latter case , @xmath43 is the only vertex which may have its count in @xmath406 affected , so @xmath406 changes by at most 1 under this perturbation . by symmetry , it remains to consider the case when @xmath421 changes its label from @xmath422 to @xmath19 . by construction , only the partner edge @xmath446 of @xmath418 in @xmath423 could suffer from changing @xmath418 s label away from @xmath422 ; this could potentially lose only @xmath198 in the count of @xmath406 , so @xmath406 would decrease by at most 1 . similarly , only the partner edge of @xmath418 in @xmath424 could benefit from changing @xmath418 s label to @xmath19 , and this could only increase @xmath406 by at most 1 . we conclude that @xmath406 is 1-lipschitz over its product space of dimension @xmath447 , so the azuma - hoeffding inequality implies that the probability that @xmath406 exceeds its expectation by more than @xmath448 is at most @xmath449 this finishes the case when ( i ) holds . the case when ( ii ) holds follows from a directly analogous argument . * proof in situation ( iii ) . * let @xmath402 be the set of @xmath436 edges specified in ( iii ) . we expose the labels in the same three stages as before : first @xmath402 , then the other non - partner edges , and finally the partner edges @xmath411 . again , after the second stage we would be at the following situation : every edge @xmath407 has a label @xmath408 , and a vertex @xmath43 will only be counted toward @xmath406 if the partner edge @xmath439 receives the label @xmath443 , and @xmath450 also receives @xmath451 . however , this time there is an additional complication , because it can happen that one or both of these partner edges have already had their labels exposed . the only way this could happen is if either @xmath452 or @xmath453 . fortunately , we are only seeking an upper bound on @xmath406 , so we only need to determine when this first round exposure already forces a vertex @xmath43 to contribute to @xmath406 . a moment s thought reveals that the only way this can happen is if the first round gave both @xmath77 and @xmath78 the same label @xmath454 , and furthermore , the permutation @xmath455 has @xmath73 as the successor of @xmath66 . now we use the assumption that at most 5 of the permutations @xmath455 satisfy this property . let @xmath456 be the set of the corresponding indices @xmath454 . we can circumvent this issue , by observing that in the first exposure , the number of @xmath43 for which both @xmath77 and @xmath78 receive the same label @xmath457 is stochastically dominated by @xmath458}$ ] . since @xmath459 , the chernoff bound then implies that with probability @xmath404 , the first round has at most @xmath460 `` bad '' @xmath43 which have the potential of being automatically included in the count for @xmath406 . the second round exposure is essentially irrelevant , so we may now condition on the result of the second round satisfying the property in the previous paragraph . as we only need to upper bound @xmath406 , it remains to consider only the `` good '' @xmath43 . from our previous discussion , if a good @xmath43 has , say , the label of @xmath439 already exposed to be @xmath454 then @xmath73 is not the successor of @xmath66 in permutation @xmath454 and so this @xmath43 can not contribute to @xmath406 . also note that the only way for @xmath439 to equal @xmath450 is if both are @xmath461 . if @xmath462 then @xmath248 and if @xmath463 then @xmath464 and so we must have @xmath465 . hence it is not possible for @xmath461 to simultaneously match both labels , and @xmath43 can not contribute to @xmath406 . so , if we let let @xmath147 be the subset of good vertices @xmath428 for which @xmath466 , and neither label has yet been exposed , it remains to control the number @xmath112 of vertices in @xmath147 which contribute to @xmath406 . since each @xmath467 has @xmath468 and their labels are independent , we immediately have @xmath469 } \leq |y| / \kappa^2 $ ] . we show that @xmath112 is concentrated by using the hoeffding - azuma inequality on the third round exposure product space of dimension @xmath470 . consider the effect of changing the label of an edge @xmath471 from @xmath422 to @xmath19 . there are three types of edges in @xmath411 : case 1 : @xmath472 . : : these @xmath418 can only arise as partner edges of some @xmath77 ( possibly several ) . however , @xmath418 has a unique partner edge @xmath77 in @xmath423 , so changing its label away from @xmath422 can only reduce @xmath112 at @xmath43 . similarly , changing it to @xmath19 can only grow @xmath112 at @xmath198 , where @xmath446 is the unique partner edge of @xmath418 in @xmath424 . therefore , @xmath112 can only change by at most 1 . case 2 : @xmath473 . : : by the same argument as case 1 , this can only change @xmath112 by at most 1 . case 3 : @xmath474 . : : these @xmath418 can only arise as partners of either @xmath77 or @xmath475 . clearly , changing @xmath476 can only affect whether @xmath43 or @xmath198 are counted in @xmath112 , so it has an effect of at most 2 . thus @xmath112 is 2-lipschitz , and the hoeffding - azuma inequality shows that the probability it exceeds its expectation by @xmath477 is at most @xmath478 since we conditioned on there being at most @xmath460 bad vertices , we have @xmath479 . putting everything together , we conclude that with probability @xmath480 , @xmath481 , completing the proof . @xmath174 the previous two lemmas now enable us to prove that the pseudo - random properties of the digraph are still maintained after deleting the @xmath369 . [ lem : digraph - delete ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath482 . let @xmath483 and @xmath484 . let @xmath86 be an @xmath57-uniform digraph on @xmath1 vertices ( @xmath1 even ) , and conduct procedure 2 . let @xmath485 be the subgraph of @xmath86 obtained by deleting all edges covered by any @xmath369 . then , with probability @xmath486 , @xmath485 is an @xmath487-uniform digraph with @xmath488 and @xmath489 . * by applying lemmas [ lem : gamma - cover ] and [ lem : gamma - rare - consec ] , the outcome of step 1 of procedure 2 will satisfy the following two properties with probability @xmath490 . ( the second follows since @xmath491 , and so @xmath348 . ) * every edge of @xmath86 covered @xmath379 times by the @xmath318 . * for any @xmath314 , the vertex @xmath73 is the successor of @xmath66 in at most 5 of the permutations for the @xmath349 . condition on this outcome . we will now show that in the remaining randomness of step 2 , the properties of @xmath487-uniformity are satisfied with probability @xmath319 . we establish them one at a time . * property ( i ) . * consider an arbitrary vertex @xmath66 . by @xmath82-uniformity , it has @xmath272 out - edges in @xmath86 , and @xmath492 . so , we may apply lemma [ lem : d - delete - out - nbrs ] to the set of out - edges of @xmath66 . this shows that with probability @xmath404 , the new out - degree of @xmath66 in @xmath485 , in terms of its original out - degree @xmath493 in @xmath86 , is @xmath494 \\ & = & ( 1 \pm \epsilon ) np \cdot \left ( 1 \pm \frac{1.06\epsilon}{\kappa } \right ) \left ( 1 - \frac{1}{\kappa } \right ) \\ & = & ( 1 \pm \epsilon ' ) np'.\end{aligned}\ ] ] here , we used @xmath495 by @xmath271-uniformity . taking a union bound over all @xmath401 , we obtain the desired result on out - degrees with probability @xmath496 . a similar argument controls all in - degrees . @xmath174 * property ( ii ) . * consider an arbitrary pair of distinct vertices @xmath69 , and let @xmath120 be the set of their common out - neighbors . let @xmath497 be the number of vertices in @xmath428 such that @xmath77 is covered by the @xmath369 , let @xmath498 be the number of @xmath428 such that @xmath438 is covered by the @xmath369 , and let @xmath499 be the number of @xmath428 such that both @xmath77 and @xmath438 are covered by the @xmath369 . clearly , the number of common out - neighbors of @xmath66 and @xmath73 in @xmath485 is exactly @xmath500 , where @xmath71 was the number of their common out - neighbors in @xmath86 . note that @xmath501 by @xmath271-regularity , and @xmath502 . so , lemma [ lem : d - delete - out - nbrs ] implies that with probability @xmath404 , both @xmath497 and @xmath498 are @xmath503 . on the other hand , we also have @xmath504 , so lemma [ lem : d - delete - co - out ] bounds @xmath499 by @xmath505 with probability @xmath480 . therefore , @xmath499 is within @xmath506 additive error of @xmath507 because @xmath508 . putting these bounds together , we have that the new number of common out - neighbors is @xmath509 \\ & = & ( 1 \pm \epsilon ) np^2 \cdot \left ( 1 \pm \frac{2.12\epsilon}{\kappa } \right ) \left ( 1 - \frac{1}{\kappa } \right)^2 \\ & = & ( 1 \pm \epsilon ' ) n(p')^2,\end{aligned}\ ] ] here , @xmath501 by @xmath271-uniformity . taking a union bound over all @xmath350 , we obtain the desired result on the new @xmath71 with probability @xmath510 . similar arguments control the other two parts of property ( ii ) of @xmath487-uniformity . @xmath174 * property ( iii ) . * this is a slight extension of the previous argument . consider any four vertices @xmath74 , which are all distinct except for the possibility @xmath75 . let @xmath120 and @xmath511 be the sets of vertices @xmath43 such that @xmath77 , @xmath78 , @xmath79 , @xmath80 are all in @xmath86 or @xmath485 , respectively . let @xmath497 , @xmath498 , @xmath512 , and @xmath513 be the sets of vertices @xmath428 such that @xmath77 , @xmath78 , @xmath79 , or @xmath80 are still covered by the @xmath514 . define the pairwise intersections @xmath515 . by inclusion - exclusion , we have @xmath516 as in ( ii ) , since @xmath517 ( to apply lemma [ lem : d - delete - out - nbrs ] ) and @xmath518 ( for lemma [ lem : d - delete - co - out ] ) , we have that with probability @xmath480 , each @xmath519 , and each @xmath520 . thus @xmath521 = ( 1 \pm \epsilon ) np^4 \left ( 1 - \frac{4}{\kappa } \pm \frac{4.21\epsilon}{\kappa } \right).\ ] ] here , we used @xmath522 by @xmath271-uniformity . now observe that @xmath523 however , since @xmath508 , the error term is @xmath524 . therefore , we have @xmath525 taking a union bound over all choices of @xmath74 , we obtain the desired result on the new @xmath71 with probability @xmath490 . @xmath174 now we have established control over the pseudo - random properties of the digraph after deletion , as well as over the individual bipartite graphs @xmath362 at each stage . we finally combine all of our lemmas to prove the main result of this section , that pseudo - random digraphs can be efficiently packed with hamilton cycles . * proof of theorem [ thm : digraph ] . * we will iterate procedure 2 , packing all intermediate @xmath369 at each iteration and deleting them from the digraph , until very few edges remain . the total number of iterations will be @xmath526 , so since our previous lemmas hold with probability @xmath319 , a final union bound will show that we achieve the desired packing * whp*. let @xmath527 , @xmath528 , and @xmath529 . define the sequences @xmath530 , @xmath531 via the following recursion : @xmath532 let @xmath533 be the smallest index such that @xmath534 . note that @xmath535 only increases , so @xmath536 also note that since @xmath537 we have in general that @xmath538 . therefore , @xmath539 we now iteratively construct @xmath540 , such that each @xmath541 is @xmath542-uniform . indeed , consider the ( random ) procedure 2 applied to @xmath541 with respect to @xmath543 . let @xmath544 . this produces bipartite graphs @xmath545 and directed graphs @xmath546 with all @xmath546 disjoint . let @xmath547 be the result of deleting all @xmath546 from @xmath541 . to apply lemmas [ lem : gamma - prime ] and [ lem : digraph - delete ] , we must check that @xmath548 . but this follows from our initial assumption that @xmath549 since @xmath550 and @xmath551 . therefore , as the individual failure probabilities of lemmas [ lem : gamma - prime ] and [ lem : digraph - delete ] are @xmath552 and @xmath553 , we may assume that the outcome of procedure 2 satisfies the following properties : ( i ) : : all degrees in each @xmath545 are @xmath554 . ( ii ) : : all codegrees between pairs of vertices on the same side of the bipartition of each @xmath545 are @xmath555 . ( iii ) : : @xmath547 is @xmath556-uniform . fix this outcome . we may apply proposition [ prop : graph ] to each @xmath545 because @xmath557 where we used @xmath558 , @xmath559 , and our initial assumption that @xmath560 . so , every @xmath545 can be packed with edge - disjoint perfect matchings , missing only a @xmath561-fraction of the edges . by lemma [ lem : hamilton - matching ] , these edge - disjoint perfect matchings in @xmath545 correspond to edge - disjoint hamilton cycles in @xmath546 , missing the same fraction of edges since there is a 2-to-1 correspondence between edges in @xmath546 and @xmath545 . we carry on the above procedure until we create @xmath562 . then , we will have packed hamilton cycles in @xmath563 , up to a fractional error of @xmath564 . since @xmath562 is @xmath565-uniform , it has at most @xmath566 edges . as @xmath86 was @xmath271-uniform , it had at least @xmath567 edges . thus @xmath562 itself has at most @xmath568-fraction of the total number of edges . therefore , the fraction of edges of @xmath86 that have not been covered is at most @xmath569 since @xmath570 by inequality . this completes the proof . * the above proof showed that we can pack the edges of @xmath86 with hamilton cycles , up to a fractional error of at most @xmath571 , where @xmath572 . in fact , one can prove the same result for any @xmath573 , and all of the calculations can be recovered by following our proof . indeed , we intentionally introduced a very high level of precision in lemma [ lem : gamma - cover ] . it is clear that the same argument can replace the @xmath574 with @xmath575 , etc . these smaller order errors are what accumulate into the decimal places in the final @xmath576 which appears in lemma [ lem : digraph - delete ] , and which later determines the value of @xmath288 . when @xmath576 is replaced by a constant arbitrarily close to 4 , the above argument produces a result with @xmath288 approaching @xmath577 . this section is organized so that its structure is parallel to the previous section . indeed , their themes are similar , as the objective of this section is to establish another reduction , this time from the hypergraph setting to the digraph setting . the fundamental building block is now the following procedure , which should be compared with procedure 1 . * procedure 3 . * this takes as input a 3-graph @xmath0 with an even number of vertices . 1 . generate a random permutation @xmath202 of the vertices of @xmath0 . split it into @xmath7 successive pairs @xmath578 , @xmath579 , , @xmath580 . 2 . construct a directed graph @xmath86 with @xmath7 vertices , one corresponding to each of the pairs above . place a directed edge from @xmath88 to @xmath89 if and only if both hyperedges @xmath581 and @xmath582 are present in @xmath0 . consider @xmath414 and @xmath418 to be _ partners _ , and introduce the notation @xmath583 and @xmath584 . since a directed edge in @xmath86 corresponds to two hyperedges in @xmath0 , we account for this by defining @xmath585 to be the hypergraph containing all @xmath90 , @xmath586 , for each directed edge from @xmath88 to @xmath587 in @xmath86 . as already noted , hamilton cycles in @xmath86 correspond precisely to tight hamilton cycles in @xmath0 . indeed , the hyperedges corresponding to the directed edges of a hamilton cycle are precisely those which are needed for a tight hamilton cycle in the hypergraph . [ lem : hamilton - di - hyp ] let @xmath588 be edge - disjoint hamilton cycles in @xmath86 . then their associated hamilton cycles @xmath589 in @xmath0 are also edge - disjoint . * proof . * it suffices to show that given a fixed pairing @xmath590 , , @xmath580 from step 1 , any given hyperedge @xmath591 can be associated with at most one directed edge in step 3 . indeed , by construction , some pair of vertices in the hyperedge , say @xmath306 , must be a pair above . if the pair containing @xmath288 is @xmath592 , then the directed edge must be from @xmath306 to @xmath593 , hence unique . similarly , if the pair containing @xmath288 is @xmath594 , then the only possibility is that the directed edge is from @xmath594 to @xmath306 . @xmath174 as in the previous section , our next goal is to show that the result of procedure 3 ( a digraph ) is pseudo - random . the following lemma will be a useful component of the proof . [ lem : pairs - kept ] let @xmath595 be a given set of ordered pairs of distinct vertices @xmath306 , with @xmath596 . then with probability @xmath597 , the number @xmath406 of ordered pairs in @xmath595 which appear in the list of @xmath7 ordered pairs from step 1 of procedure 3 , satisfies @xmath598 . * proof . * the probability that any given @xmath599 appears as an ordered pair in step 1 is precisely @xmath600 . this is because @xmath43 must land in one of the @xmath7 positions of the permutation which correspond to first - vertices of ordered pairs , and then @xmath198 must land in the position immediately following @xmath43 . since @xmath601 , linearity of expectation gives @xmath415 } = \big ( 1 \pm \frac{\epsilon}{2 } \big ) \frac{|s|}{2n}$ ] . for concentration , we apply fact [ fact : conc - perm ] . suppose the permutation @xmath178 is converted to @xmath285 via the single transposition of the positions of the vertices @xmath43 and @xmath198 . let @xmath602 and @xmath603 be the other vertices in their respective ordered pairs ( according to @xmath178 ) . then , the only changes in the set of ordered pairs produced by step 1 is that the ordered pairs @xmath604 and @xmath605 change to @xmath606 and @xmath607 . all other ordered pairs remain the same , so their inclusion or exclusion in @xmath406 remains invariant . therefore , the value of @xmath406 can only change by at most 2 . fact [ fact : conc - perm ] then implies that the probability that @xmath406 deviates from its mean by more than @xmath608 is at most @xmath609 completing the proof . @xmath174 . now we show that the result of procedure 3 is a pseudo - random digraph . [ lem : d ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath610 . let @xmath0 be an @xmath82-uniform 3-graph on @xmath1 vertices ( @xmath1 even ) , and randomly construct the @xmath85-vertex directed graph @xmath86 according to procedure 3 . then , with probability @xmath241 , @xmath86 is @xmath611-uniform . * proof . * we verify the properties of @xmath86 one at a time , starting with out - degrees . fix any two distinct vertices @xmath612 , and let @xmath291 be the number of ordered pairs @xmath269 produced by step 1 such that @xmath613 . it suffices to show that with probability @xmath313 , @xmath614 . ( note the factor of 2 because @xmath86 has @xmath7 vertices . ) for this , first let @xmath307 be the set of ordered pairs @xmath269 , not necessarily produced by step 1 , such that @xmath615 . this is a deterministic set , from the hypergraph @xmath0 , whose size we can estimate by applying @xmath271-uniformity to @xmath0 with various auxiliary graphs @xmath32 . indeed , by applying the property with @xmath353 ( in figure [ fig : gamma ] ) and @xmath306 , the number of choices for @xmath43 such that @xmath616 is @xmath272 . then , given such a choice for @xmath43 , applying uniformity with @xmath353 and @xmath617 shows that there are @xmath272 choices for @xmath198 such that @xmath618 is also an edge . putting these estimates together , we conclude that @xmath619 . since @xmath620 , lemma [ lem : pairs - kept ] gives @xmath621 with the desired probability . [ cols="^,^,^,^,^ " , ] the arguments for the other properties are very similar . for in - degrees , we fix a pair of distinct vertices @xmath306 , and apply uniformity with @xmath353 to show there are @xmath272 choices for @xmath198 such that @xmath622 . applying again on @xmath623 , we find a further @xmath272 choices for @xmath43 such that @xmath624 . this is exactly the same estimate as we had in the out - degree case above , so the same argument completes it . to control the common out - degrees of pairs of vertices in @xmath86 , we fix 4 distinct vertices of @xmath0 , arranged in two pairs @xmath306 and @xmath625 . applying uniformity with @xmath626 ( see figure [ fig : gamma ] ) , there are @xmath627 choices for @xmath43 such that @xmath628 . after fixing such a choice for @xmath43 , another application of uniformity with @xmath629 ( see figure [ fig : gamma ] ) shows that there are @xmath630 choices for @xmath198 such that @xmath631 . therefore , there are @xmath632 pairs @xmath269 which could serve as the terminal vertex of both directed edges from @xmath306 and @xmath625 , if only @xmath269 appeared as a pair in step 1 . since @xmath633 , we may apply lemma [ lem : pairs - kept ] , and finish as before . a similar argument controls @xmath72 in @xmath86 . to establish the last part of property ( ii ) of @xmath611-regularity for @xmath86 , fix any 4 distinct vertices in @xmath0 arranged in two pairs @xmath306 and @xmath634 . applying uniformity with @xmath353 ( see figure [ fig : gamma ] ) , there are @xmath272 choices for @xmath43 such that @xmath635 . fix such an @xmath43 . applying uniformity again with @xmath636 ( see figure [ fig : gamma ] ) , with @xmath637 corresponding to the labels in figure [ fig : gamma ] , we see that there are @xmath638 choices for @xmath198 such that all of @xmath639 . hence there are @xmath640 choices for a pair @xmath269 which would be both an in - neighbor of @xmath641 and an out - neighbor of @xmath642 , if only @xmath269 appeared as a pair in step 1 . thus lemma [ lem : pairs - kept ] finishes this case as before . for property ( iii ) of @xmath611-regularity for @xmath86 , fix any 8 distinct vertices in @xmath0 , arranged in 4 pairs @xmath306 , @xmath625 , @xmath643 , and @xmath644 . applying uniformity with @xmath626 shows that there are @xmath70 choices for @xmath43 such that both @xmath645 . then , after fixing such an @xmath43 , applying uniformity with @xmath646 ( see figure [ fig : gamma ] ) shows that there are @xmath647 choices for @xmath198 such that all @xmath648 . hence there are @xmath649 choices for a pair @xmath269 which would out - neighbors of both @xmath650 and in - neighbors of @xmath651 , if only @xmath269 appeared as a pair in step 1 . since @xmath652 , we may apply lemma [ lem : pairs - kept ] , and finish as before . @xmath174 next we prove the analogue of lemma [ lem : gamma - cover ] , establishing that repeated runs of procedure 3 will cover the edges of @xmath0 fairly uniformly . [ lem : d - cover ] suppose @xmath1 and @xmath124 satisfy @xmath315 . let @xmath0 be an @xmath82-uniform 3-graph on @xmath1 vertices ( @xmath1 even ) , and randomly and independently construct @xmath653 digraphs @xmath654 according to procedure 3 . let @xmath655 be their corresponding 3-graphs . then , with probability @xmath319 , every edge of @xmath0 is covered @xmath656 times by the @xmath655 . * proof . * fix an edge @xmath591 of @xmath0 , and let @xmath322 be the probability that it is covered by @xmath657 . there are several ways in which this can happen . first , it could be that @xmath306 is a pair from the permutation @xmath658 for @xmath657 , and @xmath592 is another pair , with @xmath659 also an edge of @xmath0 . let us bound the probability @xmath334 of this particular event happening . let @xmath41 be the number of choices for @xmath43 such that @xmath660 . in terms of @xmath41 , we have @xmath661 . to see this , first expose the position of @xmath66 ; it is in the first half of an ordered pair with probability @xmath337 . next , @xmath73 must be occupy the second position in that ordered pair , and then @xmath288 must select a first half spot in one of the remaining ordered pairs . finally , there are @xmath41 valid choices for @xmath288 s partner . but by @xmath82-uniformity applied with @xmath353 ( figure [ fig : gamma ] ) , @xmath662 . so ( using @xmath341 ) , @xmath663 . the above argument can be run with any permutation of @xmath664 . also , it could be that @xmath288 occupied the second position in its ordered pair , e.g. , if @xmath665 and @xmath666 were the pairs involved . therefore , @xmath667 . since the @xmath11 random constructions are independent , the number of times @xmath591 is covered is @xmath323}$ ] , which has expectation @xmath668 . the chernoff bound implies that the probability that the binomial deviates from its mean by more than a factor of @xmath345 is at most @xmath669 taking a union bound over all @xmath670 edges @xmath591 , we obtain the desired result . @xmath174 next , we prove a result which will serve the same purpose as lemma [ lem : gamma - rare - consec ] of the previous section . procedure 3 constructs a digraph @xmath86 on half as many vertices by pairing up the vertices of @xmath0 . let us say that a set @xmath595 of 4 vertices of @xmath0 is _ condensed _ in @xmath86 if two of those pairs above contain all 4 vertices in @xmath595 . for example , if @xmath671 , and two vertices of @xmath86 are the ordered pairs @xmath672 and @xmath673 , then @xmath595 is condensed in @xmath86 . [ lem : d - rare - close ] suppose @xmath674 . construct @xmath11 independent @xmath675 according to procedure 3 . then with probability @xmath319 , every set @xmath595 of 4 vertices in @xmath0 is condensed in at most 9 of the @xmath675 . * proof . * for a fixed set of 4 vertices @xmath676 , the probability that it is condensed in @xmath677 is precisely @xmath678 . to see this , first expose the position of @xmath66 in the permutation @xmath658 which defines @xmath677 this identifies the position which contains the other vertex in @xmath66 s pair ; expose the vertex in that spot . the probability that it is @xmath73 , @xmath288 , or @xmath289 is precisely @xmath679 . assuming this is the case , there are still two vertices of interest left , say @xmath288 and @xmath289 . then expose the position of @xmath288 . this identifies the position which contains the other vertex in @xmath288 s pair ; expose the vertex in that spot . the probability that it is @xmath289 is precisely @xmath680 . since the @xmath675 are independent , the number of them which have the above property with respect to @xmath681 is stochastically dominated by @xmath682}$ ] . since we assumed @xmath683 , the probability that this exceeds 9 is at most @xmath684 so a union bound over all @xmath685 choices for @xmath681 completes the proof . we now formulate the analogue of procedure 2 for the hypergraph setting . * procedure 4 . * this takes as input a 3-graph @xmath0 with an even number of vertices , and an integer parameter @xmath11 . 1 . independently generate the random digraphs @xmath654 according to procedure 3 , and let @xmath655 be their corresponding 3-graphs . 2 . for each edge @xmath686 let @xmath687 . if @xmath688 , independently select a uniformly random index in @xmath689 to label @xmath414 with . 3 . for each @xmath675 , define the subgraph @xmath690 by keeping each edge @xmath321 if and only if both of its corresponding 3-graph edges are labeled by @xmath251 . 4 . for each @xmath251 , let @xmath691 be the 3-graph containing all hyperedges which correspond to the directed edges left in @xmath690 . note that @xmath692 . as in procedure 2 , the final step ensures that the @xmath691 are all disjoint , so lemma [ lem : hamilton - di - hyp ] shows that we can work on each @xmath690 independently , packing hamilton cycles with theorem [ thm : digraph ] . then , we will remove all @xmath691 from @xmath0 , and repeat . the next result establishes the necessary pseudo - random properties of the @xmath690 . [ lem : d - prime ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath693 . let @xmath0 be an @xmath82-uniform 3-graph on @xmath1 vertices ( @xmath1 even ) , and conduct procedure 4 with @xmath694 . let @xmath695 . then , with probability @xmath319 , every @xmath690 is @xmath696-uniform . * proof . * our restrictions on @xmath64 allow us to apply lemma [ lem : d ] ( @xmath697 ) and lemma [ lem : d - cover ] @xmath698 , so we have that with probability @xmath319 , after step 1 of procedure 4 : ( a ) : : every @xmath675 is @xmath611-uniform . ( b ) : : every edge in @xmath0 is covered @xmath379 times by the @xmath655 . condition on the above outcome of step 1 , and consider an arbitrary @xmath690 . it suffices to show that in the randomness of step 2 , with probability @xmath313 each individual @xmath690 has the desired properties , since @xmath699 by the given restrictions on @xmath700 . we start by verifying out - degrees . a vertex @xmath701 corresponds to a pair of vertices @xmath306 from @xmath0 . for each edge of @xmath675 which is directed away from @xmath155 , there is a distinct pair of vertices @xmath269 from @xmath0 . that directed edge remains in @xmath690 if and only if both of its associated hyperedges @xmath702 receive label @xmath251 . this happens with probability @xmath703^{-2}$ ] by property ( b ) above . yet property ( a ) ensures that the number @xmath41 of such directed edges in @xmath675 is @xmath704 , so the expected out - degree of @xmath155 in @xmath690 is @xmath705 . for concentration , note that all of the @xmath706 hyperedges associated with the @xmath41 directed edges are distinct , so their labels are generated independently . therefore , by the chernoff inequality , the probability that the out - degree of @xmath155 in @xmath690 deviates from its expectation by more than a factor of @xmath389 is at most @xmath707 because @xmath708 . therefore , with probability @xmath496 , the degree of @xmath155 in @xmath690 is @xmath709 . taking a union bound over all @xmath710 establishes uniformity for out - degrees . all other properties of @xmath711-uniformity follow by a similar argument . importantly , in each case , all directed edges correspond to disjoint pairs of hyperedges , so we can always use the chernoff inequality to establish concentration . the smallest mean we ever deal with is @xmath712 , from property ( iii ) of uniformity for @xmath690 , so the largest error in concentration is at most @xmath713 where we use @xmath714 . @xmath174 we now move to show how the hypergraph @xmath0 is affected by the deletion of all hypergraphs @xmath691 . as before , we use inclusion / exclusion to sandwich the quantities in question , using an accurate `` first - order '' estimate , together with a rough `` second - order '' upper bound . [ lem : h - delete-1 ] condition on the first step of procedure 4 covering every edge of @xmath0 @xmath379 times by the @xmath655 . fix any two distinct vertices @xmath715 , and a set @xmath595 of vertices such that @xmath635 for all @xmath716 . suppose that @xmath717 . then with probability @xmath718 , the number @xmath406 of vertices @xmath716 such that @xmath719 is covered by the @xmath691 is @xmath720 , * proof . * let @xmath721 . since we are conditioning on the result of step 1 , the only randomness left is in the independent exposure of all hyperedge labels , most of which are irrelevant for @xmath406 . we define @xmath411 , the set of relevant edges , as follows . for each @xmath722 which covers some @xmath723 , there is a partner edge @xmath724 such that @xmath414 remains in @xmath725 if and only if @xmath414 and @xmath418 receive the label @xmath422 . let @xmath411 be the collection of all partner edges obtained in this way , together with all @xmath726 . since we assume every @xmath727 is covered at most @xmath728 times by the @xmath722 , we always have @xmath729 . observe that all labels outside @xmath411 have no effect on @xmath406 at all , so we may condition on an arbitrary setting of those labels , leaving only @xmath730 independent labels to be exposed . next , we show that the probability that a particular edge @xmath731 is covered by the @xmath725 is @xmath732^{-1}$ ] . to see this , note that after revealing the label @xmath422 of @xmath414 , there is a single partner edge @xmath724 , whose label must match that of @xmath414 . since @xmath418 is covered @xmath379 times by the @xmath655 , it has probability @xmath732^{-1}$ ] of receiving that label @xmath422 . therefore by linearity of expectation , @xmath415 } = ( 1 \pm 1.04\epsilon ) \frac{|s|}{\kappa}$ ] . we use the hoeffding - azuma inequality to establish the concentration of @xmath406 , as the remaining randomness is a product space of dimension @xmath733 . consider the effect of changing the label of a hyperedge @xmath734 from @xmath251 to @xmath422 . note that its old partner edge @xmath735 overlapped with @xmath414 in two vertices , but also included a new vertex , say @xmath736 . changing the label of @xmath414 away from @xmath251 could potentially remove any or all of @xmath737 from the count for @xmath406 ( if @xmath735 had received label @xmath251 ) , so this could potentially reduce @xmath406 by up to 4 . similarly , setting the label of @xmath414 to @xmath422 can potentially increase @xmath406 by up to 4 . therefore , @xmath406 is 4-lipschitz , so the hoeffding - azuma inequality implies that the probability that @xmath406 deviates from its expectation by more than @xmath738 is at most @xmath739 where we used @xmath740 . therefore , with probability @xmath718 , we have @xmath741 , as desired . @xmath174 having proved an accurate `` first - order '' estimate , we move to the `` second - order '' upper bound ( analogous to lemma [ lem : d - delete - co - out ] ) . in the 3-graph setting , this breaks into two cases , which we treat in the following two lemmas . [ lem : h - delete-2-disjoint ] condition on the first step of procedure 4 covering every edge of @xmath0 @xmath742 times by the @xmath655 . fix any distinct vertices @xmath743 and any set of vertices @xmath595 such that both @xmath628 for all @xmath716 . suppose that @xmath744 . then with probability @xmath745 , the number @xmath406 of vertices @xmath716 which still have both @xmath719 and @xmath746 covered by the @xmath691 is at most @xmath747 . * proof . * again , the only remaining randomness is in the labeling of the hyperedges in step 2 . also , since each edge is covered by at most @xmath748 of the @xmath722 , less than @xmath749 of them are relevant for @xmath406 , for the same reason as in the previous proof . so , we may condition on all labels for irrelevant edges , and focus on the remaining product space over the relevant edges @xmath411 . let us bound the probability that a particular @xmath716 remains in the count for @xmath406 . first , expose the labels for @xmath750 and @xmath751 , and suppose they are @xmath19 and @xmath454 . let their partner edges be @xmath752 and @xmath753 . importantly , @xmath754 can not equal @xmath418 , because @xmath754 must overlap with @xmath414 in two vertices . hence the label of @xmath754 has not yet been exposed . similarly , the label of @xmath755 has not yet been exposed . however , note that it may happen that @xmath756 . in order for @xmath43 to remain in the count for @xmath406 , @xmath754 must receive label @xmath19 and @xmath755 must receive label @xmath454 . first we show that if @xmath757 , then the probability of this occurring is 0 . indeed , if @xmath758 , it clearly can not happen . but otherwise , if @xmath759 , it is impossible for @xmath756 in the first place , because in @xmath760 , each edge has a distinct partner edge , so @xmath754 and @xmath755 should be distinct . on the other hand , if @xmath761 , the probability that these independent labels equal @xmath19 and @xmath454 is at most @xmath762^{-2}$ ] , because each edge is covered by at least @xmath763 of the @xmath722 . therefore , we conclude that in all cases , the probability of a certain @xmath764 remaining in the count for @xmath406 is at most @xmath765 , so @xmath415 } \leq \frac{2}{\kappa^2 } |s|$ ] . finally , the same argument that we used in the proof of lemma [ lem : h - delete-1 ] shows that @xmath406 is 4-lipschitz over the product space of dimension @xmath766 . hence the probability that @xmath406 exceeds its expectation by more than @xmath767 is at most @xmath768 since we assumed @xmath744 . therefore , with probability @xmath718 , we have @xmath769 , as desired . @xmath174 [ lem : h - delete-2-incident ] condition on the first step of procedure 4 covering every edge of @xmath0 @xmath742 times by the @xmath655 , and on the property that every set of 4 vertices of @xmath0 is condensed in at most 9 of the @xmath675 . fix any distinct vertices @xmath770 and any set of vertices @xmath595 such that both @xmath771 for all @xmath716 . suppose that @xmath744 . then with probability @xmath718 , the number @xmath406 of vertices @xmath772 which still have both @xmath719 and @xmath659 covered by the @xmath691 is at most @xmath773 . * proof . * again , the only remaining randomness is in the labeling of the hyperedges in step 2 , so we may concentrate on the product space of dimension @xmath774 , where @xmath411 is the set of all edges @xmath775 and their possible partners . to bound the probability that a particular @xmath716 remains in the count for @xmath406 , expose the labels for @xmath750 and @xmath776 , and suppose they are @xmath19 and @xmath454 . let their partner edges be @xmath752 and @xmath753 . this time , it is possible for @xmath777 or @xmath778 , but that can only happen if the 4 vertices @xmath779 are condensed in @xmath780 . furthermore , if @xmath781 , then @xmath43 is automatically excluded from the count for @xmath406 , which is not a problem because we only seek an upper bound . therefore , the probability that @xmath43 contributes to @xmath406 is at most @xmath782 , where @xmath334 is the probability that @xmath414 and @xmath418 both receive the same label @xmath19 , with @xmath783 condensed in @xmath780 , and @xmath335 is the conditional probability that @xmath754 and @xmath755 receive labels @xmath19 and @xmath454 , given the exposure of the labels of @xmath414 and @xmath418 , and @xmath784 , @xmath785 . we assumed that @xmath783 were only condensed in at most 9 of the @xmath786 , so @xmath787 . as for @xmath335 , the same argument that we used in the previous lemma shows that if @xmath756 , then we must have had @xmath781 , so it is impossible for @xmath754 to receive @xmath19 and @xmath755 to receive @xmath454 , hence @xmath43 can not contribute to @xmath406 . the only remaining case is @xmath761 , and the probability that both receive the correct label is at most @xmath788^{-2}$ ] . putting everything together , we conclude that the probability that @xmath43 contributes to @xmath406 is at most @xmath789 , so @xmath415 } \leq \frac{12}{\kappa^2 } |s|$ ] . finally , the same argument that we used in the previous proof again establishes the lipschitz constant of 4 for @xmath406 in the product space of dimension @xmath766 . hence the probability that @xmath406 exceeds its expectation by more than @xmath767 is at most @xmath768 since we assumed @xmath744 . therefore , with probability @xmath718 , we have @xmath790 , as desired . @xmath174 now we combine our first - order and second - order bounds , and show that the pseudo - random properties of @xmath0 are maintained after deleting all @xmath691 . [ lem : h - delete ] suppose @xmath1 , @xmath6 , and @xmath124 satisfy @xmath791 . let @xmath792 and @xmath695 . let @xmath0 be an @xmath82-uniform 3-graph on @xmath1 vertices ( @xmath1 even ) , and conduct procedure 4 . let @xmath793 be the subgraph of @xmath0 obtained by deleting all edges covered by any @xmath691 . then , with probability @xmath794 , @xmath793 is an @xmath487-uniform 3-graph with @xmath795 and @xmath489 . * by applying lemmas [ lem : d - cover ] and [ lem : d - rare - close ] , the outcome of step 1 of procedure 4 will satisfy the following two properties with probability @xmath319 . ( the second follows since @xmath796 , and so @xmath683 . ) * every edge of @xmath0 is covered @xmath379 times by the @xmath655 . * every set of 4 vertices of @xmath0 is condensed in at most 9 of the @xmath675 . condition on this outcome . we will now show that in the remaining randomness of step 2 , the properties of @xmath487-uniformity are satisfied with probability @xmath319 . for this , fix a set of @xmath49 vertices @xmath797 , and let @xmath32 be an arbitrary @xmath41-vertex graph with @xmath50 edges . it suffices to show that in @xmath793 , the random variable @xmath798 is @xmath799 with probability @xmath718 , because we can then take a union bound over all @xmath800 choices for the @xmath205 , and all @xmath801 possibilities for @xmath32 . let @xmath595 be the set of all @xmath43 such that @xmath36 for every edge @xmath37 . our assumed @xmath82-uniformity gives @xmath802 . we will use a similar inclusion / exclusion argument as in the proof of lemma [ lem : digraph - delete ] to estimate @xmath120 in terms of @xmath803 . let @xmath804 be the edges of @xmath32 . for each @xmath805 , define the random variable @xmath806 as follows . let @xmath293 be the endpoints of edge @xmath807 . then let @xmath806 be the number of vertices @xmath808 such that @xmath809 is covered by some @xmath810 . also , for every two distinct @xmath811 , let the random variable @xmath812 count the number of vertices @xmath716 such that both @xmath813 and @xmath814 are covered by some @xmath810 , where @xmath815 and @xmath816 are the respective endpoints of @xmath817 and @xmath818 . in terms of these random variables , the principle of inclusion / exclusion always gives @xmath819 since we noted above that @xmath820 , we have @xmath821 , so lemma [ lem : h - delete-1 ] controls all @xmath822 with probability @xmath718 . also , since @xmath823 , lemmas [ lem : h - delete-2-disjoint ] and [ lem : h - delete-2-incident ] show that every @xmath824 with probability @xmath718 . putting everything together , and using @xmath50 and @xmath825 , we have @xmath826 now observe that @xmath827 again using @xmath825 , we bound the error term by @xmath524 . therefore , we have @xmath828 = ( 1 \pm \epsilon ' ) n ( p')^s,\ ] ] as desired . @xmath174 we finally finish the proof of theorem [ thm:3graph ] . the method is essentially the same as that used to prove theorem [ thm : digraph ] in the previous section . * proof of theorem [ thm:3graph ] . * let @xmath829 , @xmath528 , and @xmath529 . define the sequences @xmath530 , @xmath531 via the following recursion : @xmath830 let @xmath533 be the smallest index such that @xmath831 . note that @xmath535 only increases , so @xmath832 also note that since @xmath833 we have in general that @xmath834 . therefore , @xmath835 we now iteratively construct @xmath836 , such that each @xmath837 is @xmath542-uniform . indeed , consider the ( random ) procedure 4 applied to @xmath837 with respect to @xmath838 . let @xmath839 . this produces digraphs @xmath840 and 3-graphs @xmath841 with all @xmath841 disjoint . let @xmath842 be the result of deleting all @xmath841 from @xmath837 . to apply lemmas [ lem : d - prime ] and [ lem : h - delete ] , we must check that @xmath843 . but this follows from our initial assumption that @xmath844 since @xmath550 and @xmath845 . therefore , the two lemmas show that with probability @xmath490 , procedure 4 results in the following properties : ( i ) : : every @xmath840 is @xmath846-uniform . ( ii ) : : @xmath842 is @xmath556-uniform . we may assume this outcome , since we will only iterate @xmath553 times . in order to apply theorem [ thm : digraph ] to each @xmath840 , we must verify that @xmath847 . indeed , this is the case : @xmath848 by our initial assumption that @xmath849 . so , every @xmath840 can be packed with hamilton cycles , missing only an @xmath850-fraction of the edges . by lemma [ lem : hamilton - di - hyp ] , these edge - disjoint hamilton cycles in @xmath840 correspond to edge - disjoint hamilton cycles in @xmath841 , missing the same fraction of edges since there is a 2-to-1 correspondence between edges in @xmath841 and @xmath840 . we carry on the above procedure until we create @xmath851 . then , we will have packed hamilton cycles in @xmath852 , up to error of @xmath853-fraction . it remains to estimate the numbers of edges in @xmath851 and @xmath0 . note that in general , by applying @xmath82-uniformity to every pair of vertices ( using @xmath353 in figure [ fig : gamma ] ) , we can estimate the number of edges in any @xmath82-uniform 3-graph to be @xmath854 , because this counts every hyperedge @xmath855 times . thus @xmath0 has at least @xmath856 edges , and @xmath851 has at most @xmath857 edges . thus @xmath851 itself has at most @xmath858-fraction of the total number of edges . therefore , the fraction of edges of @xmath0 that have not been covered is at most @xmath859 since @xmath860 by inequality . this completes the proof . @xmath174 * the above proof showed that we can pack the edges of @xmath0 with hamilton cycles , up to a fractional error of at most @xmath571 , where @xmath861 . in fact , one can prove the same result for any @xmath862 , and all of the calculations can be recovered by following our proof . just as in section [ sec : digraph ] , we intentionally introduced a very high level of precision in lemma [ lem : d - cover ] . it is clear that the same argument can replace the @xmath574 with @xmath575 , etc . these smaller order errors are what accumulate into the decimal places in the final @xmath863 which appears in lemma [ lem : h - delete ] , and which later determines the value of @xmath288 . when @xmath863 is replaced by a constant arbitrarily close to 6 , the above argument produces a result with @xmath288 approaching @xmath864 . our proof of theorem [ thm:3graph ] is only valid when @xmath1 is divisible by four , because we required a factor of 2 in each reduction step ( from hypergraphs to digraphs , and then to bipartite graphs ) . although we do not expect this condition to be necessary , removing this restriction is open . we also leave open the question of packing hamilton cycles of type @xmath20 in @xmath19-uniform hypergraphs when @xmath865 and @xmath866 . finally , another interesting direction is to streamline the sets of pseudo - random properties which appear in the statements of our packing results ( both for digraphs and for 3-graphs ) .

nonnuclear x - ray sources with luminosities higher than @xmath3 are classified as ultraluminous x - ray sources ( ulxs ) . many of them are variable and most likely powered by accretion onto black holes . however , their luminosities , assuming isotropic emission , are higher than the maximum seen in galactic black hole binaries , suggesting the presence of intermediate mass black holes with masses of @xmath4 @xmath5 @xcite or super - eddington accretion onto stellar mass black holes of about 10 @xmath5 @xcite . discovery of intermediate mass black holes would be important because they can not be formed via core collapse of a single star with normal metallicity @xcite and may shed light onto the formation of supermassive black holes in the early universe @xcite . dynamical measurement of the compact object mass for ulxs has been unfeasible to date @xcite . instead , indirect means such as x - ray spectroscopy may shed light on the black hole mass . at a fixed eddington ratio , the inner temperature of the accretion disk scales with black hole mass as @xmath6 @xcite . application of this relation requires the black hole binary to be in the thermal state @xcite , in which the disk is believed to extend all the way to the last stable orbit around the black hole . however , the thermal state is rarely found in ulxs , the only known occurrences are m82 x41.4 + 60 @xcite and m82 x37.8 + 54 @xcite during outburst . x - ray sources with soft , @xmath7 kev , thermal spectra may be good candidates for intermediate , @xmath8 , mass black holes in the thermal state . ngc 247 is a nearby spiral galaxy in the sculptor group , with a modest star formation rate estimated to be about 0.1 @xmath5 yr@xmath9 by counting bright main - sequence stars @xcite or based on h@xmath10 observations @xcite . however , infrared and radio observations suggest a star formation rate 10 or 100 times lower @xcite . the distance to ngc 247 has been measured independently by different groups using various means . among them , the recent result from @xcite using near - infrared observations of cepheids is claimed to be the most accurate . we therefore adopt that distance of 3.4 mpc in this paper . previously , ngc 247 has been observed in the x - ray band with einstein @xcite , rosat @xcite , and xmm - newton @xcite . a bright , off - nucleus x - ray source ( ngc 247 ulx = 1rxs j004704.8@xmath11204743 ) was detected in all of these observations . rosat observations revealed a very soft spectrum with @xmath12 kev @xcite . later on , the source was observed to have brightened by a factor of 2 @xcite , suggestive of an accretion powered system . in 2001 , xmm - newton detected the source at the highest flux observed to date , with an unabsorbed luminosity of @xmath13 ( at 3.4 mpc ) in the 0.3 - 10 kev band , and the spectrum was still dominated by a soft thermal component with @xmath12 kev @xcite . these results imply that the source is a good candidate to be an intermediate mass black hole in the thermal state . unfortunately , the 2001 xmm - newton observation suffered strong background flares , resulting in little usable data and poor statistics in the extracted spectrum . thus , we proposed a new xmm - newton observation of ngc 247 to obtain a better spectral measurement . in this paper , we report results from the new observation about the ulx as well as other sources in ngc 247 . the observation was performed on 2009 dec 27 ( pi : h. feng ) , lasting about 35 ks , with the prime full window for the three cameras . the thin filter was chosen for the pn and medium filter for the mos . event lists were created from the observation data files using sas with recent calibrations . to minimize background contamination , we selected good time intervals ( gtis ) where the 10 - 15 kev count rate of the whole ccd was lower than the @xmath14 upper limit of the quiescent level . this resulted in 22 ks of clean exposure for the pn and 30 ks for the mos . images in two energy bands , 0.3 - 2 kev and 2 - 10 kev , were created for each ccd with a pixel size of 4.1 from events with flag = 0 and pattern @xmath15 4 for the pn or pattern @xmath15 12 for the mos . the edetect_chain tool was used to detect point sources simultaneously on the six images . a total number of 75 sources with likelihood ( = @xmath16 , where @xmath17 is the chance probability ) over 10 were detected and are listed in table [ tab : src ] and shown in figure [ fig : img ] . the astrometry was corrected by aligning the x - ray position of the known quasi - stellar object ( qso ) phl 6625 ( source 22 ) to its optical position quoted in the usno - b1.0 catalog . + the absorbed flux of each source was calculated from the net counts and an effective exposure with a response matrix created at the source position on the ccd assuming a power - law spectrum with galactic absorption ( @xmath18 @xmath19 ; * ? ? ? * ) and a photon index of 1.7 . we note that this is more accurate than using a single conversion factor in edetect_chain as the energy band is wide . in the calculation , either the pn or the averaged mos count rate was adopted depending on which likelihood was higher . for sources 38 , 60 , and 63 , only the mos2 data was adopted as they were not covered by the mos1 . the hardness ratio was calculated using edetect_chain for each source as @xmath20 , where @xmath21 and @xmath22 are vignetting - corrected net count rates in the soft and hard bands , respectively . for 15 sources , optical or infrared sources in the usno - b1.0 or two micron all sky survey catalogs are found in a 2-radius circle around their corrected x - ray positions , and are indicated in table [ tab : src ] . most of the sources with optical / ir counterparts , 11 out of 15 , are outside of the d25 ellipse of ngc 247 . for sources with 200 or more net counts , we extracted the pn and the mos spectra and fitted them simultaneously with a power - law model subject to interstellar absorption . the lower limit of the absorption column density was set to the galactic value . for sources 4 and 67 , only the mos spectra were adopted as they lie on a gap or bad pixels on the pn . for phl 6625 , only the pn data were adopted for the same reason , and a redshifted power - law model with @xmath23 was used . the best - fit spectral parameters are listed in table [ tab : top ] , except for the ulx . among them , seven sources ( 4 , 16 , 19 , 22 , 39 , 61 , and 62 ) are coincident with an optical counterpart in usno - b1.0 , and source 67 has two possible counterparts . we calculated their x - ray to optical flux ratios as @xmath24 , where @xmath25 is the observed flux in the 0.3 - 3.5 kev band in units of and @xmath26 is the visual magnitude @xcite . the average of the second red and blue magnitudes in usno - b1.0 , if available , was used to estimate the visual magnitude , otherwise , the red or blue magnitude is taken as an approximation to @xmath26 . the x - ray to optical flux ratios are 0.52 , 0.05 , 0.24 , 0.33 , @xmath27 , @xmath28 , 0.04 , respectively for sources 4 , 16 , 19 , 22 ( phl 6625 ) , 39 , 61 , 62 . for source 67 , the ratio is 0.20 or 0.54 depending on the counterpart . these values are consistent with those found for active galactic nuclei ( agn ) and clusters of galaxies @xcite ; the ratios for sources 39 and 61 are also consistent with those of normal galaxies , and the ratios for source 4 and source 67 ( with the dimmer optical counterpart ) are also consistent with those of bl lac objects . for other sources that are associated with an optical counterpart but do not have enough photons for spectral fitting , we calculated their 0.3 - 3.5 kev fluxes from count rates and the x - ray to optical flux ratios were found in a range from @xmath29 to @xmath30 , which are consistent with those of agn or normal galaxies , but smaller than those of x - ray binaries . two of them ( source 48 and source 57 with the brighter optical counterpart , both in d25 ) with ratios smaller than @xmath31 could also be foreground m or k dwarfs in the halo of the milky way . thus , we conclude that sources associated with bright optical sources are unlikely to be members of ngc 247 , and more likely to be background or foreground objects . the cumulative x - ray luminosity function ( xlf ) for sources inside the d25 ellipse of ngc 247 is plotted in figure [ fig : lum ] . for sources with enough photons , those in table [ tab : top ] and the ulx , the flux in the 0.3 - 10 kev band from spectral fitting is adopted , otherwise , the two fluxes in table [ tab : src ] after correction for galactic absorption are summed and adopted . for comparison , two model xlfs are shown . the shallower curve is the xlf expected for high mass x - ray binaries ( hmxbs ) : a power - law with a slope of @xmath32 as found for hmxbs in nearby starburst galaxies normalized to a star formation rate of 0.1 @xmath5 yr@xmath9 @xcite . the steeper curve is the xlf for low mass x - ray binaries ( lmxbs ) , a power - law with a slope of @xmath33 @xcite , normalized to the observed number of sources at @xmath34 . comparing the curves with the data , it is clear that the x - ray binary population in ngc 247 is dominated by lmxbs . normalized to the observed number at @xmath35 . the shallower dotted line is the luminosity function for hxmbs with a slope of @xmath32 normalized to a star formation rate of 0.1 @xmath5 yr@xmath9 . [ fig : lum ] ] , brighter than a given observed flux in the 2 - 8 kev band , @xmath36 , for sources outside of the d25 ellipse of ngc 247 . the dotted line is the distribution for agn with a slope of @xmath37 normalized to the observed solid angle of about 0.16 degree@xmath38 . [ fig : logns ] ] for sources outside of the d25 ellipse , we plotted @xmath39 versus @xmath40 , where @xmath36 is the observed flux in the 2 - 8 kev band and @xmath41 is the number of sources with flux greater than @xmath36 , see figure [ fig : logns ] . for comparison , we show a power - law index of @xmath37 as expected for agn @xcite . the power - law is normalized to a solid angle of 0.16 degree@xmath38 , equal to the sky area outside the d25 ellipse of ngc 247 and in the field of view of xmm - newton . we examined short - term variability for each source using photons in the 0.3 - 10 kev band with the kolmogorov - smirnov test . only the ulx shows evidence for variability with a chance probability lower than @xmath42 . due to the soft spectrum of the ulx and the thin filter on the pn , the number of photons detected on the pn is about 3 - 4 times that on each mos . thus , we used only the pn data for spectral analysis . the energy spectrum was extracted from events in a 30-radius circular region around the source to avoid ccd gaps and with flag and pattern the same as those used in creating the image . background was subtracted from events in a nearby circular region with a radius of 45 on the same ccd and off the readout column of the ulx . to avoid spurious spectral structure caused by over - sampling of the spectrum , the energy channels were grouped with a bin size of 1/5th of the local spectral resolution ( in terms of full width at half maximum ) and to have at least 15 net counts per bin . a single power - law subject to interstellar absorption is inadequate to fit the ulx spectrum , resulting in @xmath43 for 44 degrees of freedom ( dof ) . an absorbed multicolor disk ( mcd ) model provides a better fit with @xmath44 , but is still inadequate , especially at energies above 1.4 kev where a hard excess is obvious , see the top panel in figure [ fig : res ] . therefore , we used a model with the sum of a power - law and a mcd , which significantly improved the fit with @xmath45 . to test the significance of the power - law tail , we simulated @xmath46 spectra based on the single mcd model . in none of the simulated spectra did addition of a power - law reduce the @xmath47 by the amount seen in the real data . this suggests a chance probability less than @xmath48 . the f - test gives a consistent probability of @xmath49 despite caveats about its validity in such a case @xcite . from fits to the spectrum of ngc 247 ulx with models wabs @xmath50 diskbb , wabs @xmath50 ( diskbb + powerlaw ) , wabs @xmath50 edge @xmath50 ( diskbb + powerlaw ) , respectively . [ fig : res ] ] although the mcd plus power - law model improves the fit considerably , it still does not provide an adequate fit . as one can see from the middle panel of figure [ fig : res ] , most residuals arise around 1 kev and appear like an absorption feature . thus , we then tried two absorption components , both of which gave adequate fits . an absorption edge at 1.02 kev gave @xmath51 and a gaussian absorption line with a centroid of 1.16 kev gave @xmath52 . the residuals for both models are flat throughout the passband , see the bottom panel of figure [ fig : res ] . the absorption equivalent width is 0.13 to 0.19 kev , depending on the model . similarly , we tested the significance of the absorption component using simulations , and an edge was detected as significant in only 1 out of @xmath46 trails , indicative of a chance probability of @xmath53 . the spectral parameters are listed in table [ tab : spe ] . due to the poor fit of the mcd plus power - law model , we fixed the absorption column density at the best value , @xmath54 @xmath19 , obtained from the best - fit model . consistent parameters of the continuum components were obtained for all models . the unfolded spectrum from the mcd plus power - law model with absorption edge is plotted in figure [ fig : unf ] . the mcd is the dominant component of the spectrum , with a contribution of 96% of the total unabsorbed flux in the 0.3 - 10 kev band . ] we also replaced the mcd with a blackbody and obtained similar results as shown in table [ tab : spe ] . we note that the use of a more sophisticated model for interstellar absorption , such as tbabs , leads to almost identical results as obtained with wabs due to the moderate resolution and statistics . we tried to fit the spectrum with a mcd plus comptonization model , which is found to fit for ulx spectra with high statistics @xcite . this model provides an equally good fit as the mcd plus power - law model and results in almost identical parameters for the disk component . the disk temperature is @xmath55 kev and the inner disk radius is @xmath56 km assuming a face - on orientation . a similar edge is required in the model at @xmath57 kev , with optical depth of @xmath58 . the comptonization parameters are poorly constrained due to the low number of photons in the 2 - 10 kev band . light curves of the ulx were created using the pn events in the same region as for spectral analysis with flag = # xmmea_ep and pattern @xmath59 4 . continuous intervals with low background were selected from within the gtis . the 0.3 - 2 kev light curve shown in figure [ fig : pow ] has a bin size of 200 s. , title="fig : " ] , title="fig : " ] for each continuous interval , power spectra were calculated from a light curve with a bin size equal to the ccd frame time , @xmath6073 ms , using 8192-point fourier transforms . a final power spectrum was created by averaging all the individual power spectra with frequency channels geometrically binned by a factor of 1.2 . each point in the final spectrum is the mean of at least 33 individual data points , to assure validity in modeling the spectrum using minimal @xmath47 . the final power spectrum in the 0.3 - 2 kev band normalized following @xcite is shown in figure [ fig : pow ] . a power - law plus a constant can adequately fit the power spectrum , with the constant describing the poisson noise level of @xmath61 and the power - law describing low - frequency noise with a slope of @xmath62 . the source count rate is 0.10 counts s@xmath9 in the 0.3 - 2 kev band and the background photon fraction is estimated to be 4.1% . the fractional root - mean - square ( rms ) power of the source is estimated to be 0.52 by integrating the power - law component over 0.001 - 0.1 hz , or 0.31 extrapolating the power - law to the frequency range 0.1 - 10 hz . the power spectrum in the 2 - 10 kev band is consistent with that of poisson noise , due to poor statistics . we re - analyzed the xmm - newton observation of ngc 247 obtained in 2001 in a similar way . as the background is too high to find a quiescent level adaptively for the pn data , an additional requirement that the background count rate is less than 1.5 counts s@xmath9 for the pn is imposed to generate gtis . the pn and the mos spectra are fitted simultaneously with a mcd plus power - law model , resulting in @xmath63 . the mcd component contributes about 88% of the unabsorbed luminosity in the 0.3 - 10 kev band , which is @xmath64 , with a disk inner temperature of @xmath65 kev and a face - on inner radius of @xmath66 km . the photon index of the power - law component is poorly constrained , with a 90% upper limit of 3.8 . the interstellar absorption has a column density of @xmath67 @xmath19 . the best - fit parameters are consistent with those reported in @xcite within errors . the xmm - newton observation of ngc 247 obtained in december 2009 is , to date , the deepest x - ray observation of the galaxy and allows an x - ray source population study . as shown in figure [ fig : lum ] , sources inside the d25 ellipse of ngc 247 have a luminosity function well consistent with that of lmxbs . a population of x - ray sources dominated by lmxbs is reasonable given the low star formation rate of the galaxy . for sources outside of the d25 ellipse of the galaxy , the @xmath39 versus @xmath40 distribution is roughly consistent with that expected for agn @xcite ( figure [ fig : logns ] ) at fluxes above @xmath68 . below this flux , the distribution flattens significantly with decreasing flux . this may be caused by the decrease of sensitivity due to vignetting at large off - axis angles . some of these sources have bright optical counterparts with an x - ray to optical flux ratio consistent with that of agn . therefore , we conclude that most sources outside of the d25 ellipse of ngc 247 are likely background objects . the observed x - ray spectrum of the ulx is completely dominated by a cool , thermal emission component with a temperature around 0.1 kev ; a weak power - law tail with a photon index of about 2.5 is also evident but contributes only a small fraction of the flux . this spectral shape is in contrast with that of many other ulxs whose high energy component contributes the majority of the flux ( more than 60% ) and can be explained by an optically - thick corona @xcite . in fact , gladstone et al . ( 2009 ) fits can predict an intrinsically soft - dominated spectrum , but this is a model - dependent result , while the observed spectral shapes of their sources are always dominated by the emission from a hard tail . it is unlikely that the high energy component in ngc 247 ulx arises from an optically thick corona ; otherwise , such a corona must have a tiny covering factor since most disk photons are unscattered . the spectrum , however , is consistent with that of black hole binaries in the thermal state @xcite , and may hint at a connection in nature . an important observational characteristic of the thermal state is that the disk luminosity varies roughly with the 4th power of the disk inner temperature , equivalent to a constant disk inner radius . this source has shown a consistent disk inner radius derived from two xmm - newton observations , though the constraint from the 2001 observation is loose . for a positive identification of the thermal state , one also needs timing information . this ulx displays red noise with a @xmath1 form above the white noise , which is consistent with that seen in the thermal state . however , the integrated rms power is significantly higher than that seen in the thermal state . for galactic black holes , the total rms power in the 0.1 - 10 hz band is usually less than 0.075 in the thermal state , and only in the hard state is the rms power found to be greater than 0.15 @xcite . however , the x - ray spectrum of the ulx is much too soft to be classified as in the hard state . if the ulx is in the thermal state , then the source would be an interesting candidate to be an intermediate mass black hole . the accretion disk in the thermal state is believed to extend all the way to the innermost stable circular orbit ( isco ) around the central black hole . the isco radius depends only on the black hole mass and spin , as @xmath69 for a non - spinning black hole or 6 times smaller for an extremely spinning object , where @xmath70 is the gravitational constant , @xmath71 is the black hole mass , and @xmath72 is the light speed . assuming a hardening correction of 1.7 and that the maximum temperature occurs at 2.4 times the isco radius @xcite , the isco radius can be expressed as @xmath73 , where @xmath74 is the apparent inner radius derived directly from the normalization of the mcd component . adopting the parameters from the best - fitted model , mcd plus power - law with absorption edge , and assuming a face - on disk , we have @xmath75 km in the 90% confidence region , corresponding to the isco radius of a @xmath76 schwarzschild black hole , or a more massive black hole if spinning or the disk is tilted . the observed luminosity is about a few percent of the eddington limit of such a black hole , which is consistent with eddington ratios typically found in the thermal state . similar results can be obtained with the gaussian absorption line . due to the lack of a characteristic feature in the power spectrum , we are unable to constrain the black hole mass via timing . the rms - mass relation derived for agn @xcite may have issues in extrapolation to lower masses . the faintness in the 2 - 10 kev band of the source prevents us from implementing the means proposed by @xcite to estimate the black hole mass . a few other ulxs show a dominant soft , thermal emission component , and are classified as supersoft ulxs : m101 ulx-1 @xcite , antennae x-13 @xcite , m81 uls1 @xcite , and ngc 4631 x1 @xcite . these sources also show extreme variability : the observed fluxes vary by up to a factor of a few tens , and the intrinsic luminosity changes may be as large as factors of @xmath77 @xcite . short - term variability has been detected in almost all of these sources at timescales down to a few ks @xcite . m101 ulx-1 also showed a power spectrum with a @xmath1 form at frequencies below @xmath42 hz @xcite during its outburst in 2000 march with variability more pronounced at high energies ( 0.8 - 2 kev ) than in the lower band ( 0.2 - 0.8 kev ) @xcite . for a direct comparison , we calculated the rms power of ngc 247 ulx in the same energy bands and found that the fractional rms is 0.80 in the 0.8 - 2 kev band and 0.44 in the 0.2 - 0.8 kev band . similar to ngc 247 ulx , ngc 4631 x1 also exhibits an absorption edge at 1.01 kev with an optical depth of 2.0 in its high state @xcite . these similarities suggest that ngc 247 ulx may be a new member of the supersoft , ultraluminous family . the other soft ulxs seem to keep a constant temperature during huge changes in luminosity , indicative of a dramatic change of the emitting surface area of the same order . @xcite has pointed out that this would lead to difficulty in interpreting the emission as due to accretion power release within a few schwarzschild radii of the black hole . for ngc 247 ulx , the four rosat and two xmm - newton observations have detected a change by a factor of about 3 of the observed flux in 0.1 - 2 kev , which is significantly smaller than others . therefore , the thermal state interpretation has not been ruled out for ngc 247 ulx ; new , multiple observations are the key to revealing its nature . an alternative explanation of the soft , thermal emission is that it arises from a photosphere that has a radius of @xmath2 cm , inferred from the blackbody component . the shortest timescale of variation above white noise is about @xmath0 s ( figure [ fig : pow ] ) , which places a constraint that the emitting area must be smaller than @xmath78 cm , compatible with the size obtained from the blackbody parameters . this photosphere is unlikely the surface of a white dwarf , as the observed luminosity and temperature are too high to be explained by steady nuclear burning on a white dwarf surface @xcite . plus , presence of the power - law component with a 0.3 - 10 kev luminosity of @xmath79 is unexpected in such a case . it has been suggested that supercritical accretion could drive massive outflows above the disk and form an optically - thick photosphere , with a temperature expected to be about 0.1 kev , which could shield the inner disk emission when viewed at a high inclination angle @xcite . in this picture , the supersoft ulxs and normal ulxs are the same objects with different viewing angles . this interpretation may be in accord with the absorption feature observed at @xmath601 kev , which is likely due to l shell transitions of highly ionized iron . besides this ulx and ngc 4631 x1 , less similar absorption features near this energy have been detected in the ulx ngc 1365 x1 @xcite and in the quasar pds 456 @xcite , and are explained as a spectral signature of a massive outflow . the absence of absorption from low - z elements like oxygen could be a consequence of high ionization . we conclude that the x - ray properties of the ngc 247 ulx are unlike any popular emission state known in black hole binaries . the source exhibits a spectrum like the thermal state , but shows variability much stronger than seen in the thermal state . a possible strong absorption feature is detected . this source and other supersoft ulxs may exhibit a new accretion state that is different from those seen in galactic black hole binaries and the majority of ulxs . a variety of potential thermal emission components have been suggested to be present in the energy spectrum of ulxs , including the soft excess ( 0.1 - 0.4 kev ) found in the majority of ulxs @xcite , the high - temperature ( @xmath60kev ) ulxs @xcite , and supersoft ulxs . the properties of these thermal components are summarized and compared in table [ tab : comparsion ] , with ngc 247 ulx and m82 x41.4 + 60 in the thermal state highlighted . among them , a disk interpretation is best established for m82 x41.4 + 60 because its thermal state is consistent with those of stellar - mass x - ray binaries in terms of timing and spectral properties and spectral evolution @xcite . for the others , definitive evidence is lacking for positive determination of their nature , except that the disk explanation is ruled out for other supersoft ulxs other than ngc 247 ulx . based on the timing information , it appears unlikely that the soft excess in normal ulxs and the soft ( dominant ) spectral component in supersoft ulxs have the same physical origin . phl 6625 was identified as a radio - quiet background qso at @xmath23 based on a mg ii emission line @xcite . a single power - law model with interstellar absorption provides an adequate fit to its x - ray spectrum . the parameters listed in table [ tab : top ] are obtained from fitting in the 0.3 - 10 kev energy range , and the hydrogen column density is very close to the galactic value at @xmath80 @xmath19 , indicative of negligible absorption within ngc 247 . however , if we include data down to 0.2 kev , two more energy bins , then the best - fit absorption is @xmath81 @xmath19 . this is caused by the low flux between 0.2 - 0.3 kev , and is consistent with the results obtained from rosat observations , in which a total absorption column density @xmath82 @xmath19 was derived from fitting in the 0.1 - 2 kev band , and the excess absorption mainly takes effect at energies below 0.3 kev @xcite . those authors claimed a positive detection of absorbing materials in ngc 247 assuming a galactic @xmath83 @xmath19 based on 21 cm observations of neutral hydrogen . here , we confirm this conclusion with tighter constraints , even assuming a higher galactic absorption . therefore , with future high - resolution spectroscopic observations , this qso could be used to study the interstellar medium within ngc 247 . for the 2009 data , the power - law photon index at the rest frame is @xmath84 . we extracted spectra for this object from the 2001 xmm - newton observation and found a harder spectrum with @xmath85 at a flux level of @xmath86 in 0.3 - 10 kev similar to that in 2009 . the fluxes observed with xmm - newton are about ten times lower than the rosat flux of @xmath87 in the 0.2 - 2.4 kev band @xcite . no short - term x - ray variability has been detected for this object . * source 28*. the source is hard , and its spectrum can be described by a power - law with @xmath88 , but the residuals vary systematically with energy . adding an mcd component improves the residuals and results in @xmath89 . the disk has a temperature at the inner radius of @xmath90 kev with a bolometric luminosity of @xmath91 assuming a face - on geometry , and the power - law component has a photon index @xmath92 . the disk component has a fractional luminosity of 11% in the 0.3 - 10 kev band . if this source is a black hole binary , it is possibly in the hard state . * source 34*. the source sits at the north end of the ngc 247 disk , and was bright during the 2001 xmm - newton observation with @xmath93 , @xmath94 @xmath19 , and @xmath95 in the 0.3 - 10 kev band , or @xmath96 corrected for absorption . however , it became faint in the 2009 xmm - newton observation , with a flux of @xmath97 in the 0.3 - 10 kev band , six times dimmer than in 2001 . it is possibly source 3 in @xcite from einstein observations and source 1 in @xcite from rosat observations . the rosat observations revealed a luminosity of @xmath98 corrected for galactic absorption only ( at 3.4 mpc ) in the 0.1 - 2 kev band , or @xmath99 corrected for all absorption . the luminosity in the same band from the 2001 xmm - newton observation is about @xmath100 . there is no bright optical counterpart found at the x - ray position . however , the source is associated with a possible star forming region , seen in the ultraviolet image from the 2001 xmm - newton observation , with diffuse emission and star clusters nearby . therefore , the source is likely an x - ray binary in ngc 247 , and is perhaps an accreting black hole . * source 36*. this source could be source 3 in @xcite . the best - fitted column density is consistent with the galactic value . its flux estimated from the 2001 xmm - newton observation is consistent with that in 2009 . in the x - ray image , the source seems to show extended features and may consists of multiple objects , which can only be resolved with future chandra observations . * source 61*. its x - ray spectrum can not be adequately fitted by a single power - law model , with @xmath101 ; adding a gaussian absorption line significantly improves the fit with @xmath102 . the absorption line has a central energy of @xmath103 kev and a width of @xmath104 kev with an optical depth of @xmath105 . the equivalent width is calculated to be around 0.30 kev . its observed flux in 2001 was about twice of that in 2009 . there is a bright optical source with @xmath106 and @xmath107 within 2 of the x - ray position , and the x - ray to optical flux ratio is estimated to be @xmath108 . as discussed above , this source is likely a background agn . due to the low absorption column density , the absorption line is likely produced intrinsic to the source . similar absorption has been seen in agn @xcite . * source 13*. this is the hardest source detected with xmm - newton . it is obvious in the hard band ( 2 - 10 kev ) but undetected in the soft band ( 0.3 - 2 kev ) , consistent with @xmath109 @xmath19 for a power - law spectrum with @xmath110 . the source lies about 7 west of the nucleus of ngc 247 , and is likely a highly absorbed background agn . * sources 41 , 50 , 59 , and 72*. these four sources are the softest besides the ulx and are only detected in the soft band ( 0.3 - 2 kev ) with a hardness ratio around @xmath111 , which is consistent with a spectrum with a blackbody temperature @xmath112 kev or a power - law photon index @xmath113 , assuming galactic absorption . sources 41 and 72 are outside of the d25 ellipse of ngc 247 , but sources 50 and 59 are inside . their x - ray luminosities are about @xmath114 . these sources do not appear to have bright optical counterparts . they could be canonical supersoft sources that are powered by nuclear burning on the surface of white dwarfs , or high redshift agn . llccccccccc 1 & j004609.2@xmath11204504 & 00 46 09.29 & @xmath1120 45 04.9 & 2.21 & @xmath115 & @xmath116 & @xmath117 & 13.7 & no & + 2 & j004617.0@xmath11204151 & 00 46 17.00 & @xmath1120 41 51.8 & 1.44 & @xmath118 & @xmath119 & @xmath120 & 15.9 & no & + 3 & j004623.0@xmath11204740 & 00 46 23.02 & @xmath1120 47 40.3 & 1.02 & @xmath121 & @xmath122 & @xmath123 & 73.5 & no & + 4 & j004628.7@xmath11205005 & 00 46 28.73 & @xmath1120 50 05.2 & 0.78 & @xmath124 & @xmath125 & @xmath126 & 195.4 & no & o + 5 & j004631.1@xmath11203959 & 00 46 31.14 & @xmath1120 39 59.1 & 1.89 & @xmath127 & @xmath128 & @xmath129 & 17.2 & no & + 6 & j004632.1@xmath11205044 & 00 46 32.11 & @xmath1120 50 44.2 & 1.85 & @xmath130 & @xmath131 & @xmath132 & 18.6 & no & + 7 & j004633.1@xmath11205457 & 00 46 33.16 & @xmath1120 54 57.3 & 1.40 & @xmath133 & @xmath134 & @xmath135 & 20.1 & no & + 8 & j004634.1@xmath11205411 & 00 46 34.15 & @xmath1120 54 11.2 & 0.93 & @xmath136 & @xmath137 & @xmath138 & 51.1 & no & + 9 & j004635.1@xmath11203529 & 00 46 35.17 & @xmath1120 35 29.2 & 1.25 & @xmath139 & @xmath140 & @xmath141 & 42.3 & no & + 10 & j004635.2@xmath11203648 & 00 46 35.23 & @xmath1120 36 48.2 & 1.43 & @xmath142 & @xmath143 & @xmath144 & 40.6 & no & + 11 & j004636.2@xmath11210018 & 00 46 36.29 & @xmath1121 00 18.7 & 1.00 & @xmath145 & @xmath146 & @xmath147 & 99.8 & no & o + 12 & j004638.0@xmath11203609 & 00 46 38.04 & @xmath1120 36 09.6 & 1.62 & @xmath148 & @xmath149 & @xmath150 & 32.5 & no & + 13 & j004639.6@xmath11204533 & 00 46 39.67 & @xmath1120 45 33.6 & 1.05 & @xmath151 & @xmath125 & @xmath152 & 40.4 & no & + 14 & j004642.6@xmath11204507 & 00 46 42.64 & @xmath1120 45 07.5 & 1.49 & @xmath153 & @xmath154 & @xmath155 & 19.4 & no & + 15 & j004645.8@xmath11210030 & 00 46 45.85 & @xmath1121 00 30.6 & 1.78 & @xmath156 & @xmath157 & @xmath141 & 21.7 & no & + 16 & j004646.3@xmath11204742 & 00 46 46.39 & @xmath1120 47 42.7 & 0.51 & @xmath158 & @xmath159 & @xmath160 & 327.2 & no & o + 17 & j004646.9@xmath11203658 & 00 46 46.92 & @xmath1120 36 58.0 & 1.56 & @xmath161 & @xmath162 & @xmath163 & 19.8 & no & + 18 & j004647.2@xmath11203628 & 00 46 47.23 & @xmath1120 36 28.1 & 1.78 & @xmath164 & @xmath165 & @xmath163 & 22.1 & no & + 19 & j004647.7@xmath11203837 & 00 46 47.72 & @xmath1120 38 37.3 & 0.65 & @xmath166 & @xmath167 & @xmath168 & 333.1 & no & o + 20 & j004648.8@xmath11205247 & 00 46 48.85 & @xmath1120 52 47.0 & 1.20 & @xmath169 & @xmath170 & @xmath171 & 60.6 & no & + 21 & j004650.4@xmath11203854 & 00 46 50.47 & @xmath1120 38 54.0 & 1.76 & @xmath172 & @xmath173 & @xmath174 & 17.4 & no & + 22 & j004651.8@xmath11204328 & 00 46 51.82 & @xmath1120 43 28.5 & 0.14 & @xmath175 & @xmath176 & @xmath177 & 14105.3 & no & qso + 23 & j004654.6@xmath11204129 & 00 46 54.63 & @xmath1120 41 29.4 & 0.78 & @xmath178 & @xmath179 & @xmath180 & 92.5 & yes & o + 24 & j004654.8@xmath11204100 & 00 46 54.87 & @xmath1120 41 00.9 & 0.56 & @xmath181 & @xmath182 & @xmath183 & 381.1 & yes & + 25 & j004655.7@xmath11203407 & 00 46 55.74 & @xmath1120 34 07.6 & 1.55 & @xmath184 & @xmath185 & @xmath117 & 19.5 & no & + 26 & j004655.9@xmath11205248 & 00 46 55.92 & @xmath1120 52 48.4 & 1.48 & @xmath186 & @xmath187 & @xmath188 & 18.1 & no & + 27 & j004656.4@xmath11205611 & 00 46 56.43 & @xmath1120 56 11.9 & 1.34 & @xmath189 & @xmath190 & @xmath191 & 35.7 & no & + 28 & j004659.3@xmath11204538 & 00 46 59.38 & @xmath1120 45 38.8 & 0.26 & @xmath192 & @xmath193 & @xmath194 & 1796.1 & yes & + 29 & j004659.7@xmath11204227 & 00 46 59.70 & @xmath1120 42 27.3 & 1.34 & @xmath195 & @xmath196 & @xmath197 & 38.7 & yes & + 30 & j004700.5@xmath11203359 & 00 47 00.55 & @xmath1120 33 59.7 & 3.09 & @xmath198 & @xmath199 & @xmath200 & 10.9 & no & + 31 & j004701.9@xmath11205450 & 00 47 01.98 & @xmath1120 54 50.8 & 1.62 & @xmath201 & @xmath202 & @xmath203 & 23.2 & no & + 32 & j004702.7@xmath11205302 & 00 47 02.79 & @xmath1120 53 02.9 & 1.97 & @xmath204 & @xmath205 & @xmath150 & 14.7 & yes & + 33 & j004703.2@xmath11204545 & 00 47 03.23 & @xmath1120 45 45.9 & 1.12 & @xmath136 & @xmath139 & @xmath206 & 56.2 & yes & + 34 & j004703.2@xmath11203657 & 00 47 03.28 & @xmath1120 36 57.9 & 1.64 & @xmath207 & @xmath208 & @xmath200 & 20.1 & yes & + 35 & j004703.4@xmath11204321 & 00 47 03.44 & @xmath1120 43 21.5 & 1.02 & @xmath209 & @xmath210 & @xmath211 & 42.7 & yes & + 36 & j004703.6@xmath11204015 & 00 47 03.68 & @xmath1120 40 15.6 & 0.42 & @xmath212 & @xmath213 & @xmath214 & 1184.2 & yes & + 37 & j004703.9@xmath11204743 & 00 47 03.90 & @xmath1120 47 43.8 & 0.16 & @xmath215 & @xmath216 & @xmath217 & 9827.5 & yes & ulx + 38 & j004704.5@xmath11205812 & 00 47 04.57 & @xmath1120 58 12.0 & 1.94 & @xmath218 & @xmath219 & @xmath220 & 17.8 & no & + 39 & j004705.9@xmath11205239 & 00 47 05.92 & @xmath1120 52 39.4 & 0.75 & @xmath221 & @xmath187 & @xmath222 & 152.1 & yes & o + 40 & j004705.9@xmath11204152 & 00 47 05.94 & @xmath1120 41 52.4 & 1.07 & @xmath223 & @xmath131 & @xmath224 & 69.9 & yes & + 41 & j004707.3@xmath11205749 & 00 47 07.38 & @xmath1120 57 49.3 & 1.30 & @xmath225 & @xmath226 & @xmath227 & 51.4 & no & + 42 & j004708.0@xmath11204346 & 00 47 08.01 & @xmath1120 43 46.8 & 1.57 & @xmath228 & @xmath216 & @xmath229 & 10.4 & yes & + 43 & j004709.0@xmath11205214 & 00 47 09.03 & @xmath1120 52 14.7 & 0.73 & @xmath230 & @xmath231 & @xmath232 & 172.1 & yes & + 44 & j004709.1@xmath11204622 & 00 47 09.16 & @xmath1120 46 22.8 & 1.10 & @xmath233 & @xmath234 & @xmath235 & 56.1 & yes & + 45 & j004710.0@xmath11204708 & 00 47 10.06 & @xmath1120 47 08.4 & 0.34 & @xmath236 & @xmath237 & @xmath238 & 1117.2 & yes & + 46 & j004710.3@xmath11204343 & 00 47 10.32 & @xmath1120 43 43.9 & 1.29 & @xmath239 & @xmath170 & @xmath163 & 24.8 & yes & + 47 & j004712.4@xmath11204219 & 00 47 12.48 & @xmath1120 42 19.1 & 1.45 & @xmath240 & @xmath241 & @xmath242 & 15.6 & yes & + 48 & j004712.7@xmath11203837 & 00 47 12.74 & @xmath1120 38 37.0 & 1.88 & @xmath243 & @xmath244 & @xmath132 & 14.0 & yes & o , i + 49 & j004713.7@xmath11205113 & 00 47 13.70 & @xmath1120 51 13.4 & 1.15 & @xmath245 & @xmath246 & @xmath247 & 45.1 & yes & + 50 & j004715.1@xmath11204717 & 00 47 15.17 & @xmath1120 47 17.3 & 0.96 & @xmath248 & @xmath249 & @xmath250 & 66.3 & yes & + 51 & j004715.9@xmath11205502 & 00 47 15.94 & @xmath1120 55 02.5 & 0.75 & @xmath251 & @xmath252 & @xmath253 & 235.8 & yes & + 52 & j004716.5@xmath11204814 & 00 47 16.59 & @xmath1120 48 14.4 & 0.85 & @xmath254 & @xmath241 & @xmath255 & 84.1 & yes & + 53 & j004716.6@xmath11203346 & 00 47 16.63 & @xmath1120 33 46.4 & 1.09 & @xmath256 & @xmath257 & @xmath258 & 11.1 & no & + 54 & j004717.9@xmath11205508 & 00 47 17.93 & @xmath1120 55 08.6 & 0.99 & @xmath259 & @xmath260 & @xmath261 & 97.5 & yes & + 55 & j004718.5@xmath11204739 & 00 47 18.56 & @xmath1120 47 39.8 & 0.90 & @xmath262 & @xmath139 & @xmath263 & 64.2 & yes & + 56 & j004719.7@xmath11205601 & 00 47 19.77 & @xmath1120 56 01.3 & 1.90 & @xmath130 & @xmath264 & @xmath163 & 11.8 & no & + 57 & j004720.3@xmath11205248 & 00 47 20.32 & @xmath1120 52 48.7 & 1.00 & @xmath265 & @xmath134 & @xmath266 & 110.1 & yes & o(@xmath267 ) , i + 58 & j004721.0@xmath11204932 & 00 47 21.05 & @xmath1120 49 32.3 & 1.32 & @xmath268 & @xmath216 & @xmath269 & 25.0 & yes & + 59 & j004721.4@xmath11205104 & 00 47 21.48 & @xmath1120 51 04.1 & 1.80 & @xmath270 & @xmath271 & @xmath272 & 14.8 & yes & + 60 & j004725.2@xmath11205442 & 00 47 25.23 & @xmath1120 54 42.4 & 1.28 & @xmath273 & @xmath162 & @xmath274 & 67.5 & no & + 61 & j004727.3@xmath11204721 & 00 47 27.36 & @xmath1120 47 21.6 & 0.28 & @xmath275 & @xmath276 & @xmath277 & 2204.7 & no & o + 62 & j004728.7@xmath11204301 & 00 47 28.76 & @xmath1120 43 01.7 & 0.72 & @xmath278 & @xmath122 & @xmath279 & 162.3 & no & o + 63 & j004730.6@xmath11205434 & 00 47 30.64 & @xmath1120 54 34.0 & 0.68 & @xmath280 & @xmath281 & @xmath282 & 223.0 & no & + 64 & j004733.5@xmath11204557 & 00 47 33.57 & @xmath1120 45 57.5 & 1.16 & @xmath283 & @xmath284 & @xmath220 & 27.4 & no & + 65 & j004737.4@xmath11203648 & 00 47 37.49 & @xmath1120 36 48.3 & 1.50 & @xmath285 & @xmath286 & @xmath129 & 25.3 & no & o + 66 & j004740.7@xmath11204442 & 00 47 40.77 & @xmath1120 44 42.5 & 1.87 & @xmath287 & @xmath288 & @xmath132 & 15.9 & no & + 67 & j004741.8@xmath11205231 & 00 47 41.84 & @xmath1120 52 31.5 & 0.52 & @xmath289 & @xmath290 & @xmath291 & 485.8 & no & o(@xmath267 ) + 68 & j004743.4@xmath11204941 & 00 47 43.49 & @xmath1120 49 41.4 & 0.68 & @xmath292 & @xmath293 & @xmath294 & 230.5 & no & + 69 & j004744.5@xmath11205321 & 00 47 44.53 & @xmath1120 53 21.2 & 1.03 & @xmath295 & @xmath296 & @xmath297 & 65.2 & no & o + 70 & j004752.1@xmath11204053 & 00 47 52.19 & @xmath1120 40 53.9 & 1.78 & @xmath298 & @xmath185 & @xmath299 & 22.5 & no & o + 71 & j004753.5@xmath11205435 & 00 47 53.55 & @xmath1120 54 35.3 & 1.23 & @xmath300 & @xmath301 & @xmath117 & 39.8 & no & + 72 & j004759.1@xmath11205201 & 00 47 59.19 & @xmath1120 52 01.2 & 2.02 & @xmath302 & @xmath303 & @xmath304 & 11.7 & no & + 73 & j004808.3@xmath11204553 & 00 48 08.39 & @xmath1120 45 53.3 & 3.61 & @xmath305 & @xmath306 & @xmath200 & 13.7 & no & + 74 & j004810.7@xmath11204818 & 00 48 10.75 & @xmath1120 48 18.5 & 9.07 & @xmath307 & @xmath308 & @xmath309 & 11.2 & no & + 75 & j004812.3@xmath11204755 & 00 48 12.30 & @xmath1120 47 55.6 & 2.56 & @xmath310 & @xmath311 & @xmath309 & 12.4 & no & + lllllll 4 & @xmath312 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & 8.7/10 + 16 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & 20.8/17 + 19 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & 18.3/15 + 22 & @xmath327 & @xmath328 & @xmath329 & @xmath330 & @xmath331 & 99.2/99 + 24 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 & 17.9/14 + 28 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & 57.2/56 + 36 & @xmath342 & @xmath343 & @xmath344 & @xmath345 & @xmath346 & 59.9/52 + 39 & @xmath312 & @xmath347 & @xmath348 & @xmath349 & @xmath350 & 7.3/8 + 43 & @xmath351 & @xmath352 & @xmath353 & @xmath354 & @xmath355 & 6.0/11 + 45 & @xmath356 & @xmath357 & @xmath313 & @xmath358 & @xmath359 & 46.8/46 + 51 & @xmath360 & @xmath361 & @xmath362 & @xmath363 & @xmath364 & 7.6/8 + 61 & @xmath365 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & 80.9/60 + 62 & @xmath370 & @xmath371 & @xmath372 & @xmath373 & @xmath374 & 12.2/12 + 63 & @xmath375 & @xmath376 & @xmath377 & @xmath378 & @xmath379 & 7.2/13 + 67 & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 & 14.1/13 + 68 & @xmath312 & @xmath385 & @xmath386 & @xmath387 & @xmath388 & 19.3/14 + llllcllcllllll @xmath389 & 3.2 & @xmath390 & @xmath391 & & @xmath392 & @xmath393 & & & & & @xmath394 & @xmath395 & 88.0/43 + @xmath396 & @xmath397 & @xmath398 & @xmath399 & & @xmath400 & @xmath401 & & @xmath57 & @xmath402 & & @xmath403 & @xmath404 & 36.8/40 + @xmath405 & @xmath406 & @xmath407 & @xmath408 & & @xmath409 & @xmath410 & & @xmath411 & @xmath412 & @xmath413 & @xmath414 & @xmath415 & 37.0/39 + @xmath416 & 2.7 & @xmath417 & @xmath418 & & @xmath392 & @xmath419 & & & & & @xmath420 & @xmath421 & 81.5/43 + @xmath422 & @xmath423 & @xmath424 & @xmath425 & & @xmath426 & @xmath427 & & @xmath428 & @xmath429 & & @xmath430 & @xmath431 & 39.0/40 + @xmath432 & @xmath433 & @xmath434 & @xmath435 & & @xmath436 & @xmath437 & & @xmath438 & @xmath439 & @xmath413 & @xmath414 & @xmath440 & 38.9/39 llccc ngc 247 ulx & low ( @xmath600.1 kev ) & major & high & unknown + other supersoft ulxs & low ( @xmath600.1 kev ) & major & high & no + m82 x41.4 + 60 during outburst & high ( @xmath601 kev ) & major & low & yes + other hot ulxs & high ( @xmath601 kev ) & major & unknown & maybe + soft excesses in ulxs & low ( 0.1 - 0.4 kev ) & minor & weak & rare +

in recent years , several studies have been carried out on the radio spectral properties of extra - galactic high - frequency selected radio sources . sadler et al . ( 2006 ) presented the general properties of a sample of radio sources with flux densities above 100 mjy obtained as part of the pilot 20 ghz survey ( at20 g ) undertaken with the australia telescope compact array ( ricci et al . 2004 ) . with the use of a radio `` two - colour diagram '' , first used in the radio band by kesteven et al . ( 1977 ) , they characterized the spectral properties of the high - frequency radio sources and confirmed the diversity of the radio spectra and the difficulty of predicting high - frequency properties by extrapolating the results of surveys observed at lower frequencies ( see also tucci et al . 2008 and massardi et al . 2008 ) . considering the two classical radio source populations ( extended steep - spectrum and compact flat - spectrum sources ) , sadler et al . ( 2006 ) found that roughly 87 per cent of the high - frequency selected sources are flat - spectrum and 13 per cent are steep - spectrum . tucci et al . ( 2008 ) performed a multifrequency spectral analysis of a sample of radio sources observed with the very small array ( vsa ) at 33 ghz exploiting existing data at different frequencies down to 1.4 ghz . one of the most interesting results they found is that the majority of 33 ghz - selected sources show flatter spectra going to higher frequency ( about 50 per cent of the sample ) or become inverted ( about 19 per cent of the sample ) . this contradicts the standard models for high - frequency source spectra , which predict a spectral steepening due to the ageing of high - energy electrons . the result they found is completely consistent with the `` unified model '' for agn : the steep - spectrum component , which arises from extended radio lobes , rapidly decreases with frequency , while the compact emission starts to dominate at frequency @xmath2 1 ghz if the radio jet axis lies enough close to the line of sight . the difficulty of predicting high - frequency flux densities from the extrapolation of lower frequency data is testified by the significant number of sources found at high frequency that were not predicted . the percentage of these `` unexpected '' sources is observed to steadily increase with frequency : 10 per cent at 15 ghz ( taylor et al . 2001 ) , 18 per cent at 20 ghz ( sadler et al . 2006 ) and 32 per cent at 33 ghz ( tucci et al . 2008 ) . the analysis of the radio spectral properties of 320 bright extragalactic radio sources with flux density @xmath3500 mjy at 20 ghz extracted from the at20 g was discussed in massardi et al . in this sample , spectral steepening is a common feature that becomes more prominent at higher frequencies , even if flat and inverted spectrum objects still dominate the sample . in a more recent paper , prandoni et al ( 2010 ) studied the radio properties of a faint sample ( s @xmath30.6 mjy ) , selected at lower frequency ( 5 ghz ) and associated with early - type galaxies , finding strong spectral similarities with the massardi et al . ( 2008 ) sample and suggesting that faint radio galaxies are mostly counterparts of the brighter ones , with no significant influence introduced by different flux limits and frequency selection . massardi et al . ( 2011a ) analysed the full sample release ( fsr , murphy et al . 2010 ) of the at20 g survey , which is 91 per cent complete above 100 mjy and 79 per cent complete above 50 mjy , and found that the high - frequency bright source sample is dominated by flat - spectrum sources , while the fraction of steep - spectrum sources increases with decreasing flux density . these results have also been confirmed by procopio et al . ( 2011 ) , who observed 263 sources , selected at 23 ghz in the wmap maps , at 5 , 8 and 22 ghz almost simultaneously to the observations performed between 30 and 857 ghz by the european space agency planck satellite ( tauber et al . the optical properties of the at20 g full sample presented in mahony et al . ( 2011 ) , who used supercosmos ( hambly et al . 2001 ) for optical identification and the 6df galaxy survey ( jones et al . 2009 ) or the literature for redshifts , showed that the sources which accrete cold gas have steeper radio spectral indices with respect to the sources that accrete hot gas . this behaviour suggests that the radio emission in sources with cold - mode accretion is dominated by the jets . interestingly , the fraction of flat- and steep - spectrum sources for the objects with no identification ( blank field ) is very different to that observed in the spectroscopic sample ; in fact the percentage of blank fields increases dramatically at steep spectral indices . the authors suggested that this increase could indicate a population of high - z ultra - steep - spectrum sources . in recent papers , massardi et al . ( 2011b ) and bonavera et al . ( 2011 ) presented the planck australia telescope compact array co - eval observations ( paco ) , which provided two complete samples of flux density - selected sources : a bright sample ( @xmath4 500 mjy ) and a faint sample ( @xmath4 200 mjy ) with extensive multi - frequency coverage ( 540 ghz ) . the spectral analysis showed no significant differences between the two samples . the steepening of the sources at high frequencies is confirmed . the main difference is a larger fraction of steep - spectrum sources in the faint sample , mostly at the expense of peaked- and flat - spectrum sources . in a subsequent paper , bonaldi et al . ( 2013 ) presented a detailed analysis of a complete sample of 69 sources spectrally - selected from the at20 g sample as those having inverted or upturning spectra and observed in the framework of the paco project . most of the source spectra ( 85 per cent ) are smooth and well described by a double power - law , while the remaining sources have complex spectra . the majority of the radio sources are likely blazars probably caught during a bright phase at 20 ghz . gawroski et al . ( 2010 ) also investigated the dependence of spectral index distribution on flux density for a sample of 57 sources detected at 30 ghz with the ocra prototype receiver at the 32-m toru radio telescope above a flux density limit of 5 mjy . they found that the proportion of steep spectrum sources increases with decreasing flux density . they found no evidence for an unexpected population of sources above their completeness limit of 10 mjy whose spectra rise towards higher frequencies . peel et al . ( 2011 ) measured the 30 ghz flux densities of 605 radio sources from the combined radio all - sky target eight - ghz survey ( crates ) with the same instrumentation as gawroski et al . they studied the radio spectra between 1.4 and 30 ghz and found that 75 per cent have steepened at @xmath5 compared with @xmath6 , 6 per cent of the sources show a ghz - peaked spectrum shape with a peak between 4.8 and 8.4 ghz , 10 per cent of the sources are flat or rising and another 10 per cent are inverted . kurinsky et al . ( 2013 ) reported the multi - frequency observations ( 4.843.3 ghz ) of 89 sources with 37 ghz flux density @xmath3 1 jy carried out with the vla / jvla . the sample contains a much higher fraction ( 42 per cent ) of flat - spectrum sources compared to 510 per cent of sources found in other samples selected at lower frequency . this sample extends the radio spectral energy distribution of the radio sources in the 5857 ghz regime . in this paper we present the results of the multi - frequency follow - up observations of the k - band northern wide survey ( knows ) pilot project ( righini et al . 2012 , hereafter paper i ) in a comprehensive way . in section [ sez : bright ] a brief summary of the pilot survey is provided ; while in section [ sez : lowfol ] the results of the 5/8 ghz follow - up observations are presented . in section [ sez : ocra ] we report on the 30 ghz follow - up observations and data reduction . in section [ sez : spec ] we show the radio spectra between 1.4 and 30 ghz and the optical identifications . in section [ sez : res ] we present the results of the analysis of radio spectra , and we compare them with the results from other high - frequency selected samples . in section [ sez : sum ] we summarize our conclusions . throughout this paper we use the following cosmological parameters : @xmath771 km s@xmath8 mpc@xmath8 , @xmath90.27 and @xmath100.73 ( larson et al . 2011 ) . during the years 20102011 , the medicina 32-m dish hosted the seven - feed 1826.5 ghz receiver built for the sardinia radio telescope , with the aim of performing its commissioning . this opportunity was exploited to carry out a pilot survey at 20 ghz over the sky region north of @xmath11 . this survey produced a catalogue of 73 confirmed sources down to a flux density limit of 115 mjy , which can be considered complete above 200 mjy . the 73 confirmed sources come from a selected list of 151 candidates chosen from the brighter and more reliable targets for the 20 ghz follow - up . for this reason the sample analysed in this paper is referred to as the knows 20 ghz bright sample . a more detailed description is provided in paper i. precise 20 ghz flux densities and position measurements were derived for all sources through 20 ghz follow - up observations , carried out at the 32-m medicina telescope , in the framework of the multi - frequency ( 5 , 8 , 20 , 30 ghz ) study of the sample . the medicina 20 ghz follow - up observations are described in detail in paper i. here we focus on the 5/8 ghz medicina observations and on the 30 ghz observations carried out at the toru telescope . follow - up observations of the knows bright sample were performed with the 32-m medicina telescope on dec 11@xmath12 , 12@xmath12 and 13@xmath12 2010 in the frequency ranges 55.15 ghz and 8.188.33 ghz . the target list of 73 sources was scheduled to be observed at 5 and 8 ghz in blocks of pointings based on ra range . a list of flux density calibrators ( 3c286 , 3c295 , 3c48 , 3c123 and ngc7027 ) were observed twice a day interspersed with blocks of target sources . both calibrators and target sources were observed in multiple ra and dec cross - scans centred on the positions obtained from the knows 20 ghz bright sample . the cross - scans were 6@xmath13hpbw wide ( where the half power beam width is 7.5 arcmin at 5 ghz and 4.8 arcmin at 8.3 ghz ) at the scanning speed of 3@xmath14/min and 2@xmath14/min at 5 and 8 ghz respectively . for comparison , the angular resolution at 20 ghz is 1.7 arcmin . data were recorded into fits files . they were then flagged in order to discard scans clearly affected by radio frequency interference and/or severe atmospheric disturbances . the data were flux - calibrated according to the absolute flux density scale by ott et al . the ra and dec cross - scans were then stacked for each target source in order to improve the s / n ratio of the detections by decreasing the rms noise . the stacked cross - scan was gaussian - fitted in order to obtain flux density and positional offsets in ra and dec . the positional offsets were then used to correct the flux density for the signal attenuation caused by the primary beam profile . a thorough description of the aforementioned calibration and fitting procedures ( performed with the data reduction package oscar , procopio et al . 2011 ) , together with the error analysis for the flux density measurements are provided in paper i. of the 73 sources , 60 ( 82 per cent ) and 57 ( 78 per cent ) were detected at 5 and 8 ghz respectively . non - detections at 5/8 ghz were mainly due to the limited sensitivity of the receivers , which were not properly cooled down at the time of observations . the @xmath15 flux density limit was 63 mjy / beam at 5 ghz and 85 mjy / beam at 8 ghz . the 5/8 ghz flux densities of the sources are listed in table [ tab : flux ] . the 30 ghz follow - up observations were carried out with the ocra - p receiver at the toru 32-m telescope ( poland ) . ocra - p is the prototype of the ocra - f ( one centimeter radio array - faraday ) multi - feed 30 ghz receiver developed in the framework of the ocra program ( browne et al . the nominal sensitivity of the receiver considering an overall system temperature of 40 k and a bandwidth of 6 ghz , is 6 mjy s@xmath16 . the fwhm of the antenna beam is 72 arcsec at 30 ghz . the 30 ghz observations were carried out on september 9@xmath17 , november 15@xmath17 and december 3@xmath18 2010 . the targets were observed using cross - scans or on / off measurements depending on the source flux density ( lowe et al . 2007 ; gawroski et al . the planetary nebula ngc 7027 was used as primary flux density calibrator . ngc 7027 was observed 610 times for each observing session . hafez et al . ( 2008 ) report a flux density of 5.39@xmath190.04 jy at 33 ghz at an epoch of 2003.0 with a secular decrease of @xmath20 per cent per year ( ott et al . 1994 ) . by extrapolating the flux density to 2010.0 a value of 5.32@xmath190.04 jy was obtained , which was then scaled to 30 ghz using the ngc 7027 spectral index of @xmath21 . the 30 ghz flux density of ngc 7027 thus obtained is 5.38@xmath190.04 jy . secondary flux density calibration was performed using the signal generated from a noise diode after each source observation . the flux density of ngc 7027 was measured at different elevations in order to obtain elevation - dependent gain corrections . the correction for atmospheric absorption was obtained by calculating the opacity from the system temperature measurements at zenith and at 30@xmath14 of elevation . the data reduction was performed using a custom software package described in peel et al . ( 2011 ) and in greater detail in peel ( 2009 ) . in total 295 measurements were obtained on the full list of 73 sources . data affected by poor weather conditions were discarded , as well as cross - scans in which the peak amplitude measured by the ocra - p beams was offset by more than 20 per cent . 52 targets were detected at 30 ghz corresponding to 71 per cent of the total . the 30 ghz flux densities of the detected sources are listed in table [ tab : flux ] . [ fig : ocra ] shows the scatter plot between the 30 ghz planck early release compact source catalogue ( ercsc ; planck collaboration 2011 ) flux densities and the 30 ghz ocra flux densities of the 16 targets present in both data sets . planck and ocra flux densities are in reasonable agreement especially at the bright end , indicating that our flux densities are not affected by significant systematic effects . source variability might account for a scatter that is larger than the flux density error bars for the fainter objects . it might also be useful to compare in terms of variability the planck - ocra scatter plot with the one in figure 6 of gawroski et al . ( 2010 ) where 42 ocra - p 30 ghz flux densities are matched with those from the very small array source subtractor at 33 ghz . 10 out of the 42 sources show variations greater than 2@xmath22 ( combined error ) . the radio spectra of the knows sources are shown in fig . [ fig : sed ] . integrated flux densities at 1.4 ghz were added from the nrao - vla sky survey ( nvss , condon et al . 1998 ) to extend the frequency baseline . the nvss flux densities were extracted from the nvss online catalogue from a cross - match with the knows 20 ghz follow - up positions within a search radius of 50 arcsec . in order to assess the reliability of these cross - matches , all the 1.4 ghz nvss images were retrieved from the nvss online postage - stamp server and visually inspected . all 73 sources in our sample appear to have nvss counterparts , with three sources ( knows 080817 + 731514 , knows 144830 + 760137 , knows 184218 + 794540 ) showing two nvss components and three sources ( knows 160731 + 850159 , knows 211402 + 820437 , knows 220549 + 743632 ) showing three nvss components in the nvss images . for these six sources the nvss components were summed to provide the integrated 1.4 ghz flux densities reported in table [ tab : flux ] . when available , the 30 ghz , 44 ghz and 70 ghz planck ercsc ( planck collaboration 2011 ) flux densities were also added to the radio spectra . however such measurements are not included in table [ tab : flux ] due to the sparsity of this information . we note that flux density measurements used to derive the radio spectra are not coeval . in particular the high - frequency ones ( @xmath23 5 ghz ) were taken within a time range of 1.5 years : between august 2009 and june 2010 for the planck flux density measurements ; in november 2010 for the ocra - p 30 ghz knows follow - up ; in december 2010 for the 5/8 ghz follow - up and in april 2011 for the knows 20 ghz follow - up measurements . the 1.4 ghz nvss observations were done between september 1993 and october 1996 with additional patching observations carried out during the fourth quarter of 1997 . this means that our radio spectra might be affected by variability . massardi et al . ( 2010 ) found a 56 per cent median variability on a 6 months timescale in the range 518 ghz , and up to a @xmath24 per cent variability over 9 months , so we do not expect variability to significantly affect the spectral analysis discussed in the following sections . the long time lag between the 1.4 ghz observations and the higher frequency ones ( 17 years ) is not troublesome because at lower frequencies most of the emission comes from radio lobes which are more stable than radio cores . in this paper we are interested only to the statistical behaviour of spectral indices in the sample . in this case positive and negative variations caused by short - term variability tend to cancel out . however , it must be born in mind , when using our data to interpret individual source spectra , that some sources may experience large variability also in relatively short periods . some long - term flux density monitoring projects carried out at high frequencies at ovro , metsahovi and umrao ( richards et al . 2011 , ciaramella et al . 2004 , pyatunina et al . 2005 , hovatta et al . 2007 ) show plenty of such examples . the nasa extragalactic database ( ned ; _ http://ned.ipac.caltech.edu/_ ) was then used to identify the radio source optical counterparts , to find redshifts and to obtain optical types within a search radius of 1 arcmin to the 20 ghz knows follow - up positions . in cases of multiple counterparts we chose the one which matched the nvss flux density and position information . we also extracted the information on the optical class , b@xmath25 and r@xmath26 magnitude from the supercosmos sky surveys database ( sss ; _ http://www-wfau.roe.ac.uk/sss/index.html_ ) . we again allowed for a 1-arcmin wide search box and in cases of multiple counterparts we chose the entry with the minimum distance from the follow - up 20 ghz radio position . the optical and redshift information of the knows sample is summarized in table [ tab : optid ] . according to ned we have 36 qsos ( 49 per cent ) , 17 galaxies ( 23 per cent ) and one planetary nebula , while 19 objects ( 36 per cent ) have no optical types . in supercosmos , 18 objects ( 25 per cent ) show galaxy appearance ( class 1 ) , 51 ( 70 per cent ) a stellar appearance ( class 2 ) , one ( the planetary nebula ) is noisy ( class=4 ) and 3 objects ( 4 per cent ) have no supercosmos identification . for the 19 sources that have no optical type in ned we use the supercosmos classification , when available . putting together the ned and supercosmos classifications , sources in the knows bright sample can be divided as follows : 20 ( 27 per cent ) are galaxies or with galaxy appearance and 50 ( 68 per cent ) are qsos or with stellar appearance , in agreement with the results of previous works ( see e.g. sadler et al . 2006 ) . two have no optical classification and one is a planetary nebula . 42 out of the 72 ( 58 per cent ) extragalactic objects ( we exclude here the planetary nebula ) in the knows bright sample have a redshift measurement . 30 of them are qsos and 12 are radiogalaxies according to the combined ned / supercosmos optical classification . for 34 objects ( 24 qsos and 10 radiogalaxies ) we computed the radio and optical luminosities using the scheme described in mahony et al . ( 2011 ) . to compute luminosities across a wide range of redshifts it is essential to shift the observed frequencies into the rest frame using a @xmath27correction . as qsos and galaxies have very different seds in the optical regime each optical class needs to be @xmath27corrected in a different manner . the @xmath27correction formula by de propris et al . ( 2004 ) , making use of the b@xmath28r@xmath26 colour , was applied to the radiogalaxies . for qsos a power law was instead used with a spectral index @xmath29 . the redshift and luminosity statistics for the qsos and radiogalaxies in the knows sample are summarized in table [ tab : lumz ] . as expected , qsos are found on average at higher redshifts than radiogalaxies and are intrinsically brighter both in the radio and the optical band . lrrrrrrrrrr + + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) + name & s30 & @xmath30 & s20 & @xmath31 & s8 & @xmath32 & s5 & @xmath33 & s1.4 & @xmath34 + & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) + + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) & ( 9 ) & ( 10 ) & ( 11 ) + name & s30 & @xmath30 & s20 & @xmath31 & s8 & @xmath32 & s5 & @xmath33 & s1.4 & @xmath34 + & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) & ( mjy ) + knows 001302@xmath35723123 & 167 & 5 & 325 & 18 & & & 559 & 14 & 509.1 & 11.2 + knows 001312@xmath35774854 & & & 208 & 19 & 548 & 11 & 718 & 16 & 2101.3 & 63.0 + knows 001633@xmath35791641 & & & 247 & 12 & 738 & 31 & 1123 & 14 & 3651.0 & 88.6 + knows 001716@xmath35813441 & 480 & 20 & 745 & 38 & 1362 & 19 & 1471 & 16 & 692.8 & 20.8 + knows 001948@xmath35732725 & 870 & 31 & 1283 & 64 & 1068 & 25 & 858 & 14 & 1251.8 & 37.6 + knows 020313@xmath35810622 & 546 & 17 & 174 & 9 & 277 & 12 & 172 & 3 & 211.0 & 6.3 + knows 020336@xmath35723253 & 420 & 22 & 594 & 34 & 549 & 21 & & & 229.9 & 6.9 + knows 020649@xmath35841102 & 189 & 14 & 120 & 12 & & & 244 & 9 & 100.0 & 3.0 + knows 020954@xmath35722920 & 488 & 21 & 488 & 42 & 454 & 14 & 398 & 8 & 669.9 & 20.1 + knows 021733@xmath35734923 & 2739 & 96 & 2894 & 145 & 3760 & 32 & 3829 & 22 & 2271.6 & 68.1 + knows 022459@xmath35765544 & & & 130 & 11 & 330 & 9 & 554 & 12 & 1927.2 & 57.8 + knows 035447@xmath35800918 & 313 & 13 & 294 & 21 & & & 406 & 11 & 643.7 & 19.3 + knows 041050@xmath35765649 & 962 & 36 & 1342 & 67 & 2122 & 38 & 2984 & 19 & 5620.1 & 168.6 + knows 041318@xmath35745107 & 152 & 11 & 297 & 25 & 900 & 22 & 1094 & 18 & 2847.6 & 80.0 + knows 042132@xmath35835837 & 188 & 16 & 205 & 10 & & & & & 5.2 & 0.6 + knows 050844@xmath35843202 & 140 & 12 & 301 & 15 & 315 & 13 & 359 & 19 & 294.8 & 8.9 + knows 061048@xmath35724843 & 228 & 16 & 360 & 25 & 484 & 12 & 683 & 18 & 1041.6 & 31.3 + knows 062555@xmath35820228 & 528 & 19 & 628 & 45 & 755 & 17 & 1016 & 14 & 681.0 & 20.4 + knows 063921@xmath35732454 & 1254 & 48 & 1467 & 73 & 1026 & 22 & & & 903.6 & 27.1 + knows 064132@xmath35881200 & & & 204 & 14 & & & 235 & 5 & 126.2 & 3.8 + knows 072608@xmath35791135 & 405 & 17 & 498 & 25 & 836 & 18 & 914 & 14 & 501.0 & 15.0 + knows 074713@xmath35763918 & 373 & 17 & 498 & 25 & 688 & 17 & 657 & 11 & 133.1 & 4.0 + knows 074922@xmath35742038 & 231 & 19 & 416 & 21 & 461 & 19 & 563 & 11 & 510.3 & 15.3 + knows 075039@xmath35790914 & 339 & 16 & 240 & 12 & 268 & 12 & 361 & 6 & 181.5 & 5.5 + knows 075052@xmath35824200 & 336 & 26 & 475 & 24 & 644 & 16 & 846 & 11 & 1845.1 & 55.4 + knows 080817@xmath35731514 & 74 & 3 & 320 & 16 & 378 & 13 & 431 & 12 & 376.8 & 9.3 + knows 092934@xmath35861236 & & & 187 & 9 & 255 & 11 & 483 & 10 & 142.7 & 4.3 + knows 093056@xmath35742017 & & & 240 & 15 & 400 & 11 & 369 & 10 & 211.0 & 6.3 + knows 101009@xmath35825020 & 283 & 13 & 353 & 18 & 420 & 13 & & & 503.7 & 15.1 + knows 104421@xmath35805447 & 577 & 30 & 1188 & 56 & 706 & 17 & 658 & 7 & 828.3 & 24.9 + knows 105359@xmath35863004 & & & 126 & 9 & & & 265 & 4 & 230.6 & 6.9 + knows 105812@xmath35811438 & 756 & 27 & 795 & 44 & 756 & 17 & 764 & 13 & 240.3 & 7.2 + knows 110149@xmath35722544 & 904 & 31 & 912 & 46 & 980 & 20 & 954 & 12 & 1245.6 & 37.4 + knows 110410@xmath35765859 & 189 & 9 & 278 & 14 & 545 & 13 & 773 & 15 & 2340.6 & 58.8 + knows 115311@xmath35805837 & 689 & 28 & 880 & 44 & 1215 & 23 & 1429 & 14 & 1343.4 & 40.3 + knows 120019@xmath35730054 & 507 & 23 & 750 & 38 & 1666 & 30 & 2572 & 14 & 5564.7 & 166.9 + knows 122340@xmath35804016 & 429 & 18 & 479 & 24 & 508 & 19 & 847 & 11 & 705.1 & 21.2 + knows 132143@xmath35831623 & 96 & 3 & 395 & 28 & 417 & 17 & 318 & 18 & 565.4 & 17.0 + knows 132351@xmath35794258 & & & 378 & 19 & 453 & 22 & & & 599.4 & 18.0 + knows 135324@xmath35753307 & 387 & 16 & 341 & 17 & 524 & 17 & & & 132.6 & 4.0 + knows 135756@xmath35764330 & 441 & 23 & 458 & 23 & 551 & 14 & & & 647.2 & 19.4 + knows 140638@xmath35782816 & & & 115 & 6 & 199 & 10 & & & 396.8 & 11.1 + knows 144830@xmath35760137 & 639 & 28 & 1225 & 61 & 747 & 17 & 507 & 13 & 280.8 & 6.2 + knows 152107@xmath35785837 & & & 178 & 9 & & & & & 124.9 & 3.8 + knows 155608@xmath35742107 & & & 190 & 11 & & & & & 179.9 & 5.4 + knows 160731@xmath35850159 & & & 265 & 13 & & & 320 & 8 & 819.8 & 18.0 + knows 160922@xmath35794023 & & & 287 & 22 & 408 & 12 & 522 & 15 & 1238.9 & 37.2 + knows 163235@xmath35823228 & 908 & 40 & 798 & 40 & 928 & 15 & 1109 & 13 & 801.8 & 14.8 + knows 172404@xmath35765328 & 621 & 29 & 730 & 37 & & & & & 424.0 & 12.7 + knows 180044@xmath35782812 & 2376 & 81 & 2660 & 133 & 2482 & 44 & 2180 & 15 & 2223.5 & 66.7 + knows 182316@xmath35793856 & 56 & 2 & 279 & 14 & 556 & 9 & 611 & 11 & 296.9 & 8.4 + knows 183659@xmath35750741 & 85 & 3 & 257 & 14 & 299 & 6 & 244 & 5 & 133.9 & 4.0 + knows 184218@xmath35794540 & 607 & 34 & 941 & 47 & 2773 & 48 & 3927 & 18 & 11555.9 & 280.6 + knows 185458@xmath35735129 & 266 & 13 & 263 & 14 & 331 & 13 & 419 & 13 & 465.1 & 11.7 + knows 192754@xmath35735816 & 4347 & 160 & 5165 & 258 & 4133 & 51 & 3289 & 23 & 3950.9 & 118.5 + knows 193526@xmath35813022 & & & 193 & 10 & 423 & 13 & 452 & 12 & 278.1 & 8.4 + knows 193706@xmath35744102 & & & 420 & 21 & 242 & 10 & 156 & 6 & 205.5 & 6.2 + knows 200422@xmath35735505 & & & 308 & 15 & 230 & 9 & 209 & 8 & 108.0 & 3.1 + knows 200539@xmath35775252 & 662 & 28 & 809 & 60 & 1128 & 23 & 993 & 20 & 1060.7 & 29.8 + knows 200955@xmath35722920 & 597 & 22 & 737 & 37 & 752 & 16 & 855 & 12 & 953.6 & 28.6 + knows 201716@xmath35744059 & 369 & 18 & 464 & 34 & 455 & 18 & 386 & 7 & 473.7 & 14.2 + knows 202242@xmath35761131 & 784 & 29 & 1471 & 101 & 942 & 19 & 898 & 17 & 428.9 & 12.9 + knows 204240@xmath35750802 & & & 229 & 12 & & & 424 & 9 & 208.3 & 5.8 + knows 211402@xmath35820437 & 157 & 19 & 235 & 12 & 287 & 12 & 347 & 13 & 474.9 & 11.2 + knows 213345@xmath35823904 & 167 & 17 & 302 & 15 & & & 479 & 17 & 915.1 & 27.5 + knows 220039@xmath35805844 & & & 166 & 12 & & & 145 & 9 & 176.3 & 5.3 + knows 220549@xmath35743632 & & & 201 & 10 & & & 363 & 9 & 430.5 & 8.9 + knows 230524@xmath35824232 & 107 & 4 & 243 & 17 & & & & & 116.7 & 3.5 + knows 231226@xmath35724055 & 143 & 3 & 310 & 23 & 225 & 12 & 229 & 6 & 270.6 & 8.1 + knows 231556@xmath35863130 & & & 280 & 19 & 344 & 13 & 299 & 8 & 170.5 & 4.7 + knows 232713@xmath35801236 & 176 & 13 & 213 & 16 & & & & & 138.4 & 4.2 + knows 234405@xmath35822638 & & & 309 & 15 & 755 & 16 & 1251 & 16 & 3777.4 & 113.3 + knows 235626@xmath35815255 & 715 & 35 & 664 & 34 & 652 & 17 & 531 & 9 & 520.9 & 15.6 + llllllcl + + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) + name & radio spectrum & ned i d & z & b@xmath36 & r@xmath37 & class & alt . name + + ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) & ( 5 ) & ( 6 ) & ( 7 ) & ( 8) + name & radio spectrum & ned i d & z & b@xmath36 & r@xmath37 & class & alt . name + knows 001302@xmath35723123 & flat & pn & ... & 12.744 & 12.741 & 4 & ngc0040 + knows 001312@xmath35774854 & steep & gal & 0.326 & 22.351 & & 1 & nvssj001311@xmath35774846 + knows 001633@xmath35791641 & steep & gal & 0.8404 & 21.707 & 19.731 & 2 & 3c 006.1 + knows 001716@xmath35813441 & peaked & qso & 3.366 & 19.013 & 17.432 & 2 & 0014@xmath35813 + knows 001948@xmath35732725 & flat & qso & 1.718 & 19.284 & 18.364 & 2 & 0016@xmath35731 + knows 020313@xmath35810622 & steep & no & ... & 19.952 & 18.621 & 2 & nvssj020307@xmath35810612 + knows 020336@xmath35723253 & inverted & qso & ... & 20.185 & 19.528 & 1 & nvssj020333@xmath35723254 + knows 020649@xmath35841102 & peaked & no & ... & 21.837 & & 1 & nvssj020713@xmath35841119 + knows 020954@xmath35722920 & flat & gal & 0.895 & 19.557 & 18.085 & 2 & s5 0205@xmath3572 + knows 021733@xmath35734923 & flat & qso & 2.367 & 17.090 & 15.298 & 2 & 0212@xmath35735 + knows 022459@xmath35765544 & steep & gal & ... & 20.534 & 19.086 & 2 & 4c@xmath3576.01 + knows 035447@xmath35800918 & flat & no & ... & 19.831 & & 2 & 0345@xmath357958 + knows 041050@xmath35765649 & steep & gal & 0.5985 & 21.678 & & 2 & 4c@xmath3576.03 + knows 041318@xmath35745107 & steep & gal & 0.373 & 20.811 & 18.782 & 1 & 4c@xmath3574.08 + knows 042132@xmath35835837 & & no & ... & 21.580 & 19.171 & 2 & nvssj042140@xmath35835842 + knows 050844@xmath35843202 & flat & qso & 1.340 & 20.134 & 19.116 & 2 & 0454@xmath35844 + knows 061048@xmath35724843 & flat & qso & 3.53 & 22.337 & & 2 & 4c@xmath3572.10 + knows 062555@xmath35820228 & flat & qso & 0.710 & 21.368 & 18.946 & 1 & 0615@xmath35820 + knows 063921@xmath35732454 & flat & qso & 1.85 & 18.592 & 18.172 & 2 & nvssj063921@xmath35732458 + knows 064132@xmath35881200 & flat & no & ... & 20.291 & & 1 & nvssj064206@xmath35881154 + knows 072608@xmath35791135 & peaked & no & ... & 21.908 & & 2 & s5 0718@xmath3579 + knows 074713@xmath35763918 & peaked & no & ... & 12.224 & 10.830 & 2 & s5 0740@xmath3576 + knows 074922@xmath35742038 & flat & qso & 1.629 & 19.282 & 18.990 & 2 & s5 0743@xmath3574 + knows 075039@xmath35790914 & peaked & no & ... & 21.917 & & 2 & nvssj075043@xmath35790917 + knows 075052@xmath35824200 & steep & qso & 1.991 & & & - & 0740@xmath358249 + knows 080817@xmath35731514 & flat & qso & 0.496 & 21.208 & 19.286 & 2 & s5 0802@xmath3573 + knows 092934@xmath35861236 & peaked & qso & ... & 21.563 & & 1 & s5 0916@xmath3586 + knows 093056@xmath35742017 & peaked & no & ... & 21.792 & & 2 & s5 0925@xmath3574 + knows 101009@xmath35825020 & flat & gal & 0.322 & 20.792 & 19.672 & 1 & s5 1003@xmath3583 + knows 104421@xmath35805447 & upturn & qso & 1.26 & 18.685 & 17.978 & 2 & 1039@xmath35811 + knows 105359@xmath35863004 & peaked & gal & ... & 21.253 & & 1 & nvssj105421@xmath35862936 + knows 105812@xmath35811438 & inverted & gal & 0.706 & 19.470 & 19.036 & 2 & s5 1053@xmath3581 + knows 110149@xmath35722544 & flat & qso & 1.46 & 17.612 & 17.122 & 2 & 1058@xmath35726 + knows 110410@xmath35765859 & steep & qso & 0.3115 & 15.330 & 14.978 & 2 & pg1100@xmath35772 + knows 115311@xmath35805837 & flat & qso & 1.25 & 18.787 & 18.173 & 2 & 1150@xmath35812 + knows 120019@xmath35730054 & steep & gal & 0.97 & 20.133 & 18.369 & 1 & 3c268.1 + knows 122340@xmath35804016 & flat & qso & ... & 21.975 & & 2 & s5 1221@xmath3580 + knows 132143@xmath35831623 & flat & no & 1.024 & 20.508 & 18.712 & 2 & nvssj132145@xmath35831614 + knows 132351@xmath35794258 & flat & qso & 1.970 & 21.501 & & 1 & s5 1323@xmath3579 + knows 135324@xmath35753307 & peaked & qso & 1.619 & 18.386 & 17.995 & 2 & nvssj135323@xmath35753258 + knows 135756@xmath35764330 & flat & qso & ... & 22.535 & & 2 & s5 1357@xmath3576 + knows 140638@xmath35782816 & steep & gal & ... & 20.338 & 18.655 & 2 & nvss j140636@xmath35782810 + knows 144830@xmath35760137 & inverted & qso & 0.899 & 21.996 & & 1 & s5 1448@xmath3576 + knows 152107@xmath35785837 & & no & ... & 18.996 & 18.382 & 2 & nvss j152102@xmath35785830 + knows 155608@xmath35742107 & & qso & 1.667 & 20.435 & 19.641 & 2 & nvss j155603@xmath35742057 + knows 160731@xmath35850159 & steep & gal & 0.183 & 18.072 & 16.660 & 1 & s5 1616@xmath3585 + knows 160922@xmath35794023 & steep & no & ... & & & - & 4c@xmath3579.15 + knows 163235@xmath35823228 & flat & gal & 0.02471 & 14.531 & 7.384 & 1 & ngc6251 + knows 172404@xmath35765328 & & qso & 0.680 & 18.901 & 18.428 & 2 & 1725@xmath357655 + knows 180044@xmath35782812 & flat & qso & 0.68 & 18.186 & 16.632 & 2 & 1803@xmath35784 + knows 182316@xmath35793856 & peaked & qso & 0.224 & 19.269 & 17.415 & 1 & s5 1826@xmath3579 + knows 183659@xmath35750741 & flat & no & ... & 19.798 & 20.297 & 2 & 1838@xmath357504 + knows 184218@xmath35794540 & steep & gal & 0.0561 & 12.582 & 11.015 & 2 & 3c390.3 + knows 185458@xmath35735129 & flat & qso & 0.461 & 16.521 & 16.411 & 2 & s5 1856@xmath3573 + knows 192754@xmath35735816 & flat & qso & 0.3021 & 7.445 & 16.557 & 2 & 4c @xmath3573.18 + knows 193526@xmath35813022 & peaked & gal & ... & 19.060 & 17.912 & 2 & s5 1939@xmath3581 + knows 193706@xmath35744102 & upturn & no & ... & 22.221 & & 1 & nvss j193702@xmath35744054 + knows 200422@xmath35735505 & inverted & no & ... & & & - & nvss j200417@xmath35735505 + knows 200539@xmath35775252 & flat & qso & 0.342 & 14.039 & 13.372 & 2 & 2007@xmath35777 + knows 200955@xmath35722920 & flat & qso & ... & 20.623 & 19.475 & 2 & 4c@xmath3572.28 + knows 201716@xmath35744059 & flat & qso & 2.187 & 18.464 & 18.165 & 2 & 4c@xmath3574.25 + knows 202242@xmath35761131 & inverted & qso & ... & 20.265 & 19.087 & 2 & s5 2023@xmath3576 + knows 204240@xmath35750802 & peaked & qso & 0.104 & 17.132 & 16.152 & 1 & 4c@xmath3574.26 + knows 211402@xmath35820437 & flat & gal & 0.084 & 17.137 & 15.832 & 1 & s5 2116@xmath3581 + knows 213345@xmath35823904 & steep & qso & 2.357 & 21.414 & 19.450 & 2 & s5 2136@xmath3582 + knows 220039@xmath35805844 & flat & no & ... & 21.534 & 20.664 & 2 & 2201@xmath358043 + knows 220549@xmath35743632 & flat & gal & ... & 21.532 & 20.585 & 2 & s5 2205@xmath3574 + knows 230524@xmath35824232 & & qso & ... & 18.390 & 17.336 & 2 & s5 2304@xmath3582 + knows 231226@xmath35724055 & flat & no & ... & 18.827 & 17.855 & 2 & nvssj231219@xmath35724127 + knows 231556@xmath35863130 & flat & no & ... & 20.992 & 18.721 & 2 & nvssj231549@xmath35863143 + knows 232713@xmath35801236 & & no & ... & 22.170 & 20.618 & 2 & nvssj232706@xmath35801259 + knows 234405@xmath35822638 & steep & qso & 0.735 & 21.769 & 20.233 & 2 & 2342@xmath35821 + knows 235626@xmath35815255 & flat & qso & 1.344 & 21.155 & 18.584 & 2 & s5 2353@xmath3581 + .redshift and luminosity statistics for the knows sources with redshift information . [ cols="<,^,^,^",options="header " , ] to better characterize the flat - spectrum source population , massardi et al . ( 2011a ) divided the sources of the at20 g fsr with flux densities measured at 5 , 8 and 20 ghz into five classes based on the place they occupy in a @xmath38 vs @xmath39 colour - colour plot , according to the following scheme : * flat : @xmath40 and @xmath41 * inverted : @xmath42 and @xmath43 minus flat region * peaked : @xmath42 and @xmath44 minus flat region * upturn : @xmath45 and @xmath43 minus flat region * steep : @xmath45 and @xmath44 minus flat region in fig . [ fig:3flux ] we show the colour - colour plot divided according to this five - population scheme for the 50 sources in our sample with flux density measurements at 5 , 8 and 20 ghz . vs @xmath39 ) with the five - population scheme described in the paper . the spectral index error bars have the size of the symbols . the central square confines the region of flat - spectrum sources and the diagonal dotted line corresponds to sources with single power - law radio spectra . qsos are indicated by crosses , radiogalaxies by stars and sources without classification by diamonds.,height=302 ] in table [ tab : at20 g ] we show the comparison between the statistics of our sample and the ones in the at20 g fsr catalog ( massardi et al . 2011a ) and the paco bright sample ( massardi et al . 2011b ) for which the same scheme was adopted . it is apparent that the flat - spectrum and steep - spectrum sources are the most represented in both our sample and the at20 g sample with the full sample ( s@xmath46 mjy ) providing the best match with our statistics . the comparison with the paco bright sample ( massardi et al . 2011b ) , estimated with the source selection applied here to the spectral indices calculated on their best epoch fit of the sed between 4.5 and 40 ghz , clearly shows the higher fraction of flat spectrum sources dominating the population of sources selected above 200 mjy , while at our flux density limit the steep spectrum objects become more significant . the fractions in the remaining sub - populations are comparable . it is also worth noticing that all the remaining sources in the up - turn quadrant are classified as flat in table [ tab : at20 g ] . in this paper we presented the results of the multi - frequency follow - up observations of the k - band northern wide survey ( knows ) pilot project . 73 objects were observed at 5 , 8 and 20 ghz with the 32-m medicina radio telescope and 30 ghz with the toru radio telescope . 16 objects have a counterpart in the 30 ghz planck ercsc . planck and toru 30 ghz flux densities are in reasonable agreement , indicating that our flux densities are not systematically biased . the objects were identified through ned and supercosmos as radiogalaxies ( 27 per cent ) and qsos ( 68 per cent ) in agreement with previous works ( e.g. sadler et al . two objects have no optical i d and one is a planetary nebula . 42 ( 58 per cent ) out of the 72 extragalactic objects have a redshift measurement : 30 are qsos and 12 are radiogalaxies . for 34 objects ( 24 qsos and 10 radiogalaxies ) we computed the optical and radio luminosity . as expected , qsos are found on average at higher redshift than radiogalaxies and are intrinsically brighter both in the radio and the optical band . four spectral indices in four contiguous frequency ranges were calculated : @xmath47 ; @xmath39 ; @xmath38 and @xmath48 . a clear steepening trend is visible both in the mean and median of the spectral indices when moving from lower to higher frequency ranges , a behaviour that is in agreement with what was found by massardi et al . ( 2011a ) for the at20 g full sample release . we used radio colour - colour plots to compare the radio spectral properties of radio sources of different populations with the ones found in other high - frequency selected catalogues . radiogalaxies mostly appear in the two lower quadrants ( corresponding to the steep - spectrum and gps - like classes ) while qsos are more scattered around the centre of the diagram indicating a prevalent flat radio spectrum : qsos are therefore more easily singled out in high frequency selected surveys . our sample appears to be in substantial agreement with the at20 g fsr source population statistics when the latter sample is cut to the same flux density limit ( 100 mjy ) . conversely , both our sample and the at20 g fsr statistics differ from the one of the sadler et al . ( 2006 ) sample for the up - turn , peaked and inverted spectral types even if all the samples have the same flux density threshold . this work is based on observations performed with the medicina telescope , operated by inaf istituto di radioastronomia . we gratefully thank the staff of the medicina radio telescope for the valuable support provided . rv used a grant of the estrela ( early - stage training site for european long - wavelength astronomy ) network during her phd . rv thanks the staff of the toru observatory ( poland ) for their kind help with the 30 ghz follow - up observations and data reduction . we gratefully acknowledge the financial support of the royal society paul instrument fund which allowed us to construct the 30 ghz ocra - p receiver . we are also grateful to the polish ministry of science and higher education ( grant number n n203 390434 ) . we acknowledge the fundamental support provided by the european community through the framework program fp5 ( faraday ) and the contribution of radionet in the technological development of the italian k - band multifeed receiver . 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 . this research has made use of data obtained from the supercosmos science archive , prepared and hosted by the wide field astronomy unit , institute for astronomy , university of edinburgh , which is funded by the uk science and technology facilities council . the authors thank the referee merja tornikoski for the useful comments . bonaldi a. , bonavera l. , massardi m. , de zotti g. , 2013 , mnras , 428 , 1845 bonavera l. , massardi m. , bonaldi a. , gonzlez - nuevo j. , de zotti g. ; ekers r.d . , 2011 , mnras , 416 , 559 browne i.w . , mao s. , wilkinson p.n . , kus a.j . , marecki , a. , birkinshaw m. , 2000 , in `` _ ocra : a one - centimeter receiver array _ '' , proc . spie , 4015 , 299 chhetri r. , ekers r.d . , mahony e.k . , jones p.a . , massardi m. , ricci r. , sadler e.m . , 2012 , mnras , 422 , 2274 ciaramella a. , bongardo c. , aller h.d . , aller m.f . , de zotti g. , lhteenmaki a. , longo g. , milano l. , et al . , 2004 , a&a , 419 , 485 condon j.j . , cotton w.d . , greisen e.w . , yin q.f . , perley r.a . , taylor g.b . , broderick j.j . , 1998 , aj , 115 , 1693 de propris r. , colless m. , peacock j.a . , couch w.j . , driver s.p . , balogh m.l . , baldry i.k . , baugh c.m . , et al . , 2004 , mnras , 351 , 125 gawronski m.p . , peel m.w . , lancaster k. , battye r.a . , birkinshaw m. , browne i.w.a . , davies m.l . , davis r.j . , et al . , 2010 , mnras , 406 , 1853 hambly n.c . , macgillivray h.t . , read m.a . , tritton s.b . , thomson e.b . , kelly b.d . , morgan d.h . , smith r.e . , et al . , 2001 , mnras , 326 , 1279 hafez y.a . , davies r.d . , davis r.j . , dickinson c. , battistelli e.s . , blanco f. , cleary k. , franzen t. , et al . , 2008 , mnras , 388 , 1775 hovatta t. , tornikoski m. , lainela m. , lehto h. j. , valtaoja e. , torniainen i. , aller m.f . , aller h.d . , 2007 , a&a , 469 , 899 jones , d.h . , read m.a . , saunders w. , colless m. , jarrett t. , parker q.a . , fairall a.p . , mauch t. , et al . , 2009 , mnras , 399 , 683 kesteven m.j . , bridle a.h . , brandie g.w . , 1977 , aj , 82 , 541 kurinsky n. , sajina a. , partridge b. , myers s. , chen x. , lpez - caniego m. , 2013 , a&a , 549a , 133 larson d. , dunkley j. , hinshaw g. , komatsu e. , nolta m.r . , bennett c.l . , gold b. , halpern m. , et al . , 2011 , apjs , 192 , 16 lowe s.r . , gawroski m.p . , wilkinson p.n . , , browne i.w.a . , pazderski e. , feiler r. , kettle d. , 2007 , a&a , 474 , 1093 mahony e.k . , sadler e.m . , croom s.m . , ekers r.d . , bannister k.w . , chhetri r. , hancock p.j . , johnston h.m . , et al . , 2011 , mnras , 417 , 2651 massardi m. , ekers r.d . , murphy t. , ricci r. , sadler e.m . , burke s. , de zotti g. , edwards p.g . , et al . , 2008 , mnras , 384 , 775 massardi m. , bonaldi a. , negrello m. , ricciardi s. , roccanelli a. , de zotti g. , 2010 , mnras , 404 , 532 massardi m. , ekers r.d . , murphy t. , mahony e.k . , hancock p.j . , chhetri r. , de zotti g. , sadler e.m . , 2011a , mnras , 412 , 318 massardi m. , bonaldi a. , bonavera l. , lpez - caniego m. , de zotti g. , ekers r.d . , 2011b , mnras , 415 , 1597 murphy t. , sadler e.m . , ekers r.d . , massardi m. , hancock p.j . , mahony e.k . , ricci r. , burke - spolaor s. , et al . , 2010 , mnras , 402 , 2403 jorstad s.g . , aller m.f . , aller h.d . , terasranta h. , 2005 , asp conf . , 340 , 113 odea , c.p . , 1998 , pasp , 110 , 493 ott m. , witzel a. , quirrenbach a. , krichbaum t.p . , standke k.j . , schalinski c.j . , hummel c.a . , 1994 , a&a , 284 , 331 peel m.w . , 2009 , phd thesis , university of manchester ( _ http://arxiv.org/abs/1006.2760 _ ) peel m.w . , gawronski m.p . , battye r.a . , birkinshaw m. , browne i.w.a . , davis r.j . , feiler r. , kus a.j . , et al . 2011 , mnras , 410 , 2960 planck collaboration , 2011 , a&a , 536 , 7 prandoni , i. , de ruiter h.r . , ricci r. , parma p. , gregorini l. , ekers r.d . , 2010 , a&a , 510 , a42 procopio p. , massardi m. , righini s. , zanichelli a. , ricciardi s. , libardi p. , burigana c. , cuttaia f. , et al . , 2011 , mnras , 417 , 1123 pyatunina t.b . , kudryatseva d.c . , gabuzda d.c . , ricci r. , sadler e.m . , ekers r.d . , staveley - smith l. , wilson w.e . , kesteven m.j . , subrahmanyan r. , walker m.a . et al . , 2004 , mnras , 354 , 305 richards j.l . , max - moerbeck w. , pavlidu v. , pearson t.j . , readhead a.c.s . , stevenson m.a . , healey s.e . , romani r.w . et al . , 2011 , apjs , 194 , 29 righini s. , carretti e. , ricci r. , zanichelli a. , mack k .- h . , massardi m. , prandoni i. , procopio p. , et al . , 2012 , mnras , 426 , 210 sadler e.m . , ricci r. , ekers r.d . , ekers j.a . , hancock p.j . , jackson c.a . , kesteven m.j . , murphy t. , et al . , 2006 , mnras , 371 , 898 tauber j.a . , mandolesi n. , puget j .- , banos t. , bersanelli m. , bouchet f.r . , butler r.c . , charra j. , et al . , 2010 , a&a , 520 , 1 taylor a.c . , grainge k. , jones m.e . , pooley g.g . , saunders r.d.e . , waldram e.m . , 2001 , mnras , 327 , l1 tucci m. , rubio - martin j.a . , rebolo r. , genova - santos r. , watson r.a . , battye r.a . , cleary k.a . , davies r.d . , et al . , 2008 , mnras , 386 , 1729

the source of cosmic rays with particle energies above @xmath3 ev is still unknown . models of origin , acceleration , and propagation must be evaluated in light of the observed energy spectrum and chemical composition of the cosmic rays . several experiments have attempted to determine the mean cosmic ray composition through the `` knee '' region of the spectrum , up to @xmath4ev@xcite . while the results are not in complete agreement , there is some consensus for a composition becoming heavier at energies above the knee , a result consistent with charge - dependent acceleration theories or rigidity - dependent escape models . in the region above the knee , the fly s eye experiment has reported a changing composition from a heavy mix around @xmath0ev to a proton dominated flux around @xmath5ev @xcite . this result makes this particular energy region much more interesting than the expectation from a naive rigidity model . this changing composition may imply that there may be multiple sources of cosmic rays . the agasa experiment shows broad agreement with this trend if the data are interpreted using the same hadronic interaction model as used in the fly s eye analysis @xcite . the recently reported hires / mia@xcite hybrid observation on the cosmic ray composition in a narrower energy region , @xmath6 ev , shows a general agreement with fly s eye experimental result . the new result gives a somewhat more rapid change in the composition . the reliability of these experimental results depends on how well we understand the development of all components of extensive air showers ( eas ) produced by cosmic rays , how well we understand the response of our detector to the eas , and how well we have done in the eas reconstruction . in this paper , we address all those issues in detail . we start with a general description of the techniques of cosmic ray composition measurement in the high energy region . the fly s eye and agasa experiments use different techniques to study composition . the fly s eye experiment technique is based on the assumption that the speed of the development of eas depends on the mass of the primary particle : a heavier nucleus induces earlier eas development in the atmosphere . the rapid break - up of a heavy nucleus at the early stage of the cascade in the air leads to an effectively higher multiplicity than that produced by a light nucleus or a proton at the same depth in the atmosphere . the consequence is that the eas is built up by a superposition of smaller subshowers induced by nuclear fragments . in this case , since the subshowers have lower energies , the eas will have a shower maximum higher in the atmosphere than in the case for a proton primary . the fly s eye experiment is designed to measure the size , or total number of charged particles , of an eas as a function of atmospheric depth . it is thus an ideal detector to measure the depth of maxima of showers directly . in practice , the intrinsic fluctuations in the depth of shower maximum and the detector resolution effects imply that one can not directly resolve the type of primary nucleus on an event by event basis . what one can do is to extract an average eas primary composition by comparing the data to a monte carlo simulation with a given primary composition . since the shower development is somewhat dependent on the choice of a hadronic model , this method leads to results which have some model dependence . the other method for studying composition depends on the assumption that the muon content of eas produced by a superposition of sub - showers ( as in the case of heavy nuclei ) is larger than those with fewer sub - showers . this is due to the fact that the dissociation of a heavy nucleus produces a relatively higher multiplicity in its interaction with atmospheric nuclei . the resultant sharing of the primary energy between the nuclear fragments make the secondary pions less energetic . as a consequence , those pions have a greater decay probability into muons than those produced by a lighter nucleus in the early stage of shower development . this leads to a difference in the muon content between the eas s induced by heavier and lighter nuclei , e.g. iron and protons . this difference shrinks with energy because the available path for the decay of the high energy pions decreases as the shower develops deeper in the atmosphere . according to the simulations the difference is still resolvable in the energy region 10@xmath7 @xmath8 10@xmath9 ev . in principle , one can obtain the information about the composition of the primary particle by either measuring the total number of muons in eas or a local density of shower muons at a specific distance to the core of the shower . however , as in the case of the previous method , the fluctuations are large compared with the separation between the different types of eas , so that the resolvability is not strong . the hadronic model dependence of the predicted @xmath10-content from a particular composition is also significant for this method . the hires / mia hybrid experiment is designed to combine the two methods together using two independently developed co - sited experiments . the two experiments share a trigger to simultaneously record both the longitudinal development information and the eas muon density . this results in a unique data set useful for the investigation of cosmic ray composition . results on comparison of the flourescence and muon methods have previously been published@xcite . in this paper , we use the mia array to improve the geometrical reconstruction and concentrate on the fluorescence technique for composition and spectrum studies . the hybrid timing information enhances the accuracy in the determination of the geometry of the eas . this accurate shower geometry plays a crucial role in the subtraction of the cerenkov light component of the eas and in the corrections for detector acceptance . these turn out to be the two key issues in the shower longitudinal development profile determination . this profile provides directly the size at shower maximum and its location while the integral of the profile yields the shower energy . the improvement in shower geometry determination is the main advantage of a hybrid experiment . the other advantage resulting from the coincident measurement with the surface muon array is the existence of a fully efficient triggering region in the central detector volume . this is very helpful for the cosmic ray energy spectrum measurement . one of the challenges for this experiment is that the two devices are separated by only 3.3 km . because the mia detector can not be triggered by remote showers , most of the triggered events are located @xmath8 4 km away from hires detector . this short shower - detector distance gives rise to difficulties for reconstruction of those eas s . the lateral distribution of shower electrons is no longer a negligible effect . the broadened source of light increases the uncertainty in the shower geometry determination and in the acceptance correction of the signals . the limited effective trigger area of mia largely suppresses the aperture of hires detector . the energy coverage of our data is one decade of magnitude over 3 years of observation , much smaller than that from the fly s eye experiment . the geometry of the triggered events is such that the fraction of cerenkov light is often large . this implies the need for tight criteria for event selection to reduce cerenkov contamination , which in turn may cause a bias . we must search for a balance between the tightness of the selection criteria and the minimization of bias . we describe our resolution of these problems in the following sections . in this experiment , we use a hybrid detector consisting of the prototype high resolution fly s eye ( hires ) air fluorescence detector and the michigan muon array ( mia ) . the detectors are located in the western desert of utah , usa at @xmath11 and @xmath12 . the hires detector is situated atop little granite mountain at a vertical atmospheric depth of @xmath13 g/@xmath14 . it overlooks the casa - mia arrays some 3.4 km to the northeast . the surface arrays are some 150 m below the fluorescence detector at an atmospheric depth of 870 g/@xmath14 . the hires prototype has been described in detail elsewhere @xcite . it views the night sky with an array of 14 optical reflecting telescopes . they image the eas as it progresses through the detection volume from 3@xmath15 to 70@xmath15 in elevation , 64@xmath15 in azimuthal angle at the top and 32@xmath15 at the bottom of the field of view . nitrogen fluorescence light ( in the 300400 nm band ) is emitted at an atmospheric depth @xmath16 in proportion to the number of charged particles in the eas at that depth , @xmath17 . the triggered tube directions and the time of arrival of light signals can be used to determine the shower - detector - plane and the tilt angle of the shower in this plane , denoted by @xmath18 . part of the shower development profile ( at least 250 g/@xmath14 long ) can be mapped by measuring the light flux arriving at the detector . assuming @xmath17 to be the gaisser - hillas @xcite shower developement function and correcting for cerenkov light contamination and atmospheric scattering effects one can measure the primary particle energy @xmath19 , and the depth , @xmath20 , at which the shower reaches maximum size@xcite . the mia detector @xcite , consisting of 16 patches formed with 64 scintillation counters each , covers about @xmath21 m with the active area over 2500 m@xmath22 . the patches are buried about 3 m under the surface . the data acquisition system records the identity and firing time of each counter participating in a given event . the eas muon arrival times are measured with a precision of 4 ns and all hits occurring within 4 @xmath23 of the system trigger are recorded . the average efficiency of mia counters for detecting minimum ionizing particles was @xmath24 when they were buried , and the average threshold energy for vertical muons is about 850 mev . the mia detector determines the muon density via the number and pattern of hit counters observed in the shower @xcite . the hires detector collects data on clear moonless nights . a focal plane camera , consisting of 16@xmath2516 photo - multiplier - tubes ( pmt ) , is triggered if two of its 4@xmath254 `` sub - clusters '' contain at least 3 fired tubes(two of them must be physically adjacent ) in a 25 @xmath10s interval . tubes trigger if the signal generates a voltage greater than a threshold , set at approximately 4 @xmath26 above nightsky background noise . this yields a mirror trigger rate of about 30 to 120 per minute . once a trigger is formed , hires sends a xenon light flash to mia as a confirming trigger for a coincident event . mia has a 100% duty cycle and a trigger rate of about 1.5hz formed by requiring at least 6 patches fired ( with at least 3 hits found in each patch ) . however , a coincident event is not selected until either it is confirmed by a hires light flash communication signal if it is received within 50@xmath10s , or the event triggers casa ( a surface scintillation detector array for eas electron observation@xcite ) simultaneously and is coincident with a hires event within @xmath273 ms according to the gps clocks in each site . more details about the trigger formation and coincident event matching can be found in @xcite . during the lifetime of this hybrid experiment between aug . 23 , 1993 and may . 24 , 1996 the total coincident exposure time was 2878 hours corresponding to a duty cycle of 11.9% . 4034 coincident events were recorded . for events passing a set of coincidence assurance cuts the shower trajectory , including arrival direction and core location for each event , was obtained in an iterative procedure using the information from both hires and mia @xcite . 2491 events survive this reconstruction procedure . further cuts are performed in order to maintain high resolutions in energy and shower maximum which are essential to the composition analysis . the criteria are based on a thorough monte carlo simulation of the detectors as described below . after quality cuts , 891 events are employed in this composition study and energy spectrum measurement . accurate knowledge of the shower geometry is important in the reconstruction of the shower profile in the atmosphere . the first step is to use the pointing directions of triggered pixels in the hires detector to determine the shower - detector - plane . this is a simple linear fit weighted by the amount of light received by each tube . the shower - detector - plane is determined quite precisely because the hires detector records showers with an average of 36 triggered pixels and the quality cut ensures that the lengths of tracks is longer than 20@xmath15 . the typical error in the horizontal position of the plane is about 0.1@xmath15 , while the error in the tilt angle of the plane is somewhat larger . the overall error in the shower - detector - plane normal direction is about 0.7@xmath15 . muon timing information from mia plays a key role in the determination of geometric parameters in the shower - detector - plane , including the shower - detector distance and shower orientation , and in the improvement of the shower - detector - plane determination . the initial trial shower arrival direction is determined by fitting the muon arrival time with a flat shower front . projecting this direction onto the shower - detector - plane yields the tilt angle of the shower , @xmath18 , defined as the angle between the shower axis and the horizon in the shower - detector - plane . once @xmath18 is known , the light arrival time , t@xmath28 , on the @xmath29-th fired hires tube is fit to the timing formula @xmath30 in which @xmath31 refers to the elevation angle of the @xmath29-th fired tube . @xmath32 and @xmath33 are two parameters indicating the time as the shower front passes the detector and the perpendicular distance from the shower axis to the detector respectively , while @xmath34 refers to the speed of light ( see the sketch in fig.[sketch ] ) . the core location can be derived by using the shower - detector - plane normal vector , @xmath18 and @xmath33 . the shower front shape can now be more accurately represented as a cone with the delay parameter @xmath35 , where @xmath36 refers to the delay of a conical muon front relative to the original flat front at a perpendicular distance , @xmath37 , to the shower axis . this procedure is iterated after additional corrections , including correction to the shower - detector - plane direction . this iteration stops when the difference between the core parameters is less than 10 m . the details of this iterative procedure can be found in @xcite . the distributions of those parameters are shown in fig.[rp_dis ] and fig.[core_dis ] . the hires tube signal , consisting mainly of the fluorescence light produced by charged particles from the shower , can be used to reconstruct the shower development , i.e. to calculate the shower size at the corresponding depth in the atmosphere . for this purpose , the raw tube signal must be corrected for several effects . if a triggered tube has a center that is not exactly in the shower - detector plane , its signal requires correction for a number of effects . these include the finite transverse width of the shower due to multiple scattering of shower electrons , the finite size of the optical spot on the face of the focal plane camera , the response function of the pmt cathode , gaps between the pixels , and the effective light collecting area of mirrors . all these effects can be taken into account by performing a `` ray tracing '' procedure , namely we trace the photons from the source direction , which can be wider than a line source due to the lateral distribution of shower electrons , all the way down to the face of the pmt via the spherical mirror surface . the response function of the pmt cathode is folded into the ray tracing . the pixel signals are then re - organized into a series of longitudinal `` bin signals '' with a 1@xmath15 bin size along the shower axis . the signals are recalculated in units of `` number of photon - electrons per unit angle along the track per unit collecting mirror area '' . @xcite provides the details of this correction and binning procedure for interested readers . the fluorescence light yield ( photons per meter ) of a single charged particle varies slightly with the atmospheric pressure and temperature and has an energy dependence given by the de / dx energy loss curve@xcite . these effects are taken into account in our shower reconstruction . the energy dependence of the fluorescence light yield is folded in by taking an average over the shower electrons at age @xmath38 , defined as @xmath39 , with an energy distribution of the shower electrons at age @xmath38 being extracted from the simulation results using the corsika package@xcite . this energy distribution is consistent with measurements by richardson @xcite and a parameterization from hillas @xcite . in addition to the corrections associated with those issues , the cerenkov light component of the bin signal must be subtracted as well , because only the pure fluorescence light is proportional to the size of shower at a given depth . this gives rise to some complications . first of all , the cerenkov light component is composed of the cerenkov light produced by the shower electrons directly illuminating the detector and the light scattered into the direction of the detector . because the cerenkov light is very forward along the direction of the electrons , the angular distribution of cerenkov light in a shower is narrowly beamed and falls exponentially with the angular distance from the shower axis . the average angular scale has been measured to be @xmath40@xcite . therefore , when the shower points towards the detector , the estimate of the cerenkov light becomes very sensitive to the geometry of shower . in this case , the signal is also dominated by direct cerenkov light . we avoid this sort of events in this analysis by a cut on the minimal viewing angle from the detector to the track , illustrated as the angle @xmath41 in fig.[sketch ] . the scattered cerenkov light , which we still have to deal with , includes components due to the rayleigh scattering from the atmospheric molecules and mie scattering from aerosols . rayleigh scattering is well understood in terms of the distribution of scattering centers and their fluctuations , angular distribution of light , frequency response and the overall extinction length for ultraviolet ( uv ) light . this scattering is used to estimate the attenuation of cerenkov light along the shower , how much of this light is scattered into the pmt direction , and the attenuation of the light during the propagation from the source to the detector . the scattering of uv light by aerosols is more uncertain . the distribution of aerosols depends on weather conditions . because of the variation in the size distribution of aerosols , the scattering phase function can vary . there are several models for this function which lead to different estimates of the amount of scattered light . we use the `` standard desert aerosol model '' with an exponential increase of the aerosol extinction length with height above a mixing layer of height @xmath42 below which the aerosol extinction length is constant . the change in extinction length with height above this is governed by a `` scale height '' parameter @xmath43 while below @xmath42 , the extinction length is given by the `` horizontal attenuation length '' , @xmath44 . we monitor the aerosol variation by the use of several xenon flashers which shoot light pulses at different angles into the atmosphere at different distances to the detector . by detecting the scattered light from these flashers with the hires detector , we monitor the variation of aerosols@xcite and partially constrain the range of those parameters . this will be discussed further in sec . iv . after subtracting the cerenkov light components , the bin signals , now proportional to the size of the shower , are fit to a function describing the longitudinal development of eas suggested by gaisser and hillas@xcite , ( g - h function ) : @xmath45 in which @xmath46 and @xmath47 refer to the size of the shower and its maximum , @xmath16 represents the atmospheric depth where the shower front passes the specific angular bin , and @xmath48 and @xmath20 the depths where the shower starts and reaches its maximum , respectively . the depth @xmath16 is calculated based on the shower geometric parameters and the pointing directions of each effective angular bin . the choice of this eas longitudinal development curve has recently been experimentally confirmed @xcite as accurate based on the same data set . the g - h function is one of the best parameterizations of longitudinal development according to this study . an example of shower reconstruction is shown in fig.[example ] . in a ) , the four components of light contributing to the best fit results are plotted : fluorescence light ( thick solid line ) , direct cerenkov light ( thin solid line ) , cerenkov light from rayleigh scattering(dotted line ) and cerenkov light through aerosol scattering ( dashed line ) . in b ) , we show the fit of the sum of all the components to the bin signals(dots ) . this shower reaches its maximum size of 1.6@xmath2510@xmath49 at a depth of 630 g/@xmath14 . note that the parameter @xmath48 is the point at which size @xmath50 according to ( [ ghfunc ] ) . we are not sensitive to light from early shower development and we fix @xmath48 at -20 g/@xmath14 in our fitting procedure . the justification for this will be given below . once we know the longitudinal development profile of an eas , we can integrate over its depth to calculate the total path length of all shower charged particles and calculate the total deposited energy @xmath51 by the charged particles in this shower , i.e. , @xmath52 where the critical energy and the radiation length of electrons in air are @xmath53 and @xmath54 , respectively . a recent study@xcite based on the monte carlo simulation package corsika verifies this formula and re - evaluates the constant @xmath55 as 2.19 . since some of the primary energy is carried away by neutrinos and muons penetrating the ground , a correction for this effect must be applied . in that study , the authors establish a new empirical formula for the converting @xmath51 into total energy of the shower , @xmath56 . it reads @xmath57 where the parameter @xmath58 , @xmath59 and @xmath60 . these parameters are determined by taking an average between proton and iron initiated showers , since it is impossible to know the primary particle mass in advance of the reconstruction . this causes a systematic uncertainty in @xmath56 of less than 10% . the energy distribution of our data set is displayed in fig.[e_dis ] . the vertical axis represents the number of events within a bin of @xmath61e . the figure shows that the threshold of our detector is about 5@xmath2510@xmath62 ev . the hybrid detector approaches a fully efficient operation above 4@xmath2510@xmath7 ev . in order to test this complex eas reconstruction procedure and evaluate the resolution in shower geometry , shower depth of maximum and energy based on this reconstruction scheme , we have developed a monte carlo code to simulate the eas shower and detector . we have made this as realistic as possible both for shower development in the atmosphere and for the response of our detector to the shower . in this section , we will address how this event generator is driven with a full monte carlo simulation of eas , how the production and propagation of light through the atmosphere is treated , how the acceptance and response of the detector to the light is simulated , and what the final resolution functions and their relationship with the event quality cuts are . the driver of the shower simulation is a series of parameterizations of the results from a full eas simulation using the corsika package@xcite . this is one of the most modern and complete simulation codes for eas development . it traces shower particles from very high energy at top of the atmosphere down to the threshold energy of 100 kev . a `` thinning '' technique is used in the simulation to reduce the size of the calculation . only one secondary particle is traced if the interaction energy falls below the thinning threshold , e.g. 10@xmath63 of the shower total energy . a weight is assigned to this traced particle to represent those not being traced . depending on the degree of realism in the fluctuations that is required , the user can set an appropriate thinning threshold if the cpu time limit allows . another advantage of the corsika code is that the user can switch between several optional hadronic interaction models . the authors have made efforts to test the program for primary particle energies up to 10@xmath62 ev , but do not claim reliability for energies higher than 10@xmath7 ev . however , it is one of the best eas models currently available . low energy shower particles , down to the tens of kev level , are treated carefully by employing the well known egs package @xcite etc . in our simulation with corsika ver . 5.624 , the thinning threshold is set to be 10@xmath63 of the shower energy and the qgsjet@xcite and sibyll@xcite hadronic interaction models are selected . the number of eas electrons as a function of depth and eas muon information , including arrival direction , time and energy for every muon above 870/cos@xmath41 mev at 870 g/@xmath14 , are recorded . we simulated 500 events for each of the 5@xmath254 grid points in energy from 3@xmath2510@xmath62 ev to 5@xmath2510@xmath9 ev and zenith angle from 0@xmath15 to 60@xmath15 . the same number of events were generated for proton and iron induced showers and under different hadronic interaction model assumptions . based on this large simulated event data set , we parameterized all the distributions of eas parameters such as the first interaction depth @xmath64 , shower decay constant @xmath65 , shower maximum @xmath47 , its position in depth @xmath20 , and the correlations between them . as an example , the @xmath20-distribution is shown in fig.[xm_mc_dis ] . it is clear that the proton induced showers possess larger fluctuation in @xmath20 than iron initiated ones . though they overlap each other , the means for each distribution are about 100 gm/@xmath14 apart , which is resolvable if sufficient statistics are available . the model dependence appears to be significant , but is smaller than the proton - iron separation . the comparison between the histogram for simulated data and curves for parametrized results shows that the parameterization faithfully represents the fluctuation in @xmath20 . a similar situation is found for the other parameters . among the parameters , we find that @xmath65 and @xmath47 are correlated and we put this correlation into our generator . we find that the parameter @xmath48 is quite insensitive to the type of primary particle and energy if we use the g - h function ( [ ghfunc ] ) to fit the longitudinal development of simulated showers . the fitting quality remains quite good by fixing it at a value of -20 g/@xmath14 . the other parameter @xmath65 is found to have a slow variation with energy and mass of the primary particle , with a central value of 70g/@xmath14 . we fix both at the values suggested here in our reconstruction procedure for real events . one of the benefits of fixing those relatively insensitive parameters is to reduces the chances of the parameter search being trapped at a local minimum of @xmath66 . once the shower parameters are determined as a function of energy , the number of electrons can be calculated by using the g - h function ( [ ghfunc ] ) at depth @xmath16 . the electrons are distributed laterally according to the nkg function at corresponding age @xmath38 of the shower . the fluorescence and cerenkov light and the corresponding signal appearing at the hires detector can then be generated . the simulation of muons in an eas is much more complex . the dependence on both zenith angle and energy is important since the observation is done at a fixed altitude . in order to simulate the mia trigger correctly , we generate the muon density , @xmath67 at a distance @xmath68 to the core according to the muon lateral distribution . @xmath67 and its fluctuation behavior are parameterized based on simulations . the muon density generated based on our parameterization is plotted in the fig.[muon ] a ) . as a comparison , the agasa muon lateral density function(ldf)@xcite and the griesen ldf are plotted in the same figure . our simulation agrees with the agasa ldf well except in the small core distance area in which our simulation is closer to the griesen function . the arrival time of the eas muon is essential in the simulation for mia triggering . we parameterize the distribution of arrival time for each muon in the shower disk within different annular rings at a distance @xmath68 from the core , and at all zenith angles and energies in the grid mentioned above . the shape of the arrival time distribution changes quite rapidly with @xmath68 as can be seen in fig.[muon]c ) . note the vertical scales are different for the different cases . we use a single function @xmath69 to describe all these distributions . the parameters @xmath70 , @xmath71 and @xmath72 are tabulated as functions of @xmath68 , energy and zenith angle . the parameters are generated with due regard to correlations if they exist and the muon arrival time are generated individually depending on how many muons are generated at @xmath68 and for a specific direction . the median time for muons generated in an annular ring at @xmath68 reveals the curvature of the shower front . in figure b ) , we plot two examples at zenith angle @xmath73 and @xmath74 . the small differences between proton and iron induced showers are also shown in the figure . the lines represent the results directly from corsika simulation and the dots are from our proton shower generator . once the shower is generated by the corsika based driver , fluorescence and cerenkov light contributions are calculated for each 0.04@xmath15 angular bin along the axis of the shower . cerenkov light generated in the previous bins is accumulated taking account of the attenuation between the bins due to the scattering of the light on atmospheric molecules and aerosols . the propagation and attenuation of light through the air and acceptance of the light by the detector are simulated in detail . the effective area and reflectivity of the mirror , filter transmission on the face of the focal plane camera , quantum efficiency of the pmt cathode , the gain of pmt and all electronics triggering and charge integration are realistically simulated . the variation of all these effects with wavelength is considered by tracking the light in 16 different uv wavelength bands covering from 300 to 400 nm . finally , the simulated signal in each pixel is built up by summing all the angular bins involved and the 16 wavelength bands . the night sky background light is also added to the signal according to results of an on site measurement@xcite . the trigger time slewing ( i.e. late triggering for small pulses ) is also taken into account for the hires sample - and - hold electronics . the mia signal is generated by sampling the number of muons for each counter at the perpendicular distance @xmath68 to the shower axis , then run over all muons in each counter to generate the arrival time for each of them . the pulse from a given counter is built up by passing all the muon signals through the electronics sequentially . dead counters , counter efficiencies , trigger formation , time windows for accepting counter hits and noise muons are taken into account in the simulation . both the hires and mia signals are written in the same format as for the real data . 8000 proton and 4000 iron induced showers are generated with a spectrum of trial energies from @xmath75 to @xmath76ev . the differential spectral index is set as -3.0 . all the simulated events are passed through the same reconstruction procedure as real data and all geometrical and shower development parameters are determined . events must pass the quality cuts " defined below . we compare the overall distributions from the simulated events with data . this is plotted in fig.[rp_dis],[core_dis],[e_dis ] . the solid(open ) squares represent the showers induced by protons(iron nuclei ) in those figures . the consistency between the data and simulation builds up our confidence in the simulation , and hence in the resolution functions we now present . since we know the input parameters for every event , we can study the detector response and the corresponding resolution function on an event by event basis . in fig.[resolutions ] , the resolution functions in shower arrival direction , core location , energy and @xmath20 are plotted for iron induced showers . in a ) , we plot the distribution of the `` opening angle '' between the `` real '' shower axis vector given by the input shower and the `` reconstructed '' shower axis vector which is determined by using of the timing information from both hires and mia . similarly in b ) , we plot the distribution of the distance between the input shower core and the reconstructed core position . those two geometrical resolution functions peak at zero but the long tails imply that some events are measured poorly . we now move to a more quantitative discussion of parts c ) and d ) in the figure , the resolution functions in energy of primary particle and shower maximum depth , respectively . these are the most important results for the energy spectrum measurement and the composition study described in this paper . the better the geometrical parameters of a shower are determined , the better the shower development profile can be extracted . of all geometrical parameters , the shower - detector - plane is the most crucial and depends strongly on how many tubes are triggered and how long the track formed by those tubes is . the number of muons detected by mia is the other contributor to precise time fitting . in order to locate the shower maximum , it and a good fraction of the rest of the profile must bee seen by the detector . moreover , as mentioned before , we must avoid those events which are dominated by cerenkov light . poorly fitted events should also be rejected . the set of quality cuts listed in table.[cut_tab ] addresses these issues . .event criteria . @xmath77(@xmath78 ) refers to the depth of the highest(lowest ) section of the shower track seen by the detector . @xmath79 is the estimated error in @xmath20 . [ cols="^,^",options="header " , ] as mentioned in the introduction , the distribution of @xmath20 can be used with the help of monte carlo simulations to extract an average composition of primary cosmic rays . the results depend on the hadronic interaction model , the eas simulation model and resolution of the detector in energy and @xmath20 . in this paper , we compare data to the predictions of two hadronic models : qgsjet and sibyll . other , lower multiplicity models have been shown to be inconsistent with any normal composition of cosmic rays . according to the simulation , the average @xmath20 for showers induced by protons is separated from the average for iron by about 100 g/@xmath14 and this separation is almost independent of energy in the range from 10@xmath7ev to 10@xmath9ev . since the resolution of the detector is 44g/@xmath14 , we can tell if the data is closer to one than the other . the absolute position of @xmath81 for a given composition assumption is model dependent ( about 25 g/@xmath14 shift ) . at any given energy , a measurement of @xmath81 implies a particular composition which is thus systematically uncertain . on the other hand , an apparent departure of the data points from either of the predictions based on proton or iron showers as a function of energy will reveal information on the change in the composition of cosmic rays . the rate of this change can be much more reliably determined than the absolute composition itself . the variation of the separation between the measured and simulated pure composition @xmath20 can be quantitatively evaluated using the so called `` elongation rate '' . based on models , it is found that the average @xmath20 increases with energy logarithmically over the energy range of interest . the `` elongation rate '' is symbolized by @xmath70 in this paper and defined by @xmath82 it is remarkable that the elongation rates are almost the same for proton and iron induced showers and nearly independent of the interaction models . they are @xmath83g/@xmath14 per decade of energy for proton showers and @xmath84g/@xmath14 per decade of energy for iron showers according to the qgsjet model . any other values for e.r . observed from the data will indicate a change in composition . an important issue associated with measuring the elongation rate is possible existence of a bias caused by tight cuts . to test this , we compare the average @xmath20 as a function of energy using the simulated events . one set of data is based on all the corsika sampled events and the other is based only on those that trigger the detector , are able to be reconstructed and pass the tight cut criteria . as shown in fig.[er ] , no matter what interaction model is used , we do not see any bias using the cuts set in the table.[cut_tab ] . the thin lines in the figure show the sampled events and the circles and the squares show the reconstructed results ( see the figure legends for details ) . the hires / mia experimental data as shown in the fig.[er ] demonstrates an unambiguous change in average @xmath20 with energy . this indicates a change towards a lighter mix of nuclei in the average composition from @xmath0 to @xmath1ev . this indication of change in composition can be evaluated by using the elongate rate measured in the experiment , i.e. @xmath85 where the number in parentheses represents the systematic error discussed below . in comparison with the predicted number for a unchanging or pure composition mentioned above , the difference is larger than any known uncertainties . the uncertainty in predicted elongation rate due to hadronic model dependence is small . we can estimate the change of the composition in the form of the average logarithm of atomic number of the primary nuclei , @xmath86 , as @xmath87 over the energy range covered by the data . this number is quite model independent assuming equal elongation rate for all different pure compositions . the systematic uncertainty in @xmath70 is included here . the absolute value of @xmath88 is strongly model dependent as implied by fig.[er ] . uncertainties in @xmath20 come both from choice of theoretical models and detector resolution . we first look into the uncertainty in the predictions . in fig.[er ] , the average difference in @xmath20 at any given energy between the predictions based on qgsjet and sibyll models is about 25g/@xmath14 . both are compatible with the data and both lead to the same qualitative conclusion of a lightening in the composition . however , the value of the average @xmath89 at any given energy is model dependent . from the experimental point view , we have made a detailed effort to understand the systematic error in shower @xmath20 and energy . for @xmath20 , we have considered systematic errors in the atmospheric transmission of light and in the production of cerenkov light . these are related since scattered cerenkov light can masquerade as fluorescence light if not accounted for properly . for atmospheric scattering , there is uncertainty in the aerosol concentration and its vertical distribution . the uncertainty , equivalent to one standard deviation with respect to the mean , is expressed as a range of possible horizontal extinction lengths for aerosol scattering at 350 nm ( taken as 11 km to 17 km based on measurements using xenon flashers , @xcite ) and a range of scale heights for the vertical distribution of aerosol density above the mixing layer ( taken as 0.6 km to 1.8 km ) . for cerenkov light production , we have varied the angular scale for the cerenkov emission angle over a one standard deviation equivalent . at ground level , we take the distribution as an exponential function of the angle from the shower axis , with a scale of @xmath40@xcite . those uncertainties are shown by the shaded area in fig.[er ] . the systematic error in the energy is about 25% and comes from fluorescence efficiency uncertainty@xcite , detector calibration uncertainty@xcite and the atmospheric corrections@xcite . the first two are intrinsically independent of the primary particle energy over this range . the fluorescence efficiency has been measured with an error of 10% . the percentage atmospheric corrections are also independent of energy because the sample of showers is restricted to core locations within 2 km of the mia detector center . therefore there is no significant atmospheric path length difference between an eas and the detector for different energies . an energy independent systematic fractional error in energy has no effect on the measured elongation rate . the magnitude of the systematic error in energy due to atmospheric attenuation can be estimated by varying the atmospheric parameters over the range described above . it is not greater than 10% . the detector calibration systematics is less than 5% . in fig.[xm_mc_dis ] , we can see that the fluctuations about the average @xmath20 for simulated proton showers is larger than that for iron showers . the fluctuations for both proton and iron induced showers are too large to allow us to distinguish one from the other on an event by event basis . however , we can determine the gross properties of the composition statistically . in fig.[xm_dis ] we plot the predicted distributions of @xmath20 for proton and iron showers together with dotted and dashed lines , respectively . the detector response has been folded into those distributions . by studying those distributions in different energy ranges as shown in the figure , one can compare the data to pure proton and pure iron composition distributions . the data clearly requires a mixed composition of light and heavy particles to account for the width and peak value of the @xmath20 distribution . the cosmic ray energy spectrum measured by all modern experiments are summarized in the fig.[spe_all ] covering the whole energy range from @xmath90 to @xmath91 ev . the consistency between this experiment and the fly s eye stereo data has been discussed previously . we have shown the marginal confirmation of the break in the spectrum at @xmath92 ev there . by comparing with the observations @xcite in the `` knee '' region , we see that both the intensity and spectrum index imply a good continuity from the results at energy lower than @xmath93 ev . we especially note that the change in cosmic ray intensity around @xmath92 ev is comparable in power law index with the change that occurs around the `` knee '' . a confirmation of this break with better statistics and similar energy resolution is important . all the other experimental results are consistent with the akeno result : the spectrum follows a single index power law between @xmath94 and @xmath0 ev . the only existing experimental result based on direct measurements of shower longitudinal development is that from fly s eye experiment . as a successor of that experiment , the hires / mia experimental result qualitatively supports old fly s eye s result , i.e. there exists a trend in the composition of cosmic rays to a lighter mix with energy . quantitatively , they are consistent with each other , taking into account the systematic errors in the original fly s eye result of about 25 gm/@xmath14 on individual @xmath20 measurements . the elongation rate measured in hires / mia experiment is marginally larger than that in fly s eye@xcite which is @xmath95g/@xmath14/decade where the quoted error is statistical only . we should also note that experimentally measured elongation rates are not corrected for acceptance . differences in acceptance for two experiments could introduce differences in elongation rates . the safest method is to compare each experiment to its own simulated proton and iron data sets . the conclusion on the composition of cosmic rays based on this kind of comparison is meaningful because the detector effects are counted in exactly the same way for both real and simulated events . for the present experiment , based on our simulation , we believe that the detector biases for the elongation rate are minimal . in summary , when all the errors are taken into account , the results on @xmath20 distribution and elongation rates , from the two experiments are consistent , in spite of the differences in the analysis . the other existing result on elongation rate in the same energy range is from the yakutsk experiment@xcite . the systematic error is not provided and the method using a ground array experiment is more indirect than the present experiment . nevertheless , the results are marginally in agreement with the result reported here . there are several measurements of elongation rate at lower energies , between @xmath3 and few @xmath94 ev . these results and ours are shown in fig.[er_all ] together . the trend of a changing cosmic ray composition shows a pattern correlated with breaks in the energy spectrum . it can be characterized qualitatively as follows . there is a rather clear break around @xmath96 ev which is related to the `` knee '' in the energy spectrum . this break seems to be confirmed by several experiments @xcite . the elongation rate shows an increasingly heavy composition around this knee . above @xmath92 ev , the composition changes to a lighter mix . it seems to be correlated to the spectral break observed by fly s eye experiment which is marginally confirmed by this experiment . those experiments imply a relatively unchanging region between @xmath94 and @xmath0 ev but no measurements of elongation rate exist between @xmath97 and @xmath0 ev . the hires / mia hybrid experiment has measured the cosmic ray energy spectrum between @xmath0 and @xmath91 . the spectral index and intensity are given in table.[spectab ] . the result is in agreement with the fly s eye experiment . this result marginally supports the fly s eye stereo observation of a break in the energy spectrum at @xmath2 ev . the hires / mia hybrid experiment confirms the fly s eye experimental result that the elongation rate is different from simulation with an unchanging composition . modern hadronic interaction models and improved detector resolution in energy and @xmath20 do not change the original conclusion . within errors , the elongation rate observed in this experiment , @xmath98g/@xmath14/decade , is consistent with previous experiments , such as fly s eye and yakutsk@xcite . while the conclusion regarding the absolute value of @xmath88 of the primary composition depends on the interaction model used , this study shows that the elongation rate is stable with respect to choice of models . in the light of this , the amount of the change in the average composition , i.e. @xmath99 , is largely model independent , no matter what value of @xmath88 the change starts from . putting all experimental results together from @xmath90 to @xmath91 ev , we note that there seems to be a correlated patterns in the energy spectrum and elongation rate , @xmath20 vs. energy , plot . both measurements in energy spectrum and @xmath20 imply a continuity from lower to higher energies , with a flat bridge between @xmath94 and @xmath0 ev . we note that following the break , the fly s eye experiment @xcite reports a hardening of the spectrum near @xmath76 ev . this has been interpreted as evidence for the emergence of an extragalactic component above a softer galactic component @xcite . a change from a heavy to a light composition in this energy region also gives support to a changing origin for those cosmic rays . the lack of a strong galactic anisotropy at the highest energies would also rule out galactic sources for energetic protons @xcite . a number of new experiments , such as hires , the pierre auger project , the telescope array , euso and owl will address this issue . we acknowledge the assistance of the command and staff of dugway proving ground . this work is supported by the national science foundation under contract no . phy-93 - 21949 , phy-93 - 22298 and u.s . department of energy and the australia research council . 000 l.f.fortson _ et al . , proceedings of the 26th international conference ( icrc ) , salt lake city , 1999 _ , edited by d.kieda _ et al._,(univ . of utah , salt lake city,1999),vol.3 , p.125 ; k.h.kampert _ et al . _ , _ ibid . _ , vol.3 , p.159 ; f. arqueros _ et al . _ , astro - ph/9908202 ; m.a.k . et al . _ , * 12 * , 1 , ( 1999 ) . d.j.bird _ et al . _ , lett . * 71 * , 3401 , ( 1993 ) . n. hayashida _ et al . _ , j.phys.*g21 * , 1101 , ( 1995 ) . b.r.dawson , r. meyhandan and k.m.simpson , astropart . phys . * 9 * , 331 , ( 1998 ) . et al . _ , lett . * 84*,4276,(2000 ) . _ , the prototype high resolution fly s eye cosmic ray detector " , to be published in nucl.instrum.methods * a*. t.gaisser and a.m.hillas , _ proceedings of the 15th international conference , plovdiv , 1977 _ , ( bulgarian academy of sciences , plovdiv , bulgaria , 1977 ) , vol.8 , p.353 . r. m. baltrusaitis _ et al . _ , nucl . methods * a240 * 410 ( 1985 ) ; r. m. baltrusaitis _ et al . _ , _ et al . , proceedings of the 19th international conference , lajolla,1985 _ , ( nasa , washington d.c.,1985 ) vol.7 , p.159 a.borione _ et al . . meth . * a 346 * , 329 ( 1994 ) . k.d.green , in `` high energy gamma ray astronomy '' , aip conf . proc.,220 , 184 ( 1990 ) . d.j.bird _ et al._,``casa - mia - hires : a hybrid detector for measuring multiple properties of @xmath0 ev extensive airshowers '' , submitted to nucl.instrum.methods * a * ( 1999 ) . f.kakimoto _ et al . _ , . meth . * a 372 * , 527 ( 1996 ) . et al . , preprint of institut fur kernphys . , univ . of karlsruhe _ , fzka-6019 , feb . , 1998 ( kernforschungszentrum , karlsruhe , 1998 ) . c. song _ et al . _ , astropart . phys . * 14 * , 7 , ( 2000 ) . j. a. richards and l. w. nordheim , phys . rev . * 74 * 1106 , ( 1948 ) . m. hillas , j. phys . * g : 8 * , 1461 , ( 1982 ) . , , proceedings of the 25th international conference , durban , 1997_,(potchefstroomse univ . , potchefstroomse , 1997 ) , vol.5 , p. 345 . _ , `` a measurement of longitudinal development of the cosmic ray air shower between 10@xmath7 and 10@xmath9ev '' , to be published in astroparticle phys . , astro - ph/0008206 . a.f.bielajew , `` photon monte carlo simulation '' , report pirs-0393 , national research council of canada ( 1993 ) ; a.f.bielajew and d.w.o.rogers , `` electron monte carlo simulation '' , report pirs-0394 , national research council of canada ( 1993 ) . n.n.kalmykov , s.s.ostapchenko & a.i.pavlov , nucl.phys.*b52 * ( proc . suppl . ) 17 ( 1997 ) . r.s.fletcher _ et al . _ , phys . rev . * d50*,5710 , ( 1994 ) . tokyo workshop on techniques for the extremely high energy cosmic rays , tokyo , japan , 1993 _ , edited by m. nagano , ( institute of cosmic ray research , tokyo , 1993 ) p. 232 . et al._,astrophys.*j 424 * , 491 , ( 1994 ) . , proceedings of the 26th icrc(ref.[1 ] ) _ , vol.5 , p.429 . m. nagano _ _ , j. phys . g.*10 * , 1295 , ( 1984 ) ; j.w.fowler _ et al . _ , astro - ph/0003190 ; m.a.k.glasmacher _ et al . _ , astroparticle phys . * 10 * , 291 , ( 1999 ) ; s.p.swordy and d.b.kieda , astro - ph/9909381 ; m.amenomori _ et al . _ , astrophys . j. , * 461 * , 408 , ( 1996 ) . afanasiev b.n._et al . , proc . tokyo workshop on techniques for the extremely high energy cosmic rays , tokyo , japan , 1993 _ , edited by m. nagano , ( institute of cosmic ray research , tokyo , 1993 ) p. 35 . bird et al . , ap.j . * 511 * , 739 , ( 1999 ) .

calorons are instantons at finite temperature . for a long time the influence of a background polyakov loop on the properties of these topological excitations has been neglected . solutions were constructed long ago @xcite and were studied in detail in the semi - classical approximation @xcite . in all these studies the polyakov loop at spatial infinity ( also called the holonomy ) was trivial , i.e. an element of the center of the gauge group . that the influence of the background polyakov loop on the topological excitations can be dramatic is particularly clear in the confined phase , where on average its trace vanishes . caloron solutions in such backgrounds were constructed only relatively recently @xcite and can be seen as composed of massive monopole constituents with their magnetic charges adding to zero . it was observed that the one - loop correction to the action for configurations with a non - trivial asymptotic value of the polyakov loop gives rise to an infinite action barrier , which were therefore considered irrelevant @xcite . however , the infinity simply arises due to the integration over the finite energy _ density _ induced by the perturbative fluctuations in the background of a non - trivial polyakov loop @xcite . the proper setting would therefore rather be to calculate the non - perturbative contribution of calorons ( with a given asymptotic value of the polyakov loop ) to this energy density , as was first successfully implemented in supersymmetric theories @xcite , where the perturbative contribution vanishes . it has a minimum where the trace of the polyakov loop vanishes , i.e. at maximal non - trivial holonomy . in a recent study at high temperatures , where one presumably can trust the semi - classical approximation , the non - perturbative contribution of these monopole constituents ( also called dyons ) was computed @xcite . when added to the perturbative contribution @xcite with its minima at center elements , a local minimum develops where the trace of the polyakov loop vanishes , deepening further for decreasing temperature . this gives support for a phase in which the center symmetry , broken in the high temperature phase , is restored and provides an indication that the monopole constituents are the relevant degrees of freedom for the confined phase . also lattice studies , both using cooling @xcite and chiral fermion zero - modes @xcite as filters , have now confirmed that monopole constituents do dynamically occur in the confined phase . a charge 1 caloron is seen for su(@xmath0 ) to consist of @xmath0 constituent monopoles . in the deconfined phase , due to the fact that the average polyakov loop becomes a center element , the caloron returns to the form known as the harrington - shepard solution @xcite . the latter can also be interpreted as consisting of constituent monopoles , however , with @xmath1 of them being massless . to be precise , for self - dual configurations in the background of non - trivial holonomy the masses of constituent monopoles are given by @xmath2 , with @xmath3 . the @xmath4 are related to the eigenvalues of the polyakov loop at spatial infinity , = _ |x|(_0^a_0(t , x)dt)=g^(2i(_1,_2, ,_n))g , ( this expression assumes the periodic gauge @xmath5 ) where @xmath6 is the gauge rotation used to diagonalize @xmath7 and @xmath8 is the period in the imaginary time direction , related to the inverse temperature . the eigenvalues @xmath9 are to be ordered on the circle such that @xmath10 , with @xmath11 and @xmath12 , which guarantees that the masses add up to @xmath13 , the instanton action per unit ( imaginary ) time . at higher topological charge @xmath14 , the parameter space of widely separated constituent monopoles is described by @xmath15 monopole constituents , @xmath14 of each of the @xmath0 types of abelian charges ( with overall charge neutrality ) . we established in an earlier paper @xcite that well - separated constituents act as point sources for the so - called far field ( that is far removed from any of the cores ) . when constituents of opposite charge ( @xmath0 constituents of different type ) come together , the action density no longer deviates significantly from that of a standard instanton . its scale parameter @xmath16 is related to the constituent separation @xmath17 through @xmath18 . yet , the gauge field is vastly different , as is seen from the fact that within the confines of the peak there are @xmath0 locations where two of the eigenvalues of the polyakov loop coincide @xcite , thus in some sense varying over the maximal range available ( e.g. for su(2 ) from @xmath19 to @xmath20 ) , whereas for trivial holonomy only one such location is found . on the other hand , when constituents of equal charge come together typically an extended core structure is found . this was deduced , in particular for the case of charge 2 calorons , from our ability to analytically determine the zero - mode density ( summed over the two zero - modes implied by the index theorem ) in the far field limit , neglecting exponential contributions , such that the range of the exponential contributions shrinks inversely proportional with the temperature . ] . in this limit it forms a singular distribution on a disc bounded by an ellipse , but approaches two delta functions for well - separated like - charge constituents . this zero - mode density only sees constituents of a given charge , depending on the boundary condition for the fermions in the imaginary time direction , which can be chosen to be a u(1 ) phase ( containing the physically relevant choice of anti - periodic boundary conditions for thermal field theory ) . we will show for su(2 ) that their difference for periodic and anti - periodic boundary conditions coincides _ exactly _ with the ( abelian ) charge distribution extracted from the gauge field in the same limit , making contact with an old result due to hurtubise @xcite for the asymptotic behaviour of the monopole higgs field . we found two particular parametrizations within the su(2 ) charge 2 moduli space that exhibited these extended charge distributions . the first includes as a limit arbitrary charge 2 monopoles . the second of these parametrizations contains as a limiting case the axially symmetric configurations constructed for arbitrary charge in ref . deforming away from the axial configuration the two discs overlap . describing the intricate behaviour for the non - abelian core of these configurations in this region of the parameter space requires one to find exact solutions , which are presented here . we rely on early work of nahm @xcite and panagopoulos @xcite for charge 2 magnetic monopoles , which is simplified to some extent by our formalism that uses the relation between the fourier transformation of the adhm construction ( as relevant for the finite temperature case ) and the nahm transformation , a crucial ingredient for our success to find explicit solutions @xcite . [ fig : fig1 ] gives a particular example for the action and zero - mode densities of a charge 2 caloron solution . the two dimensional zero - mode basis is chosen such that each zero - mode only localizes on one of the constituents of a given charge , showing both the zero - modes with periodic and anti - periodic boundary conditions in the imaginary time direction . this paper is organized as follows . in sect . [ sec : con ] we will outline the construction , introduce the green s function that is computed through the analogy of an impurity scattering problem , and summarize the various limits that can be formulated before explicitly solving for the green s function . in sect . [ sec : exact ] we present the method that allows one to find the exact solution for the green s function , first for the general case and then applied in more detail to that of topological charge 2 calorons . readers only interested in the results could skip sects . [ sec : lim ] and [ sec : exact ] . in sect . [ sec : plots ] we discuss the two classes of configurations in the moduli space of the charge 2 calorons and provide plots of the various quantities to illustrate the properties of the exact results . in sect . [ sec : higgs ] we discuss the relation between the algebraic tail of the gauge field and the zero - mode density . we end with some discussions . an appendix presents a new result for the limiting behaviour of the action density . there are two steps in the construction of charge @xmath14 caloron solutions . the first step involves finding a @xmath21 gauge field @xmath22 , which satisfies the self - duality equation , i.e. the nahm equation @xcite , on a circle parametrized by @xmath23 , with @xmath23 introduced through replacing the original su(@xmath0 ) gauge field @xmath24 by @xmath25 . although not affecting the field strength , this changes the holonomy to @xmath26 , revealing that @xmath23 has period @xmath27 . the index theorem guarantees the existence of @xmath14 zero - modes @xmath28 which satisfy the dirac equation , or in the two - component weyl form @xmath29 ( @xmath30 are the usual pauli matrices ) . we may remove @xmath23 from the gauge field @xmath24 by transforming the zero - mode to @xmath31 , which is at the expense of making the zero - mode only periodic up to a phase factor , @xmath32 . in a similar way we could introduce @xmath33 through @xmath34 , which replaces in @xmath24 by @xmath35 , where @xmath36 . assuming the @xmath14 zero - modes @xmath37 to be orthonormal one has a^ab_(z)=^(a)(x;z , z)^^(b)(x;z , z ) d^4x,[eq : ndual ] or equivalently ( demonstrating as well that @xmath38 does not depend on @xmath33 ) a^ab_0(z)=^(a)(x;z)^ ^(b)(x;z ) d^4x , a^ab_k(z)=2i^(a)(x;z)^ x_k^(b)(x;z)d^4x . we have shown how @xmath22 can alternatively be related to a fourier transformation of the adhm construction of instantons @xcite , that periodically repeat ( up to a gauge rotation with @xmath7 ) in the imaginary time direction , so as to turn an infinite charge instanton in @xmath39 to one of finite charge and finite temperature . the derivation of this relation will not be repeated here , see ref . @xcite for the details . the connection to the adhm construction has been useful to simplify the second step in the construction of the caloron solutions , namely how to reconstruct the original gauge field when given a solution to the nahm equation @xcite ( which is equivalent to the quadratic adhm constraint ) , a_j(z)+[a_0(z),a_j(z)]+_jk[a_k(z ) , a_(z)]=2i_m(z-_m)_m^j,[eq : nahmeq ] for convenience of notation we will henceforth use the classical scale invariance to set @xmath40 . the singularities in the nahm equation appear precisely for those values where @xmath41 has one of its eigenvalues equal to 1 , at @xmath42 . this is where some fermion field components become massless , i.e. the zero - mode becomes delocalized , whereas for generic values of @xmath23 it is exponentially localized , which has turned out to be a useful tool to pinpoint the constituent monopoles . one could apply the nahm transformation again , introducing @xmath43 ( with @xmath44 ) and find the @xmath0 chiral fermion zero - modes in the background of this gauge field . the construction is somewhat complicated due to the presence of the singularities , whose structure is determined from the matrices @xmath45 which appear in the nahm equation . not all @xmath46 are independent , e.g. integrating and tracing the nahm equation yields the conditions , @xmath47 . further constraints are implied @xcite by the fact that one may introduce ( as is most easily seen in relation to the adhm construction ) @xmath14 two - component spinors @xmath48 in the @xmath49 representation of su(@xmath0 ) , such that 2_a^p_m_b=_0s_m^ab-_m^ab,[eq : srho ] where @xmath50 are projections defined through @xmath51 ( @xmath52 will appear in ) . however , with reference to the adhm construction , there is great benefit in first finding the solution for the green s function , @xmath53 , where \{-+v(z;x)}f_x(z , z)=4 ^ 2 _ k ( z - z),[eq : green ] with & & v(z;x)4 ^ 2r^2(z;x)+2_m(z-_m)s_m , s_mg(_m)s_mg^(_m ) , + & & r_j(z;x)x_j-(2i)^-1g(z)a_j(z)g^(z ) , [ eq : vrsdef ] and @xmath54 playing the role of impurities " . this is formulated in the gauge where first we transform @xmath55 to a constant ( diagonal ) matrix , @xmath56 , as is always possible in one dimension , and then use g(z)(2i(_0-x_0_k)z),_0=0 ( when @xmath57 it is absorbed in a shift of @xmath58 ) to transform @xmath59 to zero . this is at the expense of introducing periodicity up to a gauge transformation ; although @xmath60 is periodic in @xmath23 and @xmath61 with period 1 ( for @xmath40 ) , @xmath62 no longer is plays the role of the holonomy associated to the dual nahm gauge field @xmath22 . ] . given a solution for the green s function , there are straightforward expressions for the gauge field @xcite ( only involving the green s function evaluated at the impurity " locations ) and the fermion zero - modes @xcite . for the zero - mode density this gives _ z^(a)(x)^_z^(b)(x)=-(2)^-2_^2 f_x^ab(z , z).[eq : zmdens ] in this paper we will only have need for the green s function at @xmath63 , which formally can be expressed as f_x(z , z)=-4 ^ 2((_2k-_z)^-1)_12,_z g^(1)_z^z+1dw , where the @xmath64 component on the right - hand side of the first identity is with respect to the @xmath65 block matrix structure . in particular this leads to a compact expression for the action density @xcite ( x)-f_^2(x)=-_^2_^2(ie^-ix_0(_2k-_z)),[eq : sdens ] which can be shown to be independent of the choice of @xmath23 . the formal expression for @xmath66 can be made explicit by a decomposition into the impurity " contributions @xmath67 at @xmath68 and the propagation " @xmath69 between @xmath70 and @xmath71 . for @xmath72 this gives _ z = w_m(z,_m)t_mh_m-1t_2h_1t_1g^(1)h_nt_n h_n-1t_m+1h_m w_m(_m , z ) , with t_m , w_m(z , z ) ^-1.[eq : twdeg ] the columns of the two @xmath73 matrices @xmath74 , defined for @xmath72 , form the @xmath75 solutions of the homogeneous green s function equation , ( -4 ^ 2r^2(z;x))v(z)=0 , [ eq : homg ] of which those in @xmath76 are exponentially rising and those in @xmath77 are exponentially falling . this implies @xcite @xmath78 for @xmath79 , in which @xmath80 can be arbitrary non - singular ( to ensure a complete set of solutions ) matrices . in ref . @xcite we put @xmath81 , but here we find it convenient to leave this choice open . with the impurity " scattering problem solved , constructing the exact solutions of the homogeneous green s function equation is the last step in finding analytic expressions for the higher charge calorons . nevertheless , approximate solutions can be derived , either assuming @xmath82 is far removed from any core such that the gauge field has become abelian ( which we called the far field limit , denoted by a subscript ff " ) , or assuming @xmath82 and the constituents of type @xmath83 ( belonging to the @xmath83-th interval ) are well separated from all others , but not necessarily from each other ( which was called the zero - mode limit , denoted by a subscript zm " ) . this is because we found @xcite that for @xmath72 the @xmath14 zero - modes @xmath84 only see " the constituents of type @xmath83 . for @xmath85 we have f^_x(z , z)&=&(f_m^-(z)f_m^-(_m+1)^-1-f_m^+(z ) f_m^+(_m+1)^-1z_m+1 ^ - ) + & & -3 mm ( f_m^-(_m)f_m^-(_m+1)^-1-z_m^+f_m^+(_m ) f_m^+(_m+1)^-1z_m+1 ^ -)^-1 + & & ( f_m^-(_m)f_m^-(z)^-1-z_m^+f_m^+(_m)f_m^+(z)^-1 ) r_m^-1(z),[eq : zmlimit ] up to _ exponential _ corrections in the distance to the constituents of type @xmath86 , with z_m^-_k-2_m^-1r_m-1(_m ) , & & z_m^+_k-2 _ m^-1r_m(_m),[eq : zedefin ] + r_m(z)(r_m^+(z)+ r_m^-(z)),&&_mr_m^-(_m)+r_m-1^+(_m)+s_m.and 2r_m^(z)(f_m^(z))f_m^(z)^-1 . [ eq : rmdef ] in the zero - mode limit @xmath87 and @xmath88 approach @xmath89 up to _ algebraic _ corrections . neglecting these contributions as well , e.g. by sending the constituents of type @xmath86 to infinity , leads to the so - called monopole limit ( denoted by a subscript mon " ) , further simplifying the expression for the green s function to f^_x(z , z)=-u(z,_m+1)u^-1_m(_m,_m+1)u_m(_m , z ) r_m^-1(z),[eq : zm - alg ] with satisfies the green s function equation with boundary conditions @xmath90 and @xmath91 , for @xmath92 . ] u_m(z , z)f_m^+(z)f_m^+(z)^-1-f_m^-(z)f_m^-(z)^-1.[eq : um ] in turn , from one may derive the far field limit , giving up to _ exponential _ corrections in the distance to _ all _ constituents f_x^(_m,_m)=2_m^-1 , as is relevant for the expression of the gauge field in this limit @xcite . for the zero - mode density ( ) we may use for @xmath93 ( @xmath23 strictly different from @xmath94 ) f_x^(z , z)=r_m^-1(z).[eq : zmff ] the discontinuity in this limit , @xmath95 , arises due to the zero - mode developing a massless component when @xmath23 approaches @xmath70 . it might seem that , combined with , is inconsistent with an exponential decay . however , it turns out that @xcite _ m(x)(4)^-1(r_m^-1(z ) ) [ eq : vdef ] is independent of @xmath23 in the @xmath83-th interval and harmonic everywhere ( hence giving vanishing zero - mode density ) except for some singularities in the cores of the constituents of type @xmath83 . it is this feature , and our ability to compute @xmath96 exactly for su(2 ) charge 2 calorons , that allowed us to make statements about the localization of the cores , without solving the green s function exactly . in appendix a we derive the following new result for the monopole limit of the action density ( ) . in the limit where @xmath82 and the constituent locations of type @xmath83 are well separated from all other constituents , for which both the action and zero - mode densities are static , we find ( with @xmath97 defined in ) ^(x)=-_i^2_j^2,[eq : szm - alg ] up to _ algebraic _ corrections . this is a direct generalization for the action density of a single bps monopole , @xmath98 $ ] ( located at the origin and with mass @xmath99 ) . we emphasize that this result can be used irrespective of the distance between the constituents of the same type @xmath83 . therefore gives in terms of @xmath74 a closed form expression for the multi - monopole energy density . the same holds , when using , for the zero - mode density ( ) . for the caloron it involves taking the limit where all constituents of type @xmath100 are sent to infinity , which is why this is called the monopole limit . we recall , that in the far field one should use @xcite ( see also app . a ) ^(x)=-_i^2_j^2_m=1^n(f_m^+ ( _ m+1)f_m^+(_m)^-1r_m^-1(_m)_m).[eq : sff ] finding the exact homogeneous solutions of the green s function equation , , closely follows nahm s method @xcite to construct the dual chiral zero - modes . the main advantage of our approach is that we need not worry about boundary conditions , as this is solved by the impurity " scattering formalism @xcite . in the following we restrict ourselves to a given interval @xmath72 and work in the gauge where @xmath101 . using that @xmath22 is self - dual , a consequence of the nahm equation , one easily shows that @xmath102 , such that it is natural to consider the equation d_x(z)=_d_x^(z)=(-2r(z;x))(z)=0.[eq : duald ] it follows that @xmath103 would be a homogeneous solution of the green s function equation , albeit that @xmath103 is a spinor ( with a chirality opposite to that for the zero - modes involved in the nahm transformation , cmp . ) . we follow nahm @xcite and use the ansatz @xmath104 , where @xmath105 is a @xmath14 dimensional complex vector , @xmath106 is a unit vector that does not depend on @xmath23 and @xmath107 is an arbitrary normalized constant spinor ( as long as it is not annihilated by @xmath108 ) . it then follows that @xmath109 satisfies . the unit vector @xmath106 is found from a complex vector @xmath110 which squares to 0 , @xmath111 , implying its real and imaginary parts are perpendicular and of equal length ( @xmath112 as long as @xmath113 ) , such that ( for ease of notation the @xmath82 dependence of @xmath114 and @xmath115 will henceforth be left implicit ) u = iyy^/(yy^ ) [ eq : udef ] is well defined and @xmath116 , i.e. re@xmath117 , im@xmath117 and @xmath115 form an orthogonal set of vectors . using the ansatz for @xmath103 and introducing ( z)-yr(z;x ) , ( z)-2ur(z;x ) , leads to the equations ( z)v(z)=0,v(z)=(z)v(z)[eq : yu ] for which the first one can only have a solution provided @xmath118 . it is the great beauty of nahm s formalism that @xmath119 is a conserved quantity . that is , @xmath120 $ ] ( the nahm equation restricted to an interval and in the gauge where @xmath121 ) implies @xmath122 for any choice of @xmath82 and @xmath114 on a null - cone in @xmath123 ( or rather in @xmath124 since we may rescale @xmath114 with a non - zero complex factor without changing the equations ) . the conserved quantities are generated by the symmetric traceless monomials , @xmath125 , built from @xmath126 with @xmath127 arbitrary , as one readily verifies . for example , @xmath128 is constant and defines the center of mass . a natural way to project on the traceless symmetric monomials is precisely through introducing @xmath129 on a null - cone , forming @xmath130 . it is interesting to note that @xmath131 is always a spherical harmonic of order @xmath127 , used in ref . @xcite to show through the multipole expansion of @xmath132 that it is conserved and harmonic , except for singularities in the core of the constituents . once it is established that @xmath119 is independent of @xmath23 for any choice of @xmath82 , we can look for its zeros . using the null - cone parametrization @xmath133 , and the fact that the matrix @xmath134 is @xmath14 dimensional , we obtain a polynomial equation in @xmath135 of order @xmath75 and hence there are for generic @xmath82 exactly @xmath75 solutions . it is useful to note that these solutions come in complex conjugate pairs , where the symmetry @xmath136 implies @xmath137 and @xmath138 ( giving @xmath139 up to a multiple , equivalent to @xmath139 in @xmath124 ) . given a particular zero @xmath114 , we may conveniently write a vector in the kernel of @xmath134 as @xcite @xmath140 for a fixed choice of @xmath141 , where @xmath142 is the matrix formed by the minors of @xmath134 . the nahm equation is easily seen to imply @xmath143 $ ] for _ any _ choice of @xmath114 on the null - cone . from this one derives that is such that @xmath144 , in which case @xmath145 and therefore @xmath146\y^{-1}\det\y+\y^{-1}\ddz\det\y=[\u,\adj\y ] + \y^{-1}\tr(\y^{-1}\ddz\y)=[\u,\adj\y]$ ] . observe that @xmath147 is analytic in @xmath114 , such that the result is valid also when @xmath114 leads to @xmath148 . ] @xmath149 $ ] . substituting @xmath150 into gives ( ( z))_ac=(z)((z)(z ) ) _ ac . using the fact that @xmath151 , we get ( z)=,=- , [ eq : mu ] where the equation for @xmath152 is the same for any value of @xmath153 ( it may depend on the value of @xmath141 ) . this is useful for studying the asymptotic behaviour of the solution . for large @xmath154 , @xmath155 implies that @xmath156 , such that ( cmp . ) latexmath:[$\vec u\to\pm\vec x/ guarantee that there are @xmath14 zeros @xmath158 with the sign of @xmath159 negative , leading to solutions that rise as @xmath160 . it then follows that the @xmath14 zeros @xmath161 give @xmath162 , leading to the solutions that decay as @xmath163 for large @xmath154 . hence we may put @xmath164 and @xmath165 . defining m_ab^+(z)=-(y^(b)r(z;x ) ) _ ac , m_ab^-(z)=-(y^(b)*r(z;x ) ) _ ac,[eq : mdef ] which are _ algebraic _ in @xmath166 and @xmath167 , we find f^+_ab(z)=m^+_ab(z)^(b)(z),f^-_ab(z)=m^-_ab(z)^(b+k)(z ) , where @xmath168 contains the exponential dependencies , and thus seemingly all the information about the cores of the constituents . to make this more precise we compute @xmath169 , see , using that implies ( with @xmath170 ) @xmath171 , finding that the factors @xmath172 _ drop out _ r_ad^(z)=-_b , e=1^ku^(b)r_ae ( z;x)m_eb^(z)(m^(z)^-1)_bd.[eq : alg ] this proves that @xmath173 is purely algebraic in @xmath166 and @xmath167 , as are @xmath174 and @xmath175 , which determine the far field limit for the zero - mode density and the gauge field . one might wonder how , given that the @xmath176 are of the wrong " chirality in the context of the nahm transformation , one could use these results to reconstruct the gauge field for magnetic monopoles where the relation to the adhm construction is not readily available . for this one observes that the columns of @xmath176 can be used to form a @xmath177 matrix @xmath178 . using that @xmath179 , one finds @xmath180 . thus the columns of @xmath181 give @xmath75 independent solutions for each interval , from which @xmath0 normalizable solutions @xmath182 should remain after imposing the appropriate boundary ( cq . matching ) conditions . these are then used in nahm s original construction to compute the gauge field ( cmp . ) a^pq_(x)=^(p)(z;x_0,x)^^(q)(z;x_0,x ) dz , where @xmath183 . there seems to be considerable advantage in using the green s function ( fourier transformed adhm ) method , since it can solve the matching conditions without relying on the availability of exact solutions for the normalizable dual zero - modes . to go beyond the approximations discussed in sect . [ sec : lim ] and resolve the constituent cores we need to solve for @xmath152 . this can not always be done in closed form , but it is given by an explicit integral which can be performed numerically when required . for charge 2 monopoles panagopoulos @xcite was , however , able to find the exact integral . we can use the same ingredients for the caloron case and explicitly solve for the green s function in the case of charge 2 calorons . for charge 2 the number of invariants associated to the conserved quantities of the nahm equation is 8 , of which @xmath184 are related to the center of mass for the constituents of given magnetic charge , coming from the interval under consideration . assuming now that @xmath166 is traceless , 5 invariants remain , given in terms of the symmetric traceless tensor m_ij-((a_i(z)a_j(z ) ) -_ij(a_k^2(z ) ) ) . three of its parameters are associated to the rotation @xmath185 which diagonalizes the @xmath186 matrix , @xmath187 , where @xmath185 is fixed by requiring @xmath188 . the @xmath189 add to zero and can be expressed in terms of the so - called scale ( @xmath190 ) and shape ( @xmath191 ) parameters , c_1=d^2,c_2=d^2 , c_3=d^2 . the nahm equation for the case of charge 2 can be solved completely in terms of jacobi elliptic functions @xcite , which was summarized in ref . @xcite g(z)a_j(z;a,,h , d,)g^(z)2i a_j_2 + i d_jbf_b(d(z - z_0))h^_bh,[eq : jnahm ] where @xmath192 is a global gauge parameter and f_1(z),f_2(z),f_3(z),. [ eq : fs ] with implicitly defined by the elliptic integral of the first kind @xmath193 . ] @xmath194 , @xmath195 and @xmath196 the standard jacobi elliptic functions . this does not yet address the matching of @xmath166 on the different intervals , where some difference between the monopole and caloron application appears . for the caloron , apart from the axially symmetric solutions constructed in ref . @xcite , we found two sets of non - trivial solutions that interpolate between overlapping and well - separated constituents . it is for these classes of solutions that we will resolve the cores , when constituents overlap and the non - linearity plays an important role . the next step in the construction is finding the zeros @xmath114 of @xmath197 . one has ( z)=y_iy_j(x_ix_j - x^2_ij-(2)^-2 m_ij ) , where we used @xmath198 . substituting a parametrization for this null - cone in @xmath124 , e.g. @xmath199 , gives a 4th order polynomial . however , for finding the 4 solutions we find it in this case more convenient to first diagonalize the matrix @xmath200 . introducing @xmath201 , the equation for the zeros reduces to @xmath202 , which in addition to the null - cone condition , @xmath203 , is now easily solved by y^(a)=(,(-1)^a+1 , i),y^(a+2)=(y^(a))^ * , a=1,2,[eq : ys ] where we fixed ( for generic @xmath82 ) the diagonalizing rotation @xmath204 by ordering @xmath205 . using we find for @xmath206 u^(a)=-((-1)^a,,0),u^(a+2)=-u^(a ) , a=1,2 . one easily checks that @xmath207 and @xmath208 give rise to respectively the exponentially rising and falling solutions . it is also instructive to give the explicit expressions for the matrices @xmath209 in ( choosing @xmath210 ) . we note that for a @xmath65 matrix @xmath211 , and without loss in generality , @xmath212 , @xmath192 , @xmath185 and @xmath190 by resp . translations , ( gauge ) rotations , and suitable rescalings . ] we take @xmath213 , @xmath214 , @xmath215 , @xmath216 and @xmath217 in , such that m^+(z)=- , + & & + m^-(z)=- . we checked that , @xmath218 , evaluated using is independent of @xmath23 and agrees with the result derived in ref . @xcite . we next address solving @xmath219 , which for charge 2 can be written as , since @xmath220 . for the same choice of parameters in as above , @xmath213 , @xmath221 , @xmath215 , @xmath216 and @xmath217 , this gives ( with @xmath222 and @xmath210 ) = 2(xy).[eq : mup ] although the dependence on @xmath82 is complicated , the integral over @xmath23 turns out to be manageable ( as was observed before in the context of charge 2 monopoles , although we here choose not to express the solution in terms of theta functions @xcite ) . to solve the equation we first rewrite the right - hand side of using the fact that @xmath223 ( cmp . ) , & = & + & = & + .in the last identity we used that @xmath198 , @xmath224 and the fact that @xmath155 implies @xmath225 . with @xmath226 we can now integrate , ( z)&=&2z(xy)+ ( ) + ii(z ) , + i(z)&&_(f_3 ^ -1(z),n)-_(1,n)+|z| , n , up to an irrelevant constant , where @xmath227 is the elliptic integral of the third kind . note that @xmath228 can alternatively be written as @xmath229 , from which it follows that @xmath230 , @xmath231 and @xmath232 . ] , _ ( s , n)_0^s . we now combine these ingredients to give in terms of @xmath114 the exact form for the homogeneous solution of the green s function equation , v_a(z)=(z)((z))_a2=((z)-2zu x ) or putting in all the relevant expressions v_a(z)&=&(i ) + & & ( 4)^- ( -y_1f_1(z)-(-1)^aiy_2f_2(z))^()^. substituting @xmath233 , with @xmath234 as defined in gives @xmath235 and @xmath236 , and thereby the green s function , once we specify the parameters involved in the solutions to the nahm equation . the discontinuities in @xmath166 at @xmath237 and @xmath238 implied by the nahm equation , , impose constraints which are ( like the quadratic adhm constraint ) in general difficult to solve . work is in progress to describe the full parameter space for su(2 ) and charge 2 , but in ref . @xcite we did find two non - trivial parametrizations for which we illustrate here in a number of figures how the full caloron solutions look like , using the exact green s function as constructed in the previous section . taking advantage of the possibility to work with arbitrary arithmetic precision the programme maple has been used for these calculations . the configurations are formulated in terms of the two intervals @xmath239 $ ] and @xmath240 $ ] , each associated with two constituent monopoles of equal magnetic charge , but opposite in sign from one interval to the next . apart from a shift and ( gauge ) orientation , the configuration is described by a shape ( @xmath191 ) and scale ( @xmath190 ) parameter ( see sect . [ sec : ch2 ] ) , for simplicity assumed to be the same on both intervals . we also take all constituents to be of equal mass , @xmath241 ( @xmath242 ) , most relevant for the confined phase with @xmath243 . the two periodic and two anti - periodic chiral fermion zero - modes each have support on oppositely charged constituents ( see fig . [ fig : fig1 ] ) and in the far field limit it was found that the zero - mode density ( summed over the two zero - modes implied by the index theorem ) is described by a disc singularity , bounded by an ellipse with semi - major axis @xmath244 and eccentricity @xmath245 ( where @xmath246 ) . this revealed that the core of a cluster of like - charged constituents is in general extended , unless the individual constituents are well separated . the far field only describes the ( algebraic ) abelian component of the gauge field and to resolve the structure of the core we need to determine the full non - abelian structure . for the first parametrization ( called rectangular " ) the two discs are parallel and separated in height by a distance @xmath17 . the configuration is charaterized by the two extremal points along the major axis of the disc , also called _ would - be _ constituent locations . ] , y_m^(j)=(0,(-1)^md,(-1)^j(4)^-1d),2d = df_2(d ) , up to an overall shift and orientation ( the definition of @xmath247 , which involves @xmath191 , can be found in ) . a typical example is shown in fig . [ fig : conloc ] ( left ) . apart from @xmath191 and @xmath190 , the parameters that enter for the @xmath83-th interval are @xmath248 , @xmath215 , @xmath217 and @xmath249 . in all cases discussed here we have also put @xmath250 . one verifies that the discontinuities of @xmath166 at @xmath68 are given by @xmath251 with appropriately chosen @xmath252 ( cmp . ) , as discussed in detail in ref . @xcite . for the second parametrization ( called crossed " ) the two discs are coplanar and intersect , see fig . [ fig : conloc ] ( right ) for a typical example . their relative orientations can vary between perpendicular and coinciding ( for which @xmath191 is forced to 1 ) . here we choose for the parameters in @xmath192 and @xmath185 to be non - trivial ( isospin ) rotations around the @xmath253-axis with angles @xmath254 , resp . @xmath255 , and @xmath256 , whereas @xmath257 and @xmath258 are as in the rectangular case . the would - be constituent locations are now given ( up to an overall shift and orientation ) by y_m^(j)=((-1)^j(4)^-1d,0,(-1)^m+j ( 4)^-1d-(-1)^md ) , where @xmath259 conveniently gives the orientation of each of the two discs with respect to the @xmath23-axis . the angle @xmath260 originates from the definition of @xmath252 , which through determines the discontinuity of @xmath166 . to ensure the proper matching , the following three equations need to be satisfied @xcite d()=8 d(1),df_2(d)+8d=0 . which determine @xmath261 , @xmath191 and @xmath190 for given @xmath260 , @xmath262 and @xmath17 . [ fig : conloc ] illustrates that for these crossed configurations the two discs always overlap , unlike for the rectangular case . this formed an important motivation for the present study , so as to determine in how far the overlapping discs would affect the behaviour in the core . we will illustrate , using the exact formalism , how the action and zero - mode densities for these solutions behave . for the rectangular case we are mostly interested in the monopole limit , @xmath263 . for finite @xmath17 , the density separates in two contributions , where each in the limit @xmath263 is exactly the density for a charge 2 monopole with the same values of @xmath191 and @xmath190 . note that in the limit @xmath263 the matching conditions for @xmath166 decouple " the intervals , turning into pole conditions at @xmath264 , as is appropriate for multi - monopoles @xcite . we do recover from this the known results for the charge 2 monopoles @xcite , like the doughnut structure for @xmath265 , corresponding to two superimposed monopoles . the interest in the monopole limit comes from the fact that the zero - mode densities for multi - monopoles had not been studied in detail before @xcite . in fig . [ fig : monlim ] we give the densities for @xmath266 and @xmath267 , which is intermediate between the doughnut and well - separated monopole configurations . on the other hand , for @xmath17 small ( compared to @xmath40 ) the configuration will look like two non - dissociated calorons , and in particular is no longer static . when @xmath190 remains much bigger than @xmath17 , forcing @xmath268 , these behave as two well - separated charge 1 instantons . otherwise , when @xmath190 is comparable to @xmath17 , one finds overlapping instantons @xcite . an example for the crossed configuration with @xmath269 and @xmath270 was already shown in fig . [ fig : fig1 ] . in this case @xmath190 is large enough for the like - charge constituents to be separated , as is particularly clear from the zero - modes , which are essentially no longer overlapping . but two nearest neighbour ( oppositely ) charged constituents still show appreciable overlap . the distance between these nearest neighbours is @xmath271 . as this is comparable to @xmath40 we would expect the configuration to depend on time . indeed , at the maxima of the action density its value of @xmath272 at @xmath273 is reduced by almost 50% at @xmath274 . at the center of mass , where the action density is much lower , there is still a time dependence . however , far from all cores the field becomes static ) and topological charge @xmath14 the far field limit is static , _ provided _ all constituent monopoles have a non - vanishing mass ] . increasing @xmath190 further ( which will push @xmath191 closer to 1 ) the configuration quickly turns into well separated spherically symmetric static bps monopoles . more interesting is to consider the case with smaller @xmath190 , like @xmath275 and @xmath276 , for which the disc singularities were illustrated in fig . [ fig : conloc ] . the corresponding densities are shown in fig . [ fig : close ] . we see that the constituents are now so close that they form a doughnut , but we stress this is different from the _ static _ monopole doughnut , which has @xmath265 . since the oppositely charged constituents now are as close as 0.438 , which is considerably smaller than the time extent , the solution will have a strong time dependence . when @xmath190 is decreased even further , it will turn into a charge 2 instanton localized in both space and time . in this section we make some comments on the far field limit of the gauge field . as we discussed before , the gauge field far removed from any core becomes abelian ( as well as static ) . the abelian subgroup is the one that leaves the holonomy invariant , in the periodic gauge equivalent to leaving the constant asymptotic value of the adjoint higgs field @xmath277 invariant . for definiteness , let us consider the case of su(2 ) with @xmath278 , @xmath40 and @xmath279 ( i.e. @xmath280 ) . up to exponential corrections we have a^_0(x)=2i - i(x ) , where we can express @xmath281 in terms of the far field limit of the green s function at the impurities @xcite ( x)=^-1 ^ -1_i(f^_x(_2,_2 ) _ 2^i ) . using the twistor description of magnetic monopoles hurtubise @xcite was able to explicitly compute the asymptotic higgs field for the su(2 ) magnetic monopole long ago . the function he found for this algebraic tail amazingly agrees exactly with @xmath96 , , which was introduced to describe the caloron zero - mode density ( @xmath83 denotes the interval and hence the type of constituents to which the corresponding zero - modes would localize @xcite ) . as mentioned before , from our multi - caloron results one can recover the multi - monopole results by sending the constituent monopoles with the unwanted " magnetic charge to infinity , cmp . . although this tends to be cumbersome to show , @xmath281 in the far field can be written as @xmath282 , where @xmath283 is the contribution coming from the type @xmath83 constituent monopoles and the difference in sign is due to the sign change in the magnetic charge . this is simply because the field is abelian in the far field and linear superposition preserves the self - duality . hence , @xmath284 , such that for the su(2 ) caloron ( x)=2_1(x)-2_2(x ) . we checked that this result indeed holds for the solutions discussed in sect . [ sec : plots ] , even when the two types of monopole structures are not well - separated . this relation trivially holds for the axially symmetric solutions that were introduced in ref . @xcite , where @xmath281 was explicitly shown to factorize in a sum of point charge contributions , compatible with what was found in ref . @xcite for the zero - mode densities . therefore , in the far field limit ( i.e. for the algebraic part ) the singularity structure in the zero - mode density agrees _ exactly _ with the abelian charge distribution , as given by @xmath285 . such a relation is at the heart of using chiral fermion zero - modes as a filter to isolate the underlying topological lumps from rough lattice monte carlo configurations @xcite . in this paper we have analyzed the higher charge caloron solutions and showed how to obtain exact results by suitably combining techniques developed in the context of the nahm transformation and the adhm formalism . the aim of these studies has been to establish that su(@xmath0 ) caloron solutions of charge @xmath14 can be described in terms of @xmath15 monopole constituents , and that these can be viewed as independent constituents . a natural way to get an ensemble would be to consider approximate superpositions of @xmath14 charge 1 calorons , but this would lead to an unwanted memory effect , with constituents remembering from which caloron they originated @xcite . our studies , within the context of self - dual configurations , have shown nevertheless that the constituents have an independent identity , with the only requirement that the net magnetic and electric charge of the configuration vanishes ( each of the @xmath0 types of constituents should occur with the same number ) . a recent lattice study @xcite , using the technique of over - improvement @xcite , fully confirms this picture . it is therefore reasonable to consider the constituents as the independent building blocks for constructing an ensemble of monopole constituents , something that was not questioned in ref . @xcite , but like for the instanton liquid @xcite forms an essential assumption in a semi - classical study . clearly the expectation is that semi - classical methods no longer work in the confined regime , at least for the part of the parameter space that corresponds to well - separated constituents , that is typically associated to instantons with a large scale parameter . it is not unlikely that the density of these constituents at low temperatures is so high that they form a coherent background and as such will no longer easily be recognized as lumps . with high quark densities leading to deconfinement , it may perhaps be that a high constituent monopole density will lead to confinement @xcite , although for now we have to leave this as a speculation . instantons that overlap get deformed and depending on the relative gauge orientation tend to repel " , i.e. inspecting the action density distribution they do not get closer than a certain distance @xcite . when deconstructing instantons in monopole constituents , interestingly only like - charge constituents will show this effect , manifesting itself through the extended core structure . for unlike charges , from the point of view of the abelian field , the configuration behaves as with linear superposition . if as a consequence of this all abelian charge is annihilated , it disappears through forming a small instanton ( localized in space and time ) , which in the limit of zero size describes the boundary of the moduli - space . the interaction between constituents of opposite duality is more complicated @xcite . in conclusion , calorons with non - trivial holonomy have revealed a rich structure , incorporating traditional instanton physics , but allowing for gauge fields that inherit some essential features associated to a confining background not present in the traditional formulations . the fact that the underlying constituents are monopoles opens the way to describe the confining aspects of the theory in terms of these degrees of freedom . much work remains to be done when it comes to understanding the dynamics , but we hope to have convinced the reader that a consistent picture is developing that holds considerable promise for the future . we thank conor houghton for pointing us to hurtubise s paper , and chris ford for discussions on the analytical aspects . we are grateful to christof gattringer , michael ilgenfritz , boris martemyanov and michael mller - preussker for stimulating discussions and correspondence on lattice implementations . the research of fb is supported by fom . he is grateful to chris ford and conor houghton for their hospitality while visiting trinity college in dublin , and thanks them , as well as werner nahm and samson shatashvili , for their interest . in this appendix we derive the zero - mode limit for the action density , which assumes that the distance of @xmath82 and the constituents of type @xmath83 to all constituents of type @xmath86 is large , but where @xmath82 and the constituent locations of type @xmath83 may otherwise be arbitrary . as in ref . @xcite we take @xmath286 for computing @xmath66 and use that we can write @xmath287 , where @xmath288 and @xmath289 , with & & _ m^-1t_m , + & & + & & f_m . we note the @xmath290 has _ no _ remaining dependence on the constituent locations of type @xmath83 . writing @xmath291 , with k , k , l , l . we find @xmath292 . we next use k^-1=,kk^-1= and note that in the zero - mode limit @xmath293 , @xmath294 , @xmath295 and @xmath296 are exponentially small ( cmp . ref . @xcite , app . a ) , such that ( ie^-ix_0(_2k - lk))=(e^-2ix_0k_++ ) ( l_++ ) . with the definition of @xmath297 we now find l_++=r_m+1 ^ -1(_m+1)(r_m+1 ^ -(_m+1)+s_m+1 ) u_m(r_m-1^+(_m)+s_m ) , where -8mm&&u_m=_m+1^+f_m^+(_m+1)f_m^+(_m)^-1r_m^-1 ( _ m)_m^+-_m+1 ^ -f_m^-(_m+1)f_m^-(_m)^-1r_m^-1 ( _ m)_m^- , + -8mm&&^_m=_k(r_m^- ( _ m)+s_m)^-1r_m-1^(_m ) , ^_m=_kr_m^(_m)(r_m-1^+(_m)+s_m)^-1.[eq : szmlim ] and @xmath298 contains _ all _ contributions due to the constituent locations of type @xmath83 , up to _ exponential _ corrections in the distance of these , _ and _ of @xmath82 , to the other constituents . hence @xmath299 splits into the sum of two contributions , @xmath300 and @xmath301 $ ] , where the last term only depends on the constituent locations of type @xmath86 whose contribution will decay inversely proportional to the fourth power of their distance . allowing for _ algebraic _ decay ( or in the monopole limit , sending all constituents of type @xmath86 to infinity ) such that in addition @xmath302 , one thus finds . a simple way to derive the result for the far field limit in is by noting that in this case _ all _ @xmath303 are exponentially small and @xmath304 can be approximated by @xmath305 , with @xmath306 . this therefore acts as a projection on the @xmath307 component and is thus seen to lead to @xmath308 . using the fact that @xcite @xmath309 gives the required result . harrington and h.k . shepard , phys . d17 ( 1978 ) 2122 ; ibid . d18 ( 1978 ) 2990 . gross , r.d . pisarski and l.g . yaffe , rev . 53 ( 1981 ) 43 . kraan and p. van baal , phys . lett . b428 ( 1998 ) 268 [ hep - th/9802049 ] ; nucl . phys . b533 ( 1998 ) 627 [ hep - th/9805168 ] ; phys . lett . b435 ( 1998 ) 389 [ hep - th/9806034 ] . k. lee and p. yi , phys . d56 ( 1997 ) 3711 [ hep - th/9702107 ] ; k. lee , phys . lett . b426 ( 1998 ) 323 [ hep - th/9802012 ] ; k. lee and c. lu , phys . d58 ( 1998 ) 025011 [ hep - th/9802108 ] . n. weiss , phys . rev . d24 ( 1981 ) 475 . davies , t.j . hollowood , v.v . khoze and m.p . mattis , nucl . phys b559 ( 1999 ) 123 [ hep - th/9905015 ] . d. diakonov , n. gromov , v. petrov and s. slizovskiy , `` quantum weights of dyons and of instantons with non - trivial holonomy , '' hep - th/0404042 . ilgenfritz , m. mller - preussker , and a.i . veselov , in : _ lattice fermions and structure of the vacuum _ , eds . v. mitrjushkin and g. schierholz ( kluwer , dordrecht , 2000 ) , 345 [ hep - lat/0003025 ] ; e .- m . ilgenfritz , b.v . martemyanov , m. mller - preussker , s. shcheredin and a.i . veselov , phys . rev . d66 ( 2002 ) 074503 [ hep - lat/0206004 ] . c. gattringer , phys d67 ( 2003 ) 034507 [ hep - lat/0210001 ] ; c. gattringer and s. schaefer , nucl . b654 ( 2003 ) 30 ; [ hep - lat/0212029 ] ; c. gattringer and r. pullirsch , phys . d69 ( 2004 ) 094510 [ hep - lat/0402008 ] . f. bruckmann , d. ngrdi and p. van baal , nucl . b666 ( 2003 ) 197 [ hep - th/0305063 ] . m. garca prez , a. gonzlez - arroyo , a. montero and p. van baal , jour . of high energy phys . 06 ( 1999 ) 001 [ hep - lat/9903022 ] . p. van baal , in : _ lattice fermions and structure of the vacuum _ , eds . v. mitrjushkin and g. schierholz ( kluwer , dordrecht , 2000 ) , p. 269 [ hep - th/9912035 ] . ilgenfritz , b.v . martemyanov , m. mller - preussker and a.i . veselov , phys . d69 ( 2004 ) 114505 [ hep - lat/0402010 ] . j. hurtubise , commun . 97 ( 1985 ) 381 . f. bruckmann and p. van baal , nucl . b645 ( 2002 ) 105 [ hep - th/0209010 ] . w. nahm , self - dual monopoles and calorons " , in : lect . notes in physics . 201 , eds . g. denardo , e.a . ( 1984 ) p. 189 . h. panagopoulos , phys . d28 ( 1983 ) 380 . atiyah , n.j . hitchin , v.g . drinfeld , yu . i. manin , phys . 65 a ( 1978 ) 185 ; m.f . atiyah , _ geometry of yang - mills fields _ , fermi lectures , ( scuola normale superiore , pisa , 1979 ) . m. garca prez , a. gonzlez - arroyo , c. pena and p. van baal , phys . d60 ( 1999 ) 031901 [ hep - th/9905016 ] . chernodub , t.c . kraan and p. van baal , nucl . phys . b(proc.suppl . ) 83 - 84 ( 2000 ) 556 [ hep - lat/9907001 ] . w. nahm , the algebraic geometry of multi - monopoles " , in : _ group theoretical methods in physics _ , eds . m. serdaroglu , e.a . , istanbul 1982 , p.456 . w. nahm , multi - monopoles in the adhm construction " , in : _ gauge theories and lepton hadron interactions _ , eds . z. horvath , e.a . , ( crip , budapest , 1982 ) . brown , h. panagopoulos and m.k . prasad , phys . d26 ( 1982 ) 854 ; + a.s . dancer , comm . math . phys . 158 ( 1993 ) 545 . p. forgcs , z. horvth and l. palla , nucl . b192 ( 1981 ) 141 ; m.f . atiyah , and n.j . hitchin , _ the geometry and dynamics of magnetic monopoles _ , ( princeton univ . press , 1988 ) . corrigan and p. goddard , annals phys . ( n.y . ) 154 ( 1984 ) 253 . m. garca prez , t.g . kovcs and p. van baal , phys . b472 ( 2000 ) 295 [ hep - ph/9911485 ] . f. bruckmann , m. ilgenfritz , b. martemyanov and p. van baal , hep - lat/0408004 . m. garca prez , a. gonzlez - arroyo , j. snippe and p. van baal , nucl . phys . b413 ( 1994 ) 535 [ hep - lat/9309009 ] . t. schfer and e.v . shuryak , rev . ( 1998 ) 323 [ hep - ph/9610451 ] ; d. diakonov , prog . part . 51 ( 2003 ) 173 [ hep - ph/0212026 ] . f. bruckmann , d. ngrdi and p. van baal , acta phys . polon . b34 ( 2003 ) 5717 [ arxiv : hep - th/0309008 ] . ilgenfritz , b.v . martemyanov , m. mller - preussker and a.i . veselov , eur . j. c34 ( 2004 ) 439 [ hep - lat/0310030 ] ; e.m . ilgenfritz , m. mller - preussker , b.v . martemyanov and p. van baal , phys . d69 ( 2004 ) 097901 [ hep - lat/0402020 ] .

the quantum states of particle and anti - particle excitations in a superconductor ( bogoliubov quasiparticles ) are related by a unitary transformation , which means that they can be represented by a _ real _ wave function . in this so - called majorana representation the @xmath4 scattering matrix @xmath1 at the fermi level is real orthogonal rather than complex unitary @xcite . since the orthogonal group @xmath5 is doubly connected , this immediately implies a twofold distinction of scattering problems in a superconductor : the subgroup @xmath6 of scattering matrices with determinant @xmath7 , connected to the unit matrix , is called _ topologically trivial _ , while the disconnected set @xmath8 of scattering matrices with determinant @xmath9 is called _ topologically nontrivial_. in mathematical terms , the experimental search for majorana bound states can be called a search for systems that have @xmath10 . this search has been reviewed , from different perspectives , in refs . . if the scattering is chaotic the scattering matrix becomes very sensitive to microscopic details , and it is useful to develop a statistical description : rather than studying a particular @xmath1 , one studies the probability distribution @xmath11 in an ensemble of chaotic scatterers . this is the framework of random - matrix theory ( rmt ) @xcite . the ensemble generated by drawing @xmath1 uniformly from the unitary group @xmath12 , introduced by dyson in the context of nuclear scattering @xcite , is called the circular unitary ensemble ( cue ) . superconductors need a new ensemble . a natural name would have been the circular orthogonal ensemble ( coe ) , but since that name is already taken for the coset @xmath13 , the alternative name circular real ensemble ( cre ) is used when @xmath1 is drawn uniformly from @xmath5 . the rmt of the cre , and the physical applications to majorana fermions and topological superconductors , have been reviewed recently @xcite . the uniformity of the distribution requires ideal coupling of the scattering channels to the continuum , which physically means that the discrete spectrum of a quantum dot is coupled to metal electrodes by ballistic point contacts . if the point contact contains a tunnel barrier , then @xmath11 is no longer uniform but biased towards the reflection matrix @xmath14 of the barrier . the modified distribution @xmath15 is known @xcite , it goes by the name `` poisson kernel '' and equals @xmath16 in the cre @xcite . ) with superconducting boundaries , connected to metal electrodes by a pair of point contacts ( supporting a total of @xmath17 propagating modes ) . a magnetic insulator introduces a tunnel barrier in each point contact ( transmission probability @xmath18 per mode ) . a magnetic vortex may introduce a majorana bound state in the quantum dot . ] in the present work we apply this result to the scattering ( andreev reflection ) in a superconducting quantum dot ( andreev billiard ) , see fig . [ fig_layout ] . we focus in particular on the effect of a bound state at the fermi level ( @xmath19 ) in the quantum dot , a so - called majorana zero - mode or majorana bound state . in addition to the scattering matrix , which determines the thermal and electrical conductance , we consider also the time - delay matrix @xmath2 . the eigenvalues of @xmath3 are positive numbers with the dimension of time , that govern the low - frequency dynamics of the system ( admittance and charge relaxation @xcite ) . moreover , the trace of @xmath3 gives the density of states and @xmath3 and @xmath1 together determine the thermopower @xcite . the joint distribution of @xmath1 and @xmath3 is known for ballistic coupling @xcite , here we generalize that to tunnel coupling . the effect of a tunnel barrier on the time - delay matrix has been studied for complex scattering matrices @xcite , but not yet for real matrices . one essential distinction is that the tunnel barrier has no effect on the density of states in the cue and coe , but it does in the cre . the outline of the paper is as follows . the next two sections formulate the scattering theory of the andreev billiard and the appropriate random - matrix theory . our key technical result , the joint distribution @xmath0 , is given in sec . [ sect_ps0q ] . we apply this to the simplest single - channel case ( @xmath20 ) in sec . [ taudos ] , where obtain a remarkable scaling relation : for a high tunnel barrier ( transmission probability @xmath21 ) the distribution @xmath22 of the density of states at the fermi level is described by a one - parameter scaling function @xmath23 : @xmath24 the average density of states in the multi - channel case is calculated in sec . [ sec_dos ] . by relating the ensemble averages of @xmath3 and @xmath1 we derive the relation @xmath25\right),\label{rhoaveragegeneral}\ ] ] for a mode - independent tunnel probability @xmath18 . in the cue and coe the average scattering matrix @xmath26 is just equal to @xmath14 , so @xmath27 remains equal to its ballistic value @xmath28 , but the cre is not so constrained . applications to the thermal conductance @xmath29 and the electrical ( andreev ) conductance @xmath30 follow in secs . [ transmissiong ] and [ electrcond ] . for ballistic coupling it is known that @xmath31 is the same with or without the majorana bound state @xcite . ( this also holds for @xmath32 @xcite . ) in the presence of a tunnel barrier this is no longer the case , but we find that the majorana bound state remains hidden if even a single scattering channel has @xmath33 . the distribution of @xmath30 , in contrast , is sensitive to the presence or absence of the majorana bound state even for ballistic coupling @xcite . the way in which @xmath34 changes as we tune the system through a topological phase transition , at which a majorana bound state emerges , is calculated in sec . [ phasetransition ] . we conclude in sec . [ conclude ] . in the main text we focus on the results and applications . details of the calculations are moved to the appendices . these also contain more general results for other rmt ensembles , with or without time - reversal and/or spin - rotation symmetry . ( both symmetries are broken in the cre . ) fig . [ fig_layout ] shows the scattering geometry , consisting of a superconducting quantum dot ( andreev billiard ) on the surface of a topological insulator , connected to normal metal electrodes by point contacts . the hamiltonian @xmath35 of the quantum dot is related to the energy - dependent scattering matrix @xmath36 by the mahaux - weidenmller formula @xcite , @xmath37 the @xmath38 matrix @xmath39 couples the @xmath40 energy levels in the quantum dot ( mean level spacing @xmath41 ) to a total of @xmath42 propagating modes in the point contact . we assume that degeneracies are broken by spin - orbit coupling in the topological insulator in combination with a magnetic field ( perpendicular to the surface ) . all degrees of freedom are therefore counted separately in @xmath43 and @xmath40 , as well as in @xmath41 . the electron - hole degree of freedom is also included in the count , but we leave open the possibility of an unpaired majorana fermion a coherent superposition of electron and hole quasiparticles that does not come with a distinct antiparticle . an odd level number @xmath40 indicates the presence of a majorana bound state in the quantum dot , produced when a magnetic vortex enters @xcite . an odd mode number @xmath43 signals a propagating majorana mode in the point contact , allowed by a @xmath44-phase difference between the superconducting boundaries @xcite . the @xmath43 modes have an energy - independent transmission probability @xmath45 $ ] per mode . if we choose a basis such that the coupling matrix @xmath39 has only nonzero elements on the diagonal , it has the explicit form @xcite @xmath46 notice that the tunnel probability @xmath47 determines the reflection amplitude @xmath48 $ ] up to a sign . the conventional choice is to take @xmath49 , when @xmath50 can be written as @xmath51 alternatively , if @xmath52 one has @xmath53 given by @xmath54 the two choices are equivalent for ballistic coupling , @xmath55 , but for a high tunnel barrier @xmath56 one has @xmath57 while @xmath58 . the sign change of @xmath59 is a topological phase transition @xcite , which we will analyze in section [ phasetransition ] . for now we take @xmath49 for all @xmath60 , so @xmath50 . from the scattering matrix we can obtain transport properties , such as the electrical and thermal conductance , and thermodynamic properties , such as the density of states . if we restrict ourselves to properties at the fermi level , @xmath19 , we need the matrix @xmath61 and the derivative @xmath62 the unitarity of @xmath36 implies that @xmath3 is hermitian , so it has real eigenvalues @xmath63 with the dimension of time . the @xmath63 s are called ( proper ) delay times and @xmath3 is called the wigner - smith time - delay matrix @xcite . the fermi - level density of states @xmath64 is obtained from @xmath3 via the birman - krein formula @xcite , @xmath65 for the thermal conductance we partition the modes into two sets , @xmath17 , each set connected to a different terminal , and decompose the scattering matrix into reflection and transmission subblocks , @xmath66 a small temperature difference @xmath67 between the two terminals , at average temperature @xmath68 , drives a heat current @xmath69 . the thermal conductance @xmath70 in the low - temperature linear - response limit @xmath71 is given by @xmath72 the quantum @xmath73 is a factor - of - two smaller than in systems without superconductivity @xcite , due to our separate counting of electron and hole degrees of freedom that allows to account for the possibility of propagation via an unpaired majorana mode . if we keep the two terminals at the same temperature but instead apply a voltage difference , we can drive an electrical current . we consider a situation where both terminal 2 and the superconductor are grounded , while terminal 1 is biased at voltage @xmath74 . the current @xmath75 from terminal 1 to ground is then given by the andreev conductance @xmath76 in the zero - temperature , zero - voltage limit . in the last equality we used the particle - hole symmetry relation @xmath77 at @xmath19 , where the @xmath78 pauli matrices act on the electron ( @xmath79 ) and hole ( @xmath80 ) degree of freedom . for a statistical description we consider an ensemble of quantum dots , each with its own random hamiltonian @xmath35 . the mean level spacing @xmath41 and coupling matrix @xmath39 are kept fixed . if the wave dynamics in the quantum dot is chaotic , the ensemble is fully characterized by the presence or absence of certain fundamental symmetries . this is the universal framework of random - matrix theory . superconducting systems are characterized by particle - hole symmetry , @xmath81 the pauli matrices @xmath82 can be removed from the symmetry relation by a unitary transformation @xmath83 after which we simply have @xmath84 in this so - called majorana basis the hamiltonian is real antisymmetric , @xmath85 with @xmath86 . if no other symmetries are imposed on the hamiltonian we have the class - d ensemble of random - matrix theory @xcite , with gaussian probability distribution @xmath87 the eigenvalues of the antisymmetric @xmath88 matrix @xmath35 come in @xmath89 pairs , hence if @xmath40 is odd there must be a nondegenerate eigenvalue @xmath19 at the fermi level , in the middle of the superconducting gap . this so - called majorana bound state is the hallmark of a topologically nontrivial superconductor @xcite . if @xmath40 is even there is no level pinned to @xmath19 , and the superconductor is called topologically trivial . it is helpful to encode the distinction in a topological quantum number @xmath90 that counts the number of majorana bound states , so @xmath90 equals 0 or 1 if the superconductor is topologically trivial or non - trivial , respectively . in the scattering matrix the presence of a majorana bound state is signaled by the sign of the determinant , @xmath91 this can be seen directly from the definition in the majorana basis : for @xmath40 even the matrix @xmath85 is invertible , so we have @xmath92 since @xmath93 if @xmath94 . for @xmath40 odd the bound state contributes to the determinant a factor @xmath95 for some vector @xmath96 , so @xmath10 . the class - d ensemble of scattering matrices thus consists of two disjunct sets : the special orthogonal group @xmath97 of orthogonal matrices with determinant @xmath7 in the topologically trivial case , and the complement @xmath8 of orthogonal matrices with determinant @xmath9 in the topologically nontrivial case . for ballistic coupling ( @xmath98 for all @xmath60 ) the matrices @xmath1 and @xmath3 are statistically independent @xcite , so they can be considered separately . the class - d ballistic scattering matrix is uniformly distributed in @xmath99 uniformity being defined with respect to the haar measure @xcite . this is the circular real ensemble ( cre ) , the analogue for real orthogonal matrices of the circular unitary ensemble ( cue ) for complex unitary matrices @xcite . the class - d ballistic time - delay matrix has probability distribution @xcite , @xmath100 where @xmath101 is the heisenberg time and @xmath102 restricts @xmath3 to positive definite real symmetric matrices . this constraint can be implemented more directly by defining @xmath103 with @xmath104 the set of @xmath105 matrices with real elements . the distribution is then equivalent to @xmath106 multiply by the jacobians @xmath107 , for @xmath108 and @xmath109 . ] to the wishart distribution @xcite @xmath110 remarkably , there is no dependence on the topological quantum number for ballistic coupling : @xmath3 has the same distribution irrespective of the presence or absence of a majorana bound state . tunnel coupling is described by a reflection matrix @xmath14 ( from outside to outside ) and transmission matrix @xmath111 ( from outside to inside ) . in the majorana basis these are real matrices , parameterized by @xmath112 . \end{split } \label{rbdef}\ ] ] as we derive in app . [ app_ps0q ] , the matrix product @xmath113 determines the joint distribution [ ps0q ] @xmath114 as a check , we can integrate out the time - delay matrix to obtain the marginal distribution of the scattering matrix , @xmath115 the jacobian evaluates to @xcite @xmath116 and we recover the class - d poisson kernel to indicate an unspecified normalization constant . the poisson kernel is normalized , @xmath117 with @xmath118 the haar measure on @xmath119 , so we use @xmath120 instead of @xmath121 . ] @xcite , @xmath122 the joint distribution tells us that @xmath1 and @xmath3 become correlated in the presence of a tunnel barrier . however , @xmath1 remains independent of the matrix product @xmath123 so that the joint distribution of @xmath1 and @xmath124 factorizes , @xmath125 the transformation from @xmath3 to @xmath124 removes the effect of the tunnel barrier on the time - delay matrix ( see app . [ app_ps0q ] ) . we calculate the joint distribution @xmath0 of scattering matrix and time - delay matrix in the presence of a tunnel barrier , starting from the known distribution @xmath241 without a barrier @xcite . the application in the main text concerns symmetry class d , but for the sake of generality and for later reference we give results for all four altland - zirnbauer @xcite symmetry classes d , diii , c , ci , as well as for the three wigner - dyson @xcite symmetry classes a , ai , aii . the symmetry indices that distinguish the ensembles are listed in table [ table_az ] , see ref . for an overview of this classification . the unitary matrix @xmath1 and the hermitian matrix @xmath3 are real in class d , complex in class a , and quaternion in class c. we consider these three symmetry classes without time - reversal symmetry first , and then include the constraints of time - reversal symmetry in classes diii , ci , ai , aii . do not exhaust the tenfold way classification of random - matrix theory : there are three more chiral classes @xcite ( labeled aiii , bdi , cii ) that require separate consideration @xcite . ] .parameters that appear in the distribution of the scattering matrix and time - delay matrix , for each of the altland - zirnbauer and wigner - dyson symmetry classes . notice that a different set of indices @xmath242 , @xmath243 govern the energy level statistics @xcite . the degeneracy factor @xmath244 refers to the kramers degeneracy of the scattering channels and the delay times , ignoring uncoupled spin bands . ( the energy levels may have a different degeneracy . ) [ cols="^,^,^,^,^,^,^,^,^,^,^ " , ] without the barrier @xmath245 is independent of @xmath124 and uniformly distributed , @xmath246 the differential @xmath247 indicates the haar measure for the unitary matrix @xmath245 and the euclidean measure for the hermitian matrix @xmath124 . the ballistic time - delay matrix distribution is given by @xcite [ pq0result ] @xmath248 degenerate eigenvalues of @xmath124 are counted only once in @xmath249 and @xmath250 . in terms of the degeneracy factor @xmath244 from table [ table_az ] this can be written as @xmath251 the channel number @xmath43 also does not include degeneracies , so the total number of eigenvalues of @xmath124 is @xmath252 . the characteristic time @xmath253 differs from the heisenberg time @xmath254 by a numerical coefficient , , with @xmath41 the mean spacing of _ nondegenerate _ levels . the ratio @xmath255 then equals the degeneracy of energy levels divided by the degeneracy of delay times @xcite . it is unity in all symmetry classes except c and ci , where the delay times have a kramers degeneracy that the energy levels lack @xcite . ] see table [ table_az ] . insertion of the barrier , with unitary scattering matrix @xmath256 transforms @xmath245 into @xmath257 variations of @xmath1 and @xmath245 are related by @xcite @xmath258 the differentials @xmath259 are anti - hermitian matrices , @xmath260 . the number of degrees of freedom of the off - diagonal elements are given by @xmath261 and the number of degrees of freedom of the diagonal elements by @xmath262 . as summarized in table [ table_alphabeta ] , real matrices ( class d ) have @xmath263 , @xmath264 , complex matrices ( class a ) have @xmath265 , @xmath266 , and quaternion matrices ( class c ) have @xmath267 , @xmath268 . these parameters determine the jacobian @xcite @xmath269 eq . also implies the relation between the time - delay matrices , @xmath270 the off - diagonal elements of the hermitian matrix @xmath3 have @xmath261 degrees of freedom , the diagonal elements have one single degree of freedom in each symmetry class . the jacobian is then given by @xmath271 the joint probability distribution @xmath0 now follows upon division of @xmath241 by the product of jacobians , @xmath272 substituting @xmath273 into eq . we thus arrive at the joint distribution @xmath274 the class - d result from the main text follows for @xmath263 , @xmath264 , @xmath275 , @xmath276 , @xmath277 . time - reversal symmetry equates the scattering matrix to its transpose @xmath278 in class ai and ci and to its dual @xmath279 in class aii and diii . ( the dual of a matrix is @xmath280 . ) we use a unified notation @xmath281 , where the tilde indicates the transpose or the dual , whichever is appropriate for that symmetry class . the symmetry @xmath281 allows for the `` square root '' factorization @xmath282 with unitary @xmath283 . the time - delay matrix in these symmetry classes is constructed such that it satisfies the same symmetry , @xmath284 this redefinition of @xmath3 differs from eq . by a unitary transformation , so the delay times are not affected . the ballistic @xmath124 and @xmath245 are again independent @xcite , distributed according to eqs . and with the appropriate values of @xmath285 and @xmath261 from table [ table_az ] . these numbers now count the diagonal and off - diagonal degrees of freedom of the symmetrized differential @xmath286 constrained by @xmath287 , @xmath288 . the matrix elements of @xmath289 are imaginary in class ai ( @xmath265 , @xmath264 ) , @xmath290 times a quaternion , with four real coefficiients @xmath291 . the matrix @xmath292 is the @xmath198 unit matrix . ] in class aii ( @xmath265 , @xmath268 ) , and of the form @xmath293 with @xmath294 imaginary in class ci ( @xmath295 , @xmath266 ) and @xmath294 real in class diii ( @xmath263 , @xmath266 ) . the elements of the hermitian matrix @xmath3 are real in class ai , quaternion in class aii , and of the form @xmath296 with @xmath294 real in both classes ci and diii . the off - diagonal elements of @xmath3 have the same number of @xmath261 degrees of freedom as @xmath289 , but the diagonal elements have only a single degree of freedom irrespective of @xmath285 . all of this is summarized in table [ table_alphabeta ] . the symmetrization of the differential modifies the relation , which now reads @xmath297 the relation between @xmath3 and @xmath124 is similarly modified by the symmetrization , @xmath298 because the matrices @xmath283 , @xmath299 are unitary , the jacobians and are unchanged , @xmath300 we thus obtain the joint distribution @xmath301 the marginal distribution of the scattering matrix resulting from the jacobians and is @xmath302 including the normalization constant . this formula combines the known expressions for the poisson kernel . in the altland - zirnbauer ensembles the average scattering matrix differs from @xmath14 , see ref . and app . [ app_saverage ] . ] in the wigner - dyson ensembles @xcite and in the altland - zirnbauer ensembles @xcite . the present analysis confirms that eq . holds without modification in the two symmetry classes d and diii that support majorana zero - modes , depending on the sign of the determinant @xmath303 in class d and the sign of the pfaffian @xmath304 in class diii @xcite . as a check , we can take @xmath20 , when @xmath305 in class d and @xmath306 in class diii . the @xmath307 sign determines the presence or absence of a majorana bound state ( twofold degenerate in class diii ) . since there is only a single element in the ensemble we should have @xmath308 , which is indeed what eq . gives for @xmath20 , @xmath263 . for ballistic coupling the distribution implies that the eigenvalues @xmath126 of @xmath127 have the @xmath90-independent distibution @xcite @xmath128 the unit step function @xmath129 ensures that @xmath130 for all @xmath131 . in the single - channel case @xmath20 we can use the joint distribution to immediately extend this result to arbitrary tunnel probability @xmath132 . the scalar @xmath1 is pinned to @xmath133 , hence @xmath134 [ this definition of @xmath135 corresponds to @xmath136 from eq . . ] since @xmath135 then appears only as a scale factor , we conclude that the single eigenvalue @xmath137 of @xmath127 for @xmath20 and any @xmath138 $ ] has distribution @xmath139 the single - parameter scaling @xmath140 is tested numerically in fig . [ fig_singlechannel ] , by drawing random hamiltonians from the gaussian class - d ensemble . the excellent agreement serves as a check on our analytics . for a single - channel chaotic scatterer , without ( @xmath141 ) or with ( @xmath142 ) a majorana bound state . the histograms are numerical results obtained by generating random hamiltonians ( of size @xmath143 ) with distribution . the scattering matrix , and hence the delay time , then follows from eqs . and . the different curves correspond to different transmission probability @xmath18 of the tunnel barrier . rescaling with a factor @xmath144 , with @xmath135 defined in eq . , makes all histograms collapse onto a single curve , in agreement with the analytical result . ] with @xmath145 the distribution gives the scaling form of the density of states distribution @xmath22 from the introduction , in the tunneling regime @xmath21 when @xmath146 . for ballistic coupling , integration of @xmath147 with distribution gives the average density of states at the fermi level @xcite , @xmath148 for @xmath149 . the ensemble average diverges for @xmath150 . to calculate the effect of a tunnel barrier we write @xmath151 , and then use the fact that @xmath124 is independent of @xmath1 and hence independent of @xmath152 . the average of @xmath124 with distribution @xmath153 is proportional to the unit matrix , @xmath154 so the average density of states ( still for @xmath149 ) is given by @xmath155\right).\label{rhoaverage}\end{aligned}\ ] ] ( this is eq . from the introduction . ) it remains to calculate the average of @xmath1 with the poisson kernel . in the wigner - dyson symmetry classes this average is just @xmath14 , but as pointed out in ref . this no longer holds in the altland - zirnbauer class d. a simple result for @xmath26 is possible for mode - independent tunnel probabilities , @xmath156 for all @xmath60 , see app . [ app_saverage ] : @xmath157 where again the @xmath158 sign corresponds to @xmath141 ( without a majorana bound state ) and the @xmath159 sign to @xmath142 ( with a majorana bound state ) . we thus arrive at the average density of states , @xmath160\right),\label{rhoaverageresult}\ ] ] plotted in fig . [ fig_rhoaverage ] . in the ballistic limit @xmath161 the dependence on the majorana bound state drops out , while in the tunneling limit @xmath162 we obtain @xmath163 the @xmath164 divergence of the density of states for @xmath142 corresponds to the delta - function contribution from the majorana bound state in the closed system . without the majorana bound state ( @xmath141 ) the density of states at the fermi level remains finite in the @xmath162 limit , but it does remain above the normal - state value of @xmath165 . this midgap spectral peak is characteristic for a class - d superconductor @xcite . while in a closed system the peak is simply a factor of two , in the weakly coupled open system it is a larger factor @xmath166 , which only tends to 2 in the large-@xmath43 limit . the fact that the ensemble of open systems does not reduce to an ensemble of closed systems in the limit @xmath162 is due to statistical fluctuations that remain important for small @xmath43 . ) or in the presence ( @xmath142 ) of a majorana bound state . the curves are calculated from the analytical expression . ] we consider the thermal conductance in the simplest case @xmath167 of a quantum dot with single - mode point contacts . these are majorana modes , carrying heat but no charge . the scattering matrix @xmath168 is parameterized by @xmath169 the haar measure equals @xmath170 the same for both @xmath171 and @xmath172 . for the tunnel barrier we take the reflection matrix @xmath173 , with @xmath174 . the poisson kernel then has the explicit form @xmath175 the dimensionless thermal conductance has distribution @xmath176 where as before , @xmath177 applies to @xmath141 and @xmath178 to @xmath142 . the distribution @xmath179 is the result @xcite for ballistic coupling ( @xmath98 , @xmath180 ) . for identical tunnel barriers , @xmath181 , this reduces to @xmath182 quite remarkably , the distribution of the thermal conductance for two identical single - mode point contacts is unaffected by the presence of a tunnel barrier in the topologically nontrivial case . [ fig_thermalcond ] is a numerical check of this analytical result . , see eq . , for a chaotic scatterer having two single - mode point contacts with identical tunnel probabilities @xmath18 . the data points are numerical results for a random hamiltonian ( @xmath183 ) , the curves are the analytical result . in the presence of a majorana zero - mode ( @xmath142 ) the distribution is independent of @xmath18 . ] notice that the distribution becomes independent of @xmath90 if @xmath184 . this is a special case of a more general result , valid for any @xmath185 : @xmath186 in words : the probability distribution of the thermal conductance becomes independent of the presence or absence of a majorana bound state if the quantum dot is coupled ballistically to at least one of the scattering channels . in other words , _ ballistic coupling to a propagating majorana mode hides the majorana bound state . _ the proof is straightforward : if @xmath187 for one of the indices @xmath188 , then the poisson kernel is unchanged if we multiply @xmath189 , with @xmath190 . [ the orthogonal matrix @xmath191 is defined in eq . . ] the haar measure remains unchanged as well , and so does the thermal conductance . since @xmath192 , so @xmath171 is mapped onto @xmath172 , we conclude that @xmath193 . this proof for the poisson kernel extends to the entire joint distribution of @xmath1 and @xmath3 : the transformation @xmath189 has no effect on the matrices @xmath3 and @xmath152 , so @xmath0 remains unchanged . it follows that the probability distribution of the density of states is the same with or without a majorana bound state if @xmath98 for at least one of the scattering channels . because a majorana mode is charge - neutral , no electrical current can be driven for @xmath167 . a nonzero current @xmath75 is possible for @xmath194 , when terminal 1 biased at voltage @xmath74 has a distinct electron and hole mode . we investigate the effect of a tunnel barrier for @xmath194 , @xmath195 . in the majorana basis the expression for the andreev conductance reads @xmath196 with @xmath197 a @xmath198 real matrix . the scattering matrix @xmath199 can be conveniently parameterized using three euler angles @xcite , @xmath200 where we have defined @xmath201 the haar measure on @xmath202 is given by @xcite @xmath203 because @xmath204 commutes with @xmath205 , the dimensionless conductance depends only on the euler angle @xmath206 , @xmath207 in point contact 1 we take a tunnel probability @xmath208 , the same for the electron and hole mode , while in point contact 2 we have tunnel probability @xmath209 for the unpaired majorana mode . evaluation of the poisson kernel with @xmath210 gives the conductance distribution @xmath211}{[(1\pm r_2(g_{\rm a}-1))^2-(g_{\rm a}-1\pm r_2)^2(1-\gamma_1)]^{3/2}},\;\;0<g_{\rm a}<2,\;\;r_2=\sqrt{1-\gamma_2},\label{ppmg1}\ ] ] plotted in fig . [ fig_electrcond ] for @xmath141 ( @xmath177 ) , @xmath142 ( @xmath178 ) and two values of @xmath212 . , , for @xmath194 , @xmath195 , @xmath213 . the data points are numerical results for a random hamiltonian ( @xmath183 ) , the curves are the analytical result . for ballistic coupling ( @xmath33 ) the distribution is the same with or without a majorana bound state . in the limit @xmath162 the distribution becomes sharply peaked at @xmath214 . ] in the limit @xmath215 , when terminal 2 is decoupled from the quantum dot , we recover the result @xcite @xmath216 independent of @xmath208 . the conductance in this case is uniquely determined by the topological quantum number . in the opposite limit @xmath217 the distribution becomes independent of @xmath90 , @xmath218}{[1-(1-\gamma_1)(g_{\rm a}-1)^2]^{3/2}},\;\;{\rm if}\;\;r_2=0.\label{ppmgr20}\ ] ] this is a special case of a more general result , for any @xmath185 , @xmath219 where @xmath220 projects onto the modes coupled to terminal 2 . ballistic coupling , even for a single mode , to terminal 2 therefore removes the dependence on the topological quantum number . the proof of eq . proceeds along the lines of the proof of eq . , with the difference that the transformation @xmath189 should leave the upper - left block @xmath197 of @xmath1 unaffected otherwise the andreev conductance would change . this also explains why the @xmath90-dependence of @xmath34 remains for ballistic coupling to terminal 1 @xcite . , while the other is coupled to electrode @xmath221 via a point contact . ] the appearance of a majorana bound state is a topological phase transition . there is a search for this transition in a nanowire geometry , see fig . [ fig_nanowire ] , where it has been predicted to occur when the zeeman energy of a magnetic field ( parallel to the wire axis ) exceeds the gap induced by the proximity to a superconductor @xcite . because the majorana bound states emerge pairwise at the two ends of the nanowire , the topological quantum number @xmath90 of the entire structure remains @xmath222 and the determinant @xmath223 of the full scattering matrix remains @xmath7 through the transition . what changes is the sign of the determinant @xmath224 of the reflection submatrix . at the topological phase transition @xmath225 , implying a perfectly transmitted mode and a quantized peak in the thermal conductance @xcite . we study the effect of the phase transition on the statistics of the electrical conductance , measured by contacting one end of the nanowire ( terminal 1 ) to a metal electrode at voltage @xmath74 , while the superconductor and the other end of the nanowire ( terminal 2 ) are at ground . terminal 1 is connected to the nanowire via a point contact , thus creating a confined region ( quantum dot ) with chaotic scattering . the minimal dimensionality of the scattering matrix @xmath1 of the quantum dot is @xmath226 : one electron and one hole mode connected to terminal 1 and one majorana mode connected to terminal 2 . to minimize the number of free parameters we assume ballistic coupling through the point contact , so the matrix @xmath14 in the poisson kernel is @xmath227 . the reflection amplitude @xmath228 at terminal 2 is tuned through zero by some external control parameter @xmath229 , typically magnetic field or gate voltage . near the transition ( conveniently shifted to @xmath230 ) this dependence can be conveniently parameterized by @xcite @xmath231 ( the width @xmath232 of the transition is system dependent . ) the corresponding coupling constant @xmath233 in eq . then has an exponential @xmath229-dependence , @xmath234 the probability distribution of the andreev conductance ( in units of @xmath235 ) follows from eq . , @xmath236^{-2}\nonumber\\ & = \tfrac{1}{2}[\cosh(\xi/\xi_0)+(g_{\rm a}-1)\sinh(\xi/\xi_0)]^{-2},\nonumber\\ & \qquad\qquad\;\;0<g_{\rm a}<2.\label{ppmg2}\end{aligned}\ ] ] the delta - function limits are reached for @xmath237 ( keeping @xmath141 , because the entire system is topologically trivial ) . right at the transition , at @xmath238 , the distribution is uniform in the interval @xmath239 . the average conductance varies through the transition as @xmath240 see fig . [ fig_gatransition ] . , as the nanowire is driven through a topological phase transition ( controlled by a parameter @xmath229 , which can be thought of as the deviation of the magnetic field from the critical field strength ) . the curve is the result for a single - channel ballistic point contact . ] in conclusion , we have investigated a variety of observable consequences of the fact that the scattering matrix of majorana fermions is real orthogonal rather than complex unitary . of particular interest is the identification of observables that can detect the sign of the determinant , since @xmath10 signifies the presence of a majorana bound state . the obvious signal of such a zero - mode , a midgap peak in the density of states @xcite , is broadened by tunnel coupling to the continuum . we find that the peak remains hidden if the coupling is ballistic ( unit transmission ) in even a single scattering channel . the thermal conductance is likewise insensitive to the presence or absence of a majorana bound state , but the electrical conductance retains this sensitivity when the coupling is ballistic . these results for the effect of a tunnel barrier on the midgap spectral peak are derived from the distribution @xmath0 of scattering matrix and time - delay matrix under the assumption of chaotic scattering , due to disorder or due to irregularly shaped boundaries . the appropriate ensemble in the absence of time - reversal and spin - rotation symmetry has symmetry class d in the altland - zirnbauer classification @xcite . chiral symmetry would change this to class bdi , in which multiple zero - modes can overlap without splitting @xcite . the effect of chiral symmetry on the joint distribution @xmath0 is known for ballistic coupling @xcite but not yet for tunnel coupling . this seems a worthwhile project for future research . this research was supported by the foundation for fundamental research on matter ( fom ) , the netherlands organization for scientific research ( nwo / ocw ) , and an erc synergy grant . according to eq . , the effect of a tunnel barrier on the average density of states follows directly once we know the average scattering matrix . simple expressions can be obtained if we assume that the tunnel probabilities are mode - independent , @xmath156 for @xmath131 . the scattering matrix of the barrier has the polar decomposition @xmath309 with @xmath310 . the block structure corresponds to eqs . and , in particular , @xmath311 . from eq . we obtain the scattering matrix @xmath1 of the quantum dot , @xmath312o_2,\label{stildes0relation}\ ] ] in terms of a matrix @xmath313 that is uniformly distributed in @xmath99 . because the average of @xmath314 for any power @xmath315 is proportional to the identity matrix , we may write the average of @xmath1 in the form of a power series , @xmath316 if the average of @xmath283 would be over the entire unitary group , then all terms in the power series would vanish and we would simply have @xmath317 . but averages over orthogonal matrices do not vanish , in the nontrivial way calculated instead of @xmath318 ) . ] by rains @xcite : @xmath319 we substitute eq . into eq . and sum the geometric series , to arrive at the average scattering matrix @xmath320 used in sec . [ sec_dos ] to obtain the average density of states in class d. + + + + in a similar way we can derive the average scattering matrix in other symmetry classes . we give the result for class c. the matrix @xmath283 then varies over the unitary symplectic group @xmath321 . ref . gives the required average : @xmath322 eq . still holds with the factor @xmath323 replaced by @xmath324 . we thus find the average scattering matrix in class c , @xmath325\right).\label{saverageuc}\ ] ] for @xmath20 this gives @xmath326 , in agreement with ref . . the average density of states in class c follows from eq . , where we again account for the doubling of the dimensionality @xmath327 : @xmath328\right)\nonumber\\ & = \frac{n}{(n+1)\delta_0}\,\left(1-\frac{1-\gamma}{n\gamma}[1-(1-\gamma)^n]\right).\label{rhoaveragec}\end{aligned}\ ] ] in the second equation we have substituted the ballistic class - c result from ref . . the tunneling limit @xmath162 gives a vanishing density of states , @xmath329 consistent with the class - c result for a closed system @xcite . 24 a. altland and m. r. zirnbauer , phys . b * 55 * , 1142 ( 1997 ) . j. alicea , rep . . phys . * 75 * , 076501 ( 2012 ) . m. leijnse and k. flensberg , semicond . * 27 * , 124003 ( 2012 ) . t. d. stanescu and s. tewari , j. phys . matt . * 25 * , 233201 ( 2013 ) . c. w. j. beenakker , annu . con . mat . * , 113 ( 2013 ) . s. das sarma , m. freedman , and c. nayak , arxiv:1501.02813 ( to be published in npj quantum inf . ) . e. p. wigner , siam review * 9 * , 1 ( 1967 ) . m. l. mehta , _ random matrices _ ( elsevier , amsterdam , 2004 ) . p. j. forrester , _ log - gases and random matrices _ ( princeton university press , 2010 ) . f. j. dyson , j. math . * 3 * , 140 ( 1962 ) . c. w. j. beenakker , arxiv:1407.2131 ( to be published in rev . mod . phys . ) . t. j. krieger , ann . * 42 * , 375 ( 1967 ) . p. a. mello , p. pereyra and t. seligman , ann . phys . * 161 * , 254 ( 1985 ) . w. a. friedman and p. a. mello , ann . * 161 * , 276 ( 1985 ) . p. w. brouwer , phys . b * 51 * , 16878 ( 1995 ) . b. bri , phys . rev . b * 79 * , 214506 ( 2009 ) . v. a. gopar , p. a. mello , and m. bttiker , phys . * 77 * , 4974 ( 1996 ) . p. w. brouwer and m. bttiker , europhys . lett . * 37 * , 441 ( 1997 ) . m. bttiker , in _ time in quantum mechanics _ , edited by j. g. muga , r. sala mayato , and i. l. egusquiza ( springer , berlin , 2002 ) . p. w. brouwer , s. a. van langen , k. m. frahm , m. bttiker , and c. w. j. beenakker , phys . lett . * 79 * , 913 ( 1997 ) . s. f. godijn , s. mller , h. buhmann , l. w. molenkamp , and s. a. van langen , phys . lett . * 82 * , 2927 ( 1999 ) . p. w. brouwer , k. m. frahm , and c. w. j. beenakker , phys . 78 , 4737 ( 1997 ) . m. marciani , p. w. brouwer , and c. w. j. beenakker , phys . b * 90 * , 045403 ( 2014 ) . h. schomerus , m. marciani , and c. w. j. beenakker , phys . lett . * 114 * , 166803 ( 2015 ) . d. v. savin , y. v. fyodorov , and h .- j . sommers , phys . e * 63 * , 035202 ( 2001 ) . sommers , d. v. savin , and v. v. sokolov , phys . . lett . * 87 * , 094101 ( 2001 ) . j. p. dahlhaus , b. bri , and c. w. j. beenakker , phys . b * 82 * , 014536 ( 2010 ) . c. w. j. beenakker , j. p. dahlhaus , m. wimmer , and a. r. akhmerov , phys . rev . b * 83 * , 085413 ( 2011 ) . c. mahaux and h. a. weidenmller , phys . 170 * , 847 ( 1968 ) ; _ shell - model approach to nuclear reactions _ ( north - holland , amsterdam , 1969 ) . l. fu and c. l. kane , phys . * 100 * , 096407 ( 2008 ) . m. titov , a. ossipov , and c. w. j. beenakker , phys . rev . b * 75 * , 045417 ( 2007 ) . j. j. m. verbaarschot , h. a. weidenmller , and m. r. zirnbauer , phys . rep . * 129 * , 367 ( 1985 ) . a. r. akhmerov , j. p. dahlhaus , f. hassler , m. wimmer , and c. w. j. beenakker , phys . . lett . * 106 * , 057001 ( 2011 ) . e. p. wigner , phys . rev . * 98 * , 145 ( 1955 ) . f. t. smith , phys . rev . * 118 * , 349 ( 1960 ) . y. v. fyodorov and h .- j . sommers , j. math . phys . * 38 * , 1918 ( 1997 ) . m. sh . birman and m. g. krein , soviet math . dokl . * 3 * , 740 ( 1962 ) . e. akkermans , a. auerbach , j. e. avron , and b. shapiro , phys . 66 * , 76 ( 1991 ) . n. lehmann , d. v. savin , v. v. sokolov , and h .- j . sommers , physica d * 86 * , 572 ( 1995 ) . a. f. andreev , sov . jetp * 19 * , 1228 ( 1964 ) . k. schwab , e. a. henriksen , j. m. worlock , and m. l. roukes , nature * 404 * , 974 ( 2000 ) . m. bocquet , d. serban , and m. r. zirnbauer , nucl . b * 578 * , 628 ( 2000 ) . d. a. ivanov , j. math . phys . * 43 * , 126 ( 2002 ) ; arxiv : cond - mat/0103089 . a. m. mathai , _ jacobians of matrix transformations and functions of matrix argument _ ( world scientific publishing , 1997 ) . d. bagrets and a. altland , phys . lett . * 109 * , 227005 ( 2012 ) . j. li , g. fleury , and m. bttiker , phys . b * 85 * , 125440 ( 2012 ) . b. bri , phys . b * 79 * , 245315 ( 2009 ) . r. m. lutchyn , j. d. sau , and s. das sarma , phys . lett . * 105 * , 077001 ( 2010 ) . y. oreg , g. refael , and f. von oppen , phys . 105 , 177002 ( 2010 ) . f. j. dyson , j. math . phys . * 3 * , 1199 ( 1962 ) . j. j. m. verbaarschot and t. wettig , ann . nucl . part . 50 * , 343 ( 2000 ) . i. c. fulga , f. hassler , a. r. akhmerov , and c. w. j. beenakker , phys . b * 83 * , 155429 ( 2011 ) . e. m. rains , probab . theory relat . fields * 107 * , 219 ( 1997 ) .

linear positive systems @xcite have been recently the subject of an increasing attention because of their natural ability to represent many real - world processes such as , among others , communication networks @xcite , biological networks @xcite , epidemiological networks @xcite , disease dynamics @xcite , etc . besides their applicative potential , linear positive systems have been shown to exhibit a number of interesting theoretical properties of independent interest . for instance , it is now well - known that linear copositive lyapunov functions can be used in order to formulate exact stability conditions taking the form of linear programs @xcite . the design of structured and bounded state - feedback controllers @xcite and certain classes of static output feedback controllers @xcite are known to be convex and hence easily tractable . the @xmath0-gains for @xmath1 can be exactly computed using convex programming and these gains are identical to the @xmath2-norm of the static matrix - gain of the system @xcite . the famous kalman - yakubovich - popov lemma has been shown to admit a linear formulation in this setting @xcite . robust analysis results also nicely extends and simplifies in this context , and often becomes necessary and sufficient criteria for stability @xcite . their generalization to delay - systems with discrete - delays also led to the surprise that the system is stable if and only if the system with zero delay is stable @xcite . design - wise , it has been shown in @xcite that the state - feedback controller that minimizes the @xmath3-gain of the closed - loop system is independent of the matrices driving the disturbance input to the state and the output of the system . symmetrically , it is shown in @xcite that the observer - gain that minimizes the @xmath4-gain of map between the disturbance input and the observation error is independent of the output matrices of the error system . extensions to deterministically @xcite or stochastically @xcite switched systems have also been considered . we consider here the case of linear positive impulsive systems , a class of systems that seems to have been quite overlooked until now as only very few results can be found ; see e.g. @xcite . such systems can be used to represent certain classes biochemical , population or epidemiological models having deterministic jumps in their dynamics . they can also be used to represent processes that can be represented as linear positive switched systems ; see e.g. @xcite for some examples including epidemiology @xcite , traffic congestion models @xcite , etc . impulsive systems are also well known to be able to exactly represent sampled - data systems as emphasized in @xcite . the goal of this paper is hence the derivation of novel stability and stabilization conditions for linear positive impulsive systems using the concepts of arbitrary , constant , minimum , maximum and range dwell - times . the concept of _ minimum dwell - time _ has been introduced by morse in @xcite in order to formulate stability conditions for general ( i.e. not necessarily positive ) switched systems . the concept of _ average dwell - time _ has been proposed in @xcite in order to obtained less conservative conditions that by using minimum dwell - time conditions . since then , a large body of the literature has been focusing on these concepts as a way to efficiently characterize the stability of switched systems or , more generally , the stability of hybrid systems ; see e.g. @xcite . the notion of minimum dwell - time has been revisited in @xcite where novel sufficient lmi conditions derived from mode - dependent quadratic lyapunov functions were proposed . based on a theoretical result proved in @xcite , these conditions were later extended and made necessary and sufficient in @xcite through the consideration of mode - dependent homogeneous lyapunov functions . unfortunately , these conditions were inapplicable to uncertain systems and to control design because of their complex nonlinear dependency in the matrices of the system . this problem motivated the introduction of the so - called _ looped - functionals _ , a particular class of indefinite ( i.e. not necessarily positive definite ) functionals having the advantage of reformulating the complex conditions of @xcite into conditions being linear in the matrices of the system ; see e.g. @xcite , thereby extending the scope of the conditions to uncertain and nonlinear systems . yet , these conditions were difficult to apply in the context of control design because of the presence of multiple products between decision matrices and the matrices of the system ; see e.g. @xcite . _ clock - dependent conditions _ have been shown to provide an essential framework for solving this latter problem as they produce stability conditions that are linear / convex in the matrices of the system and can be used for design purposes . their computational complexity has also been shown to be much lower than that of looped - functionals @xcite . since then , clock - dependent conditions have been used for the analysis and control of switched , impulsive , sampled - data and lpv systems ; see e.g. @xcite . the first part of the paper is similar to the ones in @xcite where stability conditions are formulated in terms of the decrease of a lyapunov function of a given type . unlike in the previous references where quadratic lyapunov functions are involved , we exploit here the positivity of the system and consider a linear copositive lyapunov function @xcite . the resulting conditions are stated in terms of finite - dimensional linear programs or semi - infinite dimensional linear programs , which are then relaxed into clock - dependent conditions using a lifting approach similar to that of @xcite . since linear copositive lyapunov functions are used here , the clock - dependent conditions consist of infinite - dimensional linear programs . this has to be contrasted with the fact that when quadratic lyapunov functions are used , clock - dependent conditions take the form of infinite - dimensional semidefinite programs , which may be harder to solve that their linear counterpart . three possible ways for efficiently checking these conditions are then proposed . the first one relies on a discretization approach which is largely inspired from @xcite and where the infinite - dimensional decision variable is assumed to be continuous and piecewise linear . this method has also been considered , in turn , in @xcite . by doing so , the infinite - dimensional program becomes finite - dimensional and can be solved using conventional algorithms such as interior point methods ; see e.g. @xcite . the second method is based on handelman s theorem @xcite which characterizes the positivity of a given polynomial on a compact polytope by formulating it as a nonnegative linear combination of products of the ( affine ) basis functions that describe the polytope . this result has been applied in various contexts @xcite and , notably , for characterizing the robust stability of uncertain linear positive systems in @xcite . an important property of this approach is that the obtained characterization for the positivity of the polynomial can be exactly formulated as a finite - dimensional linear program , which can again be solved using well - known approaches . finally , the last one is based on putinar s positivstellensatz @xcite which characterizes the positivity of a given polynomial on a compact semialgebraic set by formulating it as a weighted linear combination of the basis functions that describe the set and where the weights are sum of squares polynomials . the resulting problem takes , in this case , the form of a finite - dimensional semidefinite program @xcite that can be solved using standard semidefinite programming solvers such as sedumi @xcite or sdpt3 @xcite used in conjunction with the package sostools @xcite . it is notably emphasized that these relaxations are asymptotically exact meaning that when the discretization order , the number of products of basis functions or the degree of the sum of squares weights are sufficiently large , then the relaxed problem is feasible if the original one is . several examples are considered in order to demonstrate the practicality of the relaxed conditions and to compare them in terms of number of variables and solving time . the results are then extended to control design by considering the clock - dependent conditions and the dual impulsive system @xcite . by finally exploiting the possibility of formulating a switched system as an impulsive system , we derive a number of stability conditions for linear positive switched systems . notably , we recover stability conditions similar to those in @xcite which are the positive counterpart of those obtained in @xcite in the context of general linear switched systems while some other ones seem to be new . novel stabilization conditions under arbitrary , minimum and mode - dependent range dwell - time conditions are then obtained and illustrated through few examples . + * outline . * the structure of the paper is as follows : in section [ sec : preliminary ] preliminary definitions and results are given . section [ sec : stab ] is devoted to the dwell - time stability analysis of linear positive impulsive systems while section [ sec : stabz ] addresses their stabilization . these results are then applied to switched systems in section [ sec : switched ] . + * notations . * the cone of positive ( nonnegative ) vectors of dimension @xmath5 are denoted by @xmath6 ( @xmath7 ) . the set of diagonal matrices of dimension @xmath5 is denoted by @xmath8 and that of diagonal matrices with positive diagonal elements by @xmath9 . for a set of matrices @xmath10 of compatible dimensions , we define @xmath11 to be the block diagonal matrix with the @xmath12 s as diagonal blocks and by @xmath13 the matrix consisting of the @xmath12 s stacked in column . the @xmath5-dimensional vector of ones is denoted by @xmath14 . let us consider here the following class of linear impulsive system : @xmath15 where @xmath16 are the state of the system and its initial condition , respectively . the notation @xmath17 is a shorthand for @xmath18 , i.e. the trajectories are assumed to be left - continuous . the sequence of impulse instants @xmath19 is assumed to be strictly increasing and to grow unboundedly . the sequence hence does not admit any accumulation point , thereby excluding any zeno behavior for the system trajectories . note that this assumption is not restrictive for the current paper as we only focus here on dwell - time results . we define the dwell - times as the values @xmath20 , @xmath21 . the following result states necessary and sufficient conditions for the impulsive system to be positive : the following statements are equivalent : a. the system is positive ; i.e. for any @xmath22 , we have that @xmath23 for all @xmath24 . b. the matrix @xmath25 is metzler ( all its off - diagonal entries are nonnegative ) and @xmath26 is nonnegative . the following result establishes conditions for which a metzler matrix is hurwitz stable : [ prop : hurwitz ] let @xmath27 be a metzler matrix . then , the following statements are equivalent : a. @xmath25 is hurwitz stable ; b. @xmath25 is nonsingular and @xmath28 ; c. there exists a @xmath29 such that @xmath30 ; d. there exists a @xmath31 such that @xmath32 ; similarly , the following result establishes conditions for which a nonnegative matrix is schur stable : [ prop : schur ] let @xmath33 be a nonnegative matrix . then , the following statements are equivalent : a. @xmath34 is schur stable ; b. @xmath35 is hurwiz stable ; c. there exists a @xmath29 such that @xmath36 ; d. there exists a @xmath31 such that @xmath37 ; the following generic stability result will be important in the sequel : [ th : equivalence ] let the sequence @xmath38 for the system be given and assume that it satisfies the strict increase and unboundedness conditions . then , the following statements are equivalent : a. [ item : equi : a ] the impulsive system is asymptotically stable . b. [ item : equi : a2 ] the dual impulsive system @xmath39 is asymptotically stable . c. [ item : equi : b ] the discrete - time system @xmath40 is asymptotically stable . d. [ item : equi : c ] the discrete - time system @xmath41 is asymptotically stable . e. [ item : equi : d ] the discrete - time system @xmath42 is asymptotically stable . f. [ item : equi : e ] the discrete - time system @xmath43 is asymptotically stable . * proof of @xmath44 . * immediate . noindent**proof of @xmath44 . * * to prove this implication , we need to consider two cases . the first one is the case where the impulses are not persistent , i.e. there exists a @xmath45 such that @xmath46 for all @xmath47 and @xmath48 . in this case , the system is asymptotically stable if and only if @xmath25 is hurwitz stable . if @xmath48 , then @xmath49 for all @xmath50 . assuming that @xmath51 , this is equivalent to saying that @xmath25 is hurwitz stable . let us now consider the case where the jumps are persistent , i.e. for any sequence @xmath38 we choose , there exists a @xmath52 such that @xmath53 for all @xmath54 . note first that independently of the chosen sequence , we have that @xmath55 for some @xmath56 and where the norm is any vector norm . this then implies that @xmath57 since the embedded discrete - time system is asymptotically stable , then @xmath58 as @xmath59 . this then implies that @xmath60}||x(t_k+s)||\to0\ \textnormal{as}\ k\to\infty\ ] ] which proves the asymptotic stability of the system in the persistent jump case . the proof is complete . * proof of @xmath61 . * note that while the system describes the evolution of the sequence @xmath62 , the system describes the evolution of the sequence @xmath63 ; i.e. @xmath64 . using the fact that @xmath65 , then we get that @xmath66 hence , clearly , we have that @xmath67 as @xmath59 if and only if @xmath68 as @xmath59 . the proof is complete . * proof of @xmath61 . * let assume first that the sequence @xmath38 is such that @xmath69 $ ] for some @xmath70 and define the following set of matrices @xmath71\right\}.\ ] ] clearly , this set is bounded since the interval @xmath72 $ ] is compact . we know that the system is asymptotically stable if and only if the joint spectral radius condition @xmath73 holds ; see e.g. @xcite . it is known that we have @xmath74 where the latter mean that we consider transposes of the matrices in @xmath75 ; see e.g. @xcite . then the conclusion follows . assuming now that we have no upper - bound for the dwell - time , we can assume that the matrix @xmath25 is hurwitz stable ( see the beginning of the proof that ( b ) implies ( a ) ) , which then implies that @xmath76 is compact and the same arguments as before apply . the proof is complete . * proof of @xmath61 . * this follows from the same arguments . * proof of @xmath61 . * this follows from the same arguments as for the proof that @xmath61 . the meaning of the above result is that we can choose the most convenient system to work with in order to derive stability and stabilization conditions for the original impulsive system . this result is also important as it will be used to demonstrate that despite the systems can be all equivalent in terms of stability , the obtained stability conditions will not be necessarily equivalent . this point is discussed for the case of positive switched systems in @xcite and the same discussion remains valid in the case of linear positive impulsive systems . notably , the difference between the results obtained using the system and the swapped " version will be emphasized in section [ sec : imp : examples ] and in section [ sec : sw : examples ] . the interest for using the dual system and its swapped " version will be emphasized in the sections related to stabilization ; i.e. section [ sec : stabz ] and section [ sec : switched ] . note also that when the sequence @xmath38 is such that @xmath77 for all @xmath78 , @xmath79 , then all the discrete - time systems in theorem [ th : equivalence ] become time - invariant after time @xmath80 . hence , the stability conditions reduce to standard stability conditions for linear time - invariant discrete - time systems and will all be equivalent with each other . this result can be easily be understood through the fact that @xmath81 , where @xmath82 denotes the spectral radius . we derive here several stability results for linear positive impulsive systems . the case of arbitrary dwell - time is considered first in section [ sec : imp : arbitrary ] and is followed by stability results under constant dwell - time ( @xmath83 ) , minimum ( @xmath84 ) , maximum ( @xmath85 ) and range dwell - time ( @xmath86 $ ] ) in section [ sec : imp : constant ] , section [ sec : imp : minimum ] , section [ sec : imp : maximum ] and section [ sec : imp : range ] , respectively . computational results are given in section [ sec : imp : computation ] together with some discussions regarding their conservatism . comparative examples are finally given in section [ sec : imp : examples ] . let us consider first the stability under arbitrary dwell - time ( @xmath87 ) , with the additional condition that the sequence @xmath88 grows unboundedly . we then have the following result : [ th : arbdt ] the following statements are equivalent : a. there exists a @xmath89 such that @xmath90 b. we have that @xmath91 moreover , when one of the above statements holds , then the system is asymptotically stable under arbitrary dwell - time ; i.e. @xmath87 . the first condition in is equivalent to saying that @xmath92 for some @xmath93 whereas the second one is equivalent to saying that @xmath94 for some @xmath95 and all @xmath54 . these conditions , all together , imply that @xmath96 as @xmath97 regardless the impulse sequence @xmath38 . to prove the equivalence with ( ii ) , it is enough to remark that the conditions in statement ( i ) are identical to stability conditions for positive switched systems using a common copositive linear lyapunov functions @xcite . the result follows from the latter paper . [ rem : dual_arbdt][rem : arbdt2 ] dual conditions can be obtained by substituting the matrices of the dual system inside the conditions of theorem [ th : arbdt ] . in such a case , the conditions of statement ( a ) become @xmath98 whereas that of statement ( b ) become @xmath99 [ rem : arbdt1_flow ] in the case of persistent flowing , i.e. the flow will never stops , the conditions in theorem [ th : arbdt ] , ( a ) , can be relaxed to @xmath100 and those in remark [ rem : arbdt2 ] , ( a ) , to @xmath101 the example below illustrates the discussion below theorem [ th : equivalence ] about the non - equivalence between the stability conditions obtained from the impulsive system and its dual : let us consider the system with matrices @xmath102 there is no vector @xmath29 such that the conditions hold since @xmath30 implies that @xmath103 while @xmath104 implies that @xmath105 , yielding then a contradiction . on the other hand , we can readily see that @xmath106 solves the conditions , emphasizing then the discrepancy between the conditions of theorem [ th : arbdt ] and remark [ rem : arbdt2 ] . we consider now the case of constant dwell - times the case where @xmath107 , for some @xmath108 and for all @xmath21 or , in other words , the case where jumps occur periodically . we have the following result : [ th : cstdt2 ] the following statements are equivalent : a. [ item : th1:0 ] the impulsive system is asymptotically stable under constant dwell - time @xmath109 . b. [ item : th1:1 ] the copositive linear form @xmath110 , @xmath29 is a discrete - time lyapunov function for the @xmath109-periodic impulsive system in the sense that the inequality @xmath111 holds for some @xmath93 , all @xmath112 and all @xmath21 . c. [ item : th1:2 ] there exists a vector @xmath89 such that the inequality @xmath113<0\ ] ] holds or , equivalently , the matrix @xmath114 is schur stable . d. [ item : th1:3 ] there exists a vector @xmath89 such that the inequality @xmath115\lambda<0\ ] ] holds or , equivalently , the matrix @xmath116 is schur stable . e. [ item : th1:4 ] there exist a differentiable vector - valued function @xmath117\mapsto\mathbb{r}^n$ ] , @xmath118 , and a scalar @xmath119 such that the inequalities @xmath120 and @xmath121 hold for all @xmath122 $ ] . f. [ item : th1:5 ] there exist a differentiable vector - valued function @xmath123\mapsto\mathbb{r}^n$ ] , @xmath124 , and a scalar @xmath119 such that the lmis @xmath125 and @xmath126 hold for all @xmath122 $ ] . * proof of @xmath127 :* this follows from theorem [ th : equivalence ] and proposition [ prop : schur ] . * proof of @xmath127 :* it is immediate to see that the left - hand side of is given by @xmath128^\t x(t_k).\ ] ] the equivalence between the two statements immediately follows . * proof of @xmath127 :* this follows from proposition [ prop : hurwitz ] or proposition [ prop : schur ] . * proof of @xmath127 :* the proof between the two statements follows from the definition that @xmath129 . alternatively , similar arguments as in the proof that @xmath127 can also be considered . the main advantages of the conditions of statements and in the above results are the following . first of all , the conditions are affine in the matrices @xmath25 and @xmath26 of the system , allowing then for an immediate extension to uncertain matrices and to control design ( the latter requiring some additional steps that will be detailed in section [ sec : stabz ] ) . a particularity of the approach is that we only need the vectors @xmath130 and @xmath131 to be positive while most of the lyapunov approaches would require the whole vector - valued functions @xmath132 and @xmath133 to be positive on their domain ( see e.g. @xcite ) , a constraint that is computationally way more complex . the explanation for that in spite of the fact that the conditions in statements and statement look like continuous - time stability conditions , they are in fact lifted discrete - time stability condition for which only the behavior of the inequalities at the extremal points of the domain of the lifting variable ( i.e. the points @xmath134 and @xmath135 ) are actually meaningful . the same discussion is made in @xcite in the context of looped - functionals and in @xcite in the context of clock - dependent conditions . the main drawback of this procedure lies in the increase of the computational complexity as the conditions of statement and statement are infinite - dimensional linear programming conditions which are in general hard to solve . note , however , that they should be less complex than the infinite - dimensional semidefinite programming conditions obtained in @xcite . verifying these conditions in an efficient way will be the topic of section [ sec : imp : computation ] . [ sec : imp : minimum ] let us consider now the minimum dwell - time case . in this setup , the dwell - times are assumed to satisfy the condition @xmath84 , @xmath21 . even though we do not assume here the existence of an upper - bound for the dwell - times a priori , we will restrict ourselves to the case of persistent impulses ; i.e. for any chosen sequence @xmath136 , there exists an upper - bound for the dwell - times . hence , we exclude the case that @xmath48 for some @xmath137 , which is not restrictive , as in such a case , the system would become a standard continuous - time linear positive system . the following result can be seen as a positive systems " counterpart of the results in @xcite and provides a stability condition in terms of the discrete - time system : [ th : mindt2 ] the following statements are equivalent : a. the linear form @xmath110 , @xmath29 is a lyapunov function for the system in the sense that @xmath138 and @xmath139 hold for some @xmath140 , for all @xmath141 and for any sequence @xmath19 verifying @xmath84 , @xmath21 . b. there exists a @xmath89 such that the inequalities @xmath142 and @xmath143 hold for all @xmath144 . c. there exists @xmath89 such that the conditions @xmath142 and @xmath145 hold . d. the pair @xmath146 admits a common linear copositive lyapunov function . e. we have that @xmath147 f. there exist a matrix function @xmath117\mapsto\mathbb{r}^n$ ] , @xmath118 , and a scalar @xmath119 such that the inequalities @xmath148 @xmath149 and @xmath150 hold for all @xmath122 $ ] . g. there exist a matrix function @xmath123\mapsto\mathbb{r}^n$ ] , @xmath124 , and a scalar @xmath119 such that the inequalities @xmath151 @xmath152 and @xmath153 hold for all @xmath122 $ ] . moreover , when one of the above statements holds , the impulsive system ( [ eq : mainsyst ] ) is asymptotically stable under minimum dwell - time @xmath154.@xmath155 the proof that ( a ) is equivalent to ( b ) follows from the same arguments as in the proof of theorem [ th : mindt2 ] . the proof that ( b ) implies ( c ) is immediate . to prove the converse , we assume that @xmath156 , which implies that @xmath157 where we have used the fact that @xmath158 for all @xmath159 . note also that since @xmath160 , then we have that @xmath161 for all @xmath159 . adding finally the latter inequality combined to implies that @xmath162 holds for all @xmath159 , which proves the implication . the rest of the proof can be carried as for theorem [ th : mindt2 ] . the above results interestingly connect the problem of finding a common linear copositive lyapunov functions for a linear positive switched system with two subsystems to the problem of establishing whether a linear positive impulsive system is stable under some minimum dwell - time constraint . indeed , in the context of theorem [ th : mindt2 ] , the matrix of the first subsystem is @xmath25 and that of the second subsystem is @xmath163 . when the swapped system is considered ( see remark [ rem : swapped_minimum ] ) , the matrix of the first subsystem is also @xmath25 but that of the second subsystem is @xmath164 . in this regard , characterizing the existence of a linear copositive lyapunov function can be achieved using existing methods ; such as those developed in @xcite . however , these conditions are difficult to consider when the matrix @xmath25 is uncertain or when design is the main goal . the statements ( f ) and ( g ) provide alternative equivalent conditions that are affine in @xmath25 and @xmath26 and can then be hence considered for uncertain systems and for design purposes . a last interesting point is that the results for arbitrary dwell - time can be retrieved by letting @xmath165 in the above theorem as also noticed in @xcite . we consider now the maximum dwell - time case . in this setup , the dwell - times are assumed to satisfy a maximum dwell - time condition ; i.e. @xmath85 , @xmath21 . the following result can be understood as being the positive systems " version of the results in @xcite and is the maximum dwell - time counterpart of theorem [ th : mindt2 ] : [ th : maxdt2 ] the following statements are equivalent : a. the linear form @xmath110 , @xmath29 is a lyapunov function for the system in the sense that @xmath166 and @xmath167 hold for some @xmath140 , all @xmath141 and any sequence @xmath168 . b. there exists @xmath89 such that the inequalities @xmath169 and @xmath170 hold for all @xmath144 . c. there exists @xmath89 such that the conditions @xmath169 and @xmath171 hold . d. the pair @xmath172 admits a common linear copositive lyapunov function . e. we have that @xmath173 f. there exist a matrix function @xmath117\mapsto\mathbb{r}^n$ ] , @xmath118 , and a scalar @xmath119 such that the inequalities @xmath174 @xmath149 and @xmath150 hold for all @xmath122 $ ] . g. there exist a matrix function @xmath123\mapsto\mathbb{r}^n$ ] , @xmath124 , and a scalar @xmath119 such that the inequalities @xmath175 @xmath152 and @xmath153 hold for all @xmath122 $ ] . moreover , when one of the above statements holds , the impulsive system ( [ eq : mainsyst ] ) is asymptotically stable under maximum dwell - time @xmath154.@xmath155 the proof follows from the same lines as the proof of theorem [ th : mindt2 ] . [ sec : imp : range ] we address now the range dwell - time case , that is the case when @xmath86 $ ] , @xmath21 , for some prescribed bounds @xmath176 verifying @xmath70 . we then have the following result : [ th : rangedt ] the following statements are equivalent : a. [ item : th3:1 ] the copositive linear form @xmath110 , @xmath29 is a discrete - time lyapunov function for the impulsive system in the sense that the inequality @xmath177 holds for some @xmath93 , all @xmath112 , @xmath86 $ ] and all @xmath21 . b. [ item : th3:2 ] there exists a vector @xmath89 such that the inequality @xmath178<0\ ] ] holds for all @xmath179 $ ] . c. [ item : th3:4 ] there exist a differentiable vector - valued function @xmath117\mapsto\mathbb{r}^n$ ] , @xmath180 , and a scalar @xmath119 such that the inequalities @xmath181 and @xmath182 hold for all @xmath183 $ ] and @xmath179 $ ] . it is readily seen that the lyapunov condition is equivalent to the inequality . the proof of the equivalence between the statements ( b ) and ( c ) follows from exactly the same arguments as in the proof of theorem [ th : cstdt2 ] . it is thus omitted . it seems important to carefully address the problem of verifying the infinite - dimensional conditions of the previous results . we detail here three methods for accurately and efficiently verifying them . only the relaxation of the conditions in theorem [ th : cstdt2 ] , are considered here for simplicity . however , relaxed finite - dimensional results for the other conditions can be easily obtained using the same procedure . this approach is based on the piecewise linear approximation of the function @xmath184 or @xmath185 in the results . this method has been initially proposed in @xcite in order to relax infinite - dimensional lmi conditions arising in the analysis of switched systems into finite - dimensional lmi conditions . below is the approximation of the conditions in theorem [ th : cstdt2 ] , statement : [ prop : discretization ] the following statements are equivalent : a. for some @xmath186 , the function @xmath187,\ i=0,\ldots , d_g-1\ ] ] verifies the conditions of theorem [ th : cstdt2 ] , statement . b. there exist vectors @xmath188 , @xmath189 , such that the following conditions @xmath190 @xmath191 and @xmath192 hold for all @xmath193 . the proof follows from direct substitutions . this approach is based on putinar s positivstellensatz @xcite and is solved using semidefinite programming methods , as initially proposed in @xcite . before stating the main result , we need first to define some terminology . a multivariate polynomial @xmath194 is said to be a sum - of - squares ( sos ) polynomial if it can be written as @xmath195 for some polynomials @xmath196 . a polynomial vector @xmath197 is said to _ componentwise sum - of - squares _ ( csos ) if each of its components is an sos polynomial . checking whether a polynomial is sos can be exactly cast as a semidefinite program @xcite that can be easily solved using semidefinite programming solvers such as sedumi @xcite or sdpt3 @xcite . the package sostools @xcite can be used to formulate the sos program in a convenient way . this approach has been used for the analysis and control of impulsive systems @xcite , switched systems @xcite , pseudo - periodic systems with impulses @xcite and stochastic impulsive systems @xcite . below is the approximation of the conditions in theorem [ th : cstdt2 ] , statement : [ prop : sos ] let @xmath198 , @xmath119 and @xmath199 be given and assume that there exists a polynomial vectors @xmath200 and @xmath201 of degree @xmath202 such that a. @xmath203 is csos , b. @xmath204 ( or is csos ) , c. @xmath205 is csos and d. @xmath206 ( or is csos ) . then , the conditions of theorem [ th : cstdt2 ] , statement hold with the same @xmath132 and the system is asymptotically stable under constant dwell - time @xmath154 . clearly , the condition in ( ii ) is equivalent to saying that @xmath207 and that in ( iv ) is equivalent to the condition . we then consider the condition in ( iii ) , which implies that @xmath208 for all @xmath209 . since @xmath203 is csos , then we have that @xmath210 for all @xmath209 and , therefore , this implies that @xmath211 for all @xmath212 $ ] which is exactly the condition . the proof is complete . this approach is based on handelman s theorem @xcite that states that if a polynomial is positive on a compact polytope of the form @xmath213 , then it can be expressed as a nonnegative linear combinations of products of @xmath214 . this elegant result has been considered in @xcite for deriving finite - dimensional linear programs establishing the stability of linear uncertain positive systems or in @xcite for the construction of lyapunov functions and the verification of polynomials optimization problems . below is the approximation of the conditions in theorem [ th : cstdt2 ] , statement : [ prop : handelman ] let @xmath215 , @xmath119 and @xmath199 be given and assume that there exists a vector - valued polynomial @xmath200 of degree @xmath216 and parameters @xmath217 , @xmath218 such that a. @xmath219 for all @xmath220 , b. @xmath221 , c. the polynomial @xmath222 has nonnegative coefficients and d. @xmath206 . then , the conditions of theorem [ th : cstdt2 ] , statement hold with the same @xmath132 and the system is asymptotically stable under constant dwell - time @xmath154 . the proof follows from the same line as for proposition [ prop : sos ] with the difference that we consider here handelman s theorem in place of putinar s positivstellensatz . following similar arguments as in the proof of proposition [ prop : sos ] , if the conditions in statement ( i ) and ( iii ) are satisfied then we have that @xmath211 for all @xmath212 $ ] . the fact that the condition is stated in terms of the nonnegativity of the coefficients ( as opposed to the nonnegativity of the eigenvalues of the gram matrix for the sum of squares relaxation ) comes from the fact that handelman s theorem deals with an equality between the supposedly positive polynomial and a nonnegative sum of products of the basis functions of the polytope however , we are not interested here in finding an equality but finding a positive lower bound for the polynomial that can be expressed as a nonnegative sum of products of basis functions . the polynomial in statement ( iii ) can be rewritten in the form @xmath223 where @xmath223 . using now the fact that @xmath224 , then we have that@xmath223 for all @xmath224 if @xmath225 . it is also important to stress that the third condition could have also been formulated in terms of an sos condition instead of in terms of a condition on the coefficients on some polynomial . however , the former would lead to semidefinite programming conditions as opposed to linear programming ones for the latter . note also that the latter condition shares similarities with the polya s theorem @xcite . we illustrate here the efficiency obtained results and compare them with existing ones . comparisons between the proposed relaxation results are also made for completeness . let us consider the system with the matrices @xcite @xmath226 clearly , the matrix @xmath25 is hurwitz stable whereas the @xmath26 is schur anti - stable . simple calculations show that the matrix @xmath227 is schur stable provided that @xmath228 . using the minimum dwell - time stability result of theorem [ th : mindt2 ] , ( a ) , we find the value 0.2311 for the estimated minimum dwell - time value . as this value coincides with the lower - bounded obtained for the constant dwell - time case , then we can conclude that the estimated minimum dwell - time is exact . this proves that the proposed method is way less conservative than the one proposed in @xcite where the value 1.0986 is found for the minimum dwell - time . let us consider the system with the matrices @xmath229 when @xmath230 , the conditions of theorem [ th : mindt2 ] , ( a ) and of remark [ rem : swapped_minimum ] both yield the value @xmath231 for the minimum dwell - time . this value was also found in @xcite using conditions based on quadratic lyapunov functions . this value also coincides with the minimal value obtained in the constant dwell - time case , emphasizing then the exactness of the computed minimum dwell - time . on the other hand , when @xmath232 , the conditions of remark [ rem : swapped_minimum ] yield the value 0.4290 while the value 0.3615 is obtained using the conditions of theorem [ th : mindt2 ] , ( a ) . the latter value is also found when using the conditions based on the use of quadratic lyapunov functions stated in @xcite and coincides with the value obtained in the constant dwell - time case . hence exactness of the estimate of the minimum dwell - time is again proved . moreover , the discrepancy between the results obtained with the conditions in theorem [ th : mindt2 ] and in remark [ rem : swapped_minimum ] demonstrate the non - equivalence of the conditions and in the necessity of considering both in order to obtain less conservative result . note that this phenomenon does not occur when considering quadratic lyapunov function ; see e.g. @xcite . a comparison of the different computational methods described in section [ sec : imp : computation ] is performed and the results are summarized in table [ table : mindt1 ] where we can see that all the approaches are able to approach the value obtained using the condition of theorem [ th : mindt2 ] , ( b ) , when the discretization order @xmath233 or the polynomial degrees @xmath234 or @xmath216 increase . however , this comes at the price of an increase of the computational complexity and the solving time . .comparison of the numerical results obtained for the minimum dwell - time using the discretization approach ( section [ sec : discretization ] ) , the sum of squares approach ( section [ sec : sos ] ) and the handelman - based approach ( section [ sec : handelman ] ) . for each method , we give the estimate for the minimum dwell - time @xmath154 , the number of primal / dual variables of the optimization problem and the overall solving time in seconds . for fairness , the solver sedumi is used for solving all the optimization problems . [ cols="^,^,^,^,^,^,^",options="header " , ] let us consider the system with the matrices @xmath235 note that both subsystems are unstable . the goal is then to design a controller of the form such that the closed - loop system is positive and stable under minimum dwell - time @xmath236 . we consider theorem [ th : mindt_switched_2 ] and solve the conditions using the sos relaxation with @xmath237 ( degree of @xmath238 and @xmath239 is one while the additional polynomials are of degree 2 ) . the obtained semidefinite program has 180 primal variables and 71 dual variables . solving the conditions yields the controller matrices @xmath240 for simulation purposes , we generate a random sequence of dwell - times satisfying the minimum dwell - time constraint and we obtain the trajectories depicted in figure [ fig : mindt_switched_stabz ] where we can see observe the stabilizing effect of the controller . let us consider the system with the matrices @xmath241 as in the previous examples , both subsystems are unstable and we seek for a controller of the form that makes the closed - loop system positive and stable under range dwell - time @xmath242 and @xmath243 . we consider theorem [ th : range_switched_2 ] and solve the conditions using the sos relaxation with @xmath237 ( degree of @xmath238 and @xmath239 is one while the additional polynomials are of degree 2 ) . the semidefinite program to solve has 229 primal variables and 84 dual variables . solving the conditions takes 1.5875 seconds and returns the controller matrices @xmath244 for simulation purposes , we generate a random sequence of dwell - times satisfying the range dwell - time constraint and we obtain the trajectories depicted in figure [ fig : rangedt_switched_stabz ] where we can see observe the stabilizing effect of the controller . several stability and stabilization conditions for linear positive impulsive systems and linear positive switched systems have been obtained using the notion of dwell - times . discrete - time stability conditions have been obtained and reformulated using a lifting approach in order to get conditions that can be applied for design purposes . because of their infinite - dimensional nature , several relaxation approaches have been proposed to make the lifted conditions finite - dimensional . several examples have been given for illustration . 95 [ 1]#1 [ 1]`#1 ` urlstyle [ 1]doi : # 1 c. briat , e. a. yavuz , h. hjalmarsson , k .- h . johansson , u. t. jnsson , g. karlsson , and h. sandberg . the conservation of information , towards an axiomatized modular modeling approach to congestion control . _ ieee transactions on networking _ , 23(3 ) , 2015 . c. briat and m. khammash . computer control of gene expression : robust setpoint tracking of protein mean and variance using integral feedback . in _ 51st ieee conference on decision and control _ , pages 35823588 , maui , hawaii , usa , 2012 . v. jonsson , a. rantzer , and r. m. murray . a scalable formulation for engineering combination therapies for evolutionary dynamics of disease . in _ american control conference _ , pages 27712778 , portland , usa , 2014 . w. m. haddad and v. chellaboina . stability and dissipativity theory for nonnegative dynamic systems : a unified analysis framework for biological and physiological systems . _ nonlinear analysis : real world applications _ , 6:0 3565 , 2005 . c. briat . robust stability and stabilization of uncertain linear positive systems via integral linear constraints - @xmath245- and @xmath246-gains characterizations . _ international journal of robust and nonlinear control _ , 23(17):0 19321954 , 2013 . c. briat robust stability analysis of uncertain linear positive systems via integral linear constraints - @xmath245- and @xmath246-gains characterizations . in _ 50th ieee conference on decision and control _ , pages 31223129 , orlando , florida , usa , 2011 . y. ebihara , d. peaucelle , and d. arzelier . @xmath245 gain analysis of linear positive systems and its applications . in _ 50th conference on decision and control , orlando , florida , usa _ , pages 40294034 , 2011 . m. ait rami . stability analysis and synthesis for linear positive systems with time - varying delays . in _ positive systems - proceedings of the 3rd multidisciplinary international symposium on positive systems : theory and applications ( posta 2009 ) _ , pages 205216 . springer - verlag berlin heidelberg , 2009 . y. ebihara , d. peaucelle , and d. arzelier . optimal @xmath249-controller synthesis for positive systems and its robustness properties . in _ american control conference _ , pages 59925997 , montreal , canada , 2012 . c. briat and m. khammash . interval peak - to - peak observers for continuous- and discrete - time systems with persistent inputs and delays . _ to appear in automatica _ , 2016 . url http://arxiv.org/abs/1511.07888 . a. zappavigna , p. colaneri , j. c. geromel , and r. h. middleton . stabilization of continuous - time switched linear positive systems . in _ american control conference _ , pages 32753280 , baltimore , maryland , usa , 2010 . a. zappavigna , p. colaneri , j. c. geromel , and r. shorten . dwell time analysis for continuous - time switched linear positive systems . in _ american control conference _ , pages 62566261 , baltimore , maryland , usa , 2010 . f. mazenc and m. malisoff . stability analysis for time - varying systems with delay using linear lyapunov functionals and a positive systems approach . _ ieee transactions on automatic control _ , 61(3):0 771776 , 2016 . n. sivashankar and p. p. khargonekar . characterization of the @xmath251-induced norm for linear systems with jumps with applications to sampled - data systems . _ siam journal on control and optimization _ , 32(4):0 11281150 , 1994 . g. chesi , p. colaneri , j. c. geromel , r. middleton , and r. shorten . a nonconservative lmi condition for stability of switched systems with guaranteed dwell - time . _ ieee transactions on automatic control _ , 57(5):0 12971302 , 2012 . c. briat and a. seuret on the necessity of looped - functionals arising in the analysis of pseudo - periodic , sampled - data and hybrid systems . _ international journal of control _ , 88((12):0 25632569 , 2015 . c. briat theoretical and numerical comparisons of looped functionals and clock - dependent lyapunov functions - the case of periodic and pseudo - periodic systems with impulses . _ international journal of robust and nonlinear control _ , 26:0 22322255 , 2016 . c. briat . convex lifted conditions for robust @xmath252-stability analysis and @xmath252-stabilization of linear discrete - time switched systems with minimum dwell - time constraint . _ automatica _ , 50(3):0 976983 , 2014 . a. papachristodoulou , j. anderson , g. valmorbida , s. prajna , p. seiler , and p. a. parrilo . _ sostools : sum of squares optimization toolbox for matlab v3.00 _ , 2013 . url http://www.cds.caltech.edu/sostools/sostools.pdf . r. kamyar , c. murti , and m. m. peet . constructing piecewise - polynomial lyapunov functions for local stability of nonlinear systems using handelman s theorem . in _ 53rd ieee conference on decision and control _ , pages 54815487 , los angeles , usa , 2014 . g. plya . ber positive darstellung von polynomen vierteljschr , naturforsch . ges . zrich , vol . 73 , pp . 141145 , 1928 . volume 2 of _ collected papers ( ed . r. p. boas ) _ , pages 309313 . mit press , cambridge , ma , 1974 . t. tanaka and c. langbort . emma for internally positive systems and a tractable class of distributed h - infinity control problems . in _ american control conference _ , pages 62386243 , baltimore , maryland , usa , 2010 . t. tanaka and c. langbort . symmetric formulation of the s - procedure , kalman - yakubovich - popov lemma and their exact losslessness conditions . _ ieee transactions on automatic control _ , 58(6):0 14861496 , 2013

hopf algebras are robust generalization of groups . recently , many authors have studied much more general hopf - like structures : weak hopf algebras , hopf monads , @xmath2-hopf algebras , compact quantum groups to name a few . in this article , we will mainly be interested with hopf algebroids and hopf categories . in the literature , there are plenty of inequivalent notions of a hopf algebroid . for the exposition on these notions , see bhm @xcite . batista et al . @xcite introduced the notion of a hopf category which is the natural categorification of a hopf algebra . motivated by a fundamental related to his phd thesis , the author tries to describe the geometry of hopf algebroids over @xmath3 . this geometric description necessitates a structure closely related to a hopf category , but which has not appeared in the literature as far as the author s knowledge . we will define such structures in section ( [ s3.0 ] ) . we will recall in section ( [ s2.0 ] ) definitions and properties of hopf algebroids . for completeness , we will also give a short exposition on the representation theoretic and galois theoretic aspects of hopf algebroids . most of section ( [ s2.0 ] ) follows @xcite except for the definition of morphisms of hopf algebroids and the definition of a coupled hopf algebra . in section ( [ s3.0 ] ) , we will define what topological hopf categories are and we will also define coupled hopf categories and their topological version . we will end that section with a formulation of galois theory for hopf categories and all the related variant we will introduce in that section . one of the main result of this paper is theorem ( [ t4.1 ] ) . it gives a bijective correspondence between finitely - generated projective hopf algebroids over @xmath3 and topological coupled hopf categories of finite type . using algebraic geometric and spectral theoretic methods , spanning the entirety of section ( [ s4.0 ] ) , we will prove this result . the second main result is theorem ( [ t5.1 ] ) , which states that , not only is there a bijection between hopf algebroids and topological hopf categories , their galois theories also matched in a bijective manner . following david hilbert s statement : 2cm2 cm _ `` the art of doing mathematics consists in finding that special case which contains all the germs of generality . '' we will discuss a very important example in section ( [ s3.2 ] ) which completely illustrates the general situation . * acknowledgement . * i would like to thank my phd supervisor ryszard nest for guiding me through my studies in non - commutative geometry and for the valuable discussions that help me write this article . i would also like to thank dsf grant , up diliman and the support of the danish national research foundation through the centre for symmetry and deformation ( dnrf92 ) . there are several inequivalent notions of a hopf algebroid . we will briefly present here the one defined in bhm @xcite . an @xmath4-_ring _ is a monoid object in the category of @xmath4-bimodules . explicitly , an @xmath4-ring is a triple @xmath5 where @xmath6 and @xmath7 are @xmath4-bimodule maps satisfying the associativity and unit axioms similar for algebras over commutative rings . a morphism of @xmath4-rings is a monoid morphism in category of @xmath4-bimodules . it is important to note that there is a bijection between @xmath4-rings @xmath5 and @xmath8-algebra morphisms @xmath7 . similar to the case of algebras over commutative rings , we can define modules over @xmath4-rings . for an @xmath4-ring @xmath5 , a @xmath5-_module _ is an algebra for the monad @xmath9 ( resp . @xmath10 ) on the category @xmath11 ( resp . @xmath12 ) of right ( resp . left ) modules over @xmath4 . we can dualize all the objects we have defined in the previous paragraph . an @xmath4-_coring _ is a comonoid in the category of @xmath4-bimodules , i.e a triple @xmath13 where @xmath14 and @xmath15 are @xmath4-bimodule maps satisfying the coassociativity and counit axioms dual to those axioms satisfied by the structure maps of an @xmath4-ring . a morphism of @xmath4-corings is a morphism of comonoids . given an @xmath4-coring @xmath13 , similar to coalgebras over commutative rings , we define a _ right _ _ left _ ) @xmath13-_comodule _ as a coalgebra for the comonad @xmath16 ( resp . @xmath17 ) on the category @xmath11 ( resp . @xmath12 ) . [ d2.1 ] a _ right _ ( resp . _ left _ ) @xmath4-_bialgebroid _ @xmath18 is an @xmath19-ring @xmath20 and an @xmath4-coring @xmath21 satisfying : 1 . @xmath22 and @xmath23 are @xmath8-algebra maps with commuting images defining the @xmath19-ring structure on @xmath18 which is compatible to the @xmath4-bimodule structure as an @xmath4-coring thru the following relation : @xmath24 2 . with the above @xmath4-bimodule structure on @xmath18 one can form @xmath25 . the coproduct @xmath26 is required to corestrict to a @xmath8-algebra map to @xmath27 respectively , @xmath28 3 . the counit @xmath29 extends the right ( resp . left ) regular @xmath4-module structure on @xmath4 to a right ( resp . left ) @xmath30-module . a _ morphism _ of @xmath4-bialgebroids is a morphism of @xmath31-rings and @xmath4-corings . [ r2.1 ] 1 . 2 . the @xmath8-algebra maps @xmath32 and @xmath33 define a @xmath8-algebra map @xmath34 . as we have noted , such @xmath8-algebra uniquely determines an @xmath19-ring structure on @xmath18 . the maps @xmath32 and @xmath33 are called the _ source _ and _ target _ maps , respectively . the @xmath8-submodule @xmath35 ( resp . @xmath36 ) of @xmath25 is a @xmath8-algebra with factorwise multiplication . this is called the _ takeuchi product_. the map @xmath37 , @xmath38 is easily seen to be a @xmath8-algebra morphism and hence , @xmath35 is an @xmath19-ring . the corestriction of @xmath26 is an @xmath19-bimodule map . hence , @xmath26 is an @xmath31-ring map . the same is true for @xmath36 . 4 . the source map @xmath32 is a k - algebra map and so it defines a unique @xmath4-ring structure on @xmath18 . the right version of condition ( c ) explicitly means that @xmath39 , @xmath40 defines a right @xmath30-action on @xmath4 . [ d2.2 ] let @xmath8 be a commutative , associative unital ring and let @xmath41 and @xmath4 be associative @xmath8-algebras . a _ hopf algebroid _ @xmath42 is a triple @xmath43 . @xmath44 and @xmath45 are bialgebroids having the same underlying @xmath8-algebra @xmath46 . specifically , @xmath44 is a left @xmath41-bialgebroid with @xmath47 and @xmath48 as its underlying @xmath49-ring and @xmath41-coring structures . similarly , @xmath45 is a right @xmath4-bialgebroid with @xmath50 and @xmath51 as its underlying @xmath19-ring and @xmath4-coring structures . let us denote by @xmath52 ( resp . @xmath53 ) the multiplication on @xmath54 ( resp . @xmath55 ) . @xmath56 is a ( bijective ) @xmath8-module map @xmath57 , called the _ antipode_. the compatibility conditions of these structures are as follows . 1 . the sources @xmath58 , targets @xmath59 and counits @xmath60 satisfy + @xmath61 2 . the left- and right - regular comodule structures commute , i.e. + @xmath62_-{\delta_{l } } \ar[r(1.7)]^-{\delta_{r } } & & h{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[d(1.6)]^-{\delta_{l}{\mathbin{\mathop{\otimes}\limits_{r } } } id}\\ & & \\ h{\mathbin{\mathop{\otimes}\limits_{l } } } h \ar[r(1.6)]_-{id { \mathbin{\mathop{\otimes}\limits_{l}}}\delta_{r } } & & h{\mathbin{\mathop{\otimes}\limits_{l } } } h{\mathbin{\mathop{\otimes}\limits_{r } } } h } \hspace{0.75 in } \xymatrix { h \ar[d(1.7)]_-{\delta_{r } } \ar[r(1.7)]^-{\delta_{l } } & & h{\mathbin{\mathop{\otimes}\limits_{l } } } h \ar[d(1.6)]^-{\delta_{r}{\mathbin{\mathop{\otimes}\limits_{l } } } id}\\ & & \\ h{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[r(1.6)]_-{id { \mathbin{\mathop{\otimes}\limits_{r}}}\delta_{l } } & & h{\mathbin{\mathop{\otimes}\limits_{r } } } h{\mathbin{\mathop{\otimes}\limits_{l } } } h } \ ] ] 3 . for all @xmath63 and for all @xmath64 we have @xmath65 , 4 . @xmath56 is the convolution inverse of the identity map i.e. , the following diagram commute + @xmath66 ^ -{s{\mathbin{\mathop{\otimes}\limits_{l } } } i d } & & & & h{\mathbin{\mathop{\otimes}\limits_{l } } } h \ar[rrd]^-{\mu_{l } } & & \\ h \ar[rru]^-{\delta_{l } } \ar[rrrr]^-{\epsilon_{r } } & & & & r \ar[rrrr]^-{s_{r } } & & & & h \\ h \ar[rrd]_-{\delta_{r } } \ar[rrrr]_-{\epsilon_{l } } & & & & l \ar[rrrr]_-{s_{l } } & & & & h \\ & & h{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[rrrr]_-{id{\mathbin{\mathop{\otimes}\limits_{r } } } s } & & & & h{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[rru]_-{\mu_{r } } & & \\ } \ ] ] [ r2.2 ] 1 . 2 . in the constituent bialgebroids @xmath45 and @xmath44 , the counits @xmath67 and @xmath68 extend the regular module structures on the base rings @xmath4 and @xmath41 to the @xmath4-ring @xmath55 and to the @xmath41-ring @xmath54 , respectively . equivalently , the counits extend the regular module structures on the base rings @xmath4 and @xmath41 to the @xmath69-ring @xmath70 and to the @xmath71-ring @xmath72 . this particularly implies that the maps @xmath73 , @xmath74 , @xmath75 and @xmath76 are idempotents . this means that the images of @xmath77 and @xmath78 coincides in @xmath46 . same is true for the images of @xmath79 and @xmath80 . notice that for condition ( b ) to make sense , apart from being an @xmath41-bimodule map , @xmath81 has to be an @xmath4-bimodule map . this is the case using remark ( 1 ) . similarly , @xmath82 is an @xmath41-bimodule map . 4 . we can equip @xmath46 with two @xmath83-bimodule structures one using @xmath80 and @xmath78 and the other using @xmath77 and @xmath79 . condition ( c ) relates these two @xmath83-bimodules structures via the antipode @xmath56 which in turn makes the diagram in condition ( d ) defined . 5 . a most convenient way to summarize the property of the antipode of a hopf algebra is to express it as the inverse of the identity map in the convolution algebra of endomorphisms of that hopf algebra . for hopf algebroids , the antipode is the inverse of the identity map in the appropriate category , called the _ convolution category _ of @xmath42 . as before , @xmath4 and @xmath41 are @xmath8-algebras . let @xmath84 and @xmath85 be @xmath8-modules such that @xmath84 has an @xmath4-coring @xmath86 and an @xmath41-coring @xmath87 structures and @xmath85 has an @xmath88-ring structure with multiplications @xmath89 and @xmath90 . define the convolution category @xmath91 to be the category with two objects labelled @xmath4 and @xmath41 . for @xmath92 , a morphism @xmath93 is a @xmath94-@xmath95-bimodule map @xmath96 . for @xmath97 and morphisms @xmath98 and @xmath99 , we define the composition @xmath100 to be the following convolution + @xmath101 + the antipode @xmath56 of a hopf algebroid @xmath42 is the inverse of the identity map @xmath102 viewed as an arrow in @xmath103 . let us note that condition ( c ) in the definition of a bialgebroid implies that @xmath104 is the identity . similarly , @xmath105 is also the identity . using condition ( a ) in the definition of a hopf algebroid , we see that the following compositions define pairs of inverse @xmath8-algebra maps . + @xmath106 ^ -{\epsilon_{r}\circ s_{l } } & & r^{op } \ar[rr]^-{\epsilon_{l}\circ t_{r } } & & l } \hspace{.5 in } \xymatrix { r \ar[rr]^-{\epsilon_{l}\circ s_{r } } & & l^{op } \ar[rr]^-{\epsilon_{r}\circ t_{l } } & & r}\ ] ] + this is particular implies that @xmath4 and @xmath41 are anti - isomorphic @xmath8-algebras . 7 . since there are two coproducts involved in a hopf algebroid , namely @xmath81 and @xmath82 , we will use different sweedler notations for their corresponding components . we will write @xmath107}\otimes_{l}h_{[2]}$ ] and @xmath108}\otimes_{r}h^{[2]}$ ] for @xmath64 . 8 . with a fixed bijective antipode @xmath56 , the constituent left- and right - bialgebroids of a hopf algebroid determine each other , see for example the article @xcite . in view of this and the fact that @xmath41 and @xmath4 are anti - isomorphic , in the sequel where we will be mainly interested with hopf algebroids with bijective antipodes we will simply call @xmath42 a hopf algebroid _ over _ @xmath4 instead of explicitly mentioning @xmath41 . [ d2.3 ] let @xmath109 and @xmath110 be hopf algebroids over @xmath4 . an _ algebraic morphism _ @xmath111 of hopf algebroids is a pair @xmath112 of a left - bialgebroid morphism @xmath113 and a right - bialgebroid morphism @xmath114 for which the following diagrams commute @xmath115_-{\varphi_{l } } \ar[rr]^-{s } & & h_{r } \ar[dd]^-{\varphi_{r}}\\ & & \\ h_{l}^ { ' } \ar[rr]_-{s^ { ' } } & & h_{r}^ { ' } } \hspace{0.75 in } \xymatrix { h_{r } \ar[dd]_-{\varphi_{r } } \ar[rr]^-{s } & & h_{l } \ar[dd]^-{\varphi_{l}}\\ & & \\ h_{r}^ { ' } \ar[rr]_-{s^ { ' } } & & h_{l}^ { ' } } \ ] ] and composition of such a pair is componentwise . let @xmath4 and @xmath116 be @xmath8-algebras and @xmath109 and @xmath117 be hopf algebroids over @xmath4 and @xmath116 , respectively . in view of remark ( [ r2.2 ] ) ( 7 ) above , denote by @xmath118 and @xmath119 . a _ geometric morphism _ @xmath120 of hopf algebroids is a pair @xmath121 of @xmath8-algerba maps @xmath122 and @xmath123 , where @xmath124 denote the underlying @xmath8-algebra structures of the hopf algebroids under consideration . these two maps satisfy the following compatibility conditions . 1 . @xmath125 and @xmath126 intertwines the source , target and counit maps of the left - bialgebroid structures of @xmath42 and @xmath127 , i.e. + @xmath128_-{\phi } \ar[rr]^-{\epsilon^{h}_{l } } & & l \ar[dd]^-{f}\\ & & \\ k \ar[rr]_-{\epsilon^{k}_{l } } & & l^ { ' } } \hspace{0.15 in } \xymatrix { l \ar[dd]_-{f } \ar[rr]^-{t^{h}_{l } } & & h \ar[dd]^-{\phi}\\ & & \\ l^ { ' } \ar[rr]_-{t^{k}_{l } } & & k } \hspace{0.15 in } \xymatrix { l \ar[dd]_-{f } \ar[rr]^-{s^{h}_{l } } & & h \ar[dd]^-{\phi}\\ & & \\ l^ { ' } \ar[rr]_-{s^{k}_{l } } & & k. } \ ] ] + same goes for the source , target and counit maps of the right - bialgebroid structures . 2 . in view of condition @xmath129 , the @xmath8-bimodule map @xmath130 defines @xmath8-bimodule maps + @xmath131 ^ -{\phi\prescript{}{f}{\otimes } \ \phi } & & k \prescript{}{l^{'}}{\otimes } \ k , } \hspace{.25 in } \xymatrix { h \otimes_{r } \ h \ar[rr]^-{\phi\ \otimes_{f } \phi } & & k \otimes_{r^ { ' } } \ k. } \ ] ] + we then require that the following diagrams commute + @xmath131 ^ -{\phi\prescript{}{f}{\otimes } \ \phi } \ar[dd]_-{\mu^{h}_{l } } & & k \prescript{}{l^{'}}{\otimes } \ k \ar[dd]^-{\mu^{k}_{l } } \\ & & \\ h \ar[rr]_-{\phi } & & k } \hspace{.25 in } \xymatrix { h \otimes_{r } \ h \ar[rr]^-{\phi\ \otimes_{f } \phi } \ar[dd]_-{\mu^{h}_{r } } & & k \otimes_{r^ { ' } } \ k \ar[dd]^-{\mu^{k}_{r } } \\ & & \\ h \ar[rr]_-{\phi } & & k}\ ] ] 3 . also by of condition @xmath129 , the @xmath8-bimodule maps @xmath132 and @xmath133 of condition @xmath134 further define @xmath8-bimodule maps + @xmath135 ^ -{\phi\prescript{}{f}{\times } \ \phi } & & k \prescript{}{l^{'}}{\times } \ k , } \hspace{.25 in } \xymatrix { h \times_{r } \ \times_{f } \phi } & & k \times_{r^ { ' } } \ k. } \ ] ] + we then require that the following diagrams commute . + @xmath136 ^ -{\phi } \ar[dd]_-{\delta^{h}_{l } } & & k \ar[dd]^-{\delta^{k}_{l } } \\ & & \\ h \prescript{}{l}{\times } \ h \ar[rr]_-{\phi\prescript{}{f}{\times } \ \phi } & & k \prescript{}{l^{'}}{\times } \ k } \hspace{.25 in } \xymatrix { h \ar[rr]^-{\phi } \ar[dd]_-{\delta^{h}_{r } } & & k \ar[dd]^-{\delta^{k}_{r } } \\ & & \\ h \times_{r } \ h \ar[rr]_-{\phi\ \times_{f } \phi } & & k \times_{r^ { ' } } \ k } \ ] ] 4 . @xmath126 intertwines the antipodes of @xmath42 and @xmath127 , i.e. @xmath137 . [ r2.3 ] 1 . 2 . for a @xmath8-algebra @xmath4 , let us denote by @xmath138 the category whose objects are hopf algebroids over @xmath4 and morphisms are algebraic morphisms . for a fixed @xmath8 , let us denote by @xmath139 the category whose objects are hopf algebroids over @xmath8-algebras and morphisms are geometric morphisms . the existence of these two naturally defined categories reflect the fact that hopf algebroids are generalization of both hopf algebras and groupoids . 3 . equip @xmath140 with the hopf algebroid structure defined in example 5 of the next section . let @xmath109 be a hopf algebroid over @xmath4 . then the unit maps @xmath141 together with the identity map on @xmath4 define geometric morphisms @xmath142 and @xmath143 . [ e2.1 ] * hopf algebras*. a hopf algebra @xmath46 over the commutative unital ring @xmath8 gives an example of a hopf algebroid . here , we take @xmath144 as @xmath8-algebras , take @xmath145 to be the source and target maps , set @xmath146 to be the counits , and @xmath147 to be the coproducts . [ e2.2 ] * coupled hopf algebras*. it might be tempting to think that hopf algebroids for which @xmath144 must be hopf algebras . this is not entirely the case . we will give a general set of examples for which this is not true . two hopf algebra structures @xmath148 and @xmath149 over the same @xmath8-algebra @xmath46 are said to be _ coupled _ if 1 . there exists a @xmath8-module map @xmath150 , called the _ coupling map _ such that + @xmath151 ^ -{c\otimes i d } & & & & h\otimes h \ar[rd(1.6)]^-{m } & & \\ & & & & & & & & \\ h \ar[ru(1.8)]^-{\delta_{1 } } \ar[rd(1.8)]_-{\delta_{2 } } \ar@<1ex>[rrrr]^-{\epsilon_{2 } } \ar@<-1ex>[rrrr]_-{\epsilon_{1 } } & & & & k \ar[rrrr]^-{\eta } & & & & h \\ & & & & & & & & \\ & & h\otimes h \ar[r(3)]_-{id\otimes c } & & & & h\otimes h \ar[ru(1.6)]_-{m } & & \\ } \ ] ] + commutes , and 2 . the coproducts @xmath152 and @xmath153 in @xmath46 commutes . coupled hopf algebras give rise to hopf algebroids over @xmath8 . the left @xmath8-bialgebroid is the underlying bialgerba of @xmath154 while the right @xmath8-bialgebroid is the underlying bialgebra of @xmath155 . the coupling map plays the role of the antipode . let us give examples of coupled hopf algebras . connes and moscovici constructed _ twisted _ antipodes in @xcite . let us show that such a twisted antipode is a coupling map for some coupled hopf algebras . let @xmath156 be a hopf algebra . take @xmath157 as hopf algebras . let @xmath158 be a character . define @xmath159 by @xmath160 . take @xmath161 . define @xmath162 by @xmath163 . note the sweedler - legs of @xmath164 appearing in the definition of @xmath165 is the one provided by @xmath26 and not by @xmath153 . then , @xmath166 is a hopf algebra coupled with @xmath154 by the coupling map @xmath167 defined by @xmath168 . [ e2.3 ] * groupoid algebras*. given a small groupoid @xmath169 with finitely many objects and a commutative unital ring @xmath8 , we can construct what is called the groupoid algebra of @xmath169 over @xmath8 , denoted by @xmath170 . for such a groupoid @xmath169 , let us denote by @xmath171 its set of objects , @xmath172 its set of morphisms , @xmath173 the source and target maps , @xmath174 the unit map , @xmath175 the inversion map , @xmath176 the set of composable pairs of morphisms , and @xmath177 the partial composition . the groupoid algebra @xmath170 is the @xmath8-algebra generated by @xmath172 subject to the relation @xmath178 for @xmath179 . the groupoid algebra @xmath170 is a hopf algebroid as folows . the base algebras @xmath4 and @xmath41 are both equal to @xmath180 and the two bialgebroids @xmath45 and @xmath44 are isomorphic as bialgebroids with underlying @xmath8-module @xmath181 . the partial groupoid composition @xmath182 dualizes and extends to a multiplication @xmath183 which then factors through the canonical surjection @xmath184 to give the product @xmath185 . the source and target maps @xmath186 of the groupoid give the source and target maps @xmath187 , respectively . the unit map gives the counit map @xmath188 . finally , the inversion map gives the antipode map @xmath189 . note that the underlying bimodule structures of the right and the left bialgerboid is related by the antipode map . [ e2.4 ] * weak hopf algebras*. another structure that generalize hopf algebras , called weak hopf algebras , also are hopf algebroids . explicitly , a weak hopf algebra @xmath46 over a commutative unital ring @xmath8 is a unitary associative algebra together with @xmath8-linear maps @xmath190 ( weak coproduct ) , @xmath191 ( weak counit ) and @xmath192 ( weak antipode ) satisfying the following axioms : 1 . @xmath26 is multiplicative , coassociative , and weak - unital , i.e. @xmath193 2 . @xmath194 is counital , and weak - multiplicative , i.e. for any @xmath195 @xmath196 3 . for any @xmath64 , @xmath197 and @xmath198 let us sketch a proof why a weak hopf algebra @xmath46 is a hopf algebroid . consider the maps @xmath199 , @xmath200 and @xmath201 , @xmath202 . by @xmath8-linearity and weak - multiplicativity of @xmath194 , @xmath203 and @xmath204 are idempotents . multiplicativity and coassiociativity of @xmath26 and counitality of @xmath194 implies that for any @xmath64 , @xmath205 now , using these relations and coassiociativity of @xmath26 we get @xmath206 @xmath207 thus , the first tensor factor of the left - hand side of the first equation above is in the image of @xmath203 . similarly , the last tensor factor of the left - hand side of the second equation above is in the image of @xmath204 . clearly , @xmath208 . hence , the images of @xmath203 and @xmath204 are unitary subalgebras of @xmath46 . denote these subalgebras by @xmath4 and @xmath41 , respectively . by the weak - unitality of @xmath26 we see that these subalgebras are commuting subalgebras of @xmath46 . taking the source map @xmath32 as the inclusion @xmath209 and the target map as @xmath210 , @xmath211 equips @xmath46 with an @xmath19-ring structure . taking @xmath212 and @xmath82 as the composition @xmath213 ^ -{\delta } & & h\otimes_{k}h \ar@{->>}[rr ] & & h\otimes_{r}h } \ ] ] equips @xmath46 with an @xmath4-coring structure @xmath51 . the ring and coring structures just constructed gives @xmath46 a structure of right @xmath4-bialgebroid @xmath45 . using @xmath69 in place of @xmath4 in the above construction , we get a left @xmath69-bialgebroid @xmath214 . together with the right @xmath4-bialgebroid constructed and the existing weak antipode @xmath56 , we get a hopf algebroid @xmath215 . in this section , we will look at representations of hopf algebroids . towards the end of the section , we will look at the descent theoretic aspect of a special class of modules over hopf algebroids , the so called relative hopf modules . let @xmath43 be a hopf algebroid with underlying @xmath8-module @xmath46 . @xmath46 carries both a left @xmath41-module sctructure and a left @xmath4-module structure via the maps @xmath79 and @xmath80 , respectively . a _ right _ @xmath216 is a right @xmath41-module and a right @xmath4-module together with a right @xmath45-coaction @xmath217 and a right @xmath44-coaction @xmath218 such that @xmath219 is an @xmath44-comodule map and @xmath220 is an @xmath45-comodule map . for the coaction @xmath219 , let us use the following sweedler notation : @xmath221}{\mathbin{\mathop{\otimes}\limits_{r } } } m^{[1]}\ ] ] and for the coaction @xmath220 , let us use the following sweedler notation : @xmath222 } { \mathbin{\mathop{\otimes}\limits_{l } } } m_{[1]}.\ ] ] with these notations , the conditions above explicitly means that for all @xmath223 , @xmath224 and @xmath225 we have @xmath226}{\mathbin{\mathop{\otimes}\limits_{r}}}(m\cdot l)^{[1]}=\rho_{r}(m\cdot l)=m^{[0]}{\mathbin{\mathop{\otimes}\limits_{r } } } t_{l}(l)m^{[1]}\ ] ] @xmath227}{\mathbin{\mathop{\otimes}\limits_{l}}}(m\cdot r)_{[1]}=\rho_{l}(m\cdot r)=m_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } m_{[1]}s_{r}(r).\ ] ] we further require that the two coactions satify the following commutative diagrams @xmath228_-{\rho_{r } } \ar[r(1.7)]^-{\rho_{l } } & & m{\mathbin{\mathop{\otimes}\limits_{l } } } h \ar[d(1.6)]^-{\rho_{r}{\mathbin{\mathop{\otimes}\limits_{l } } } id}\\ & & \\ m{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[r(1.6)]_-{id { \mathbin{\mathop{\otimes}\limits_{r}}}\delta_{l } } & & m{\mathbin{\mathop{\otimes}\limits_{r } } } h{\mathbin{\mathop{\otimes}\limits_{l } } } h } \hspace{0.75 in } \xymatrix { m \ar[d(1.7)]_-{\rho_{l } } \ar[r(1.7)]^-{\rho_{r } } & & m{\mathbin{\mathop{\otimes}\limits_{r } } } h \ar[d(1.6)]^-{\rho_{l}{\mathbin{\mathop{\otimes}\limits_{r } } } id}\\ & & \\ m{\mathbin{\mathop{\otimes}\limits_{l } } } h \ar[r(1.6)]_-{id { \mathbin{\mathop{\otimes}\limits_{l}}}\delta_{r } } & & m{\mathbin{\mathop{\otimes}\limits_{l } } } h{\mathbin{\mathop{\otimes}\limits_{r } } } h } \ ] ] we will denote by @xmath229 the category of right @xmath42-comodules . symmetrically , we can define left @xmath42-comodules and we denote the category of a such by @xmath230 . comodules over hopf algebroids are comodules over the constituent bialgebroids . thus , one can speak of two different coinvariants , one for each bialgebroid . for a given right @xmath42-comodule @xmath216 , they are defined as follows : @xmath231 @xmath232 in the general case , we have @xmath233 . but in our case , where we assume @xmath56 is bijective these two spaces coincide . this will be important in the formulation of galois theory for hopf algebroids . to see that these coinvariants coincide , consider the following map @xmath234 @xmath235 here , @xmath46 acts on the right of @xmath236 through the second factor . if @xmath237 , then we have @xmath238}{\mathbin{\mathop{\otimes}\limits_{r } } } m^{[1 ] } ) = \rho_{l}(m^{[0]})\cdot s(m^{[1]})\\ \nonumber & = & ( m^{[0]}_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } m^{[0]}_{[1]})\cdot s(m^{[1 ] } ) = m^{[0]}_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } m^{[0]}_{[1]}s(m^{[1]})\\ \nonumber & = & m_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } m^{[0]}_{[1]}s(m^{[1]}_{[1]})= m_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } s_{l}(\epsilon_{l}(m_{[1]}))\\ \nonumber & = & m_{[0]}s_{l}(\epsilon_{l}(m_{[1]})){\mathbin{\mathop{\otimes}\limits_{l } } } 1 = m{\mathbin{\mathop{\otimes}\limits_{l } } } 1\\ \nonumber\end{aligned}\ ] ] this shows the inclusion @xmath233 . to show the other inclusion , one can run the same computation but using the inverse of @xmath239 which is the following map @xmath240 @xmath241 in this case , we can simply write @xmath242 for @xmath243 and refer to it as the @xmath42-coinvariants of @xmath216 instead of distinguishing the @xmath45- from the @xmath44-coinvariants , unless it is necessary to do so . let us now discuss monoid objects in @xmath229 . they are called @xmath42-comodule algebras . a right @xmath42-_comodule algebra _ is an @xmath4-ring @xmath244 such that @xmath216 is a right @xmath42-comodule and @xmath245 and @xmath246 are @xmath42-comodule maps . using sweedler notation for coactions , this explicitly means that for any @xmath247 we have @xmath248}{\mathbin{\mathop{\otimes}\limits_{r}}}(mn)^{[1]}=\rho_{r}(mn)=m^{[0]}n^{[0]}{\mathbin{\mathop{\otimes}\limits_{r } } } m^{[1]}n^{[1]},\ ] ] @xmath249}{\mathbin{\mathop{\otimes}\limits_{l}}}(mn)_{[1]}=\rho_{l}(mn)=m_{[0]}n_{[0]}{\mathbin{\mathop{\otimes}\limits_{l } } } m_{[1]}n_{[1]},\ ] ] @xmath250}{\mathbin{\mathop{\otimes}\limits_{r}}}1_{m}^{[1]}=\rho_{r}(1_{m})=1_{m}{\mathbin{\mathop{\otimes}\limits_{r } } } 1_{h},\ ] ] @xmath251}{\mathbin{\mathop{\otimes}\limits_{l}}}(1_{m})_{[1]}=\rho_{l}(1_{m})=1_{m}{\mathbin{\mathop{\otimes}\limits_{l } } } 1_{h}.\ ] ] let @xmath43 be a hopf algebroid with underlying @xmath8-module @xmath46 . a @xmath8-algebra extension @xmath252 is said to be ( _ right _ ) @xmath45-_galois _ if @xmath18 is a right @xmath45-comodule algebra with @xmath253 and the map @xmath254 ^ -{\mathfrak{gal}_{r } } & & b{\mathbin{\mathop{\otimes}\limits_{r}}}h } \ ] ] @xmath255}{\mathbin{\mathop{\otimes}\limits_{r}}}b^{[1]}\ ] ] is a bijection . the map @xmath256 is called the galois map associated to the bialgebroid extension @xmath252 . symmetrically , the extension @xmath252 is ( _ right _ ) @xmath44-_galois _ if @xmath18 is a right @xmath44-comodule algebra with @xmath257 and the map @xmath254 ^ -{\mathfrak{gal}_{l } } & & b{\mathbin{\mathop{\otimes}\limits_{l}}}h } \ ] ] @xmath258}b{\mathbin{\mathop{\otimes}\limits_{l}}}a_{[1]}\ ] ] is a bijection . we say that a @xmath8-algebra extension @xmath252 is @xmath42-_galois _ if it is both @xmath45-galois and @xmath44-galois . it is not known in general if the bijectivity of @xmath256 and @xmath259 are equivalent . however , if the antipode @xmath56 is bijective ( which is part of our standing assumption ) then @xmath256 is bijective if and only if @xmath259 . to see this , note that @xmath260 where @xmath261 is the map defined in the previous section for @xmath262 . since @xmath56 is bijective , @xmath261 is an isomorphism which gives the desired equivalence of bijectivity of @xmath256 and @xmath259 . thus , the extension @xmath252 is @xmath42-galois if it is a bialgebroid galois extension for any of its constituent bialgebroids . batista et al . @xcite introduced the notion of a hopf category over an arbitrary strict braided monoidal @xmath263 . in this section , we will introduce its topological version . for this purpose , we specialize @xmath263 as the category of complex vector spaces whose braiding is the usual flip of tensor factors . also , we will assume that the underlying categories of such hopf categories are small . we will be primarily interested with _ finite - type _ @xmath263-enriched categories , by which we mean the hom - sets are finite - dimensional vector spaces . before giving the definition of a hopf category , let is introduce some notation first . for two @xmath263-enriched categories @xmath264 and @xmath265 with the same set of objects @xmath84 , we define @xmath266 to the the @xmath263-enriched category with @xmath84 as the set of objects and for @xmath267 , the hom - set of arrows from @xmath268 to @xmath269 is the vector space @xmath270 we call @xmath266 the _ tensor product _ of @xmath264 and @xmath265 . with this @xmath271 , the category of @xmath263-enriched categories over @xmath84 becomes a strict monoidal category whose monoidal unit , denoted by @xmath272 , is the category over @xmath84 such that for any @xmath267 we have @xmath273 . [ d3.1 ] a _ hopf category _ @xmath274 over @xmath84 is a @xmath263-enriched category satisfying the following conditions . 1 . there are functors + @xmath275 ^ -{\delta } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } } , \hspace{.5 in } \xymatrix{\mathscr{h } \ar[rr]^-{\epsilon } & & \mathbb{1}^{x}}\ ] ] + called the _ coproduct _ and _ counit _ , respectively , such that @xmath26 is _ coassociative _ and _ counital _ with respect to @xmath194 , i.e. the diagram of functors + @xmath276 ^ -{\delta } \ar[dd]_-{\delta } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[dd]|-{id\otimes_{x}\delta } \\ & & \\ \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rr]_-{\delta\otimes_{x } i d } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \\ } \hspace{.25 in } \xymatrix { \mathscr{h } \ar@{=}[rr ] \ar@{=}[dd ] \ar[rrdd]^-{\delta } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathbb{1}^{x } \\ & & \\ \mathbb{1}^{x}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[ll]^-{\epsilon\otimes_{x } i d } \ar[uu]_-{is\otimes_{x}\epsilon } \\ } \ ] ] + commute . 2 . there is a functor @xmath277 , called the _ antipode _ , satisfying + @xmath278 ^ -{s\otimes_{x } i d } & & \mathscr{h}^{op}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rrd]^-{\circ } & & \\ \mathscr{h } \ar[rru]^-{\delta } \ar[rrd]_-{\delta } \ar[rrr]^-{\epsilon } & & & \mathbb{1}_{x } \ar[rrr]^-{\eta } & & & \mathscr{h } \\ & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rr]_-{id\otimes_{x } s } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h}^{op } \ar[rru]_-{\circ } & & \\}\ ] ] + here , @xmath279 denotes the bifunctor induced by the categorical composition in @xmath274 and @xmath280 is the functor that send @xmath281 to the identity element of @xmath282 . [ r3.1 ] functoriality of @xmath26 and @xmath194 means that for any @xmath267 , we have linear maps @xmath283 ^ -{\delta_{x , y } } & & \mathscr{h}_{x , y}\otimes \mathscr{h}_{x , y } } \hspace{.5 in } \xymatrix{\mathscr{h}_{x , y } \ar[rr]^-{\epsilon_{x , y } } & & \mathbb{c}}\ ] ] where @xmath284 is coassociative and counital with respect to @xmath285 in the usual sense . this implies that @xmath282 is a coalgebra . if we denote by @xmath286 the category of coalgebras on @xmath263 , another way to package part @xmath129 of definition ( [ d3.1 ] ) is to say that @xmath274 is enriched over @xmath287 . for the main results of this paper , we will be mostly interested with the case @xmath84 is a topological space . in such a case , it makes sense to reflect _ continuity _ on the functors @xmath26 , @xmath194 and @xmath56 along with the categorical structure maps . this calls for the following definition . [ d3.2 ] let @xmath84 be a topological space and let @xmath288 be the sheaf of continuous complex - valued functions on @xmath84 . a _ topological hopf category _ @xmath274 over @xmath84 is a hopf category together with a sheaf @xmath289 over @xmath290 ( with the product topology ) of @xmath288-bimodules satisfying the following conditions . 1 . denote by @xmath291 the projection onto the first and second factor , respectively . over an open set @xmath292 , for any @xmath293 , @xmath294 and @xmath295 we have + @xmath296 + for any @xmath297 . @xmath282 is the fiber of @xmath289 at @xmath298 . @xmath279 , @xmath280 , @xmath26 , @xmath194 and @xmath56 are the induced maps on global sections of the following map of sheaves + @xmath299 ^ -{\circ^{sh } } & & h^{sh } } , \hspace{.5 in } \xymatrix{\mathcal{o}_{x } \ar[rr]^-{\eta^{sh } } & & h^{sh } } , \ ] ] + @xmath300 ^ -{\delta^{sh } } & & h^{sh}{\mathbin{\mathop{\otimes}\limits_{\mathcal{o}_{x } } } } h^{sh } } , \hspace{.5 in } \xymatrix{h^{sh } \ar[rr]^-{\epsilon^{sh } } & & \mathcal{o}_{x}},\ ] ] + @xmath301 ^ -{s^{sh } } & & \left(h^{sh}\right)^{op } } \ ] ] + respectively . here , @xmath302 is the pullback of the sheaf @xmath289 along the map @xmath303 flipping the factors . [ r3.2 ] the bimodule tensor product @xmath304 used in part @xmath305 for @xmath306 of definition ( [ d3.2 ] ) is the tensor product of the appropriately modified @xmath288-bimodule @xmath289 , one in which we have @xmath307 for any @xmath308 . for the bimodule tensor product @xmath304 used for @xmath309 is the one with @xmath310 for any @xmath311 . the following , which will play an important role in our formulation of the main result , is the categorification of a coupled hopf algebra . [ d3.3 ] a _ coupled hopf category _ @xmath274 is a @xmath263-enriched category with two @xmath287-enrichments , denoted by @xmath312 and @xmath313 , with coproducts @xmath314 and counits @xmath315 , respectively ; and a functor @xmath277 , called the _ coupling _ functor , such that the following conditions are satisfied : 1 . the following diagrams , indicating the _ coupling condition _ , commute . + @xmath316 ^ -{s\otimes_{x } i d } & & \mathscr{h}^{op}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rrd]^-{\circ } & & \\ \mathscr{h } \ar[rru]^-{\delta^{l } } \ar[rrr]^-{\epsilon^{r } } & & & \mathbb{1}^{x } \ar[rrr]^-{\eta } & & & \mathscr{h } \\ \mathscr{h } \ar[rrd]_-{\delta^{r } } \ar[rrr]_-{\epsilon^{l } } & & & \mathbb{1}^{x } \ar[rrr]_-{\eta } & & & \mathscr{h } \\ & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rr]_-{id\otimes_{x } s } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h}^{op } \ar[rru]_-{\circ } & & \\ } \ ] ] 2 . the coproducts @xmath317 and @xmath318 commute , i.e. + @xmath319_-{\delta^{l } } \ar[rr]^-{\delta^{r } } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[dd]^-{\delta^{l}{\mathbin{\mathop{\otimes}\limits_{x } } } id}\\ & & \\ \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rr]_-{id { \mathbin{\mathop{\otimes}\limits_{x}}}\delta^{r } } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } } \hspace{.25 in } \xymatrix { \mathscr{h } \ar[dd]_-{\delta^{r } } \ar[rr]^-{\delta^{l } } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[dd]^-{\delta^{r}{\mathbin{\mathop{\otimes}\limits_{x } } } id}\\ & & \\ \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } \ar[rr]_-{id { \mathbin{\mathop{\otimes}\limits_{x}}}\delta^{l } } & & \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h}{\mathbin{\mathop{\otimes}\limits_{x } } } \mathscr{h } } \ ] ] [ r3.3 ] 1 . 2 . coupled hopf categories are almost the categorification of coupled hopf algebras . while the constituent bialgebras of a coupled hopf algebra is a hopf algebras in itself , the constituent categories @xmath312 and @xmath313 of a coupled hopf category @xmath274 need not be hopf categories . 3 . just like hopf categories , we can also _ topologize _ coupled hopf categories . we can take definition ( [ d3.2 ] ) : assert the existence of a sheaf @xmath289 over @xmath290 of @xmath288-bimodules , take conditions @xmath129 and @xmath134 as they are , and replace condition @xmath305 by 1 . @xmath317 , @xmath318 , @xmath320 , @xmath321 and @xmath56 are the induced maps on global sections of the following map of sheaves + @xmath300 ^ -{(\delta^{l})^{sh } } & & h^{sh}{\mathbin{\mathop{\otimes}\limits_{\mathcal{o}_{x } } } } h^{sh } } , \hspace{.5 in } \xymatrix{h^{sh } \ar[rr]^-{(\epsilon^{l})^{sh } } & & \mathcal{o}_{x}},\ ] ] + @xmath300 ^ -{(\delta^{r})^{sh } } & & h^{sh}{\mathbin{\mathop{\otimes}\limits_{\mathcal{o}_{x } } } } h^{sh } } , \hspace{.5 in } \xymatrix{h^{sh } \ar[rr]^-{(\epsilon^{r})^{sh } } & & \mathcal{o}_{x}},\ ] ] + @xmath301 ^ -{s^{sh } } & & \left(h^{sh}\right)^{op } } \ ] ] + respectively , making the following diagram + commute for any @xmath292 . here , @xmath322 and @xmath323 denote the maps induced by the composition and unit maps of @xmath324 . the maps @xmath325 and @xmath326 denote @xmath303 , @xmath327 and @xmath303 , @xmath328 , respectively . in this section , we will look at a very important example of a hopf category . this example will also be an example of our main result . this is a special case of proposition 7.1 of @xcite . consider a finite set @xmath84 whose elements are conveniently labelled as @xmath329 . equipped @xmath84 with the discrete topology . consider the category @xmath330 whose set of objects is @xmath84 and define @xmath331 . the category @xmath330 is obviously a hopf category . by proposition 7.1 of @xcite , @xmath332 is a weak hopf algebra . using the arguments in example 4 of section ( [ s2.2 ] ) , @xmath42 is a hopf algebroid over @xmath333 . the hopf algebroid @xmath42 has a more familiar form . it is isomorphic , as a hopf algebroid , the algebra @xmath334 over its diagonal @xmath335 . with the @xmath336-bimodule structure on @xmath334 defined as @xmath337 the coproduct @xmath82 and the counit @xmath67 are given as @xmath338 where @xmath126 is the linear functional defined by @xmath339 for all @xmath340 . with the usual matrix multiplication and unit , @xmath82 and @xmath67 constitutes a right @xmath336-bialgebroid structure on @xmath334 . for completeness , let us define the structure maps of the left @xmath336-bialgebroid structure of @xmath334 . consider the @xmath336-bimodule structure on @xmath334 defined as @xmath341 the coproduct @xmath81 and the counit @xmath68 are defined as @xmath342 where @xmath126 is the same linear functional used to defined @xmath67 . the antipode @xmath56 of this hopf algebroid is defined as @xmath343 . as a weak hopf algebra , @xmath126 is the counit of @xmath42 . the coproducts @xmath81 and @xmath82 are the extension of the weak coproduct @xmath26 to @xmath344 relative to the @xmath336-bimodule structure used . as we will see in section ( [ s4.0 ] ) , this is not a coincidence . this is in fact a special case of a more general result which we shall prove at the end of that section . formulation of galois theory for hopf category is straightforward . recall that in the case of hopf algebras , only the underlying bialgebra structure is relevant . in the coaction picture , the coalgebra is used to make sense of a coaction while the algebra structure is used to make sense of the galois map . all these ingredients are already present in the case of a hopf category . we will discuss the situation for topological hopf categories . the case for hopf categories follow almost immediately by dropping any manifestation of topology . before giving the definition of the categorical analogue of a comodule algebra , let us first discuss what a topological category is , at least for our purpose . a @xmath263-enriched category @xmath345 over a space @xmath84 is a _ topological category _ if there is a sheaf @xmath346 of @xmath288-bimodules such that conditions @xmath129 , @xmath134 and the relevant part of condition @xmath305 of definition ( [ d3.2 ] ) hold . [ d3.4 ] let @xmath274 be a topological hopf category with space of objects @xmath84 , coproduct @xmath26 , counit @xmath194 and antipode @xmath56 with associated sheaf @xmath289 . 1 . a topological category @xmath345 over @xmath84 enriched over @xmath263 , with associated sheaf @xmath346 , is a _ right _ @xmath274-_comodule _ if there is a functor @xmath347 such that the following conditions hold . 1 . @xmath348 is coassociative with respect to @xmath26 and counital with respect to @xmath194 , i.e. the diagrams of functors + @xmath349 ^ -{\rho } \ar[dd]_-{\rho } & & \mathscr{m}\otimes_{x}\mathscr{h } \ar[dd]|-{id\otimes_{x}\delta } \\ & & & \\ \mathscr{m}\otimes_{x}\mathscr{h } \ar[rr]_-{\rho\otimes_{x } i d } & & \mathscr{m}\otimes_{x}\mathscr{h}\otimes_{x}\mathscr{h } \\ } \hspace{-.5 in } \xymatrix { \mathscr{m } \ar@{=}[rr ] \ar[rdd]_-{\rho } & & \mathscr{m}\otimes_{x}\mathbb{1}^{x } \\ & & \\ & \mathscr{m } \otimes_{x } \mathscr{h } \ar[ruu]|-{id\otimes_{x}\epsilon } & \\}\ ] ] + commute , and 2 . the functor @xmath348 is the map induced by the map of sheaves @xmath350 where the tensor product is the same as the first one we described in remark ( [ r3.2 ] ) . @xmath274-_comodule _ can be symmetrically defined . 2 . a _ morphism _ @xmath351 of right @xmath274-comodules is a functor that commutes with the right coactions , i.e. one which makes the following diagram commute + @xmath352 ^ -{\rho^{m } } \ar[dd]_-{\phi } & & & \mathscr{m}\otimes_{x}\mathscr{h } \ar[dd]|-{\phi\otimes_{x}id } \\ & & & \\ \mathscr{n } \ar[rrr]_-{\rho^{n } } & & & \mathscr{n}\otimes_{x}\mathscr{h}. \\ } \ ] ] + here , @xmath353 and @xmath354 are the coactions of @xmath274 on @xmath345 and @xmath355 , respectively . 3 . a right @xmath274-comodule @xmath345 is a _ right _ @xmath274-_comodule - category _ if in addition , the composition map @xmath356 is a map of right @xmath274-comodules , where @xmath357 is equipped with the diagonal coaction . the _ coinvariants _ of a right @xmath274-comodule - category @xmath345 is the subcategory @xmath358 whose space of objects is @xmath84 and whose hom - sets are defined as + @xmath359 + for any @xmath267 . [ r3.4 ] a hopf category is the categorification of a hopf algebra with categorical composition corresponding to the algebra product . a right @xmath274-comodule @xmath345 is in particular a category , it already has a composition . this means that we only need to impose requirement @xmath360 in definition ( [ d3.4 ] ) to get a categorification of the notion of a comodule - algebra . in the classical set - up , one has to require the existence of a product and assert its compatibility with the comodule structures . in the set - up of hopf - galois theory with respect to hopf algebras , there is a well - understood notion for extensions of @xmath8-algebras @xmath252 to be @xmath46-galois for a hopf algebra @xmath46 even if @xmath361 . this is because @xmath362 makes sense as a @xmath8-module . all that is left to do is require @xmath363 and that the map @xmath364 , @xmath365 is bijective . on the other hand , in the situation of a hopf category @xmath274 and extensions of comodule - categories @xmath366 with @xmath367 , we can only make sense of the product @xmath368 in the case @xmath264 is the subcategory of @xmath345 whose hom - sets @xmath369 are all zero except when @xmath370 , in which case @xmath371 . in this case , we identify @xmath368 with @xmath357 . let us call such a category the _ trivial linear category _ over @xmath84 , and denote by @xmath372 . there might be a way to consider galois extensions by hopf categories in which the subcategory of coinvariants is strictly larger than @xmath372 , but at present it is not clear to the author how to make sense of it . fortunately , for our purpose of proving theorem ( [ t5.1 ] ) it is enough to have @xmath372 as the subcategory of coinvariants . [ d3.5 ] a right @xmath274-comodule - category @xmath345 is a @xmath274-_galois extension _ of @xmath372 provided 1 . @xmath373 , and 2 . the functor + @xmath374 @xmath375}\right)\otimes \beta_{[1]}\ ] ] + called the _ galois morphism _ , is fully faithful . [ r3.5 ] 1 . 2 . we are using sweedler notation for the legs of the coaction + @xmath376 + in other words , for any @xmath267 and @xmath377 , we have @xmath378}\otimes\alpha_{[1]}$ ] , where @xmath379}\in\mathscr{m}_{x , z}$ ] and @xmath380}\in\mathscr{h}_{z , y}$ ] for some @xmath381 . this , in particular , tells us that the map @xmath382 above make sense . galois extension by a coupled hopf category @xmath383 means simultaneous galois extensions of the constituent @xmath287-enriched categories @xmath312 and @xmath313 . in this section , we will consider hopf algebroids @xmath42 over a commutative unital @xmath1-algebra @xmath384 . we will restrict to the case where @xmath42 is finitely - generated and projective as a left and a right @xmath384-module . with the underlying assumption that the antipode is bijective , by @xcite , finitely - generated projectivity of any of the @xmath384-module structures of @xmath42 coming from the source and target maps are all equivalent . note that even though @xmath384 is commutative , its image under the source or the target map need not be central in @xmath42 . we will deal with this general situation and specialize in the case when we have centrality . let @xmath43 be a hopf algebroid over @xmath384 , a commutative unital @xmath1-algebra . assume that @xmath44 is finitely - generated and projective as a left- and a right-@xmath384-module via the source and target maps . with our standing assumption , @xmath45 has the same properties . let us first consider the left bialgebroid @xmath44 . the gelfand duality implies that @xmath385 for some compact hausdorff space @xmath84 . the serre - swan theorem applied to the left @xmath384-module @xmath44 gives us a finite - rank vector bundle @xmath386 such that @xmath387 as left modules , where the left @xmath3-module structure on @xmath388 is by pointwise multiplication , i.e. @xmath389 for all @xmath390 and @xmath391 . by the bimodule nature of @xmath44 , the right @xmath384-module structure of @xmath388 commutes with the left @xmath384-module which implies that we have a representation @xmath392 of @xmath3 into the endomorphism bundle of @xmath386 . since @xmath3 is abelian and @xmath348 is a @xmath393-morphism , @xmath394 lands in a maximal abelian subalgebra @xmath395 of @xmath396 . choose a finite collection of open sets @xmath397 that cover @xmath84 over which @xmath398 is trivializable . choose a system of coordinates such that @xmath398 trivial over each @xmath399 , i.e. @xmath400 , where @xmath401 is a finite - dimensional vector space.choosing a basis @xmath402 one has @xmath403 where @xmath404 is the rank of @xmath398 . commutativity of @xmath3 implies that up to unitaries @xmath405 , we have @xmath406 ^ -{\rho } & & c(u_{i } , diag(n))}\ ] ] where @xmath407 denotes the subalgebra of diagonal matrices on @xmath334 and @xmath408 for each @xmath409 , choosing a set of central orthogonal idempotents @xmath410 gives @xmath404 projections @xmath411 given by the following composition @xmath406 ^ -{\rho_{i } } & & c(u_{i } , diag(n))\cong \bigoplus\limits_{k=1}^{n}c(u_{i } ) \ar[rr]^-{proj_{j } } & & c(u_{i } ) } \ ] ] these projections are in particular continuous @xmath1-morphisms . hence , they give , for each @xmath409 , ( possibly non - distinct ) @xmath404 continuous injective maps @xmath412 , @xmath413 . geometrically , the situation is depicted figure ( [ fg4.1 ] ) . .,width=510 ] let us describe the nature of the set @xmath414 over the intersections @xmath415 . over @xmath416 we get a unitary @xmath417 which gives @xmath404 central orthogonal idempotents and up to ordering of such idempotents , one gets the sets @xmath418 . the union @xmath419 does not depend on the ordering of these idempotents . thus , over @xmath415 one gets unitaries @xmath417 and @xmath420 which simultaneously diagonalize @xmath394 . thus , we have @xmath421 from which we get that @xmath422 that is , the sets @xmath419 agree on the intersections . a subset @xmath423 is called _ transverse _ if @xmath424 are homeomorphisms , where @xmath425 and @xmath426 denotes the projection onto the first and second factor , respectively . in particular , @xmath427 is homeomorphic to @xmath84 . using the above argument , we have the following proposition . [ p4.1 ] for every @xmath409 , @xmath428 the set @xmath429 extends to a transverse subset of @xmath290 completely contained in @xmath430 . in particular , @xmath430 is the union of @xmath404 ( possibly overlapping ) transverse subsets of @xmath290 . this means that the curves in figure ( [ fg4.1 ] ) overlap . [ r4.1 ] another way to see why the closed subset @xmath431 is the union of transverse subsets of @xmath290 is by the fact the we can run the construction of the sets @xmath418 described in the beginning of this section in a symmetric fashion , one for each factor of @xmath290 . the whole picture ( [ fg4.1 ] ) is a decomposition of @xmath290 into @xmath432 , @xmath433 . the graphs of @xmath434 are labelled accordingly . note that each @xmath435 , @xmath436 are diagonalizable since they commute with their adjoint @xmath437 . and since such operators commute with each other , the collection @xmath438 is simultaneously diagonalizable . over a point @xmath439 , the fiber @xmath440 decomposes into joint eigenspaces of @xmath438 . the dimension of these eigenspaces are determined by the number of intersections of the vertical dotted line through @xmath439 with the graphs of @xmath441 . using this eigenspace decomposition , we have the following proposition which describes geometrically the right @xmath3-module structure of @xmath44 . [ p4.2 ] given @xmath391 and @xmath436 the section @xmath442 is given as @xmath443 where @xmath444 and @xmath445 . [ r4.2 ] 1 . 2 . in case @xmath3 is central in @xmath44 , the above picture reduce to @xmath446 the trivial cover and @xmath447 is the identity , i.e. the graph in the above picture is the diagonal of @xmath290 . the action defined by equation ( [ eq4.1 ] ) then reduces to pointwise multiplication which then coincides with the left @xmath3-module structure of @xmath387 . 3 . one can understand the right action above as _ pointwise - eigenvalue - scaled _ action . compared to the central case , every @xmath436 acts on a @xmath448 in a way that @xmath449 acts diagonally on @xmath450 , i.e. @xmath440 constitutes a single eigenspace for the operator @xmath449 corresponding to the eigenvalue @xmath451 . in the noncentral case , the action is still pointwise . however , the operator @xmath449 no longer has a single eigenspace . the eigenspaces are labelled by the points @xmath452 where @xmath444 and the eigenvalues of @xmath449 are @xmath453 , @xmath413 . [ p4.3 ] as a @xmath3-bimodule , @xmath454 where @xmath455 is a sheaf of complex vector spaces over @xmath290 supported on a closed subset @xmath431 . the @xmath3-bimodule structure on @xmath456 is defined as @xmath457 for @xmath458 and @xmath459 . the @xmath460 is dense in @xmath461 thus we can extend the @xmath460-module structure of @xmath44 to a @xmath461-module structure . consider the annihilator of @xmath44 , @xmath462 then , there is an open set @xmath463 such that @xmath464 . then @xmath465 , the support of the bimodule @xmath44 . [ p4.4 ] the subset @xmath466 is completely determined by the @xmath3-bimodule structure of @xmath44 . moreover , @xmath430 is the support of @xmath467 . by proposition ( [ p4.1 ] ) , @xmath430 is the union of transverse subsets of @xmath290 which is individually are unions of graphs of @xmath434 . let @xmath468 be the fiber of @xmath455 over @xmath469 , where @xmath470 denotes the eigensubspace of @xmath440 over the point @xmath471 . this defines a sheaf of vector spaces on @xmath290 supported on @xmath430 . a section of @xmath472 defines a section @xmath473 whose value at a point @xmath474 is @xmath475 where @xmath476 denotes the projection @xmath477 . conversely , any section @xmath478 defines a section @xmath479 by @xmath480 now , given @xmath481 we have @xmath482 for some @xmath483 . for any @xmath484 we have @xmath485 which shows that @xmath486 , @xmath487 is a bimodule map whose inverse is the map @xmath488 , @xmath489 . using proposition ( [ p4.1 ] ) , we can relate the vector bundles the serre - swan theorem gives when applied to the left and right @xmath3-module structure of @xmath44 as follows . [ p4.5 ] let @xmath490 and @xmath491 be the vector bundles given by the serre - swan theorem applied to the finitely - generated projective left and right @xmath3-module @xmath44 , respectively . then @xmath492 and @xmath493 are the direct - images of the sheaf @xmath455 along @xmath494 and @xmath495 , respectively . first , the direct - image of @xmath455 along @xmath494 is easily seen to be a vector bundle and the space of sections @xmath496 is easily seen to be isomorphic as left @xmath3-modules to the left @xmath3-module @xmath456 . by proposition ( [ p4.3 ] ) , @xmath497 as left @xmath3-modules . thus , by corollary 2.8 of @xcite we see that @xmath492 and @xmath498 are isomorphic as vector bundles . similar argument works for @xmath493 . let us say more about the nature of the eigenspaces @xmath499 , @xmath267 in relation to the subset @xmath430 . [ p4.6 ] 1 . 2 . @xmath500 3 . @xmath501 is the number of transverse subsets of @xmath290 contained in @xmath430 passing through @xmath474 , with multiplicities . 4 . @xmath502 for any @xmath503 . the previous section describes the geometry of @xmath44 using its bimodule structure over @xmath3 . but @xmath44 has more structure than just being a bimodule . in particular , it is a @xmath3-ring via the left source map @xmath504 . in this section , we will look at what this additional structure contributes to the geometry of @xmath44 . we will keep the notations of the previous section . the @xmath3-ring structure on @xmath505 via the source map @xmath79 consists of a pair of @xmath3-bimodule maps @xmath506 @xmath507 satisfying the associativity and unitality conditions . for brevity we will write @xmath508 . the unit map @xmath280 gives an element @xmath509 satisfying @xmath510 for all @xmath436 . since @xmath84 is hausdorff , if @xmath511 then we can find an @xmath436 such that @xmath512 and @xmath513 . thus , for @xmath511 we have @xmath514 thus , the source map @xmath515 is implemented by @xmath516 , @xmath517 . this means that @xmath518 and choosing @xmath125 such that @xmath519 and @xmath520 we see that @xmath521 is a nonzero element . thus , the diagonal @xmath522 of @xmath290 is in @xmath430 . see figure ( [ fg4.3 ] ) . of the bimodule @xmath18.,width=510 ] note that @xmath523 where @xmath524 is the sheaf of vector spaces whose fiber at a point @xmath525 is the vector space @xmath526 due to the balancing condition @xmath527 for @xmath528 and @xmath436 . notice that all but finitely many summands above are zero . specifically , only those @xmath503 for which @xmath474 and @xmath529 are both in @xmath430 contribute nontrivially . let us denote these @xmath503 as @xmath530 . by proposition ( [ p4.3 ] ) , @xmath531 as @xmath3-bimodules . since @xmath532 is a fully faithful functor by corollary 2.8 of @xcite , we can convert the global ring structures @xmath533 and @xmath280 into something fiber - wise . in particular , the product map @xmath533 induces a map @xmath534 ^ -{\mu _ { * } } & & e_{(x , z ) } } \end{aligned}\ ] ] illustrated in figure ( [ fg4.2 ] ) . by the universal property of direct sums , there are maps @xmath535 ^ -{\mu^{y_{i } } _ { * } } & & e_{(x , z ) } } \ ] ] one for each @xmath536 . the collection of these maps satisfy a set of conditions which , though derivable from associativity , is complicated to write down . see ( 3 ) of the remark below for these conditions . however , for the maps @xmath537 these conditions are precisely the associativity condition . likewise , the map @xmath280 induces maps @xmath538 which is nonzero when @xmath370 and zero otherwise . the map @xmath539 together with @xmath540 makes the vector space @xmath541 a unital algebra , whose dimension depend on the multiplicity of the associated eigenvalue . the following proposition is then immediate from these arguments . [ p4.7 ] let @xmath542 be the @xmath3-sub - bimodule of @xmath456 supported on the diagonal @xmath522 . then @xmath542 is an @xmath384-subring of @xmath44 where the multiplication is pointwise . moreover , @xmath542 is the centralizer of @xmath384 in @xmath44 . [ r4.3 ] 1 . 2 . using abuse of notation , let us identify @xmath384 with its image in @xmath44 . in case @xmath384 is central in @xmath44 , the fibers of the vector bundle @xmath543 are algebras . these algebras correspond to @xmath541 together with the maps @xmath544 and @xmath545 since in the central case , @xmath546 . thus , @xmath547 in the central case which is not surprising at all knowing that @xmath542 is the centralizer of @xmath384 . the maps @xmath548 are only restricted by the associativity of @xmath533 . since @xmath531 and @xmath532 is known to be a fully faithful functor by corollary 2.8 of @xcite , we have + @xmath549 ^ -{\left(\bigoplus\limits_{i}\mu^{i}_{*}\right)\otimes i d } \ar[dd]_-{id\otimes \left(\bigoplus\limits_{j}\mu^{j}_{*}\right ) } & & \bigoplus\limits_{y_{j}}\left(e_{(x , y_{j } ) } \otimes e_{(y_{j},z)}\right ) \ar[dd]^-{\bigoplus\limits_{j}\mu^{j } _ { * } } \\ & & \\ \bigoplus\limits_{y_{i}}\left(e_{(x , y_{i } ) } \otimes e_{(y_{i},z)}\right ) \ar[rr]_-{\bigoplus\limits_{i}\mu^{i } _ { * } } & & e_{(x , z ) . } } \ ] ] + universal property of direct sums gives us + @xmath550 ^ -{\mu^{i}_{*}\otimes i d } \ar[dd]_-{id\otimes \mu^{j } _ { * } } & & e_{(x , y_{j } ) } \otimes e_{(y_{j},z ) } \ar[dd]^-{\mu^{j } _ { * } } \\ & & \\ e_{(x , y_{i } ) } \otimes e_{(y_{i},z ) } \ar[rr]_-{\mu^{i } _ { * } } & & e_{(x , z)}. } \ ] ] + this justifies the argument before proposition ( [ p4.7 ] ) . we can also use this to say more about the fibers of @xmath455 which we state in the next proposition . [ p4.8 ] @xmath499 is a left @xmath551bimodule for every @xmath267 . [ r4.4 ] using remark ( [ r4.3 ] ) above , we can construct a small category @xmath312 enriched over the category of complex vector spaces . the set of objects of @xmath312 is @xmath84 . for every @xmath267 , we define @xmath552 we will call @xmath312 the _ associated category _ of the left @xmath384-bialgebroid @xmath44 . in the next section , we will see the additional properties of @xmath312 coming from the @xmath384-coring structure of @xmath44 . on a different note , let us give a complete geometric description of the @xmath384-ring structure of @xmath44 . [ p4.9 ] denote by @xmath553 , @xmath554 and @xmath555 . the product of @xmath528 takes the form @xmath556 for all @xmath525 . this follows immediately from equation ( [ eq4.2 ] ) . notice the resemblance of this formula to the one for matrix multiplication . this should remind the reader of an example we discussed in section ( [ s3.2 ] ) . one can view a @xmath3-ring to be a `` matrix '' of vector spaces whose entries are indexed by @xmath290 and what sits in entry @xmath474 is the vector space @xmath499 . as we have defined after proposition ( [ p4.4 ] ) , the vector space @xmath499 is the zero vector space if @xmath557 . for matrix algebras @xmath334 , @xmath84 would be an @xmath404-element set and the vector spaces @xmath499 would all be @xmath558 . there are a plethora of algebraic structures package into a bialgebroid let alone in a hopf algebroid . before we end this section , let us take a detour to describe the relationships among the structures of @xmath46 : being a @xmath558-algebra , the @xmath384-ring and the @xmath559-ring structures being a left - bialgebroid over @xmath560 . for the purpose of this discussion , let us denote by @xmath561 the @xmath558-algebra structure of @xmath46 and recall that @xmath562 and @xmath563 denote the relevant @xmath384-ring and @xmath559-ring structures of @xmath46 , respectively . as we mentioned in section ( [ s2.1 ] ) , for a @xmath8-algebra @xmath4 , @xmath4-ring structures are in bijection with @xmath8-algebra maps @xmath564 . thus , the complex algebra structure of @xmath46 is uniquely determined by the unit map @xmath565 . similarly , the @xmath384-ring and the @xmath559-ring structures are determined by the @xmath558-linear maps @xmath79 and @xmath566 . these maps satisfy the following commutativity relations . @xmath567 ^ -{\eta_{\mathbb{c } } } \ar[dd ] & & h\\ & & \\ a \ar[rr ] \ar[rruu]|-{s_{l } } & & a^{e } \ar[uu]_-{\eta_{l } } } \hspace{.5 in } \xymatrix{h\otimes h \ar@{->>}[rr ] \ar[rrdd]_-{\mu_{\mathbb{c } } } & & h{\mathbin{\mathop{\otimes}\limits_{a}}}h \ar@{->>}[rr ] \ar[dd]_-{\mu_{l } } & & h{\mathbin{\mathop{\otimes}\limits_{a^{e}}}}h . \ar[lldd]^-{\mu_{a^{e } } } \\ & & & & \\ & & h & & } \ ] ] in terms of the local eigenspace decomposition , the map @xmath568 induces maps @xmath569 & & e_{(x , y ) } } \ ] ] while , by ( [ eq4.2 ] ) , we have maps @xmath570 & & e_{(x , y ) } } .\ ] ] on the other hand , because the @xmath461-bimodule structure of @xmath46 is given as follows , @xmath571 for any @xmath572 , @xmath573 , and @xmath267 , the product @xmath574 induces maps @xmath575 & & e_{(x , x ) } } .\ ] ] another way of seeing this is by noting that the product @xmath52 uses the tensor product @xmath576 which kills products @xmath577 & & e_{(x , y ) } } $ ] for which @xmath578 . likewise , the tensor product @xmath579 kills products @xmath580 & & e_{(x , y ) } } $ ] for which @xmath511 . in this section , using the techniques and results we have developed in sections ( [ s4.1 ] ) and ( [ s4.2 ] ) we will describe what the coring structure of @xmath44 contributes to the geometry of @xmath455 . we will keep the notations of the previous two sections . the @xmath3-bimodule structure of the underlying @xmath384-coring structure of @xmath44 is related to the @xmath3-bimodule structure of the underlying @xmath384-ring via @xmath581 for @xmath459 , @xmath458 , and @xmath267 . the left - hand side of equation ( [ eq4.3 ] ) concerns the bimodule structure one has for the underlying @xmath384-coring of @xmath44 while the right - hand side concerns its @xmath384-ring structure . this , in particular , implies that if we run the construction we have in section ( [ s4.1 ] ) for the bimodule structure of the @xmath384-coring of @xmath44 , we will get the same sheaf @xmath455 supported over the same closed subset @xmath430 . the coproduct @xmath81 of @xmath44 , @xmath582 , uses a different @xmath384-bimodule structure from the @xmath384-bimodule structure involved in the @xmath384-ring structure . thus , @xmath583 means different from the @xmath583 we have in the product @xmath533 . with this , let us denote by @xmath584 this new tensor product . thus , we have @xmath585 ^ -{\delta_{l } } & & \gamma(z,\mathcal{e})\boxtimes_{c(x)}\gamma(z,\mathcal{e } ) } .\ ] ] however , using the relation ( [ eq4.3 ] ) the codomain of @xmath81 can be expressed as @xmath586 where @xmath587 is the sheaf of vector spaces whose fiber at @xmath588 is @xmath589 using the same argument we used in the previous section , the map @xmath81 induces a map @xmath590 of sheaves over @xmath430 . over point a @xmath469 , we have a map @xmath591 ^ -{\left(\delta_{l}\right)_{*}^{(x , y ) } } & & \bigoplus\limits_{z',z''\in x}\left(e_{(x , z')}\otimes e_{(x , z'')}\right ) } \ ] ] meanwhile , the counit @xmath592 induces a map @xmath593 of sheaves over @xmath430 and @xmath522 , respectively . here , @xmath594 is the map @xmath595 for any @xmath469 and @xmath596 is the subsheaf of @xmath455 where the fiber of @xmath596 at @xmath474 is @xmath597 unless @xmath370 , to which the fiber is @xmath558 viewed as the one - dimensional subalgebra of @xmath541 spanned by its unit @xmath598 . hence , over a point @xmath469 we have @xmath599 . counitality of @xmath81 with respect to @xmath68 implies that for fixed but arbitrary @xmath267 we have @xmath600 ^ -{\bigoplus\limits_{z^{'}}id\otimes\left(\epsilon_{l}\right)_{*}^{(x , z^ { ' } ) } } \\ & & \\ & & \\ & & \\ & & \\ e_{(x , y ) } \ar[rruuuuu]^-{\left(\delta_{l}\right)_{*}^{(x , y ) } } \ar@{=}[rr ] & & \bigoplus\limits_{z}\left(e_{(x , z)}\otimes\mathbb{c}\right ) \\ v \ar@{|->}[rr ] & & v\otimes 1 \\ } \hspace{.15 in } \xymatrix@r=2 mm { & & \bigoplus\limits_{z , z^{'}}\left(e_{(x , z)}\otimes e_{(x , z^{'})}\right ) \ar[ddddd]^-{\bigoplus\limits_{z}\left(\epsilon_{l}\right)_{*}^{(x , z)}\otimes i d } \\ & & \\ & & \\ & & \\ & & \\ e_{(x , y ) } \ar[rruuuuu]^-{\left(\delta_{l}\right)_{*}^{(x , y ) } } \ar@{=}[rr ] & & \bigoplus\limits_{z^{'}}\left(\mathbb{c}\otimes e_{(x , z^{'})}\right ) \\ v \ar@{|->}[rr ] & & 1\otimes v \\}\ ] ] the bottom isomorphisms imply that @xmath601 and @xmath602 are nonzero maps for @xmath603 and @xmath604 . since @xmath268 and @xmath269 are arbitrary to start with , we have the following proposition . [ p4.10 ] for any @xmath469 , we have @xmath605 . another thing we can infer from the diagrams ( [ eq4.6 ] ) , using the isomorphisms in the bottom and the fact that @xmath269 is among the @xmath606 and @xmath607 that appears as indices , is that the image of @xmath608 is contained in @xmath609 we will show in the next section that more can be said . in fact , the image of @xmath608 is completely contained in @xmath610 . in this section , we will complete our description of the geometry of the hopf algebroid @xmath42 over @xmath3 . in doing so , we will be able to illustrate the main point of this article . that to such a hopf algebroid , one can associate a highly structured category . so far , we have considered only the constituent left bialgebroid @xmath44 of @xmath42 . running the arguments we have presented in sections ( [ s4.1 ] ) and ( [ s4.2 ] ) for @xmath45 , we see that there is a sheaf of vector spaces @xmath596 over @xmath290 such that @xmath611 . let us denote by @xmath612 the support of @xmath45 under the isomorphism @xmath611 . the following proposition relates these two sheaves . [ p4.11 ] for @xmath613 and @xmath611 as @xmath3-bimodules as constructed in sections ( [ s4.1 ] ) and ( [ s4.2 ] ) , where @xmath455 and @xmath596 are sheaves of vector spaces supported on @xmath614 , we have 1 . @xmath615 . 2 . @xmath616 as sheaves over @xmath430 . condition ( c ) of the definition of a hopf algebroid implies that the antipode @xmath56 of @xmath42 flips the @xmath3-bimodule structure used for the @xmath3-ring structure of @xmath44 to that of the @xmath3-bimodule structure used for the @xmath3-ring structure of @xmath45 . likewise , @xmath56 flips the bimodule structures of the underlying @xmath3-coring structures of @xmath44 and @xmath45 . in particular , this tells us that @xmath56 induces a map @xmath617 which on fibers does @xmath618 for any @xmath469 . symmetrically , we also have a map denoted the same , @xmath619 , which on fibers does @xmath620 for any @xmath621 . this proves proposition ( [ p4.11 ] ) . @xmath622 in view of proposition ( [ p4.11 ] ) , we have @xmath623 . similar to equation ( [ eq4.5 ] ) , @xmath82 induces maps @xmath624 ^ -{\left(\delta_{r}\right)_{*}^{(x , y ) } } & & \bigoplus\limits_{z',z''\in x}\left(e_{(z',y)}\otimes e_{(z'',y)}\right ) } \ ] ] for @xmath469 . as we promised at the end of section ( [ s4.3 ] ) , @xmath608 maps @xmath499 into @xmath610 , for any @xmath469 . same holds for @xmath625 . let us summarize these statements into the following proposition . [ p4.12 ] for every @xmath469 , 1 . @xmath499 is a coalgebra with coproduct @xmath608 and counit @xmath626 , and 2 . @xmath499 is a coalgebra with coproduct @xmath625 and counit @xmath627 . we will only prove part @xmath628 . the proof for part @xmath629 is similar . the second commutation relation of @xmath81 and @xmath82 in part @xmath134 of the definition ( [ d2.2 ] ) gives the following diagram @xmath630 ^ -{\left(\delta_{l}\right)_{*}^{(x , y ) } } \ar[ddd]_-{\left(\delta_{r}\right)_{*}^{(x , y ) } } & & & & \bigoplus\limits_{z^{'},z^{''}}\left(e_{(x , z^{'})}\otimes e_{(x , z^{''})}\right ) \ar[dd]^-{\bigoplus\limits_{z^{'}}\left(\delta_{r}\right)_{*}^{(x , z^{'})}\otimes id}\\ & & & & \\ & & & & \bigoplus\limits_{z^{'},z^{''}}\bigoplus\limits_{\alpha^{'},\alpha^{''}}\left(e_{(\alpha^{'},z^{'})}\otimes e_{(\alpha^{''},z^{'})}\otimes e_{(x , z^{''})}\right ) \ar@{=}[d ] \\ \bigoplus\limits_{\beta^{'},\beta^{''}}\left(e_{(\beta^{'},y)}\otimes e_{(\beta^{''},y)}\right ) \ar[rrrr]_-{\bigoplus\limits_{\beta^{''}}id\otimes\left(\delta_{l}\right)_{*}^{(\beta^{''},y ) } } & & & & \bigoplus\limits_{\beta^{'},\beta^{''}}\bigoplus\limits_{\gamma^{'},\gamma^{''}}\left(e_{(\beta^{'},y)}\otimes e_{(\beta^{''},\gamma^{'})}\otimes e_{(\beta^{''},\gamma^{''})}\right)\\ } \ ] ] for fixed but arbitrary @xmath469 . in the composite @xmath631 , the third leg lands in @xmath632 . on the other hand , the third leg of the composite @xmath633 lands in @xmath634 . this implies that for @xmath635 , @xmath636 . from our last statement in section ( [ s4.3 ] ) , @xmath637 is contained in @xmath638 counitality of @xmath81 with respect to @xmath68 , implemented locally by diagram ( [ eq4.6 ] ) , gives @xmath639 ^ -{\cong } & \bigoplus\limits_{f^{'}}\left(id\otimes\left(\epsilon_{l}\right)_{*}^{(\beta^{''},f^{'})}\right)\left(\delta_{l}\right)_{*}^{(\beta^{''},y)}(e_{(\beta^{''},y ) } ) \ar@{=}[r ] & \left\{0\right\}. \\ v \ar@{|->}[r ] & v\otimes 1 & \\}\ ] ] by assumption , @xmath640 are nontrivial . this is a contradiction unless the summands corresponding to @xmath635 of the direct sum in the lower left corner of diagram ( [ eq4.8 ] ) do not intersect the image of @xmath625 . using the first commutation relation in part @xmath134 of the definition ( [ d2.2 ] ) , we get a diagram similar to diagram ( [ eq4.8 ] ) . inspecting that resulting diagram tells us that the image of @xmath625 does not intersect those summands of the direct sum in the lower left corner of diagram ( [ eq4.8 ] ) corresponding to @xmath641 . this shows that , indeed , @xmath642 the coassociativity of @xmath625 follows from coassociativity of @xmath82 and its counitality with respect to @xmath627 follows from counitality of @xmath82 with respect to @xmath67 . this proves part @xmath628 of the above proposition . exchanging the roles of @xmath81 and @xmath82 with minor modifications proves part @xmath629 . @xmath622 following the arguments in sections ( [ s4.1 ] ) , ( [ s4.2 ] ) and ( [ s4.3 ] ) for @xmath45 , we see that we can similarly associate a category @xmath313 enriched over @xmath263 . denoting by @xmath287 by the category of coalgebras on @xmath263 , we have the following proposition . [ p4.13 ] the categories @xmath312 and @xmath313 are enriched over @xmath287 . these categories are strongly related . by proposition ( [ p4.11 ] ) , we have the following corollary . [ p4.14 ] the @xmath287-enriched categories @xmath312 and @xmath313 have isomorphic underlying @xmath263-enriched categories . note that the underlying @xmath263-enriched category of @xmath312 and @xmath313 only depends on the @xmath3-ring structures of @xmath44 and @xmath45 , respectively . another way to prove corollary ( [ p4.14 ] ) is to use the fact that @xmath44 and @xmath45 have the isomorphic @xmath3-ring structures . to see why @xmath44 and @xmath45 have isomorphic @xmath3-ring structures , note that the source map of @xmath44 is the target map of @xmath44 while the target map of @xmath44 is the source map of @xmath45 . in the general definition of a hopf algebroid , one can either use the source or the target map to select a particular ring structure to consider , see for example @xcite . using the general fact that for a general @xmath8-algebra @xmath4 , @xmath4-rings @xmath5 corresponds uniquely to @xmath8-algebra maps @xmath280 , we see that @xmath44 and @xmath45 are isomorphic as @xmath3-rings . [ r4.5 ] another way to see why @xmath44 and @xmath45 are isomorphic as @xmath3-rings is the fact that general hopf algebroids @xmath42 with bijective antipode over a _ commutative _ ring @xmath643 is a coupled @xmath643-hopf algebra . unlike the ring structures , the @xmath3-coring structures of @xmath44 and @xmath45 can vary wildly as illustrated by coupled hopf algebras . this implies that the @xmath287-enrichments @xmath312 and @xmath313 need not be isomorphic . however , they form a topological coupled hopf category . the coupling functor is the one induced by the antipode @xmath56 of the hopf algebroid @xmath42 . we formalize this in the following theorem . [ t4.1 ] given a finitely - generated projective hopf algebroid @xmath42 over @xmath3 with bijective antipode , one can associate a topological coupled hopf category @xmath274 via the construction we presented in sections ( [ s4.1 ] ) and ( [ s4.2 ] ) . conversely , to any topological coupled hopf category @xmath274 , the space of sections @xmath644 of the associated sheaf @xmath46 of @xmath274 is a hopf algebroid over @xmath3 . the proof of the first statement is basically the breadth of section ( [ s4.0 ] ) . for the second statement , one can consider the bimodule structures presented in proposition ( [ p4.3 ] ) . the rest of the structures are given by the rest of the structure maps of @xmath274 . the above theorem is a generalization of the example in @xcite where they constructed out of a @xmath8-linear category with finitely many objects a weak hopf algebra . the theorem not only recovers an inverse to the construction they presented but it also work for weak hopf algebra as long as the subalgebra spanned by the left and the right units are commutative . the above theorem is the generalization of the example we discussed in section ( [ s3.2 ] ) . although @xmath3 is commutative , it may not be central in @xmath42 . let us look at the special case when @xmath3 is central in @xmath42 , by which we mean that the images of the source and target maps are central in the relevant @xmath3-ring structure of @xmath42 . for simplicity , we will blur the disctinction between @xmath3 at its images under the source maps of @xmath42 . let us consider first the constituent left bialgebroid @xmath44 of @xmath42 . by proposition ( [ p4.7 ] ) , @xmath44 is supported along the diagonal @xmath645 . this means that the sheaf @xmath455 coincides with the vector bundle @xmath543 . we can simply identify the diagonal @xmath522 with @xmath84 . with this , the multiplication @xmath52 in @xmath44 via the identification @xmath387 is pointwise , i.e. the fibers of the vector bundle @xmath543 are ( possibly nonisomorphic ) unital complex alegrbas @xmath646 , where @xmath647 and @xmath648 are the maps induced by @xmath52 and @xmath79 on the fiber @xmath440 . by proposition ( [ p4.12 ] ) , the coproduct @xmath81 and counit @xmath68 of @xmath44 also descends into a coproduct @xmath649 and a counit @xmath650 for the fibers @xmath440 , @xmath651 , making them coalgebras . using condition @xmath134 in the definition of a bialgebroid , we see that @xmath649 is multiplicative for any @xmath651 . meanwhile , using condition @xmath305 of the definition of a bialgebroid we see that @xmath650 is multiplicative for any @xmath651 . this gives us the following proposition . [ p4.15 ] if @xmath3 is central in @xmath44 , then for any @xmath651 , @xmath652 is a bialgebra . moreover , the bialgebroid @xmath44 is a bundle of bialgebras via @xmath387 . similar statement holds for the constituent right bialgebroid @xmath45 . since for very @xmath651 the maps @xmath653 and @xmath654 induced by the source maps @xmath79 and @xmath77 are the same , the multiplications @xmath655 and @xmath656 coincide . assuming mild nondegeneracy conditions for @xmath657 and @xmath658 , we get the following proposition . [ p4.16 ] let @xmath43 be a hopf algebroid over @xmath560 where @xmath384 is central in both @xmath44 and @xmath45 . denote by @xmath46 the underlying complex algebra of @xmath42 . suppose that the maps @xmath659 ^ -{\mathfrak{gal}_{l } } & & h{\mathbin{\mathop{\otimes}\limits_{a}}}h \\ a{\mathbin{\mathop{\otimes}\limits_{a}}}b \ar@{|->}[rr ] & & ab_{[1]}{\mathbin{\mathop{\otimes}\limits_{a}}}b_{[2]}\\ } , \hspace{.5 in } \xymatrix@r=2mm{h{\mathbin{\mathop{\otimes}\limits_{a}}}h \ar[rr]^-{\mathfrak{gal}_{r } } & & h{\mathbin{\mathop{\otimes}\limits_{a}}}h \\ a{\mathbin{\mathop{\otimes}\limits_{a}}}b \ar@{|->}[rr ] & & ab^{[1]}{\mathbin{\mathop{\otimes}\limits_{a}}}b^{[2]}\\}\ ] ] are bijections . then 1 . @xmath46 is a coupled hopf algebra with constituent hopf algebras @xmath44 and @xmath45 and coupling map @xmath56 . 2 . each fiber @xmath440 is a hopf algebra and @xmath387 as hopf algebras , where the structure maps of @xmath388 are all pointwise . same is true for @xmath45 . 3 . @xmath42 is a bundle of coupled hopf algebras over @xmath84 such that the constituent hopf algebras at a point @xmath651 are the fiber hopf algebras of @xmath44 and @xmath45 . centrality of @xmath384 in both @xmath44 and @xmath45 implies that @xmath44 and @xmath45 are in fact bialgebras over @xmath384 ( not just bialgebroids ) . the nondegeneracy conditions assumed in the proposition implies that @xmath46 is a galois extension for both bialgebras @xmath44 and @xmath45 . by @xcite , the bialgebras @xmath44 are @xmath45 are in fact hopf algebras , i.e. the identity maps @xmath660 and @xmath661 are invertible in the respective convolution algebras associated to the bialgebras @xmath44 and @xmath45 . the rest of the conditions for @xmath42 to be a hopf algebroid imply that @xmath44 and @xmath45 are coupled hopf algebras with coupling map @xmath56 , the antipode of @xmath42 . this proves part @xmath662 . to prove part @xmath663 , we argue that the maps @xmath259 and @xmath256 are @xmath384-bimodule maps . thus , there descend into fiberwise bijections . using the same argument we did for part @xmath662 , see that the fibers are coupled hopf algebras . part @xmath664 readily follows from the proofs of parts @xmath662 and @xmath663 . in this section , we will see that the correspondence between hopf algebroids and coupled hopf categories we established in theorem ( [ t4.1 ] ) persists to their corresponding galois theories . to be precise , we will prove the following theorem . [ t5.1 ] let @xmath43 be a hopf algebroid over @xmath560 for some compact hausdorff space @xmath84 . let @xmath274 be the corresponding topological coupled hopf category of @xmath42 . then @xmath42-galois extensions of @xmath384 corresponds bijectively to @xmath274-galois extensions of @xmath272 . before proving the above theorem , let us comment on what we mean by galois extension by a ( topological ) coupled hopf category @xmath383 . by this , we mean an inclusion of categories @xmath665 which is simultaneously @xmath312-galois and @xmath313-galois in the sense of section ( [ s3.3 ] ) . note that by definition ( [ d3.3 ] ) , we are not requiring @xmath312 and @xmath313 to be hopf categories ( individually , they are only @xmath287-enriched categories ) . in particular , they do not necessarily have antipodes . fortunately , galois extension in the sense described in section ( [ s3.3 ] ) does not really make use of the antipode . let @xmath18 be a ( left ) @xmath42-galois extension of @xmath384 . in particular , @xmath18 is an @xmath384-ring . note that the arguments we used in sections ( [ s4.1 ] ) and ( [ s4.2 ] ) only use the @xmath384-ring structure of the hopf algebroid @xmath42 . using the same arguments , @xmath666 where @xmath667 is a sheaf of vector spaces over @xmath290 . by the galois condition , we see that @xmath667 has the same support @xmath466 as the sheaf @xmath455 we get from either @xmath44 or @xmath45 . similar to remark ( [ r4.4 ] ) , we get a small category @xmath265 over @xmath84 enriched over @xmath263 whose associated sheaf is @xmath667 . the ( right ) @xmath44-coaction @xmath668 induces a map @xmath669 of sheaves of @xmath288-bimodules over @xmath290 . by definition , @xmath18 is a right @xmath384-module and a right @xmath670-module . using this , the @xmath384-bimodule structure on @xmath18 is as follows : @xmath677 ^ -{\left(\rho_{l}\right)_{*}^{(x , y ) } } \ar[ddd]_-{\left(\rho_{r}\right)_{*}^{(x , y ) } } & & & & \bigoplus\limits_{z^{'},z^{''}}\left(b_{(x , z^{'})}\otimes e_{(x , z^{''})}\right ) \ar[dd]^-{\bigoplus\limits_{z^{'}}\left(\rho_{r}\right)_{*}^{(x , z^{'})}\otimes id}\\ & & & & \\ & & & & \bigoplus\limits_{z^{'},z^{''}}\bigoplus\limits_{\alpha^{'},\alpha^{''}}\left(b_{(\alpha^{'},z^{'})}\otimes e_{(\alpha^{''},z^{'})}\otimes e_{(x , z^{''})}\right ) \ar@{=}[d ] \\ \bigoplus\limits_{\beta^{'},\beta^{''}}\left(b_{(\beta^{'},y)}\otimes e_{(\beta^{''},y)}\right ) \ar[rrrr]_-{\bigoplus\limits_{\beta^{''}}id\otimes\left(\delta_{l}\right)_{*}^{(\beta^{''},y ) } } & & & & \bigoplus\limits_{\beta^{'},\beta^{''}}\bigoplus\limits_{\gamma^{'},\gamma^{''}}\left(b_{(\beta^{'},y)}\otimes e_{(\beta^{''},\gamma^{'})}\otimes e_{(\beta^{''},\gamma^{''})}\right)\\ } \ ] ] @xmath678 ^ -{\bigoplus\limits_{z^{''}}id\otimes\left(\epsilon_{l}\right)_{*}^{(z , y ) } } \\ & & \\ & & \\ & & \\ & & \\ b_{(x , y ) } \ar[rruuuuu]^-{\left(\rho_{l}\right)_{*}^{(x , y ) } } \ar@{=}[rr ] & & \bigoplus\limits_{z^{'}}\left(b_{(x , z^{'})}\otimes\mathbb{c}\right ) \\ v \ar@{|->}[rr ] & & v\otimes 1 \\}\ ] ] these tell us that @xmath265 is a right @xmath312- and a right @xmath313-comodule . the composition @xmath279 in @xmath265 is induced by the @xmath384-product on @xmath18 . by equations ( [ eq2.2 ] ) to ( [ eq2.5 ] ) , this composition @xmath279 is a map of right @xmath312- and a right @xmath313-modules . thus , @xmath265 is a right @xmath312- and a right @xmath313-comodule - category . it is not hard to see that the right coactions of @xmath312 and @xmath313 on @xmath265 are both galois whose subcategories of coinvariants are both the same as @xmath372 . these imply that @xmath265 is a galois extension of @xmath372 by the topological coupled hopf category @xmath383 . the inverse of this correspondence is easily seen as the the one that associates to an @xmath680-galois extension @xmath681 the @xmath109-galois extension @xmath252 where @xmath682 and @xmath384 are the space of global sections of the associated sheaves to @xmath683 and @xmath372 , respectively . the compatibility conditions in the categorical side precisely correspond to the analogous compatibility conditions in the algebraic side . @xmath622

nature is filled with fluid flow , from small scales to large scales . our daily life on a human scale constantly experiences currents of air , water , and the like . the earth sustains the ocean flows and atmospheric circulation , leading to an abundance of life . in space , stars and galaxies flow , forming many large - scale structures . most flow in nature is actually turbulent . as the velocity increases , flow generally changes from laminar to turbulent . turbulence has been a mystery for a long time . it has been investigated not only in basic science , including physics and mathematics research , but also in applied sciences , such as fluid engineering and aeronautics . as feynman said , turbulence is a problem `` that is common to many fields , that is very old , and that has not been solved . '' @xcite this is chiefly because turbulence is a complicated dynamical phenomenon with strong nonlinearity far from an equilibrium state . `` turbulence '' reminds one of some sketches by leonardo da vinci . he observed the turbulent flow of water and drew pictures showing that turbulence has a structure comprised of vortices of different sizes ( fig . vortices may be a key to understanding turbulence . however , vortices are not well defined for a classical fluid , and the relationship between turbulence and vortices remains unclear . independently of these studies in classical fluid dynamics , quantum fluids such as superfluid helium and atomic bose - einstein condensates ( becs ) have been investigated . these systems are subject to quantum restrictions ; the appearance of order parameters makes the rotational motion exist only in the presence of quantized vortices . a quantized vortex is a stable topological defect with quantized circulation . such vortices give rise to quantum turbulence ( qt ) . since quantized vortices are well defined as elements composing a turbulent flow , qt is expected to be an easier system to study than classical turbulence ( ct ) and has a simpler model of turbulence . this article addresses recent developments in quantum fluid dynamics and qt in superfluid helium and atomic becs . in sec . 2 , basic concepts such as superfluidity , bose - einstein condensation , and quantized vortices are briefly reviewed . section 3 is devoted to quantized vortices in a rotating bec , addressing typical phenomena of quantum hydrodynamics . section 4 describes present important topics of qt in superfluid helium . applying these ideas , qt in atomic becs is discussed in sec . section 6 is devoted to conclusions . quantum fluid dynamics has been studied in connection with superfluid helium for over 50 years . liquid @xmath0he enters a superfluid state below the @xmath1 point ( 2.17 k ) through bose einstein condensation of the atoms @xcite . the characteristics of superfluidity were discovered experimentally in the 1930s by kapitza _ the hydrodynamics of superfluid helium is well described by the two - fluid model , in which the system consists of an inviscid superfluid ( of density @xmath2 ) and a viscous normal fluid ( of density @xmath3 ) with two independent velocity fields @xmath4 and @xmath5 . the mixing ratio of the two fluids depends on temperature . as the system is cooled down below the @xmath1 point , the ratio of the superfluid component increases , and the entire fluid becomes superfluid below about 1 k. the bose - condensed system exhibits the macroscopic wavefunction @xmath6 as an order parameter . the superfluid velocity field representing the potential flow is given by @xmath7 with boson mass @xmath8 . since the macroscopic wavefunction must be single - valued for space coordinate @xmath9 , the circulation @xmath10 for an arbitrary closed loop within the fluid is quantized in terms of the quantum value @xmath11 . a vortex with such quantized circulation is called a quantized vortex . any rotational motion of a superfluid is only sustained by quantized vortices . a quantized vortex is a topological defect characteristic of a bose einstein condensate and is different from a vortex in a classical viscous fluid . first , its circulation is quantized and conserved , in contrast to a classical vortex whose circulation can have any value and is not conserved . second , a quantized vortex is a vortex of inviscid superflow . thus , it can not decay by the viscous diffusion of vorticity that occurs in a classical fluid . third , the core of a quantized vortex is very thin , on the order of the coherence length , which is only a few angstroms in superfluid @xmath0he and submicron in atomic becs . since the vortex core is thin and does not decay by diffusion , it is possible to identify the position of a quantized vortex in the fluid . these properties make a quantized vortex more stable and definite than a classical vortex . early experimental studies of superfluid turbulence focused primarily on thermal counterflow , in which the normal fluid and superfluid flow in opposite directions . the flow is driven by an injected heat current , and it is found that the superflow becomes dissipative when the relative velocity between the two fluids exceeds a critical value @xcite . feynman proposed that it is a superfluid turbulent state consisting of a tangle of quantized vortices @xcite . vinen later confirmed feynman s proposal experimentally by showing that the dissipation arises from mutual friction between vortices and the normal flow @xcite . the quantization of circulation was also observed by vinen using a vibrating wire technique @xcite . subsequently , many experimental studies have examined superfluid turbulence ( st ) in thermal counterflow systems , revealing a variety of physical phenomena @xcite . the dynamics of quantized vortices are nonlinear and nonlocal , so it has not been easy to quantitatively understand the experimental results on the basis of vortex dynamics . a breakthrough was achieved by schwarz , who clarified the picture of st consisting of tangled vortices by a numerical simulation of the quantized vortex filament model in the thermal counterflow @xcite . however , since the thermal counterflow has no analogy in conventional fluid dynamics , these experimental and numerical studies are not helpful in clarifying the relationship between st and classical turbulence ( ct ) . superfluid turbulence is often called quantum turbulence ( qt ) , which emphasizes the fact that it is comprised of quantized vortices . comparing qt and ct reveals definite differences . turbulence in a classical viscous fluid appears to be comprised of vortices , as pointed out by da vinci . however , these vortices are unstable , repeatedly appearing and disappearing . moreover , the circulation is not conserved and is not identical for each vortex . quantum turbulence consists of a tangle of quantized vortices that have the same conserved circulation . looking back at the history of science , _ reductionism _ , which tries to understand the nature of complex things by reducing them to the interactions of their parts , has played an extremely important role . the success of solid state physics owes much to _ reductionism_. in contrast , conventional fluid physics is not reducible to elements , and thus does not enjoy the benefits of _ reductionism_. however , quantum turbulence is different , being reduced to quantized vortices . thus _ reductionism _ is applicable to quantum turbulence . consequently , qt should lead to a simpler model of turbulence than ct . based on these considerations , research into quantum fluid dynamics has opened up new directions since the mid 1990s . one new direction has occurred in the field of low temperature physics by studying superfluid helium . it started with the attempt to understand the relationship between qt and ct @xcite . the energy spectrum of fully developed classical turbulence is known to obey the kolmogorov law in the inertial range . recent experimental and numerical studies support a kolmogorov spectrum in qt . following these studies , qt research on superfluid helium has moved to important topics such as the dissipation process at very low temperatures , qt created by vibrating structures , and visualization of qt @xcite . another new direction is the realization of bose - einstein condensation in trapped atomic gases in 1995 , which has stimulated intense experimental and theoretical activity @xcite . as proof of the existence of superfluidity , quantized vortices have been created and observed in atomic becs , and numerous efforts have been devoted to a number of fascinating problems @xcite . atomic becs have several advantages over superfluid helium . the most important is that modern optical techniques enable one to directly control condensates and visualize quantized vortices . because a series of experiments on becs clearly show the properties of quantum fluid dynamics @xcite , we shall start with this topic . what happens if we rotate a cylindrical vessel with a classical viscous fluid inside ? even if the fluid is initially at rest , it starts to rotate and eventually reaches a steady rotation with the same rotational speed as the vessel . in that case , one can say that the system contains a vortex that mimics solid - body rotation . a rotation of arbitrary angular velocity can be sustained by a single vortex . however , this does not occur in a quantum fluid . because of quantization of circulation , superfluids respond to rotation , not with a single vortex , but with a lattice of quantized vortices . feynman noted that in uniform rotation with angular velocity @xmath12 the curl of the superfluid velocity is the circulation per unit area , and since the curl is @xmath13 , a lattice of quantized vortices with number density @xmath14 ( `` feynman s rule '' ) arranges itself parallel to the rotation axis @xcite . such experiments were performed for superfluid @xmath0he : packard _ et al . _ visualized vortex lattices on the rotational axis by trapping electrons along the cores @xcite . this idea has also been applied to atomic becs . several groups have observed vortex lattices in rotating becs @xcite . among them , madison _ et al_. directly observed nonlinear processes such as vortex nucleation and lattice formation in a rotating @xmath15rb bec @xcite . by sudden application of a rotation along the trapping potential , an initially axisymmetric condensate undergoes a collective quadrupole oscillation to an elliptically deformed condensate . this oscillation continues for a few hundred milliseconds with gradually decreasing amplitude . then the axial symmetry of the condensate is recovered and vortices enter the condensate through its surface , eventually settling into a lattice configuration . this observation has been reproduced by a simulation of the gross - pitaevskii ( gp ) equation for the macroscopic wavefunction @xmath6 in two - dimensional @xcite and three - dimensional @xcite spaces . the corresponding gp equation in a frame rotating with frequency @xmath16 is given by @xmath17 \psi({\bf x},t ) . \label{gpeqrot}\ ] ] here @xmath18 is a trapping potential , and @xmath19 is the angular momentum along the rotational axis . the interparticle potential @xmath20 is approximated by a short - range interaction @xmath21 , where @xmath22 is a coupling constant characterized by the s - wave scattering length @xmath23 . the term @xmath24 indicates phenomenological dissipation . in this simulation @xcite . ] figure 2 shows the profile of the condensate density @xmath25 and the phase @xmath26 when there is a quantized vortex in a trapped bec . the density has a hole representing the vortex core . the phase has a branch cut between 0 and @xmath27 , and the edge of the branch cut corresponds to the vortex core around which the phase rotates by @xmath27 as the superflow circulates . one can therefore clearly identify the vortex both in the density and the phase . a typical two - dimensional numerical simulation of eq . ( [ gpeqrot ] ) @xcite for the vortex lattice formation is shown in fig . 3 , where the condensate density and the phase are displayed together . the trapping potential is @xmath28,\ ] ] where @xmath29 hz , and the parameters @xmath30 and @xmath31 describe small deviations from axisymmetry corresponding to experiments @xcite . we first prepare an equilibrium condensate trapped in a stationary potential ; the size of the condensate cloud is determined by the thomas - fermi radius @xmath32 . when we apply a rotation with @xmath33 , the condensate becomes elliptic and performs a quadrupole oscillation [ fig . 3(a ) ] . then , the boundary surface of the condensate becomes unstable and generates ripples that propagate along the surface [ fig . as stated previously , it is possible to identify quantized vortices in the phase profile also . as soon as the rotation starts , many vortices appear in the low - density region outside of the condensate [ fig . 3(a ) ] . since quantized vortices are excitations , their nucleation increases the energy of the system . because of the low density in the outskirts of the condensate , however , their nucleation contributes little to the energy and angular momentum . since these vortices outside of the condensate are not observed in the density profile , they are called `` ghost vortices '' . their movement toward the thomas - fermi surface excites ripples [ fig . it is not easy for these ghost vortices to enter the condensate , because that would increase both the energy and angular momentum . only some vortices enter the condensate cloud to become `` real vortices '' wearing the usual density profile of quantized vortices [ fig . 3(d ) ] , eventually forming a vortex lattice [ fig . 3 ( e ) and ( f ) ] . the number of vortices forming a lattice is given by `` feynman s rule '' @xmath34 . the numerical results agree quantitatively with these observations . note the essence of the dynamics . the initial state has no vortices in the absence of rotation . the final state is a vortex lattice corresponding to rotational frequency @xmath12 . in order to go from the initial to the final state , the system makes use of as many excitations as possible , such as vortices , quadruple oscillation , and surface waves . we refer the readers to ref . 26 for details . these experimental and theoretical results demonstrate typical behavior of quantum fluid dynamics in atomic becs . 63 ms , ( b ) 305 ms , ( c ) 350 ms , ( d ) 410 ms , ( e ) 450 ms , and ( f ) 850 ms after the start of the rotation . the phase varies continuously from 0 ( red ) to @xmath27 ( blue ) . ] this section reviews research on qt since the mid 1990s . before considering qt , we begin with classical fluid dynamics and statistical properties of ct @xcite . classical viscous fluid dynamics is described by the navier stokes equation , @xmath35 where @xmath36 is the velocity of the fluid , @xmath37 is the pressure , @xmath38 is the density of the fluid , and @xmath39 is the kinematic viscosity . the flow can be characterized by the ratio of the second term on the left - hand side of eq . ( [ eq - navier - stokes ] ) , referred to as the inertial term , and the second term on the right - hand side , called the viscous term . this ratio defines the reynolds number @xmath40 , where @xmath41 and @xmath42 are a characteristic velocity and length scale , respectively . when @xmath41 increases and the reynolds number exceeds a critical value , the system changes from a laminar to a turbulent state , in which the flow is complicated and contains eddies . such turbulent flow shows characteristic statistical behavior @xcite . we assume a steady state of fully developed turbulence of an incompressible classical fluid . energy is injected into the fluid at a rate @xmath43 in an energy - containing range . in an inertial range , this energy is transferred to smaller length scales without dissipation . in this range , the system is locally homogeneous and isotropic , leading to energy spectral statistics described by the kolmogorov law , @xmath44 the energy spectrum @xmath45 is defined by @xmath46 , where @xmath47 is the kinetic energy per unit mass and @xmath48 is the wavenumber from the fourier transform of the velocity field . the spectrum of eq . ( [ eq - kolmogorov ] ) is derived by assuming that @xmath45 is locally determined only by the energy flux @xmath43 and by @xmath48 . the energy transferred to smaller scales is dissipated at the kolmogorov wavenumber @xmath49 in an energy - dissipative range via the viscosity of the fluid at a dissipation rate @xmath43 in eq . ( [ eq - kolmogorov ] ) , which is equal to the energy flux @xmath50 in the inertial range . the kolmogorov constant @xmath51 is a dimensionless parameter of order unity . the kolmogorov spectrum has been confirmed experimentally and numerically in turbulence at high reynolds numbers . the inertial range is thought to be sustained by a self - similar richardson cascade in which large eddies break up into smaller eddies through vortex reconnections . in ct , however , the richardson cascade is not completely understood because it is impossible to definitely identify each eddy . as described in sec . 2 , most experimental studies on superfluid turbulence have examined thermal counterflow . however , the nonlinear and nonlocal dynamics of vortices delayed progress in the microscopic understanding of the vortex tangle . schwarz overcame these difficulties @xcite by developing a direct numerical simulation of vortex dynamics connected with dynamical scaling analysis , enabling the calculation of physical quantities such as the vortex line density , anisotropic parameters , and mutual friction force . the observable quantities calculated by schwarz agree with typical experimental results . two formulations are generally available for studying the dynamics of quantized vortices . one is the vortex filament model and the other is the gross pitaevskii ( gp ) model . as discussed in sec . 1 , a quantized vortex has quantized circulation . the vortex core is extremely thin , usually much smaller than other characteristic length scales of the vortex motion . these properties allow a quantized vortex to be represented as a vortex filament . in classical fluid dynamics , the vortex filament model is an idealization . however , the vortex filament model is accurate and realistic for a quantized vortex in superfluid helium . the vortex filament model represents a quantized vortex as a filament passing through the fluid , having a definite direction corresponding to its vorticity . except for the thin core region , the superflow velocity field has a classically well - defined meaning and can be described by ideal fluid dynamics @xcite . the velocity at point @xmath52 due to a filament is given by the biot savart expression , @xmath53 where @xmath54 is the quantum of circulation . the vortex moves with the superfluid velocity . at finite temperatures , mutual friction occurs between the vortex core and the normal flow . a numerical simulation method on this model has been described in detail elsewhere @xcite . a vortex filament is represented by a single string of points separated by distance @xmath55 . the vortex configuration at a given time determines the velocity field in the fluid , thus causing the vortex filaments to move . vortex reconnection needs to be included when simulating vortex dynamics . a numerical study of a classical fluid shows that close interaction of two vortices leads to their reconnection , chiefly because of the viscous diffusion of vorticity . schwarz assumed that two vortex filaments reconnect when they come within a critical distance of each other and showed that statistical quantities such as the vortex line density are not sensitive to how the reconnections occur @xcite . even after schwarz s study , it remained unclear as to whether quantized vortices can actually reconnect . however , koplik and levine directly solved the gp equation to show that two closely quantized vortices reconnect even in an inviscid superfluid @xcite . more recent simulations have shown that reconnections are accompanied by the emission of sound waves having wavelengths on the order of the healing length @xcite . schwarz numerically studied how remnant vortices develop into a statistical steady vortex tangle under thermal counterflow @xcite . the tangle is self - sustained by the competition between the excitation due to the applied flow and the dissipation by mutual friction . the numerical results were quantitatively consistent with experimental results , confirming feynman s original picture that qt is a tangle of quantized vortices . the following two quantities are characteristic of a vortex tangle . the line density @xmath56 is defined as the total length of vortex cores per unit volume . the mean spacing @xmath57 between vortices is given by @xmath58 . in a weakly interacting bose system , the macroscopic wavefunction @xmath59 appears as the order parameter of the bose einstein condensate , obeying the gross pitaevskii ( gp ) equation @xcite , @xmath60 writing @xmath61 , the squared amplitude @xmath62 is the condensate density and the gradient of the phase @xmath63 gives the superfluid velocity @xmath64 , corresponding to frictionless flow of the condensate . this relation causes quantized vortices to appear with quantized circulation . the only characteristic scale of the gp model is the coherence length defined by @xmath65 , which determines the vortex core size . the gp model can explain not only the vortex dynamics but also phenomena related to vortex cores , such as reconnection and nucleation . however , the gp equation is not applicable to superfluid @xmath0he , which is not a weakly interacting bose system . but it is applicable to bose einstein condensation of a dilute atomic bose gas @xcite , as discussed in sec . most older experimental studies on qt were devoted to thermal counterflow . since this flow has no classical analog , these studies did not significantly contribute to the understanding of the relation between ct and qt . since the mid 1990s , important experimental studies have been published on qt in the absence of thermal counterflow , differing significantly from previous studies @xcite . the first important contribution was made by maurer and tabeling @xcite , who confirmed experimentally the kolmogorov spectrum in superfluid @xmath0he . a turbulent flow was produced in a cylinder by driving two counter - rotating disks . the authors measured the local pressure fluctuations to obtain the energy spectrum . the experiments were made at three different temperatures : 2.3 k , 2.08 k , and 1.4 k. both above and below the @xmath1 point , the kolmogorov spectrum was confirmed . next was a series of experiments on grid turbulence performed for superfluid @xmath0he above 1 k by the oregon group @xcite . flow through a grid is usually used for generating turbulence in classical fluid dynamics @xcite . at sufficient distance behind the grid , the flow displays homogeneous isotropic turbulence . this method has also been applied to superfluid helium . in the oregon experiments , the helium was contained in a channel along which a grid was pulled at constant velocity . a pair of second - sound transducers was implanted in the walls of the channel to observe vortex tangles . in combining their observations with the decay of the turbulence , the authors had to make several assumptions . the analysis is too complicated to be described in detail here . the important point is that the coupling between the superfluid and the normal fluid by mutual friction makes a quasiclassical flow appear at length scales much larger than @xmath57 and causes the fluid to behave like a one - component fluid @xcite . the line density is found to decay as @xmath66 . a simple analysis shows that the @xmath66 decay can be derived from a quasiclassical model with a kolmogorov spectrum @xcite . subsequently many experimental , theoretical , and numerical works appeared . the present areas of study of qt are @xcite : the energy spectra and the dissipation mechanism at zero temperature ; qt created by vibrating structures ; and visualization of qt . the pioneering experiments by maurer _ et al . _ and by the oregon group were performed at relatively high temperatures where the normal fluid component was significant , and so the situation is complicated because of the two - fluid behavior . here we confine ourselves to the case of pure superfluid at zero temperature . what happens to qt at zero temperature is not trivial @xcite . the first problem is the nature of the energy spectrum of the turbulence for the pure superfluid component . the second problem is the dissipation in this system . since there is no normal fluid component , any dissipative mechanism does not work at large scales . however , some dissipative mechanism should operate at small length scales . one possibility is acoustic emission during vortex reconnections . in classical fluid dynamics , vortex reconnections cause acoustic emission . in quantum fluids , numerical simulations of the gp model show acoustic emission at every reconnection event @xcite . however , this mechanism is thought not to be important in superfluid helium because of its short coherence length . another possible mechanism is the radiation of sound ( phonons ) by the oscillatory motion of the vortex cores . the third problem is how energy is transferred from large to small scales at which the dissipative mechanism operates . if the energy spectrum obeys the kolmogorov law , the energy should be transferred through a richardson cascade of quantized vortices in contrast to the classical case . vortex reconnections sustaining a richardson cascade should be less effective at size scales shorter than the mean vortex spacing @xmath57 . no experimental studies have directly addressed the energy spectra of qt at zero temperature , although three numerical studies have been made . the first study was done by nore _ _ using the gp model @xcite . they numerically solved the gp equation starting from taylor green vortices and followed their time development . the quantized vortices became tangled and the energy spectra of the incompressible kinetic energy appeared to obey the kolmogorov law for a short period , though eventually deviated from it . the second study was done by the vortex filament model @xcite . made a vortex tangle arising from taylor green vortices and obtained an energy spectrum consistent with the kolmogorov law . the third used the modified gp model by kobayashi and tsubota @xcite . the kolmogorov spectra were confirmed for both decaying @xcite and steady @xcite qt by the modified gp model . the normalized gp equation is @xmath67\phi(\mathbf{x},t),\label{eq - gp}\ ] ] which determines the dynamics of the macroscopic wavefunction @xmath68 $ ] . the hydrodynamics in the gp model are compressible . the total number of condensate particles is @xmath69 and the total energy is @xmath70\phi(\mathbf{x},t),\ ] ] as represented by the sum of the interaction energy @xmath71 , the quantum energy @xmath72 , and the kinetic energy @xmath73 @xcite , @xmath74 ^ 2 , e_{kin}(t)=\frac{1}{n}\int d\mathbf{x}\:[f(\mathbf{x},t)\nabla\phi(\mathbf{x},t)]^2.\label{eq - energy - definition } \end{aligned}\ ] ] the kinetic energy is further divided into a compressible part @xmath75 due to compressible excitations and an incompressible part @xmath76 due to vortices . the kolmogorov spectrum is expected for @xmath76 . the failure to obtain a kolmogorov law in the pure gp model @xcite is attributable to the following . the simulations showed that @xmath76 decreases and @xmath75 increases while the total energy @xmath77 is conserved because many compressible excitations are created through vortex reconnections @xcite and disturb the richardson cascade of quantized vortices . _ overcame these difficulties and obtained a kolmogorov spectrum in qt that revealed an energy cascade @xcite . they made numerical calculations of the fourier - transformed gp equation with dissipation , @xmath78\tilde{\phi}(\mathbf{k},t ) + \frac{g}{v^2}\sum_{\mathbf{k}_1,\mathbf{k}_2}\tilde{\phi}(\mathbf{k}_1,t)\tilde{\phi}^\ast(\mathbf{k}_2,t ) \times\tilde{\phi}(\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2,t ) . \label{eq - fourier - gp1 } \end{aligned}\ ] ] here @xmath79 is the spatial fourier component of @xmath80 and @xmath81 is the system volume . the healing length is @xmath82 . the dissipation has the form @xmath83 with step function @xmath63 , which only dissipates excitations smaller than @xmath84 . this form of dissipation can be justified by a coupled analysis of the gp equation and the bogoliubov de gennes equations for thermal excitations @xcite . kobayashi _ et al . _ confirmed the kolmogorov spectra for decaying turbulence @xcite . to obtain a turbulent state , they started the calculation from an initial configuration in which the density was uniform and the phase of the wavefunction had a random spatial distribution . the initial state was dynamically unstable and soon developed turbulence with many vortex loops . the spectrum @xmath85 was then found to obey the kolmogorov law . a more elaborate analysis was made for steady qt by introducing energy injection at large length scales in addition to energy dissipation at small scales @xcite . energy injection at large scales was effected by moving a random potential @xmath86 that had a characteristic spatial scale @xmath87 . once steady qt occurs as a balance between energy injection and dissipation , it has an energy - containing range @xmath88 , an inertial range @xmath89 , and an energy - dissipative range @xmath90 . a typical simulation of steady turbulence was performed for @xmath91 . the dynamics started from the uniform wavefunction . the moving random potential nucleates sound waves as well as vortices , but the incompressible kinetic energy @xmath76 due to vortices becomes dominant in the total kinetic energy @xmath73 . the energies soon become almost constant , and steady qt is obtained . the energy cascade can be confirmed quantitatively by checking that the energy dissipation rate @xmath43 of @xmath76 is comparable to the energy flux @xmath92 via a richardson cascade in the inertial range . the energy flux is found to be approximately independent of @xmath48 and comparable to @xmath43 . as shown in fig . [ fig - steady - kolmogorov ] ( b ) , the energy spectrum is quantitatively consistent with the kolmogorov law in the inertial range @xmath89 , equivalent to @xmath93 . for qt . the plotted points are from an ensemble average of 50 randomly selected steady states . the solid line is the kolmogorov law @xcite.,title="fig : " ] + ( a ) for qt . the plotted points are from an ensemble average of 50 randomly selected steady states . the solid line is the kolmogorov law @xcite.,title="fig : " ] + ( b ) the above arguments are chiefly limited to length scales larger than the mean spacing @xmath57 of a vortex tangle , in which the richardson cascade is effective . what happens at smaller scales where the richardson cascade is less effective ? the most probable scenario is the kelvin - wave cascade . a kelvin wave is a deformation of a vortex line into a helix , with the deformation propagating as a wave along the vortex @xcite . kelvin waves were first observed by making torsional oscillations in uniformly rotating superfluid @xmath0he @xcite . the approximate dispersion relation for a rectilinear vortex is @xmath94 with a constant @xmath95 and a cutoff parameter @xmath96 equal to the vortex core size . here @xmath48 is the wavenumber of an excited kelvin wave , which is different than the wavenumber used for the energy spectrum above . at a finite temperature , a significant fraction of the normal fluid contains damped kelvin waves because of mutual friction . at very low temperatures , however , mutual friction does not occur and the only possible mechanism of dissipation is the radiation of phonons @xcite . phonon radiation becomes effective only when the frequency is very high , typically of the order of ghz ( @xmath97 nm@xmath98 ) . therefore , a mechanism is required to transfer energy to such high wavenumbers for kelvin - wave damping . an early numerical simulation based on the vortex filament model showed that kelvin waves are unstable to the buildup of sidebands @xcite . consequently nonlinear interactions between different kelvin wavenumbers can transfer energy from small to large wavenumbers , which constitutes a kelvin - wave cascade . this idea was first suggested by svistunov @xcite and later developed and confirmed through theoretical and numerical work by kivotides _ @xcite , vinen _ et al . _ @xcite , and kozik and svistunov @xcite . consider the nature of the transition between the richardson and the kelvin - wave cascades . turbulence at scales larger than the mean vortex spacing obeys the kolmogorov law with a richardson cascade . in this regime , the nature of individual vortices is not relevant , and the large - scale velocity field created collectively by the vortices mimics classical turbulent behavior . this regime may therefore be called classical . on the other hand , at length scales smaller than the mean vortex spacing the energy is transferred along each vortex through a kelvin - wave cascade ; this regime has no classical analog and can be called quantum . several theoretical considerations of the classical - quantum crossover have been proposed @xcite , but this topic is not yet settled and remains controversial . recently , vibrating structures including discs , spheres , grids , and wires , have been used for research into qt @xcite . despite differences between the structures , the experiments show surprisingly similar behavior . when a classical viscous fluid steadily flows past a structure , the flow changes from laminar to turbulence in the wake as the reynolds number increases . the net drag force on the structure is often expressed in terms of the drag coefficient @xmath99 as @xmath100 , where @xmath101 is the projected area of the structure onto a plane normal to the flow . for laminar viscous flow , the drag is approximately proportional to @xmath102 so that @xmath103 , while turbulent flow with high reynolds number yields a drag coefficient @xmath99 of order unity . this drastic reduction in the drag force is called the drag crisis , and is closely related to the formation of the boundary layer @xcite . a boundary layer in a high velocity gradient produces eddies . however , such boundary layers do not appear in the superfluid component . another classical aspect is the effect of oscillations . the oscillatory case is more complicated than that of steady flow because a second length scale appears in addition to the linear size of the structure , namely the viscous penetration depth @xmath104 where @xmath105 is the oscillation frequency of the structure @xcite . owing to the importance of the topic in marine engineering , a number of experimental studies have been published on oscillatory classical flow past a cylinder . as laminar flow becomes unstable , the drag coefficient @xmath99 reduces to a value of order unity . however , the values of @xmath99 oscillate within the range 0.5 to 2.0 , showing a more complicated transition to turbulence than in the case of steady flow . visual observations show that vortices are formed in the wake of the oscillating cylinder . a few experimental studies have reported classical flow around an oscillating sphere . a typical experiment shows that a single vortex ring is generated and shed during each half period of oscillation @xcite . typical behavior appeared in the pioneering observation of qt for an oscillating microsphere in superfluid @xmath0he by jger _ the sphere they used had a radius of approximately 100 @xmath106 m . it was made from a strongly ferromagnetic material and was magnetically levitated in superfluid @xmath0he . its response to an alternating drive was observed . at low drives , the velocity response @xmath102 was proportional to the drive @xmath107 , taking the `` laminar '' form @xmath108 with a temperature - dependent coefficient @xmath109 . at high drives , the response changed to the `` turbulent '' form @xmath110 above a critical velocity @xmath111 . at low temperatures , the transition from laminar to turbulent response was accompanied by hysteresis . subsequently , several groups have experimentally investigated the transition to turbulence in superfluid @xmath0he and @xmath112he - b by using grids , wires , and tuning forks @xcite . these experimental studies reported some common behaviors independent of the details of the structures , including its type , shape , and surface roughness . the observed critical velocities were in the range from 1 to 200 mm / s . since the velocity was lower than the landau critical velocity of approximately 50 m / s , the transition to turbulence should come not from intrinsic nucleation of vortices but from extension or amplification of remnant vortices.he at rest usually contains remnant vortices @xcite . they are metastably pinned between the walls of the container and are established when the liquid is cooled down through the @xmath1 point to the superfluid phase . ] such behavior is seen in the numerical simulation using the vortex filament model @xcite . figure 5 shows how the remnant vortices initially attached to a sphere develop into turbulence under an oscillating flow . many unresolved problems remain , such as the nature of the critical velocity and the origin of the hysteresis in the transition between the laminar and turbulent response . m in an oscillating superflow of 150 mm / s at 200 hz . [ hnninen , tsubota , and vinen , phys . rev . b * 75 * ( 2007 ) 064502 , reproduced with permission . copyright 2007 by the american physical society . ] ] there has been little direct experimental information about flow in superfluid @xmath0he because standard flow visualization techniques are not applicable to it . however , the situation is changing @xcite . for qt , one can seed the fluid with tracer particles in order to visualize the flow field . quantized vortices are observable by appropriate optical techniques . a significant contribution was made by zhang and van sciver @xcite . using the particle image velocimetry ( piv ) technique with 1.7-@xmath106m - diameter polymer particles , they visualized a large - scale turbulent flow both in front of and behind a cylinder in counterflowing superfluid @xmath0he at finite temperatures . in classical fluids , such turbulent structures are seen downstream of objects such as cylinders , and the structures periodically detach to form a vortex sheet . in the case of @xmath0he counterflow , however , the locations of the large - scale turbulent structures were relatively stable ; they did not detach and move downstream , although local fluctuations in the turbulence were evident . do the tracer particles follow the normal flow , the superflow , or a more complex flow ? _ studied this question theoretically and numerically to show that the situation changes depending on the size and mass of the tracer particles @xcite . another important contribution was the visualization of quantized vortices by bewley _ _ @xcite . in their experiments , liquid helium was seeded with solid hydrogen particles smaller than 2.7 @xmath106 m in diameter at a temperature slightly above @xmath113 , following which the fluid was cooled to below @xmath113 . when the temperature is above @xmath113 , a homogeneous cloud of the particles was seen that disperses throughout the fluid . however , on passing through @xmath113 , the particles coalesced into web - like structures . _ suggested that these structures represent decorated quantized vortex lines . they reported that the vortex lines form connections rather than remaining separated , and were homogeneously distributed throughout the fluid . the observed fork - like structures may indicate that several vortices are attached to the same particle . applying this technique to thermal counter flow , paoletti _ have directly visualized the two - fluid behavior @xcite . the long history of research into superfluid helium has uncovered two main cooperative phenomena of quantized vortices : vortex lattices under rotation and vortex tangles in qt . however , almost all studies of quantized vortices in atomic becs have been limited to vortex lattices . this section briefly discusses qt in atomic becs . kobayashi and tsubota proposed an easy , powerful method to make a steady qt in a trapped bec , by using precession @xcite . the dynamics of the wavefunction are described by the gp equation with dissipation . first , one traps a bec cloud in a weakly elliptical harmonic potential , @xmath114,\ ] ] where the parameters @xmath115 and @xmath116 exhibit elliptical deformation in the @xmath117- and @xmath118-planes . second , to transform the bec into a turbulent state , a rotation is applied along the @xmath119-axis followed by a rotation along the @xmath120-axis , as shown in fig . the rotation vector is @xmath121 where @xmath122 and @xmath123 are the frequencies of the first and second rotations , respectively . consider the case where the spinning and precessing rotational axes are perpendicular to each other . then the two rotations do not commute and can not be represented by their sum . this form of precession is also used for turbulence in water @xcite . starting from a stationary solution without rotation or elliptical deformation , kobayashi _ et al . _ numerically calculated the time development of the gp equation by turning on the rotation @xmath124 and elliptical deformation @xmath125 . by monitoring the kinetic energy and anisotropic parameters , the system eventually becomes statistically steady and isotropic . in the steady state , the spectrum @xmath85 of the incompressible kinetic energy per unit mass is consistent with the kolmogorov law [ fig . the inertial range which sustains the kolmogorov law is determined by the thomas - fermi radius @xmath32 and the coherence length @xmath84 . the application of a combined precession around three axes enables one to obtain more isotropic qt @xcite . the velocity field in a bec cloud can be observed by bragg spectroscopy @xcite , enabling one to obtain the energy spectrum . is the thomas - fermi radius amd @xmath126 is the characteristic length scale of the trap . ] there are several advantages of studying qt in atomic becs compared to in superfluid helium . first , one can observe the vortex configuration , probably even the richardson cascade process of vortices . in fact one can directly study the relation between the real - space richardson cascade and the wavenumber - space cascade ( the kolmogorov spectrum ) . second , one can control the transition to turbulence by changing the rotational frequencies or other parameters . for example , rotation along one axis forms a vortex lattice . if one applies another rotation , it may rotate the lattice if the frequency is low . if the frequency is high , however , the second rotation can melt the lattice into a vortex tangle . that would enable one to investigate in detail the entire transition to turbulence . third , by changing the shape of the trapping potential , one can study the effect of the anisotropy on turbulence ; a typical question is how the kolmogorov spectrum is changed when the bec becomes anisotropic . this issue is related to the transition between 2d and 3d turbulence . fourth , one can create rich qt in multicomponent becs by controlling the interaction parameters . multicomponent becs have been extensively studied and found to create many kinds of vortex structures depending on the intercomponent interaction @xcite . by using such systems , we will be able to make multicomponent qt coupled through the interaction . in this article , we have reviewed recent research on quantum hydrodynamics and qt in superfluid helium and atomic becs . research on qt is currently a key branch in low - temperature physics , attracting the attention of many scientists . it is comprised of quantized vortices as definite elements , in contrast to conventional turbulence . thus , the investigation of qt may lead to a breakthrough in understanding a mystery of nature pondered since the era of da vinci . there are many related topics not addressed in this article , regarding which we refer the readers to other review articles @xcite . the author thanks kenichi kasamatsu and michikazu kobayashi for help with the manuscript . the author acknowledges a grant - in - aid for scientific research from jsps ( grant no . 18340109 ) and a grant - in - aid for scientific research on priority areas from mext ( grant no . 17071008 ) . r. p. feynman , r. b. leighton , and m. sands , _ the feynman lectures on physics _ , addison - wesley , san francisco , 2006 . d. r. tilley and j. tilley , _ superfluidity and superconductivity _ , 3rd ed . , institute of physics publishing , bristol and philadelphia , 1990 . c. j. gorter and j. h. mellink , _ on the irreversible processes in liquid helium ii _ , physica 15 ( 1949 ) , pp . r. p. feynman , _ application of quantum mechanics to liquid helium _ , in _ prog . c. j. gorter , north - holland , amsterdam , 1955 , vol . i , pp . 1753 . w. f. vinen , _ mutual friction in a heat current in liquid helium ii : i. experiments on steady heat currents _ , proc . soc . a 240 ( 1957 ) , . w. f. vinen , _ mutual friction in a heat current in liquid helium ii : ii . experiments on transient effects _ , proc . soc . a 240 ( 1957 ) , . w. f. vinen , _ mutual friction in a heat current in liquid helium ii : iii . theory of the mutual friction _ , proc . soc . a 242 ( 1957 ) , . w. f. vinen , _ mutual friction in a heat current in liquid helium ii : iv . critical heat currents in wide channels _ , proc . soc . a 243 ( 1958 ) , . w. f. vinen , _ the detection of a single quantum of circulation in liquid helium ii _ , proc . soc . a 260 ( 1958 ) , 400413 . j. t. tough , _ superfluid turbulence _ , in _ prog . low temp . _ , ed . c. j. gorter , north - holland , amsterdam , 1982 , vol viii , pp . . k. w. schwarz , _ three - dimensional vortex dynamics in superfluid 4he : line - line and line - boundary interactions _ , phys . b 31 ( 1985 ) , pp . 57825803 . k. w. schwarz , _ three - dimensional vortex dynamics in superfluid 4he : homogenous superfluid turbulence _ , phys . rev . b 38 ( 1988 ) , pp . 23982417 . w. f. vinen and j. j. niemela , _ quantum turbulence _ , j. low temp . ( 2002 ) , pp . . w. f. vinen , _ an introduction to quantum turbulence _ , j. low temp . 145 ( 2006 ) , pp . , eds . w. p. halperin and m. tsubota , elsevier , amsterdam , 2008 , vol . m. tsubota , _ quantum turbulence _ , j. phys . 77 ( 2008 ) , 111006 . c. j. pethick and h. smith , _ bose - einstein condensation in dilute gases _ , cambridge university press , cambridge , 2002 . k. kasamatsu and m. tsubota , _ quantized vortices in atomic bose - einstein condensates _ , in _ prog . w. p. halperin and m. tsubota , elsevier , amsterdam , 2008 , vol . 16 , pp . 351403 . j. r. abo - shaeer , c. raman , j. m. vogels , and w. ketterle , _ observation of vortex lattices in bose - einstein condensates _ , science 292 ( 2001 ) , pp . 476479 . k. w. madison , f. chevy , w. wlhlleben , and j. dalibard , _ vortex formation in a stirred bose - einstein condensate _ , lett . 84 ( 2000 ) , pp . . k. w. madison , f. chevy , w. wlhlleben , and j. dalibard , _ statonary states of a rotating bose - einstein condensate : routes to vortex nucleation _ , phys . 86 ( 2001 ) , pp . 44434446 . g. a. williams and r. e. packard , _ photographs of quantized vortex lines in rotating he ii _ 33 ( 1974 ) , pp . e. j. yarmchuck and r. e. packard , _ photographic studies of quantized vortex lines _ , j. low temp 46 ( 1982 ) , pp . 479515 . m. r. matthews , b. p. anderson , p. c. haljan , d. s. hall , c. e. wieman , and e. a. cornell , _ vortices in a bose - einstein condensate _ , 83 ( 1999 ) , pp . m. tsubota , k. kasamatsu , and m. ueda , _ vortex lattice formation in a rotating bose - einstein condensate _ , a65 ( 2002 ) , 023603 . k. kasamatsu , m. tsubota , and m. ueda , _ nonlinear dynamics of vortex lattice formation in a rotating bose - einstein condensate _ , phys . a67 ( 2003 ) , 033610 . k. kasamatsu , m. machida , n. sasa and m. tsubota , _ three - dimensional dynamics of vortex - lattice formation in bose - einstein condensate _ phys . a71 ( 2005 ) , 063616 . u. frisch , _ turbulence _ , cambridge university press , cambridge , 1995 . a. n. kolmogorov , _ the local structure of turbulence in incompressible viscous fluid for very large reynolds number _ , dokl . nauk sssr 30 ( 1941 ) , pp . 299 - 303 [ reprinted in proc . soc . a 434 ( 1991 ) , a. n. kolmogorov , _ on degeneration ( decay ) of isotropic turbulence in an incompressible viscous liquid _ , nauk sssr 31 ( 1941 ) , pp . 538 - 540 [ reprinted in proc . a 434 ( 1991 ) , pp . 15 - 17 ] . m. tsubota , t. araki , and s. k. nemirovskii , _ dynamics of vortex tangle without mutual friction in superfluid @xmath0he _ , phys . rev . b 62 ( 2000 ) , pp . 11751 - 11762 . j. koplik and h. levine , _ vortex reconnection in superfluid helium _ , phys . 71 ( 1993 ) , pp . 1375 - 1378 . m. leadbeater , t. winiecki , d. c. samuels , c. f. barenghi , and c. s. adams , _ sound emission due to superfluid vortex reconnections _ , 86 ( 2001 ) , pp . 1410 - 1413 . s. ogawa , m. tsubota , and y. hattori , _ study of reconnection and acoustic emission of quantized vortices in superfluid by the numerical analysis of the gross - pitaevskii equation _ , j. phys . soc . jpn . 71 ( 2002 ) , pp . 813 - 821 . j. maurer and p. tabeling , _ local investigation of superfluid turbulence _ , europhys . lett . 43 ( 1998 ) , pp . s. r. stalp , l. skrbek , and r. j. donnelly , _ decay of grid turbulence in a finite channel _ , phys . 82 ( 1999 ) , pp . 4831 - 4834 . l. skrbek , j. j. niemela , and r. j. donnelly , _ four regimes of decaying grid turbulence in a finite channel _ , phys . 85 ( 2000 ) , pp . 2973 - 2976 . l. skrbek and s. r. stalp , _ on the decay of homogeneous isotropic turbulence _ , phys . fluids 12 ( 1997 ) , pp . 1997 - 2019 . s. r. stalp , j. j. niemela , w. f. vinen , and r. j. donnelly , _ dissipation of grid turbulence in helium ii _ , fluids 14 ( 2002 ) , pp . 1377 - 1379 . w. f. vinen , _ classical character of turbulence in a quantum liquid b 61 ( 2000 ) , pp . 1410 - 1420 . c. nore , m. abid , and m. e. brachet , _ kolmogorov turbulence in low - temperature superflows _ , phys . rev . 78 ( 1997 ) , pp . 3296 - 3299 . c. nore , m. abid , and m. e. brachet , _ decaying kolmogorov turbulence in a model of superflow _ , phys . fluids 9 ( 1997 ) , pp . 2644 - 2669 . t. araki , m. tsubota and s. k. nemirovskii , _ energy spectrum of superfluid turbulence with no normal - fluid component _ , phys . ( 2002 ) , 145301 . m. kobayashi and m. tsubota , _ kolmogorov spectrum of superfluid turbulence : numerical analysis of the gross - pitaevskii equation with a small - scale dissipation _ 94 ( 2005 ) , 065302 . m. kobayashi and m. tsubota , _ kolmogorov spectrum of quantum turbulence _ , j. phys . 74 ( 2005 ) , pp . 3248 - 3258 . m. kobayashi and m. tsubota , _ thermal dissipation in quantum turbulence _ , 97 ( 2006 ) , 145301 . j. thomson , _ vibrations of a columnar vortex _ , phil mag . 10 ( 1880 ) , pp . 155 - 168 . h. e. hall , _ an experimental and theoretical study of torsional oscillations in uniformly rotating liquid helium ii _ , proc . london a 245 ( 1958 ) , pp . . h. e. hall , _ the rotation of helium ii _ , phil . mag . suppl . 9 ( 1960 ) , pp . w. f. vinen , _ decay of superfluid turbulence at a very low temperature : the radiation of sound from a kelvin wave on a quantized vortex _ , b 64 ( 2001 ) , 134520 . d. c. samuels and r. j. donnelly , _ sideband instability and recurrence of kelvin waves on vortex cores _ , phys . 64(1990 ) , pp . 1385 - 1388 . b. v. svistunov , _ superfluid turbulence in the low - temperature limit , _ phys . b 52 ( 1995 ) , pp . 3647 - 3653 . d. kivotides , j. c. vassilicos , d. c. samuels , and c. f. barenghi , _ kelvin waves cascade in superfluid turbulence _ 86 ( 2001 ) , pp . 3080 - 3083 . w. f. vinen , m. tsubota , and m. mitani , _ kelvin - wave cascade on a vortex in superfluid @xmath0he at a very low temperature _ , phys . 91 ( 2003 ) , 135301 . e. kozik , and b. v. svistunov , _ kelvin - wave cascade and decay of superfluid turbulence _ , 92 ( 2004 ) , 035301 . e. kozik and b. v. svistunov , _ scale - separation scheme for simulating superfluid turbulence : kelvin - wave cascade _ , phys . 94 ( 2005 ) , 025301 . e. kozik and b. v. svistunov , _ vortex - phonon interaction _ phys . b 72 ( 2005 ) , 172505 . v. s. lvov , s. v. nazarenko , and o. rudenko , _ bottleneck crossover between classical and quantum superfluid turbulence _ , b 76 ( 2007 ) , 024520 . e. kozik and b. v. svistunov , _ kolmogorov and kelvin - wave cascades of superfluid turbulence at t=0 : what lies between _ , phys . b 77 ( 2008 ) , 060502 . e. kozik and b. v. svistunov , _ scanning superfluid - turbulence cascade by its low - temperature cutoff _ , phys . ( 2008 ) , 195302 . l. skrbek and w. f. vinen , _ the use of vibrating structures in the study of quantum turbulence _ , in _ prog . low temp . w. p. halperin and m. tsubota , elsevier , amsterdam , 2008 , vol . 195 - 246 . l. d. landau and e. m. lifshitz , _ fluid mechanics _ , 2nd ed . , pergamon press , oxford , 1987 . r. j. donnelly , private communication . j. jger , b. schuderer , and w. schoepe , _ turbulent and laminar drag of superfluid helium on an oscillating microsphere _ , 74 ( 1995 ) , pp . 566 - 569 . d. d. awschalom and k. w. schwarz , _ observation of a remanent vorte - line density in superfluid helium _ , phys . 52 ( 1984 ) , pp . r. hnninen , m. tsubota , and w. f. vinen , _ generation of turbulence by oscillating structures in superfluid helium at very low temperatures _ , b 75 ( 2007 ) , 064502 . s. w. van sciver and c. f. barenghi , _ visualization of quantum turbulence _ , in _ prog . low temp . w. p. halperin and m. tsubota , elsevier , amsterdam , 2008 , vol . 247 - 303 . t. zhang and s. w. van sciver , _ large - scale turbulent flow around a cylinder in counterflow superfluid 4he ( he(ii ) ) _ , nature phys . 1 ( 2005 ) , pp . d. r. poole , c. f. barenghi , y. a. sergeev , and w. f. vinen , _ motion of tracer particles in he ii _ , phys . b 71 ( 2005 ) , 064514 . d. kivotides , c. f. barenghi , and y. a. sergeev , _ collision of a tracer particle and a quantized vortex in superfluid helium : self - consistent calculations _ b 75 ( 2007 ) , 212502 . d. kivotides , c. f. barenghi , and y. a. sergeev , _ interactions between particles and quantized vortices in superfluid helium _ , phys . b 77 ( 2008 ) , 014527 . g. p. bewley , d. p. lathrop , and k. r. sreenivasan , _ visualization of quantized vortices _ , nature 441 ( 2006 ) , 588 . m. s. paelotti , r.b . firrito , k. r. sreenivasan , and d. p. lathrop . _ visualization of superfluid helium flow _ , j. phys . 77 ( 2008 ) , 111007 . m. kobayashi and m. tsubota , _ quantum turbulence in a trapped bose - einstein condensate _ a76 ( 2007 ) , 045603 . s. goto , n. ishii , s. kida , and m. nishioka , _ turbulence generator using a precessing sphere _ fluids 19 ( 2007 ) , 061705 . m. kobayashi and m. tsubota , _ quantum turbulence in a trapped bose - einstein condensate under combined rotations around three axes _ , j. low temp . ( 2008 ) , pp . 587 - 592 . j. stenger , s. inouye , a. p. chikkatur , d. m. stamper - kurn , d. e. pritchard , and w. ketterle , _ bragg spectroscopy of a bose - einstein condensate _ 82 ( 1999 ) , pp . 4569 - 4573 . k. kasamatsu , m. tsubota , and m. ueda , _ vortices in multicomponent bose - einstein condensates _ , int . b 19 ( 2005 ) , pp . 1835 - 1904 .

at mesoscopic scale and at low temperature , electrons in metallic samples keep their phase coherence on a length @xmath2 which is bigger than the sample size . transport and thermodynamic properties of the system are then sensitive to interference between electronic wave functions and present spectacular signatures of this phase coherence . to study these effects the ring geometry is especially suitable . indeed in the presence of a magnetic flux @xmath3 through the ring the periodic boundary conditions for electronic wavefunctions acquire a phase factor @xmath4 with @xmath5 the flux quantum @xcite . as a result , the magnetoconductance of a phase coherent ring exhibits quantum oscillations which periodicity corresponds to one flux quantum through the area of the sample @xcite . the phase of the first harmonics of these oscillations is sample specific so that these harmonics do not survive ensemble average . in contrast the second harmonics have a contribution which resists this averaging . this results from the interference between time reversed paths around the ring ( weak localisation contribution ) . these @xmath6 periodic oscillations were observed in transport measurements on long cylinders or connected arrays of rings @xcite . their sign corresponds to a positive magnetoconductance in zero field . in the case of singly connected geometries , like full disks , the signature of weak localization consists in a single peak of positive magnetoconductance which width corresponds to @xmath7 through the sample @xcite . magneto - transport experiments on connected systems constitute a very sensitive and powerful probe for the investigation of sample specific signatures of quantum transport . however , because of strong coupling between the system and the measuring device , quantum corrections represent a small fraction of the conductance ( of the order of @xmath8 where @xmath9 is the dimensionless conductance expressed in @xmath10 units ) which is still dominated by the classical drude response in the diffusive regime @xmath11 . there exists a number of experiments which can address some of the electronic properties of mesoscopic samples without coupling to macroscopic wires . this is the case of ac conductance experiments where aharonov bohm rings are coupled to an electromagnetic field . in contrast with the connected case , the response of an isolated system can be dominated by quantum effects . moreover , the quasi discrete nature of the energy spectrum and the sensitivity to the statistical ensemble ( canonical or grand - canonical ) are new features of isolated mesoscopic systems . in particular it has been emphasized that the average absorbtion of isolated mesoscopic systems is determined by the energy level statistics and its sensitivity to time reversal symmetry breaking by a magnetic field . the first experiments done in this spirit were performed by coupling an array of disconnected gaas / gaalas rings to a strip - line superconducting resonator @xcite . in such a geometry the rings experience both ac magnetic and electric field . the magnetic response of the rings i.e. their orbital magnetism is directly related to persistent currents in the zero frequency limit @xcite . on the other hand the electric response of isolated metallic sample is related to the screening of the electric field inside the metal . the induced charge displacement is at the origin of the polarisability @xmath12 , defined as the ratio between the induced electric dipole @xmath13 and the applied electric field @xmath14 ( @xmath15 ) . the polarisability is known to be essentially determined by the geometry of the sample with correction of the order of @xmath16 , with @xmath17 the screening length and @xmath18 the typical size of the system @xcite . it has been recently predicted that this quantity is sensitive to phase coherence around the ring @xcite and is thus expected to present flux oscillations . the electric contribution can be in the particular case of gaas rings of the same order of magnitude as the magnetic response @xcite . to be able to distinguish between the two types of response we have designed a superconducting lc resonator which capacitive part and inductive part are physically separated . in this paper we present measurements of both magnetic and electric response of aharonov - bohm rings . note that these experiments are done on the very same array of rings for both types of response , giving us the opportunity to compare them . preliminary account of measurement of the electric response was given in reference @xcite . the paper is organised in the following way . section [ experimentalsetup ] gives a detailed presentation of the sample , an array of aharonov - bohm rings , and the resonating technic used to measure the magnetic and electric response . results on the non - dissipative part of the flux dependent electric response are presented in section [ magnetopolarisabilty ] . a comparison with theoretical predictions is given , including frequency dependence . temperature and electronic density dependence are also investigated . the next section focuses on the dissipative part of the magnetopolarisability of the rings . theoretical results for this quantity are derived and compared to the experiment . the section [ magneticresponse ] is devoted to the measurements of the magnetic response of the same rings . despite the fact that the signal is then smaller , the magnetic response of the rings is detected and compared to predictions on averaged persistent currents . we conclude by a comparison between the magnetic and electric response . [ cols="^,^,^ " , ] we have studied the electric and magnetic susceptibilities of isolated aharonov - bohm rings . our system is an array of @xmath0 2d rings etched by reactive ion etching in a high mobility algaas / gaas heterojunction . the characteristics of these rings are given in table [ tab : characteristicsoftherings ] . they are ballistic in the transverse direction and diffusive longitudinally ( @xmath19 and @xmath20 ) . it is important to perform a deep etching of the heterojunction ( down to gaas ) in order to minimize high frequency losses , which have been observed to be important in etched algaas . because of etching the electronic density is strongly depressed . however we are able to recover the nominal density of the heterojunction by illuminating the rings with a infrared diode placed close to the sample in the dilution refrigerator . for each illumination a current of @xmath21 is run through the diode during several minutes . measurements are done at least one hour after the illumination in order to ensure good stability of the sample . an upper value of the estimated illumination power coupled to the sample is 600 photons / s with a wavelength of 766 nm . with this setup we are thus able to perform measurements at different electronic density . the control on the density is rather qualitative because of the difficulty to calibrate the illumination procedure . we have checked the effect of illumination on a connected aharonov - bohm ( ab ) ring ( figure [ fig : abillum ] ) . at zero illumination time the conductance of the ring is zero . on such a sample we can follow the ab oscillations when the resistance decreases by more than an order of magnitude with illumination . as a consequence a clear effect of illuminating the ring is to increase its conductance . the fourier transform of the resistance of the ring is shown on figure [ fig : fftabillum ] . we see for each illumination an oscillation whose periodicity is consistent with a flux quantum @xmath22 in the area of the ring . however the fourier transform shows that the peak corresponding to this periodicity changes with illumination both in shape and in amplitude . the amplitude increases with illumination due to the increase of ab oscillations . the fact that the shape , and in particular the width , of the peak changes with illumination is an indication that the width of the ring increases with illumination time . to be more precise the width of the rings is multiplied by a factor 2 between the first and last curve of figure [ fig : fftabillum ] . note that the increase of the electronic density has also been shown to induce an increase of the electronic mobility @xcite . peak is strongly dependent on illumination . the curves are shifted for clarity.,width=302 ] in order to study the disorder average we have measured conductance of a single ring and a mesh , representing a two - dimensional square array , etched in the same type of heterojunction than the rings . the magnetoconductance is shown on figure [ fig : angr ] . as expected the ab effect disappears under ensemble averaging . the @xmath6 oscillations on the other hand remains on the mesh . in this case the triangular shape of the magnetoconductance is attributed to weak localisation in the wire of the mesh . signal disappears with ensemble average , so that in the mesh only the @xmath6 component remains . note the triangular shape of the magnetoconductance on the mesh . the curves are shifted for clarity.,width=302 ] to measure the electric or magnetic response of the rings we couple them to a superconducting micro - resonator and detect the changes in its properties . this resonator is made by optical lithography . it consists of a niobium strip - line deposited on a sapphire substrate . this substrate has been preferred to silicon or gaas because it induces the weakest temperature dependence of the resonance frequency and gives the best quality factor due to the quality of the niobium layer on sapphire . a schematic drawing is given in figure [ fig : resonator ] . the width of the wire constituting the resonator is 2 @xmath23 m , the thickness 1 @xmath23 m and the spacing between two adjacent wires is 4 @xmath23 m . the total length of the capacitance or the inductance is 10 or 20 cm . in this kind of resonator the inductance is physically separated from the capacitance by a distance of 300 @xmath23 m , allowing to submit the sample only to an electric field ( or to a magnetic field ) to measure its electric ( or magnetic ) response . this separation between magnetic and electric response has been checked by deposition of a paramagnetic system ( dpph ) alternatively on the capacitive and inductive part of the resonator . a magnetic spin resonance signal was only observed when dpph was on the inductive part . the resonance frequency of the bare resonator varies between 200 mhz and 400 mhz depending on the geometry . its quality factor is 10000 at 4.2 k and 200000 at 20 mk . the resonator can be modelled by an lc circuit of resistance @xmath24 , inductance @xmath25 , capacitance @xmath26 , whose resonance frequency is @xmath27 and quality factor @xmath28 . from the value of the higher resonance frequencies of the resonator we have estimated that the residual capacitance of the meander line is at least 10 times smaller than @xmath26 . due to the meissner effect , the dc field just above the resonator is strongly inhomogeneous . in order to minimize this effect , we have inserted a thin , @xmath29m - thick , mylar film between the detector and the rings . this reduces the field inhomogeneity to about @xmath30% , which is of the order of fluctuations in the lithography . in order to measure their electric response the rings are placed on top of the capacitance of the resonator . note that with this procedure the rings are not well aligned with the resonator so that they do not have the same coupling with the capacitance . this is not a problem as soon as only linear response is investigated . we note @xmath31 the polarisability averaged over disorder of a ring at the frequency @xmath32 . the impedance of the capacitance @xmath26 slightly modified by the rings reads : @xmath33 in this expression @xmath34 is the number of rings coupled to the capacitance , @xmath35 is an averaged coefficient measuring the dielectric coupling between one ring and the capacitance . the capacitance with the rings is equivalent to a capacitance @xmath36 in series with a resistance @xmath37 . hence : @xmath38 the frequency shift due to the rings is then : @xmath39 the quality factor is determined by : @xmath40 so that , with @xmath41 at resonance : @xmath42 provided that @xmath43 . the electric coupling coefficient is estimated in appendix [ electric_coupling ] . knowing this value and the number of rings coupled to the resonator , it is possible to evaluate quantitatively the polarisability of the rings by measuring the resonance frequency shift ( equation [ frequencyshift ] ) and the variation of the quality factor ( equation [ qshift ] ) . when the rings are placed on top of the inductance @xmath25 of the resonator , this inductance is shifted because of their magnetic response @xmath44 according to : @xmath45 with @xmath34 the number of rings coupled to the resonator , @xmath46 the magnetic coupling coefficient between one ring and the inductance , which has the dimension of the inverse of a volume . note that , properly defined , the coupling coefficient @xmath46 is of the same order of magnitude than @xmath35 . more precisely the estimation of @xmath35 and @xmath46 done in appendix leads to @xmath47 , as expected from reference @xcite . following the same reasoning than for the electric coupling , the properties of the resonator are modified according to : @xmath48 from previous equations it is in principle possible to measure the absolute value of @xmath12 or @xmath49 . however when a gaas sample is on the inductive or capacitive part of the resonator the modification of the resonance is dominated by the influence of the substrate . as a consequence it is very difficult to have an accurate absolute measurement . nevertheless relative measurements are possible so that the variation of the electric or magnetic response with magnetic field can be detected in a reliable way . the reflected signal of the resonator is measured with the setup of figure [ fig : reflectedsignal ] and used in a feedback loop to lock the frequency of a rf generator to the resonance frequency . the setup is summarized in figure [ fig : asservissement ] . the resonator is coupled capacitively to the external circuit using on - chip capacitances . in order to preserve the quality factor of the resonator we work in a configuration where the resonator is undercoupled . the rf power injected is sufficiently low ( @xmath50 10 pw ) so as not to heat the sample . the frequency of the rf generator is modulated at @xmath51 and the signal from the resonator is detected by a lock - in detector at the frequency of the modulation . the lock - in signal is to first approximation the derivative of the resonance peak : it gives an error signal i.e. this signal is zero at resonance , and changes sign when the frequency of the generator is higher or lower than the resonance frequency . using this signal in a feedback loop the frequency of the rf generator is locked to the resonance frequency . this way , by measuring the feedback signal , one has directly access to the shift of the resonance frequency . to enhance the accuracy we modulate the magnetic field by a 1 g ac field oscillating at 30 hz , produced by a small superconducting coil close to the sample , and detect the modulated resonance frequency with a lock - in detector . at this point we consider that the frequency of the generator is locked to the resonance frequency by the previous setup . the signal measured is the signal reflected from the resonator . as a consequence it is related to the reflexion coefficient @xmath52 , with @xmath53 the impedance of the resonator and the coupling capacitance and @xmath54 the impedance matched by the external circuit . we assume that near the resonance frequency the impedance @xmath53 reads : @xmath55 with @xmath56 the resonance frequency . in the limit @xmath57 , which correspond to a very undercoupled resonator , the reflected signal is a linear function of @xmath53 . as a consequence if the rf signal is frequency modulated at @xmath51 around the resonance frequency @xmath56 the reflected signal at @xmath58 is related to the second derivative of the real part @xmath53 , which is proportional to @xmath59 . this way by measuring the signal at @xmath58 we have access to the quality factor . however when the frequency modulation is not small compared to the width of the resonance peak or the resonator is not very undercoupled to the external circuit , the relation between the signal at @xmath58 and the quality factor is not straightforward and needs calibration . in this part we present measurements of the flux dependent polarisability of the rings , which are placed on the capacitive part of the resonator as described in section [ experimentalsetup ] . in this configuration the resonance frequency is decreased by 15 % , due to the dielectric constant of the gaas substrate . moreover the quality factor drops down to 3000 at 20 mk at zero illumination . this strong decrease is attributed to dielectric losses in the heterojunction . the derivative of the resonance frequency of the resonator with the rings is shown on figure [ fig : polarextract ] ( a ) . this signal is a straight line , on top of which small oscillations are superimposed . the straight line is due to the field dependence of the penetration length in niobium , which constitutes the resonator . this behaviour has been verified to be the same with or without the rings . the oscillating signal is on the other hand attributed to the flux dependent electric response of the rings . these oscillations are extracted by subtracting the base line ( figure [ fig : polarextract ] ( b ) ) . note their anharmonicity as well as the existence of an aperiodic signal as illustrated by the fourier transform of the data ( figure [ fig : polarextract ] ( c ) ) showing a well defined peak . in order to focus on this periodic contribution , which is the expected signature of phase coherence , a numerical high pass filter with a cut - off frequency of 0.05 g@xmath60 ( corresponding to the arrow on figure [ fig : polarextract ] ( c ) ) is applied and the signal is then numerically integrated in order to have the frequency shift due to the rings ( figure [ fig : polarextract ] ( d ) ) . this shift is proportional to the variation of polarisability versus magnetic field according to formula [ frequencyshift ] . we will return to the aperiodic signal in the section devoted to illumination effect . the frequency shift due to the rings is periodic with a period of approximately 12.5 g. from the fourier transform ( figure [ fig : polarextract ] ( c ) ) the period of the oscillation is deduced to be consistent with half a flux quantum @xmath61 in a ring with no signature of @xmath22 periodicity , as expected for an aharonov - bohm effect averaged over many rings @xcite . note the extra broadening ( by more than a factor 2 ) of this @xmath6 peak compared to the measurements on a single connected ring . we interpret this as resulting from the dispersion in circumfrences in the different rings . the sign of frequency shift is negative at low magnetic field which means according to formula [ frequencyshift ] that the magneto - polarisability is positive , i.e. @xmath62 at low magnetic field . the screening is thus better when time reversal symmetry is broken by magnetic field . the scale of the signal is given by the amplitude of the first oscillation . from figure [ fig : polarextract ] ( d ) we deduced @xmath63 g@xmath64 . note that this value means detecting a frequency shift of 100hz on a frequency of 350 mhz . with the coupling coefficient estimated in appendix [ electric_coupling ] it leads to the value of the magnetopolarisability @xmath65 , where @xmath66 is the calculated polarisability of a quasi one dimensional ( quasi-1d ) circular ring of radius @xmath67 . at different value of the parameter @xmath68 . note that the value of magnetopolarisability is zero at zero whatever the level broadening.,width=298 ] our experiment shows that there is a flux correction to the polarisability of the rings , which is _ positive _ at low field . let s now compare this result to recent theoretical predictions . since we are using a ring geometry we are going alternatively from a situation where the system presents time reversal symmetry ( at flux values of @xmath69 , with @xmath70 ) to the case where time reversal symmetry is broken by magnetic field . in the random matrix theory ( rmt ) the first case corresponds to gaussian orthogonal ensemble ( goe ) whereas the second is related to the gaussian unitary ensemble ( gue ) . so the quantity to be compared with theoretical predictions , which evaluate the variation of a physical variable @xmath71 between goe and gue , is @xmath72 defined as @xmath73 . note that since the rings are semi - ballistic , the transition with magnetic field may not be exactly from goe to gue . the polarisability of small metallic grains was studied using rmt first by gorkov and eliashberg @xcite . the sensitivity of the electrostatic properties of mesoscopic systems to quantum coherence has been emphasized by bttiker for connected geometries @xcite . the phase coherent correction to the polarisability of isolated systems was recently theoretically investigated . efetov found that it is possible to relate self consistently this correction to the flux dependence of the screened potential @xcite . two recent works have calculated this effect in the diffusive regime using linear response formalism ( noat _ et al . _ @xcite ) or supersymmetry techniques ( blanter and mirlin @xcite ) . in the grand canonical ensemble ( gce ) the chemical potential in each ring is supposed to be constant . it describes a situation where the rings are connected to a reservoir of particules . _ a priori _ this is not the case in the experiment where the rings are isolated but as the theory is simpler in gce we recall first the result in this statistical ensemble . no flux dependence for the polarisability is predicted if the rf pulsation @xmath32 is much smaller than the inverse relaxation time @xmath74 . however when @xmath75 the magnetopolarisability is related to the flux dependence of the diagonal matrix element of the screened potential : @xmath76 @xmath77 is the disorder averaged square of the diagonal matrix element of the screened potential @xmath78 at energy @xmath23 , the mean chemical potential of the rings . @xmath79 is the applied electric field . we note @xmath80 the eigenstates of the unperturbed system . this matrix element is then given by : @xmath81 from this expression it appears that the magnetopolarisability is related to the difference of correlation function of the eigenstates with and without time reversal symmetry . this correlation function has been computed in the diffusive regime within a supersymetric @xmath82-model approach @xcite : @xmath83 \left [ 1 + 2 \pi_d({\bf r_1},{\bf r_2 } ) \right ] \quad ( goe)\end{gathered}\ ] ] @xmath84 \left [ 1 + \pi_d({\bf r_1},{\bf r_2 } ) \right ] \quad ( gue)\end{gathered}\ ] ] with @xmath85 the volume of the sample , @xmath86 a short range function which decays on the length scale of the mean free path and @xmath87 the diffusion propagator . the correction due to the short range term @xmath86 has been shown to be negligible @xcite . by considering only the diffusion term the magnetopolarisability is given by : @xmath88 note that this derivation of the magnetopolarisability is equivalent to the one used by noat _ et al _ @xcite based on the following rmt argument : @xmath89 this relation can also be obtained from [ eq : corrgoe ] and [ eq : corrgue ] using the fact that : @xmath90 due to symmetry properties of the screened potential . the calculation of the magnetopolarisability using formula [ polardiffuson ] for the case of a quasi-1d ring ( appendix [ quasi1dpolar ] ) leads to : @xmath91 @xmath92 is a function related to the geometry and the dimension of the sample . using this expression and the value of table [ tab : characteristicsoftherings ] we have @xmath93 . in our experiment the rings are isolated , the number of electrons in each ring is supposed to be constant , so that the result of the canonical ensemble ( ce ) should apply . at t=0 and zero frequency the flux dependent correction to polarisability is found to be zero . however at @xmath94 the gce result is recovered . the complete frequency dependence of the magnetopolarisability in the ce has been recently derived by blanter and mirlin @xcite . following their reasoning but taking into account the level broadening @xmath74 we can write : @xmath95 with @xmath96 a function which depends only on the statistic of energy levels : @xmath97\end{gathered}\ ] ] @xmath98 and @xmath99 are respectively the two and three levels correlation function , known from rmt @xcite . by evaluating this expression versus frequency at different value of level broadening we get the results shown on figure [ fig : polarce ] . the behaviour at low value of the level broadening is in qualitative agreement with result of reference @xcite . in particular the magnetopolarisability is found to be zero at zero frequency ( in our calculation the value of @xmath100 is at least 25 time smaller than @xmath101 ) . the present experiment was performed at @xmath102 , the ce magnetopolarisability is equal at most to 50 % of the gce value ( in the limit of small level broadening ) . as a consequence the expected value for @xmath103 is then @xmath104 which is of the same order of magnitude as the experimental value . note that the measurement is not sufficiently accurate to give an estimate of the level broadening by comparing the experimental result with the curves of figure [ fig : polarce ] . a very interesting extension of the experiment would be to study the magnetopolarisability at different frequencies in order to test the theoretical predictions . this could be done by working with resonators with smaller inductances . component of the fourier transform of the signal . the fitting function used is proportional to @xmath105 with two fitting function for @xmath106 . first we took the phase coherence length measured on connected sample @xmath107 , which exhibits a @xmath108 behaviour . the other fitting function is @xmath109 . the best agreement is found with an exponential decay with a temperature scale of 90 mk . inset : temperature dependence of the frequency shift due to the rings.,width=298 ] the temperature dependence of the signal is also investigated . the magnetopolarisability decreases with temperature ( inset of figure [ fig : polartemp ] ) . theoretically the effect of temperature on magnetopolarisability has not been studied yet . we will base our analysis of the temperature dependence on the hypothesis that the amplitude of the signal is related to the phase coherence length @xmath2 in the same way as weak - localization . in this case the amplitude of the @xmath6 component of the signal is proportional to @xmath110 @xcite . in figure [ fig : polartemp ] the temperature dependence of this component is shown . we have tried to fit it using two laws for @xmath106 . first using the behaviour deduced from the measurements on connected wires @xcite we have tried the experimental value @xmath107 which exhibits a @xmath108 dependence , as expected for 1d system @xcite . it leads to a poor agreement with experimental points . using for the phase coherence time the result of electron - electron interaction in quantum dots ( od system ) @xcite @xmath111 , leading in the diffusive regime to @xmath112 , gives a better agreement . in this case the temperature scale is found to be 90 mk . we deduced from this value @xmath113 mk at 18 mk , i.e. much smaller than the level spacing . the phase coherence length deduced from this analysis is 10 times higher than the length measured on connected sample @xcite . we relate this difference between the non - connected and connected case to the fact that whereas the connected samples are one dimensional with a continuous energy spectrum due to the strong coupling with the reservoirs , the spectrum of the non - connected rings is discrete . as a concluding remark on this temperature dependance , we want to emphasize the need of a deeper theoretical analysis of the behaviour of the magnetopolarisability versus temperature . component before numerical integration , for different illumination times.,width=298 ] using the procedure described in [ sec : rings ] we are able to study the influence of electronic density on magnetopolarisability . on figure [ fig : fftpolarillum ] the fourier transform of the derivative of frequency shift , when the base line due to the resonator is removed , is shown at different illumination time . as expected the fourier transform exhibits a @xmath6 peak . note also the low frequency component which corresponds to the aperiodic signal seen on figure [ fig : polarextract ] ( b ) . the @xmath6 peak depends on electronic density . its amplitude shows first an increase and then decreases at high illumination . moreover the width of the @xmath6 peak increases showing that the rings widen with illumination . the peak becomes asymmetric suggesting that rings having long circumferences initially not populated contribute to the signal at high illumination . following the procedure described in section [ magnetopolarisabilty ] we measured the amplitude of magnetopolarisability versus illumination time . it yields figure [ fig : effetillum ] , which shows the change of the amplitude of the magnetopolarisability . at first the signal increases , then for illumination time above 1400 s the amplitude of oscillation decreases . we interpret this behaviour in the following way . before illumination the electronic density in the rings obtained after deep etching of the 2deg is strongly depressed compared to the nominal value . as a consequence an important fraction of the rings are likely to be localized and do not contribute to the magnetopolarisability . in this regime the signal is small . after illumination the number of rings contributing to the signal increases so that the frequency shift due to the rings increases . at high enough electronic density when the rings are sufficiently populated so that they contain delocalized electrons the theoretical results obtained in the diffusive regime are expected to be valid leading to a @xmath8 dependence ( formula [ da ] ) , with @xmath114 the dimensionless conductance . this is a possible explanation for the decrease of the magnetopolarisability observed at high illumination level . note that we can not exclude also a reduction of the screening length @xmath17 due to illumination . from formula [ da ] we deduce then a decrease of the signal . however we believe that the screening length changes very weakly with illumination because this length is essentially determined by the density of states at the fermi energy , which is independent of energy for a 2d system ( see table [ tab : characteristicsoftherings ] ) . we have analysed so far the @xmath6 periodic component of the magnetopolarisability signal obtained after filtering low frequency ( see figure [ fig : polarextract ] ) . on the other hand the whole integrated signal is depicted in figure [ fig : polarstep ] . one can clearly see a triangular shape dependence of the signal with magnetic field superimposed to the oscillations , very similar to the weak - localisation conductance of the connected mesh shown in figure [ fig : angr ] . note that this behaviour is only present at low temperature , it completely disappears for temperature higher than 300 mk . the amplitude of this extra signal due to the finite width of the rings strongly increases and sharpens with illumination . we think that it is reasonable to attribute this evolution to the increase of the width of the rings . note that a similar evolution has been previously observed in the ac magnetoconductance of ballistic gaas squares @xcite . versus @xmath32 in gce at different value of the parameter @xmath68.,width=298 ] , width=298 ] by measuring the quality factor of the resonator versus magnetic field , we have access to the flux dependent electric absorption ( formula [ qshift ] ) , which is related to the conductance , in the case of an electric dipole , through : @xmath115 the contribution due to the rings ( figure [ fig : fig2 ] ) exhibits the same periodicity as the frequency shift , which corresponds to half a quantum flux in a ring . the low field signal decreases . it corresponds to a negative magnetoconductance , i.e. opposite to weak - localization . this surprising sign was pointed out in the context of the magnetoconductance of rings submitted to an oscillating magnetic flux in the discrete spectrum limit @xcite . to explain this result one has to take into account the level spacing distribution in a disordered system @xcite . the sign and amplitude of the typical variation of electric absorption are understandable using the fact that level repulsion in a disordered system is higher in gue than in goe . following reference @xcite the flux dependent electric absorption in a system described by eigenvalues @xmath116 and the corresponding eigenfunctions @xmath80 could be written in a linear response regime : @xmath117 with @xmath118 . we will first consider this expression in the gce where an average over the chemical potential is computed . with this procedure @xmath119 and @xmath120 , where @xmath121 is the range over which the average over the chemical potential is done . the first term in the right hand side of equation [ absorption ] then reads : @xmath122 note that in this sum the energies @xmath123 and @xmath124 have to belong to the range @xmath125 $ ] . using this constraint we replace the sum by an integral : @xmath126 with @xmath98 the two energy level correlation function . in this expression we neglect the flux dependence of the matrix element and note its average value @xmath127 . with this approximation equation [ nondiag ] reads : @xmath128 the debye term of equation [ absorption ] is equal in the gce to : @xmath129 in the gce in the dynamical regime the flux correction to the polarisability is given by formula [ dagce ] at @xmath130 . using the following correlation function @xcite : @xmath131 \pi_d(\mathbf{r_1},\mathbf{r_2 } ) \quad ( goe)\end{gathered}\ ] ] we have @xmath132 . hence : @xmath133 which can be evaluated numerically ( figure [ fig : electricabsorption ] ) . for isolated rings we have to apply the result of ce . it is then possible to estimate the correction to electric absorption by using the same treatment as for the real part of polarisability . it leads to : @xmath134\end{gathered}\ ] ] numerical estimation of this formula at different value of the level broadening leads to the behaviour shown on figure [ fig : absorpce ] . the electric absorption is always negative at low frequency and may change sign at low value of @xmath68 . in order to compare these calculations with our experimental result we will compute the ratio @xmath135 . it is worth noting that in our experiment this quantity is given according to equations ( [ frequencyshift],[qshift ] ) by @xmath136 and is independent of the number of rings coupled to the resonator or the electric coupling coefficient @xmath35 . experimentally we find @xmath137 at illumination time zero and -0.23 after 420 s of illumination . theoretically , at @xmath138 , @xmath139 and @xmath140 so that the expected value of @xmath141 is around 1 . it corresponds to small @xmath68 . for higher value of this parameter the ratio is of the same amplitude or higher . as a consequence the predicted behaviour is consistent with the experimental value for the sign , but the theoretical amplitude is too high by more than a factor 2 . this conclusion is different from our previous statement where the frequency dependence of the real part of magnetopolarisability was not taken into account @xcite . .,width=298 ] due to the design of the resonator we have also the opportunity to investigate the magnetic response of the same aharonov - bohm rings used for the measurements of the electric response . in this case the rings are placed on top of the inductive part of the resonator . note that to do so we have to warm - up , cool down and re - illuminate the rings . as a consequence , strictly speaking , the rings are not the same than for the measurement of the electrical response because the electronic density and the disorder realisation in each ring is not exactly the same from one run to the other . nevertheless due to the fact that we are measuring an ensemble average quantity the change in disorder realisation of each ring does not modify the result of the experiment . moreover we have checked ( on the electric response measurement ) that , for the same illumination procedure , the result varies within a 15 % range from one run to the other . the flux dependent orbital magnetism at a frequency of 350 mhz is then detected . in this configuration the quality factor of the resonator is only 500 . we do not understand this strong increase of magnetic losses . this low quality factor decreases the accuracy of our measurements of the resonance frequency . moreover it prevents precise measurements of the flux dependence of the dissipative part of magnetic response of the rings . as a consequence in this part we present only the flux dependent non dissipative part of the magnetic response . note that we can not rule out the fact that the signal measured in this configuration could be partially due to electric response of the rings . however due to the small value of the residual capacitance of the meander line and the bad electric coupling in this geometry this electric component is estimated to be at least 20 times smaller than when the rings are placed on top of the capacitance . moreover the very different shape of the electric and magnetic signals is a strong evidence that we are indeed measuring essentially the magnetic response of the rings . the signal measured at zero illumination , after subtracting the base line due to the resonator , is shown in the lower part of figure [ fig : magneticsignal ] . inspired by our previous analysis we decompose the measured field dependent part of the signal into an aperiodic and a periodic part which corresponds to @xmath6 in a ring ( figure [ fig : magneticsignal ] ) . we interpret the @xmath6 component as the contribution of electronic trajectories enclosing the whole ring . on the other hand the triangular shape signal could be due to the contribution of trajectories confined in the finite width of the ring . the amplitude of the @xmath6 periodic component of the signal is @xmath142 . we deduce from formula [ dfmagnetic ] and evaluation of the magnetic coupling done in appendix [ magnetic_coupling ] that the flux dependent magnetic response of the ring is @xmath143 m@xmath144 . in the following we first assume that the main contribution to this signal is due to the flux derivative of the persistent currents @xcite and then discuss finite frequency effects . if the flux dependence of persistent currents is @xmath145 , we deduce : @xmath146 with @xmath147 the surface of the ring . we find then a _ diamagnetic _ average persistent current , the amplitude of which is : @xmath148 na . the aperiodic component of the signal corresponds on the other hand to low field paramagnetism . let s now compare our result for the average persistent currents to other experimental results and to theoretical predictions . a @xmath6 periodic diamagnetic persistent currents has been also observed in arrays of metallic rings @xcite . the expected amplitude of the averaged current due to repulsive electron - electron interactions from first order hartree - fock calculation @xcite is @xmath149 na , this value is expected to be decreased by higher order terms . considering on the other hand theoretical predictions for non interacting electrons @xcite the expected value is between @xmath150 na and @xmath151 na . in both cases the currents are predicted to be paramagnetic . the rather small difference between interacting and non interacting electrons is very specific to the gaas rings where the number of electrons is small . the measured signal is consistent for the amplitude but not for the sign ( unless assuming attractive interactions ) with theoretical predictions . it may also be important to take into account an effect of frequency for the flux dependent orbital magnetism . in fact by applying a formalism very similar to the one used for magnetopolarisability the variation of the real part ( non - dissipative ) of the susceptibility of a ring submitted to an oscillating magnetic flux in ce without interactions is given by @xcite : @xmath152 with @xmath153 the matrix element of the current operator . it is then possible to apply the same reasoning as for the real part of polarisability , and to use the fact that @xmath154 , so that : @xmath155 \right)\end{gathered}\ ] ] the evaluation of this expression is easily deduced from the evaluation of magnetopolarisability and leads to figure [ fig : magneticce ] . frequency induces a strong decrease of the magnetic signal for frequencies of the order of @xmath156 but does not seem to induce a sign change of @xmath157 . note that in strong contrast with the electric response the magnetic susceptibility is maximum at zero frequency , which corresponds to persistent currents . it would be important to investigate the effect of finite frequency on the contribution due to electron - electron interactions on persistent currents . recently it has been suggested that the measured currents may be due to a rectifying behaviour of the rings : a high frequency noise leads then to a dc current @xcite . noise also induces dephasing . a recent paper by kravtsov and altshuler @xcite predicts that those two quantities , average persistent current and dephasing measured by the saturation of the phase coherence time , are related in a simple way @xmath158 . @xmath159 is a constant giving the sign of the persistent current and @xmath160 the dephasing time at zero temperature . using the value of @xmath161 ns at 20 mk , deduced from measurements on connected sample , and considering the orthogonal case ( absences of spin orbit , then @xmath162 ) we deduce an expected value for persistent currents of -0.14 na . the predicted persistent current is then _ diamagnetic_. the sign and amplitude are then consistent with our experimental findings . on the other hand if we take the value deduced from the temperature dependance of the magnetopolarisability of non - connected rings , which is not the case considered theoretically , we deduce @xmath163 na , smaller by an order of magnitude than the experimental value . the influence of electronic density on the magnetic response of the rings has been investigated by illuminating them . different illumination times are shown in figure [ fig : magneticillum ] . we observed that the triangular envelope of the signal changes sign and increases with illumination . for each illumination the fourier transform of the signal exhibits a component which is consistent with half a flux quantum periodicity ( figure [ fig : magneticillumfft ] ) . one sees however that with illumination the shape of the @xmath6 peak in the fourier transform is modified . the peak broadens with electronic density indicating that the width of the rings increases . note that this width is always consistent with the one deduced from etching and depletion effects . the power of the fourier transform integrated in the @xmath6 zone is constant within 10 % . so the amplitude of the @xmath6 signal is constant but its shape is modified . it indicates that the amplitude of the persistent currents does not depend much on electronic density . the sign change of the low frequency part of the flux dependent magnetic response of the rings , going from low field paramagnetism to diamagnetism , is not understood . we have presented measurements of electric and magnetic responses of aharonov - bohm rings etched in a 2deg . they present a flux dependent correction to screening . this correction is positive in low field which means that screening is enhanced when time reversal symmetry is broken by a magnetic field . the sign of the effect is consistent with theory for isolated rings at finite frequency . the value of magnetopolarisability is @xmath164 , with @xmath165 the calculated polarisability of a quasi-1d circular ring of radius r. the temperature dependence of magnetopolarisability is consistent with @xmath166 . the behaviour versus electronic density is compatible with a @xmath8 dependence of magnetopolarisability . the magnetic response has been measured on the very same array of rings . the rings exhibits a signal consistent with _ diamagnetic _ average persistent currents of amplitude @xmath167 na . because the measurements are done on the same rings it is possible to compare the electric and the magnetic signal . the experimental ratio between the frequency shift due to the electric or magnetic response is around 10 , a value consistent with theoretical expectations taking into account the electric and magnetic coupling coefficient ( appendix [ magnetic_coupling ] and [ electric_coupling ] ) and the ratio between the typical matrix element of the screened potential and the current operator which leads to @xcite : @xmath168 with @xmath169 the vacuum impedance , @xmath170 the drude conductance and @xmath171 the fine structure constant . we have thus shown that the mesoscopic electromagnetic response of gaas rings is dominated by the flux dependent polarisability instead of orbital magnetism . this would not be the case in metallic rings , where , due to the very short screening length , the mesoscopic electric response is negligible . the low field diamagnetic sign of the orbital magnetism needs further investigation both on the experimental and theoretical sides . we thank b. etienne for the fabrication of the heterojunction . we acknowledge fruitful discussions with l.p . lvy , ya.m . blanter , s. guron and g. montambaux and technical help of m. nardone and p. demianozuck . in this appendix we evaluate the magnetic coupling of one square ring with the resonator in the configuration of the experiment . the inductance @xmath25 is modelled by two cylindrical wires separated by a distance of @xmath172 ( see figure [ fig : res_anneaux ] ( a ) ) . a ring is submitted to the magnetic field of those wires . let s first evaluate the mutual inductance @xmath173 between the ring and the resonator . using ampere theorem the magnetic field generated by a current @xmath174 in the the inductance is easily calculated . the flux of this magnetic field through a ring of size @xmath175 located at the ( 0,0 ) point is then : @xmath176 if the ring is located at a point ( x , y ) @xmath173 reads : @xmath177 the ring submitted to a magnetic field @xmath178 acts as a magnetic dipole @xmath179 . this dipole is equivalent to the ring with a current @xmath180 , so that the flux in the inductance is now @xmath181 . we deduce from these results that @xmath182 with @xmath183 the inductance of the meander - line . from the resonance frequency and the calculation of the capacitance we deduce @xmath184h , . the rings are not perfectly well coupled to the inductance so that they are not all located at @xmath185 . moreover because of the mylar sheet inserted between the rings and the resonator the rings are not in the plane of the resonator . to take this into account the inductance is averaged over the x position of the rings and @xmath186m@xmath187 m . within this approximations : @xmath188 m@xmath144 . is determined by the polarisability of the rings and the electric field generated by the capacitance at the ring position . , width=302 ] in this appendix we evaluate the electric coupling coefficient @xmath189 of one ring with the capacitance @xmath26 of the resonator . the capacitance is modelled by two cylindrical wires of radius @xmath24 and separated by a distance @xmath172 , one wire with a linear charge of @xmath190 , the other one @xmath191 . the electric field outside the wires is the one generated by two lines of lineic charge @xmath190 and @xmath191 separated by a distance @xmath192 determined by @xmath193 @xcite . in our case @xmath194 pf . using gauss theorem we can easily calculate the electric field in the plane of the rings in every point @xmath195 outside the wires : @xmath196 a ring submitted to this field generates an electric dipole @xmath197 with @xmath12 the polarisability of one ring , so that the rings submitted to electric field act as an ensemble of dipole . we model them by two line of lineic charge @xmath82 and @xmath198 separated by a distance @xmath175 , and such that @xmath199 ( see figure [ fig : res_anneaux ] ) . note that to do so the electric field has to be constant on the scale of the rings : this is roughly the case . evaluating the potential @xmath200 created by these two wires between each side of the capacitance , and using the relation @xmath201 we have for rings located at ( x , y ) : @xmath202 to have the capacitance shift induced by one ring we have to divide the previous result by the number of rings @xmath203 with @xmath204 the length of the capacitance . moreover as the ring are imbedded in gaas - algaas we have to divide our result by the dielectric constant of the substrate @xmath205 . we can now evaluate the electric coupling coefficient @xmath35 defined by @xmath206 by averaging over the x - position of the rings and considering that the rings are located between 1.5 @xmath23 m and 2.5 @xmath23 m in the y - direction from the resonator . within these approximations @xmath207 m@xmath144 . note that the previous result is very close to the value of the magnetic coupling coefficient @xmath46 . in this part we evaluate the magnetopolarisability given by formula [ polardiffuson ] for a quasi-1d ring . the diffusion propagator at frequency @xmath32 is given by : @xmath208 @xmath147 and @xmath156 are respectively the surface of the ring and the mean level spacing . @xmath209 and @xmath210 are the eigenvalues and eigenvectors of the diffusion equation : @xmath211 we consider a 2d ring of perimeter @xmath18 , radius @xmath67 and width @xmath212 , with @xmath213 . in this case the solutions of the diffusion equation are : @xmath214 with @xmath215 and @xmath216 . the modes corresponding to @xmath217 are given by : @xmath218 @xmath219 is the coordinate along the ring , and @xmath220 the radial coordinate . in our description the ring corresponds to @xmath221 $ ] . we consider reflecting border in the @xmath220 direction.the corresponding eigenvalue is : @xmath222\ ] ] @xmath223 is the thouless energy . the mean charge density ( average over the width of the ring ) in the ring submitted to an electric field @xmath79 is : @xmath224 note that using this density we recover the classical result for a quasi-1d ring @xmath165 . in the thomas - fermi approximation we deduce the mean screened potential : @xmath225 using this relation and the formula for the diffuson one can do the calculation analytically . because of the form of @xmath78 only the mode ( m=0,n=1 ) remains and leads to : @xmath226 with @xmath227 .

the energy loss of hard partons propagating through the soft medium created in heavy - ion collisions has long been regarded as a promising tool to gain information on the medium properties @xcite . radiative energy loss , i.e. the idea that medium - induced radiation predominantly carries away energy from a hard parent parton has been rather successful in describing not only the hadronic nuclear suppression factor for central collisions but also the effects of changing medium geometry @xcite . calculations within dynamical evolution models in various formalisms @xcite have improved on this result and show also agreement with measured hard back - to - back correlations @xcite and the measured suppression of protons @xcite . nevertheless , there are indications that a radiative energy loss picture fails to describe the suppression of heavy quarks as seen in the measurements of the single - electron spectra @xcite . energy loss due to elastic collisions with the medium @xcite has been suggested as a possible solution to this problem . such calculations indicate a large component of elastic energy loss also for light quarks and gluons . however , the pathlength dependence of elastic energy loss does not agree with the observed response of the suppression to a large pathlength bias as probed in back - to - back hadron correlations @xcite . in the light of these inconclusive findings , it would be desirable to have a baseline calculation of elastic energy loss which treats the relevant physics in as detailed manner as possible . in particular , many analytical calculations of elastic energy loss assume eikonal propagation of a hard parton , however in a more realistic description even hard partons , @xmath2 gev , can be deflected from their path by interactions with the medium . similarly , in processes like @xmath3 there is not only a change of energy between incoming and outgoing parton but also a change in parton species which influences the subsequent interaction probability of the outgoing parton . finally , in scattering processes in which the energy is distributed almost evenly among the outgoing partons , it is not justified to use the picture of energy from a hard parton being lost into the medium . clearly , one also has to account for situations in which both partons after a scattering process are hard or both become part of the medium , or in which the identity of the leading parton changes . in the present manuscript , we present work towards such a baseline computation . for this purpose , we develop a monte carlo ( mc ) simulation for the hard parton s interaction with the medium , as this is an appropriate framework to account for all the above mentioned effects . in addition , we also try to to give estimates on the uncertainties of such a computation . similar , in some aspects even more ambitious approaches are the mc models jewel ( jet evolution with energy loss ) @xcite and martini ( modular algorithm for relativistic treatment of heavy ion interactions ) @xcite , both of which consider pqcd interactions of a parton shower with a thermal medium and include both elastic and radiative energy - loss components . in jewel , one models shower evolution and medium - induced radiation in a static ( no flow , @xmath4 constant ) medium , and in martini the vacuum showering is assumed to cease when the hydrodynamic evolution starts , while our approach focuses on the leading partons . like martini , also we utilize a realistic well - tested 3-dimensional hydrodynamical description of the medium ( in our case azimuthal symmetry in central collisions , and longitudinal boost symmetry but 3-d flow ) instead of a static medium applied in jewel . in the following , we make the assumption that the medium can be characterized as a thermal gas of quasi - free quarks and gluons and that its interactions with a hard parton can be treated as incoherent @xmath5 processes in leading - order pqcd . we stress that these assumptions need not be realized for the medium produced in ultrarelativistic heavy - ion collisions any discrepancy between the calculation presented here and the data will reflect this . we model the medium by means of a hydrodynamical calculation @xcite with the initial state obtained from the saturation of the phase space with minijets @xcite . this provides the medium temperature @xmath4 at each spacetime point . for the moment , we assume that only the scatterings with a partonic medium are relevant and ignore the possibility that a hadronic medium might interact with a hard parton . thus the mean free path of hard partons becomes infinite once the medium temperature drops below the critical temperature @xmath6 . given the local @xmath4 , the distribution of scattering partners is given by the fermi - dirac ( bose - einstein ) distribution for quarks ( gluons ) respectively . we also account for the 3-d hydrodynamical flow of the fluid cell . our approach , most specifically the assumption that the relevant interactions can be described in terms of partonic pqcd interaction , shares many features with parton cascade models ( e.g. @xcite ) . however , unlike in parton cascade models where the partonic pqcd interactions are assumed to describe the whole off - equilibrium bulk medium , we assume that the bulk is essentially thermalized and use explicit pqcd interactions only to account for the non - equilibrium dynamics between hard parton and medium . in the following , we first give a more detailed account of the hydrodynamical simulation , the parton - medium scattering rates and details of the mc implementation . then we present our results for the partonic and hadronic nuclear modification factor @xmath7 and for qcd - matter tomography , and also discuss the relevant uncertainties . some technical details regarding the scattering rate calculations can be found in the appendix . in the present exploratory stage we consider only the cleanest possible case , ( nearly ) central @xmath8 collisions in the mid - rapidity region . the produced qcd matter , which evolves and flows 3-dimensionally , can to a first realistic approximation then be described by using the nonviscous , azimuthally symmetric , longitudinally boost - invariant hydrodynamic framework . the 1 + 1 d hydrodynamical framework , which we apply here for the qcd medium evolution , is presented in detail in ref . the initial energy density @xmath9 for the hydrodynamics is here obtained from pqcd - minijet production @xcite and final - state saturation as computed in the ekrt - model @xcite . the transverse profile of @xmath10 is assumed to scale with the number of binary collisions , and thermalization is assumed to take place at formation ( i.e. at saturation ) . for central au+au collisions at @xmath11 gev , we thus initialize the hydrodynamics at ( see table 1 in @xcite ) @xmath12 fm/@xmath13 , where the longitudinal proper time is defined as @xmath14 . the transverse flow is assumed to be zero at @xmath15 . the equation of state ( eos ) in our hydrodynamic background contains a high-@xmath4 qgp phase , a low-@xmath4 hadron resonance gas ( hrg ) phase , and a mixed phase between them . the qgp here is an ideal gas of gluons and three flavours of massless quarks and antiquarks obeying a bag - eos , while the hrg is an ideal gas of all hadrons and hadron resonances up to @xmath16 gev . the maxwell construction gives the mixed phase , and the bag constant @xmath17 mev fixes the phase transition temperature to @xmath18 mev . we terminate the hydrodynamic evolution at a single decoupling temperature @xmath19 mev . as discussed in ref . @xcite , the measured pion and kaon spectra and proton multiplicity in central au+au collisions at rhic can be reproduced very nicely with such hydrodynamical set - up , once the thermal hadron spectra have been computed via the cooper - frye decoupling procedure and once the strong and electromagnetic hadronic decays have been accounted for . for more details , as well as for plots of the initial energy density and of the phase boundary locations in the @xmath20 plane , we refer the reader to ref . @xcite . our simulation of energy losses of high - energy partons in the produced qcd matter rests largely on one particular physical quantity , namely the scattering rate for a high - energy parton of a type @xmath21 , @xmath22 where all possible elastic and inelastic partonic processes @xmath23 are accounted for by summing over all types of collision partners , @xmath24 in the initial state , and over all possible parton type pairs @xmath25 in the final state . in general , the scattering rate depends on the frame , and in particular on the high - energy parton s 4-momentum @xmath26 , on the flow 4-velocity @xmath27 and on the temperature @xmath28 of the fluid at each space - time location @xmath29 . in the notation below , we leave these dependences implicit but in the simulation they are fully accounted for . ignoring pauli blocking and bose enhancement , we express the scattering rate as follows ( see also ref . @xcite ) @xmath30 where @xmath31 is the spin- and colour - summed / averaged scattering amplitude which depends on the standard mandelstam variables @xmath32 , @xmath33 and @xmath34 . the distribution function for a thermal particle of momentum @xmath35 is denoted by @xmath36 , which is the bose - einstein distribution @xmath37 for gluons and the fermi - dirac distribution @xmath38 for quarks . since the initial state spins and colours have been averaged over in the squared matrix elements , the spin- and color - degeneracy factors in @xmath39 and @xmath40 are the standard ones , @xmath41 for gluons , and @xmath42 for quarks and antiquarks , while the possible flavour - degeneracy needs to be determined separately in each process . in our simulation , we evaluate the scattering rate always in the local rest - frame of the fluid , so that @xmath43 and @xmath44 is the energy of the high - energy parton @xmath21 in this frame . the singularities arising in forward and backward scatterings of massless partons appearing in the scattering amplitudes at @xmath45 need to be regularized in some physically meaningful way . for the moment , an exact field - theoretical solution to this non - equilibrium problem is not known . one possible treatment is to apply resummed thermal ( equilibrium ) propagators separately for the longitudinal and transverse components , as is done in the amy approach @xcite . since the problem is , however , still phenomenological , we choose here a more transparent approach ( tested also in jewel before ) and introduce a thermal - mass - like overall cut - off scale @xmath46 for the momentum phase space , such that @xmath47 . here @xmath48 is the strong coupling constant and @xmath49 is a parameter of the order of one but which we can vary . as we can not control the running of @xmath50 , we keep the strong coupling constant fixed with momentum scale but vary both @xmath48 and @xmath49 in different runs of the mc code to test the model sensitivity to these parameters . since @xmath51 , the cut - off conditions are equivalent to @xmath52 with the requirement @xmath53 . these are encoded in the function @xmath54 above , which is defined as @xmath55 for the evaluation of the lorentz - invariant part in eq . , it is useful to recall that the scattering rate can be expressed in terms of the scattering cross sections . for the process @xmath23 , where the lorentz - invariant cut - off conditions are imposed , and the identities of the final state particles are accounted for , the cross section is @xmath56 and we can rewrite eq . as @xmath57 where @xmath58 is the thermal particle density in the rest frame of the qcd fluid , and where we have expressed the mandelstam variable as @xmath59 , i.e. in terms of the angle @xmath60 between the hard parton and the thermal parton , and their energies in the fluid rest frame . changing the integration variable @xmath61 to @xmath32 , it is straightforward to arrive at the following result @xmath62 where @xmath63.\ ] ] the analytical results for the relevant elements of @xmath64 are listed in the appendix . the remaining @xmath65 integral is easiest to do numerically . the obtained scattering rates @xmath66 for quarks , antiquarks and gluons , along with the contributions @xmath67 from different processes , are shown in fig . [ fig : rates_vs_t ] as a function of temperature @xmath4 and in fig . [ fig : rates_vs_e1 ] as a function of the hard parton energy @xmath44 . of gluon ( above ) and quark ( below ) as a function of temperature @xmath4 for a qcd plasma at rest . flavour- and quark - antiquark -summed contributions from different processes and the analytical estimates for the total rates discussed in the text are also shown ( solid lines).,title="fig:",width=340 ] of gluon ( above ) and quark ( below ) as a function of temperature @xmath4 for a qcd plasma at rest . flavour- and quark - antiquark -summed contributions from different processes and the analytical estimates for the total rates discussed in the text are also shown ( solid lines).,title="fig:",width=340 ] of gluon ( above ) and quark ( below ) as a function of the parton energy @xmath44 , for a qcd plasma at rest at @xmath68 gev . flavour - summed contributions from different processes are shown . the analytical estimates for @xmath66 ( see text ) are shown by the solid lines . in the upper panel , the analytical estimate for the @xmath69 process ( dashed line ) is the same as for the quark - antiquark summed @xmath70 process . in the lower panel , the dashed lines show the contributions from the @xmath71 process and an individual @xmath72 process . , width=340 ] of gluon ( above ) and quark ( below ) as a function of the parton energy @xmath44 , for a qcd plasma at rest at @xmath68 gev . flavour - summed contributions from different processes are shown . the analytical estimates for @xmath66 ( see text ) are shown by the solid lines . in the upper panel , the analytical estimate for the @xmath69 process ( dashed line ) is the same as for the quark - antiquark summed @xmath70 process . in the lower panel , the dashed lines show the contributions from the @xmath71 process and an individual @xmath72 process . , width=340 ] to get a feeling of the general characteristics of the system , we consider also the following analytical estimates . in the process @xmath23 , the average of a quantity @xmath73 in the plasma at a temperature @xmath4 can be computed in the fluid rest frame as @xmath74 at the high - energy limit , @xmath75 , the relevant pqcd cross sections all behave as @xmath76 , and @xmath77 ( see appendix , take the limit @xmath78 , and @xmath79 ) . in the boltzmann approximation we thus get @xmath80 from which we see that , in this limit , @xmath81 ) . the last result in eq . indicates that the collisions of the high - energy parton with the medium partons happen on the average in an angle @xmath82 . using the same approximation , we can easily obtain analytical estimates also for the scattering rates , @xmath83 \nonumber \\ & \approx & ( g_g + 2\cdot3\cdot g_q\cdot{\textstyle \frac{4}{9 } } ) \frac{9}{2}\frac{\alpha_s^2t^3}{\pi m^2 } = 144\frac{\alpha_s^2t^3}{\pi m^2}\\ \label{analytic_gammaq } \gamma_{q_i } & \approx & \gamma_{{q_i}g\rightarrow { q_i}g } + \sum_{j}[\gamma_{{q_i}{q_j}\rightarrow { q_i}{q_j } } + \gamma_{{q_i}{\bar q_j}\rightarrow { q_i}{\bar q_j}}]\nonumber\\ & \approx & ( g_g\cdot{\textstyle \frac{4}{9 } } + 2\cdot3 \cdot g_q\cdot({\textstyle \frac{4}{9}})^2)\frac{9}{2}\frac{\alpha_s^2t^3}{\pi m^2 } = 64 \frac{\alpha_s^2t^3}{\pi m^2}\end{aligned}\ ] ] where the factors @xmath84 account for quarks and antiquarks times the number of quark flavours . these estimates are shown in figs . [ fig : rates_vs_t ] and fig . [ fig : rates_vs_e1 ] , together with the actual rates applied in our simulation ( once the boost to the local rest frame has been taken into account ) . we observe that the analytical rates describe the numerical ones ( both the total rate and the ones from the main processes ) quite well in the kinematic and temperature range of rhic . in a static homogeneous qgp , the average number of collisions experienced by the high - energy parton of an average path length @xmath85 , would , in the high - energy limit , be @xmath86 . taking @xmath87 fm and using eqs . - with @xmath88 and @xmath89 , we arrive at the range 145 collisions for gluons and 62 collisions for quarks at @xmath90 mev . this gives us a simple order - of - magnitude estimate of the expected number of elastic scatterings for a hard parton . to initiate the hard massless parton of a type @xmath21 in each event , we sample the lo pqcd single - jet production spectrum ( for more details , see @xcite ) at @xmath91 and at @xmath92 , @xmath93 \frac{1}{1+\delta_{kl } } , \end{split}\ ] ] where @xmath94 with @xmath95 containing the squared matrix elements listed in eqs . ( a-1)-(a-8 ) . for the parton distribution functions ( pdfs ) , we use the cteq6l1 set @xcite . for simplicity in this exploratory work we , however , neglect the nuclear effects to the pdfs @xcite , since these are small in comparison with the ones arising from the final state interactions with the medium . the initial rapidity @xmath96 is randomly generated in the range @xmath97 $ ] from a flat distribution . ( this is consistent on the level of the longitudinal boost - symmetry approximation made for the hydrodynamical part . ) this fixes the hard - parton energy @xmath98 and polar angle @xmath99 of its momentum vector . the azimuth angle @xmath100 is evenly distributed between @xmath101 $ ] . in order to keep track of the hard partons interactions with the medium in the coordinate space , the initial spatial position of the hard parton has to be fixed as well . we start to follow the hard parton at the initial longitudinal proper time @xmath15 of our hydrodynamical model . since in the c.m . frame all hard partons are produced in the lorentz - contracted overlap region at @xmath102 , the longitudinal position at later times ( before the first collision at @xmath103 ) is assumed to be determined by the longitudinal momentum only . the initial time and longitudinal coordinates for the hard parton are thus @xmath104 and @xmath105 . the initial position in the transverse plane is sampled from the nuclear overlap function @xmath106 where @xmath107 is the impact parameter . as we focus only on the central collisions , in the following @xmath108 . the nuclear thickness function @xmath109 is defined as usual , @xmath110 where for the nuclear density @xmath111 we use the standard woods - saxon distribution @xmath112 with @xmath113 , @xmath114 , @xmath115 fm and @xmath116 @xmath117 . the hard parton propagates through the plasma in small time steps @xmath118 , during which we propagate the parton in position space . since we consider only incoherent scatterings here , the probability for not colliding in the time interval @xmath118 is assumed to be given by the poisson distribution @xmath119 where @xmath66 is the total scattering rate for the hard parton of the type @xmath21 . hence the probability to collide at least once during the time @xmath118 is @xmath120 . for small enough @xmath118 we can assume that there will be at most one collision . however , as explained above , we calculate the scattering rates in the local rest frame of the quark - gluon plasma fluid element and we must boost also the time step @xmath118 to the same frame , @xmath121 where the lorentz factor @xmath122 appears , @xmath123 being the 3-dimensional flow velocity vector of the hydrodynamical plasma in the c.m . frame of the colliding nuclei . should a scattering happen , the probability @xmath124 for a given type of scattering process is determined by the partial scattering rates as @xmath125 we produce the 4-momentum @xmath35 of the scattering partner from the thermal medium according the energy and angular distributions defined by the differential scattering rate @xmath126 ( see eq . ) of the process . we then boost the 4-momenta of the hard parton @xmath26 and the thermal particle @xmath35 to the c.m . frame of the partonic collision , rotating the frame so that @xmath26 is on the @xmath127-axis . it is now easy to determine the scattering angle @xmath128 by sampling the distribution @xmath129 using @xmath130 , and sampling the azimuthal angle @xmath131 from a flat distribution . after determining the scattering angles @xmath128 and @xmath131 , we can construct the final state 4-momenta @xmath132 and @xmath133 and lorentz transform these back to the original c.m . frame of the colliding nuclei . the final state parton with highest energy is then chosen as the new hard parton to be propagated further , for which we repeat the procedure outlined above with the next timestep . note that the type of the hard parton may change here . ( we also note that in our code , we have the possibility to follow the recoil partons as well but at sufficiently high parton momenta ( @xmath2 gev ) , we do not see a significant effect on the observables studied here . ) this procedure continues until the surrounding matter has cooled down to the critical temperature @xmath6 . at @xmath134 the system is in this work in a mixed phase where energy density @xmath135 keeps decreasing while the temperature stays constant . it is _ a priori _ not clear how the interaction with the medium in the mixed phase should be modelled properly in pqcd . however , we will demonstrate _ a posteriori _ ( see fig . [ raapartonqgp ] ahead ) that within the assumption of an effective temperature , the effect of the mixed phase is small because the medium density is small as compared to the initial density . thus , unless there is an anomalous enhancement of parton - medium scattering in the mixed phase by non - perturbative processes , a detailed treatment of the mixed phase is not important , and we can also ignore the interactions of the hard parton in the hadron gas . in the following , we treat the interactions in the mixed phase by using an effective temperature @xmath136 where @xmath137 is the quark - gluon plasma degrees of freedom and @xmath138 is the bag constant . our simulation ends when @xmath139 and the system enters the hadron gas phase , where we assume no significant interaction between parton and medium . the outcome of the procedure described above is a medium - modified distribution of high - energy partons , @xmath140 . in order to compare with observables , we have to hadronize this distribution . in the current leading - particle case , this can be done by convoluting the obtained partonic distribution with the fragmentation function @xmath141 : @xmath142 where @xmath143 is the fraction of the final parton momentum @xmath144 available to the hadron with momentum @xmath1 , and @xmath145 is the fragmentation scale . in this work we have used the kniehl - kramer - ptter ( kkp ) fragmentation functions @xcite . in the following , we choose a lower bound of the parton transverse momentum , @xmath146 , as 5.0 gev to account for the observation that at low @xmath1 the main hadron production mechanism is not independent fragmentation . partons which fall below @xmath147 in our simulation are assumed to be thermalized and become part of the medium , and we do not follow these partons anymore . we average all observables across the rapidity window @xmath148 $ ] which corresponds to the phenix acceptance . to make sure we account properly for all the possible partons falling into this final rapidity window , we choose the initial @xmath149 . we refer to all partons above both the @xmath146 cut and within the final rapidity acceptance as punch - through partons. is shown by the dashed lines.,width=340 ] to verify the magnitude of the rapidity shifts experienced by the high-@xmath98 partons , we plot in fig . [ fig : rapidity ] the initial rapidity distribution of the punch - through partons in our default set - up with @xmath88 and @xmath150 . we can see that about 13 % of the punch - through partons originate from outside of the final rapidity interval . on the one hand , this suggests that the eikonal approximation where all partons would remain within the interval @xmath151 is a fairly good first approximation . on the other hand , if one aims at a precision analysis of the single - hadron distributions or , even more importantly , at correlations of two high-@xmath1 hadrons , the non - eikonal dynamics should be obviously taken into account . an attempt to quantitatively understand the rapidity shift seen in fig . [ fig : rapidity ] can be found in appendix [ sec : yshifts ] . also the leading - parton flavour - conversions are usually neglected in the eikonal energy - loss simulations . interestingly , this effect is not symmetric between quarks and gluons : due the collision rates , there are more flavour conversions related to gluons than to quarks , and a quark from a @xmath152 leading - parton conversion is then more likely to become a punch - through parton than a gluon from a @xmath153 conversion . our simulation shows that for a simplified homogeneous static plasma cylinder of , say , radius @xmath87 fm and @xmath154 mev , as much as 40 % of those original high-@xmath144 gluons which will become punch - through partons , will turn into quarks . in the full simulation with 3-d hydrodynamical background medium , however , only about 4% of all punch - through partons have undergone this effect . we thus conclude that in simulations for single - hadron distributions at rhic , where a realistically evolving background medium is considered , the leading - parton flavour - conversions can be neglected . nuclear modification factor both at hadron and parton level with strong coupling constant @xmath88 . phenix data is from @xcite.,width=340 ] in fig . [ raahadron ] we present the nuclear modification factor @xmath155 , @xmath156 for @xmath88 , @xmath150 , both on the hadronic level and on the partonic level , and compare it with the @xmath157 data from phenix @xcite . it is evident that for this default choice of @xmath158 and @xmath49 elastic collisions are not enough to account for the data but that they should not be neglected , either . this result is in agreement with the calculation of purely elastic energy loss in the arnold - moore - yaffe ( amy ) formalism averaged over hydrodynamics @xcite where a similar value of @xmath159 is found for @xmath160 . note that a comparison with models not using a constrained fluid - dynamics model for the bulk evolution or the same @xmath158 is not meaningful . it can further be deduced from the figure that since the suppression as a function of @xmath1 ( or partonic @xmath144 ) is flat , the convolution of the medium - modified parton spectrum with the fragmentation function does not cause a significant effect in @xmath155 . we will thus in the following focus on the partonic suppression pattern , with the implication that the partonic suppression provides a reasonable estimate for the suppression also on the level of hadrons . as a technical point , we note that in order to collect enough statistics also at high-@xmath144 , the partonic results in fig . [ raahadron ] consist of three dedicated runs , corresponding to @xmath161 gev . and @xmath150 , including only the scatterings in the qgp phase ( squares ) and in the qgp and mixed phase but not the in hadron gas ( circles).,width=340 ] in fig . [ raapartonqgp ] we demonstrate , as discussed above , that the mixed phase is in this set - up indeed causing only a small additional suppression and , consequently , that it is justified to neglect the hadron gas effects in the present study . the results shown in this and in the following figures are from runs with @xmath162 gev . for the regulator - mass parameter value @xmath163 and strong coupling constant values @xmath164.,width=340 ] in fig . [ raapartonalfa ] we show the partonic nuclear modification for three different choices of constant @xmath158 . as expected , we observe a rather strong dependence on the value of @xmath158 . the behavior is quite straightforward ; doubling the coupling approximately doubles the relative suppression . with @xmath165 we find a suppression due to elastic energy loss already below the data . on the other hand , with @xmath166 the effect is rather weak . we thus focus on @xmath88 in the following when examining other model parameters . for mass parameter values @xmath167 and strong coupling constant @xmath168.,width=340 ] as is obvious the other significant model parameter in our simulation is @xmath49 , controlling the momentum cutoff scale @xmath169 which regulates the scattering cross sections at smallest and largest angles . like @xmath158 , it has a direct effect on the scattering cross sections of a hard parton with medium partons . however , it could be expected to influence energy loss in a different way . while the total cross section can become very large if small - angle forward scattering processes are taken into account , such processes do not transfer a large amount of energy and are hence not efficient for energy loss . thus , a smaller cutoff mass is expected to increase the total cross section without affecting large - angle scattering whereas a change in @xmath158 affects both forward and large - angle scattering in equal measure . a comparison of different values of the cutoff mass in fig . [ raapartonmass ] shows , however , that this parameter has an unexpectedly large effect on the nuclear modification factor . the suppression increases with decreasing mass ; this indicates that the cutoff mass is not yet in the region where forward scattering , which is inefficient for energy loss , takes place . ( above ) @xmath168 ( middle ) and @xmath170 ( below ) . all particles are moving to the positive @xmath29-direction.,width=321 ] ( above ) @xmath168 ( middle ) and @xmath170 ( below ) . all particles are moving to the positive @xmath29-direction.,width=321 ] ( above ) @xmath168 ( middle ) and @xmath170 ( below ) . all particles are moving to the positive @xmath29-direction.,width=321 ] in fig . [ xyinit ] we show the geometry underlying the suppression in terms of the initial spatial distribution of punch - through partons for different values of @xmath158 . this figure can be considered as a tomography image , where we have rotated the position of each particle in transverse @xmath171-plane in such a way that they all move to the same direction , to the direction of positive @xmath29-axis . with @xmath172 ( upper panel ) , the quark - gluon plasma is quite transparent in transverse direction as far as elastic collisions are concerned , thus the initial distribution of punch - through partons is centered around the origin . doubling the value of @xmath158 to 0.3 ( middle panel ) makes the distribution move off - center . thus , the medium introduces a surface bias which means that hard partons from the center can not punch through the matter anymore . overall , the plasma becomes more opaque , as is expected from fig . [ raapartonalfa ] . the opaqueness is emphasized even more in the bottom panel , which corresponds to @xmath165 , which reproduces , or slightly overestimates the suppression measured at rhic . we have also checked the tomography images for different mass parameters ( @xmath49 ) which reproduce the same suppression pattern in @xmath155 as when @xmath158 is varied ( see fig . [ raapartonmass ] ) . our conclusion here is that these do not significantly differ from the ones we have shown in fig . [ xyinit ] . the results above indicate on the one hand that for our default choice of parameters @xmath168 and @xmath173 , elastic energy loss is not sufficient to account for the observed value of @xmath155 . on the other hand , we see that at least in the framework which is based on individual parton - parton scatterings instead of , e.g. , static scattering centers , the elastic energy losses give a non - negligible contribution to @xmath155 . this has also been observed by a comparable calculation within the amy formalism @xcite in which a combination of radiative and elastic energy loss was needed to account for the data . however , our results show an unfortunately strong dependence on cutoff parameters which are in essence supposed to effectively account for the physics of a dynamically running @xmath158 and thermal field - theoretical processes generating a screening mass . thus for a reasonable variation of these parameters there are pairs @xmath174 for which the @xmath155 data can be described by elastic processes alone . it is thus evident from these findings that a comparison based on the magnitude of the single hadron suppression alone can not resolve the question how significant the contribution of @xmath0 processes to energy loss actually is . to improve upon this problem , the next step is to include @xmath175 processes and coherence effects into the simulation . however , a comparison of the tomographic pictures in this paper ( fig . [ xyinit ] ) with the results of a radiative energy loss calculation @xcite indicates that the way hard partons probe the medium geometry through energy loss is quite different for radiative and elastic energy loss . thus , as outlined in a schematical way in @xcite it can be expected that pathlength - dependent observables , like the suppression as a function of the angle of the observed hadron with the reaction plane , or the suppression of back - to - back correlations , will show pronounced differences between elastic and radiative energy loss . the mc code introduced here provides an excellent tool to explore this idea , and this will be addressed in a future publication . the spin- and color summed / averaged squared parton scattering amplitudes @xmath176 for gluons and quarks are ( see e.g. @xcite ) @xmath177 inserting the expression for the cross section into the definition of the @xmath178-function we have @xmath179 . \end{split}\ ] ] several amplitudes contain identical terms . we list the integrals for each of these terms separately , as it is then easy to combine the appropriate terms to form the final expressions for @xmath180 . @xmath181 \nonumber \\ \hline \begin{array}{c } 1 \end{array } & 4e_1 e_2 -2m^2 \log{\frac{2e_1 e_2}{m^2}}-2m^2 \quad \\ \begin{array}{c } \frac{ut}{s^2 } \end{array } & \frac{2}{3 } e_1 e_2-\frac{3}{4}m^2+\frac{m^4}{4e_1 e_2}-\frac{m^6}{48e_1 ^ 2 e_2 ^ 2 } \quad \\ \begin{array}{c } \frac{su}{t^2},\frac{st}{u^2 } \end{array } & -\frac{8e_1 ^ 2 e_2 ^ 2}{m^2 } + 4e_1 e_2 \log{\left(\frac{4e_1 e_2}{m^2}-1 \right ) } + 2m^2 \quad \\ \begin{array}{c } \frac{t^2+u^2}{s^2 } \end{array } & \frac{8}{3 } e_1 e_2 -2m^2 \log{\frac{2e_1 e_2}{m^2 } } -\frac{1}{2}m^2-\frac{m^4}{2e_1 e_2}+\frac{m^6}{24e_1 ^ 2 e_2 ^ 2 } \quad \\ \begin{array}{c } \frac{t^2+u^2}{tu } \end{array } & 2(4e_1 e_2-m^2 ) \log{\left(\frac{4e_1 e_2}{m^2 } -1 \right ) } - 16e_1 e_2 + 4m^2\log{\frac{2e_1 e_2}{m^2}}+8m^2 \quad \\ \begin{array}{c } \frac{s^2+u^2}{t^2},\frac{s^2+t^2}{u^2 } \end{array } & \frac{16e_1 ^ 2e_2 ^ 2}{m^2}-8e_1 e_2 \log{\left(\frac{4e_1 e_2}{m^2}-1 \right ) } + 4e_1 e_2 -2m^2 \log{\frac{2e_1 e_2}{m^2}}-6m^2 \quad \\ \begin{array}{c } \frac{s^2+u^2}{su } \end{array } & -(4e_1 e_2-m^2 ) \log{\left(\frac{4e_1 e_2}{m^2}-1\right ) } + 2e_1 e_2 + m^2 \log{\frac{2e_1 e_2}{m^2}}-m^2 \quad \\ \begin{array}{c } \frac{s^2}{tu } \end{array } & 2(4e_1 e_2-m^2 ) \log{\left(\frac{4e_1 e_2}{m^2}-1\right ) } -8e_1 e_2 + 4m^2 \quad \\ \begin{array}{c } \frac{u^2}{st } \end{array } & -(4e_1 e_2-m^2 ) \log{\left(\frac{4e_1 e_2}{m^2}-1\right ) } + 10e_1 e_2 -3m^2\log{\frac{2e_1 e_2}{m^2}}-5m^2 \quad \end{aligned}\ ] ] here we try to understand quantitatively the magnitude of the rapidity shifts of punch - through partons shown in fig . [ fig : rapidity ] . at the high - energy limit @xmath182 , a straightforward calculation in the static plasma approximation ( see sec . [ subsec : rates ] ) gives the average scattering angle in the cms frame as @xmath183 we set the initial high-@xmath98 parton at zero rapidity and its momentum @xmath184 to point into the direction of the @xmath29-axis , while the beam direction is , as usual , the @xmath127-axis . having the maximum rapidity shift in mind , we take the medium - parton s momentum @xmath185 to reside in the @xmath186-plane , its energy to be @xmath187 and the collision ( polar ) angle relative to @xmath184 such that @xmath188 , which correspond to the averages in eq . [ averages ] . thus @xmath189 in a reverse boost from the cms frame of the parton - parton scattering , the energy and momentum of the scattered high - e parton become @xmath190 where the velocity of the cms frame is @xmath191 , @xmath192 is the lorentz gamma factor and the asterisks denote the cms - frame quantities . the momentum of the scattered high-@xmath98 parton in the cms - frame can be computed to be @xmath193 \nonumber\\ & + & { \bf \hat e_y}e^*\sin\theta_3^*\cos\phi_3^ * \\ & + & { \bf \hat e_z}[p_{1x}^*\sin\theta_3^*\sin\phi_3^ * + p_{1z}^*\cos\theta_3^ * ] , \nonumber\end{aligned}\ ] ] where @xmath194 is the scattering ( polar ) angle relative to @xmath195 and @xmath196 is the azimuthal angle around @xmath195 , measured from the @xmath197-axis , and where the cms - frame quantities are obtained in a boost to @xmath198 as @xmath199 the point here is that although in the cms frame ( where @xmath200 ) the scattering cross section does not depend on the angle @xmath201 , a @xmath201 dependence is generated in the boost back to the plasma rest frame . the rapidity shift @xmath202 is now easy to obtain numerically from the components @xmath203 and @xmath204 and we find that to a good approximation @xmath205 for a hard parton of , say , @xmath206 gev in a @xmath154 mev plasma , we obtain @xmath207 . the maximum shifts @xmath208 occur at @xmath209 and @xmath210 , i.e. when the scattering remains in the ( @xmath211)-plane . similarly , we find that @xmath212 and @xmath213 . to estimate the maximum possible total rapidity shift @xmath214 induced by @xmath215 scatterings , we may assume that at the limit of small scattering angles ( small rapidity shifts ) the maximum - shift distribution @xmath216 holds in each scattering . the rapidity shifts are additive and we can write the probability distribution for @xmath217 as follows : @xmath218 where the probability distribution for a rapidity shift in one collision , @xmath219 where @xmath220 , is obtained from the corresponding flat azimuthal angle distribution @xmath221 by a change of variables according to eq . [ deltay ] . now it is easy to obtain @xmath222 for an order of magnitude estimate , we take @xmath223 ( from our simulation , since for this static high-@xmath4 plasma test case the leading - parton flavour - coversions are in fact a non - negligible effect ) , and arrive at the average maximum total rapidity shift @xmath224 . figure [ fig : yshifts ] shows our simulation test results for the rapidity spread of punch - through partons which originate from 10 gev partons sent off from the origin into the @xmath29-direction . the width of the distribution in the static 500 mev plasma case ( squares ) is seen to be about 0.3 , which nicely confirms the order of magnitude of our upper limit estimate @xmath214 . the smaller width in the simulation follows from the fact that the simulation accounts for scatterings ( corresponding to a continuum of azimuthal incoming angles @xmath225 around @xmath184 ) , where the rapidity shift effects are smaller than in our estimate above . changing the background into a hydrodynamically evolving more realistic medium ( circles ) reduces the width of the rapidity shift by about a factor 3 . this is caused by the fact that collision rates , scattering angles @xmath194 and rapidity kicks all decrease with decreasing temperature . finally , in a full simulation , where the high-@xmath98 partons can be produced anywhere in the transverse plane , and where also the surface bias effect is present ( see fig . [ xyinit ] ) , the rapidity shifts become still somewhat smaller . from these considerations , and realizing that the rapidity spread distributions should be applied to each final - rapidity bin of fig . [ fig : rapidity ] , we conclude that the degree of non - eikonality ( @xmath197-shifts ) is understood , and also that if we wish to fully account for punch - through partons in the finite rapidity interval @xmath151 , we have to consider an initial window of @xmath226 . we thank harri niemi for providing us with the hydrodynamic simulations . j.a . gratefully acknowledges the grant from the jenny and antti wihuri foundation . this work was also financially supported by the academy of finland : k.j.e.s projects 115262 and 133005 , and t.r.s academy research fellowship with the associated project 130472 . k. j. eskola , k. kajantie and j. lindfors , nucl . b * 323 * , 37 ( 1989 ) . k. j. eskola , k. kajantie , p. v. ruuskanen and k. tuominen , nucl . b * 570 * , 379 ( 2000 ) . t. matsui , b. svetitsky and l. d. mclerran , phys . d * 34 * , 783 ( 1986 ) [ erratum - ibid . d * 37 * , 844 ( 1988 ) ] .

it is generally assumed that diffusion in macroscopic systems is a result of chaos on a microscopic scale , following einstein s 1905 explanation for brownian motion , that is , the movement of a colloidal particle due to thermal fluctuations in the surrounding fluid . we describe chaos , diffusion and related properties in detail in sec . [ s : fund ] . we define microscopic chaos ( usually shortened to just `` chaos '' ) in terms of unpredictability as quantified by positive lyapunov exponents or a positive kolmogorov - sinai entropy ; later we will find that properties based on periodic orbits also provide a useful description of chaotic aspects of the microscopic dynamics . note that we restrict the use of the term `` chaotic '' to its usual meaning , that is a positive kolmogorov - sinai entropy ; while the properties of periodic orbits are important , we do not classify systems as chaotic or nonchaotic based on these properties . we call a system diffusive when the mean square displacement of a particle is proportional to the time , or a distribution of particles satisfies the diffusion equation . a recent experiment of gaspard et al @xcite , described below , purports to show that the diffusion of a brownian particle is due to microscopic chaos . while we believe that brownian motion ( including both the brownian particle and the solvent ) is most likely chaotic , our simulations using nonchaotic models , in a brief comment @xcite and in greater detail here lead to the same results as found in the experiment , so that no experimental proof of microscopic chaos has been given in @xcite . here we explore the question of what kind of experimental measurements or data analysis might be required to identify microscopic chaotic dynamics , and the connection between microscopic chaos and diffusion . we consider generalizations of models of diffusion due to lorentz and ehrenfest ( figs . [ f : fix]-[f : lor ] ) , where a single point particle undergoes elastic collisions with a fixed arrangement of circular or square scatterers in two dimensions , respectively . collisions with the circular scatterers of the ( chaotic ) lorentz gas lead to exponential separation of nearby trajectories , that is , a positive lyapunov exponent , while collisions with the square scatterers of the ( nonchaotic ) ehrenfest model lead to at most linear separation of nearby trajectories , and the lyapunov exponents are all zero . actually , we consider models with three types of scatterers , all of which are non - overlapping : circles as in the lorentz gas , squares oriented such that their diagonals are parallel to the coordinate axes and with only four particle velocity directions as in the ehrenfest wind - tree model , and squares of arbitrary orientations with arbitrary particle velocity directions as a generalization of the ehrenfest model . ( 280,280)(-50,0 ) ( 40,10 ) ( 280,280)(-50,0 ) ( 40,10 ) ( 280,280)(-50,0 ) ( 40,10 ) for each of these three models , we consider the following three cases : an infinite number of randomly placed fixed scatterers as in the original models of lorentz and ehrenfest , a small number of fixed scatterers in an elementary cell subject to periodic boundary conditions , and an arrangement of a finite number of fixed scatterers enclosed by absorbing boundaries . note that a model with periodic boundary conditions can be viewed as an infinite periodic system . for the purposes of studying properties such as the mean free time , it is better to view the model as a finite system with periodic boundaries , but to compute the diffusion coefficient in terms of the growth of the mean square displacement with time , it is necessary to view the model as an infinite periodic system . a link between chaos and diffusion involves a fundamental question of statistical mechanics , since one has to make a connection between the microscopic and macroscopic behavior of large systems . the models we consider here are very special from a physical point of view , but very attractive from a mathematical point of view , because there is only one moving particle , rather than a large number as in , for example , brownian motion . these can be considered bona fide statistical mechanical models if the large number of scatterers are included , as long as their lack of motion is irrelevant to the questions we ask . we believe that our models incorporate the essential features needed for a discussion of microscopic chaos and diffusion . we now describe the experiment and analysis of gaspard et al . in more detail . the position of a brownian particle in a fluid was measured at regular intervals of @xmath0 . the experimental time series data was then interpreted using standard techniques of chaotic time series analysis , suggesting a positive lower bound on the kolmogorov - sinai entropy , hence microscopic chaos . the method they used , due to cohen and procaccia @xcite was adapted from an approach pioneered by grassberger and procaccia @xcite . in this method one analyzes the distribution of recurrences , _ i.e. _ instances where the system approximately retraces part of its previous trajectory in phase space for a certain length of time , leading to a determination of the kolmogorov - sinai entropy . there is detailed mathematical justification for the method ( under assumptions such as the length of the data set being sufficient to approximate infinite time limits , see sec . [ s : gp ] ) , but the idea is very simple : recurrences give useful information about the predictability of the system . if long recurrences occur very often , it is easy to predict the future of the system from previous instances similar to the recent part of the trajectory , so the system has a small or zero kolmogorov - sinai entropy , whereas a rapid decay of the frequency with length of the recurrences indicates a high degree of microscopic chaos . thus , the cohen - procaccia method used by gaspard can be used to calculate the kolmogorov - sinai entropy from a time series in principle , but certain mathematical conditions and limits apply , sometimes restricting its applicability in practice . gaspard et al . concluded from their analysis that the kolmogorov - sinai entropy of the system containing the brownian particle was positive , and hence that the microscopic dynamics was chaotic . subsequently , we showed @xcite that the same approach applied to an `` identical '' time series generated by a numerical simulation of the nonchaotic ( infinite , fixed orientation ) ehrenfest model yielded virtually identical results . this is surprising , since the collisions with the flat sides of square scatterers do not lead to positive lyapunov exponents , as described above . in ref . @xcite we attributed the discrepancy to the physical issue of time scales the time interval between measurements ( @xmath0 ) was vastly greater than the typical collision times of the brownian particle with the solvent particles in the fluid ( approximately @xmath1 ) . while this is certainly an experimental problem , it leaves open the question of whether in principle a similar analysis with a much higher resolution could ever prove from experimental data that the microscopic dynamics is chaotic or not . one aim of the current work is to shed light on this , and related questions . we perform the same cohen - procaccia analysis on all the models mentioned above , finding that the same results are obtained even for a model with a periodic array of squares . this rules out the possibility that the apparently positive value of the kolmogorov - sinai entropy is due to a loss of information associated with the randomness of the many scatterers . we discuss the relevant time scales in our models , finding a single microscopic time scale , and we can then qualitatively explain the results of the cohen - procaccia algorithm when the time step is shorter or longer than this time . we conclude that due to the rarity of recurrences in a diffusive system , the determination of chaoticity in such a system requires vastly longer time series than are practical for experimental or computational work , even when the measurements are taken at microscopic time and distance scales . at that point we return to the question of how microscopic dynamics manifests itself in a time series , as well as in the macroscopic behavior . we develop a new method which we call `` almost periodic recurrences '' that selects a particular class of recurrences different from those used in the cohen - procaccia method . our approach is designed to distinguish systems based on their periodic orbit properties , which are related to , but not equivalent to the usual definition of chaos as positive kolmogorov - sinai entropy . we find that , in contrast to the cohen - procaccia method , this method can distinguish between the chaotic lorentz model and the nonchaotic ehrenfest model and thus reveals at least one way in which microscopic chaos is manifest , although on macroscopic scales both models exhibit diffusion . the basic idea of our method is simple : almost periodic recurrences are related to periodic orbits which have very different properties in chaotic and nonchaotic systems , so a search for periodic orbits using recurrences may distinguish between the two classes of systems . the periodic orbits of chaotic systems ( such as the lorentz gas ) are exponentially unstable , so many repetitions close to a periodic orbit are unlikely . in contrast , the periodic orbits of nonchaotic systems ( for example our wind - tree models ) can be power law unstable , allowing many repetitions . the markedly different properties of periodic orbits in the lorentz and ehrenfest models leads to the following important observations about diffusion in finite geometries . solutions of the diffusion equation with absorbing boundaries exhibit exponential decay , corresponding to the probability of a particle remaining in the system for a given time . the randomly oriented wind - tree model has periodic orbits with power law escape , so that the escape from the whole system can not be exponential at long times . the fixed orientation wind - tree model has no periodic orbits at all , so the particle can not remain in the system longer than a fixed maximum time . both of these are examples of situations in which the infinite model satisfies the diffusion equation by having a gaussian distribution with the mean square displacement proportional to the time , but the corresponding finite model does not satisfy the diffusion equation . the outline of this paper is as follows : first we introduce our models ( sec . [ s : mod ] ) , and discuss in detail their relevant time scales and chaotic and diffusive properties ( sec . [ s : fund ] ) . then we discuss the cohen - procaccia method and its inability to distinguish chaotic from nonchaotic dynamics in our case ( sec . [ s : gp ] ) . finally we introduce our alternative `` almost periodic recurrence '' method and discuss the properties of the periodic orbits ( sec . [ s : isr ] ) , and their consequences for finite systems ( sec . [ s : esc ] ) . we conclude with a discussion of our results and some open questions ( sec.[s : conc ] ) . the ehrenfest wind - tree model describes a point `` wind '' particle moving in straight lines in the plane punctuated by elastic collisions with fixed square scatterers , the `` trees '' . in the original model ( fig . [ f : fix ] ) , the trees are oriented with their diagonals along the @xmath2 and @xmath3 axes , and the wind particle moves with a fixed velocity in only four possible directions , along these axes . the trees are located at random positions , with a given number density , and not overlapping . there is an overlapping version we will not consider ; it exhibits anomalous ( sub)-diffusion and non - gaussian distributions @xcite . secondly , we consider the case of randomly oriented scatterers ( fig . [ f : ran ] ) , which allows all possible wind particle directions . in both the fixed and randomly oriented cases all lyapunov exponents are zero , because a collision from a flat scatterer causes only a linear separation of nearby trajectories ; it does not cause the exponential separation associated with convex curved scatterers . thirdly , we consider the ( two dimensional ) lorentz gas ( fig . [ f : lor ] ) , that is , a model where the scatterers are circular , and we choose the area and number density of the scatterers to correspond to the wind - tree models . this model is known to have normal diffusive properties @xcite and is used here as a comparison to examine the effects of positive versus zero lyapunov exponents . for the numerical simulations , we take the velocity of the wind particle to be @xmath4 . for the wind - tree models we take the side length @xmath5 equal to @xmath6 . this is done for computational convenience ; space is divided into unit squares ( of side length 1 ) aligned with the coordinate axes , so that at most one scatterer can be contained in each unit square . for the lorentz gas the circular scatterers have a radius @xmath7 to give them the same area @xmath8 as the squares . in all cases the total area considered has a number of unit squares , @xmath9 , up to @xmath10 in both @xmath2 and @xmath3 directions , and periodic boundary conditions are used . to simulate an infinite system , a large value of @xmath9 ( up to 3500 ) is used ; for the maximum time of @xmath11 time units the particle never travels far enough to sample the periodic boundary conditions . to simulate a periodic system , a small size such as @xmath12 is used ; we always consider the case of finite horizon , so that the time between collisions is bounded . to simulate a system with absorbing boundaries , an intermediate size such as @xmath13 is used ; the absorbing boundaries are one distance unit inside the edge of the system to avoid the possibility of the particle colliding with a scatterer as well as with its periodic image . we use a number of scatterers @xmath14 equal to @xmath15 . this means that the number density @xmath16 is @xmath17 in all cases and the proportion of the total area covered by the scatterers is @xmath18 . thus we have an intermediate density of scatterers , which is most convenient from the point of view of the simulations and also the time scales . for a significantly lower density , an infinite horizon is much more likely in all but the largest systems . for a significantly higher density ( limited by a maximum of @xmath19 ) the metropolis algorithm ( see below ) used to position the scatterers is very much slower . also , both for low and high density systems , there is an additional time scale ( see section [ s : time ] ) , which complicates the analysis . we now describe the method used to position the fixed scatterers . each configuration of random positions and orientations is obtained from a version of the metropolis algorithm , used in @xcite . the square scatterers of the wind - tree models are placed initially at points belonging to a square lattice defined such that they do not overlap , that is , with their centers at integer coordinates @xmath20 such that @xmath21 is even , and their orientation as in fig . [ f : fix ] . not every lattice site is then occupied , depending on the number density . each scatterer is shifted and rotated a small random amount in turn , typically a few tenths of a distance unit and of order ten degrees respectively . if the shift causes an overlap , the move is rejected and the previous configuration is used . the procedure is repeated with different shifts and rotations . a large fixed number of shifts are attempted , sufficient for reasonable measures of the correlation between scatterers to have long converged to their final values . if the circles of the lorentz gas are placed on the same lattice as the squares they overlap slightly ; at the chosen density of @xmath22 there is enough freedom to allow them to find non - overlapping locations , so that the final configuration is non - overlapping , as we could check . at higher densities circles may not find non - overlapping locations , and a close packed triangular lattice should then be used instead . we now define the notation we will use to distinguish between the various models . there are a very large number of possible parameters that could appear in general , referring to the shape of the scatterers ( squares , circles , etc . ) , different densities and different shaped boundary conditions to name a few , but for conciseness we limit the notation to include only those models we use here , in particular keeping the areal density always at @xmath22 . using square brackets [ , ] to denote different cases , our notation is [ f , r , l][@xmath23,[p , a]@xmath9 ] . the first symbol [ f , r , l ] denotes the type and orientation of the scatterers , f for fixed oriented squares , r for randomly oriented squares and l for circles ( the lorentz gas ) . the next symbol [ @xmath23,p , a ] gives information about the boundary conditions , either @xmath23 , that is , no boundaries , or p for periodic boundary conditions or a for absorbing boundaries . the symbols p and a are followed by the size of the system , @xmath9 , which is an even integer . thus the original ehrenfest model is denoted f@xmath23 , while a periodic lorentz gas with @xmath25 ( containing @xmath26 scatterers at the density used in our simulations ) is denoted lp8 ( see fig . [ f : lor ] ) . the latter model is generated by random shifts of the scatterers as above , ensuring that periodic images of the scatterers do not overlap . when no confusion can arise , we will sometimes refer to classes of models with a simplified notation , for example we denote all randomly oriented wind - tree models as r , and all lorentz models with absorbing boundaries as la . this section collects a number of fundamental properties and results that form a basis for sections [ s : gp ] , [ s : isr ] and [ s : esc ] . first we consider the important issue of _ time scales _ , which puts the time series analysis methods of sections [ s : gp ] and [ s : isr ] into a physical context . then we consider _ chaotic properties _ , giving known results and conjectures for our models ; all of the following sections use this . after this we study _ diffusive properties _ , obtaining some new results . these are of direct importance to sec . [ s : esc ] ; the time series analysis methods are applied only to our diffusive models , . ] but the diffusion is somewhat incidental as the methods can be applied to arbitrary time series . at the end of this section we relate the chaotic and diffusive properties of our models . one of our difficulties of the gaspard et al experiment @xcite is that the interval between their measurements ( 1/60 s ) is so much greater than the relevant time scale ( @xmath1 ) of the dynamics , that is the time interval determined by collisions between the brownian particle and the other particles in the fluid . it is thus important to clarify the issue of what time scales are relevant in our models , so that the simulation results can be put in context . there are three relevant time scales here , the time taken for the particle to traverse the length of a scatterer ( defined to be of order one time unit ; see sec . [ s : comp ] ) , the mean free time between collisions @xmath27 , and the time at which the particle begins to notice the finite size @xmath9 of the system ( when it is not infinite ) . it turns out ( see sec . [ s : diff ] ) that all the random and periodic models except fp have a well defined diffusion coefficient @xmath28 , so in these cases the time taken to reach the boundary is of order @xmath29 . the mean free time can be calculated exactly ( see the appendix ) and is given by @xmath30 , @xmath31 and @xmath32 for the f , r , and l models respectively , independent of the size of the system ( ignoring the case of absorbing boundaries ) . we have checked these values numerically , and the results agree within their uncertainties of about @xmath33 . it is clear that all the time scales in these models are of order one time unit , except those defined by the boundary . this means that a time step of one time unit should be sufficient to observe effects due to the microscopic chaos in the time series analysis methods of sections [ s : gp ] and [ s : isr ] . we began with the notion of microscopic chaos as the presence of a positive lyapunov exponent arising from collisions with strictly convex scatterers , or as a positive kolmogorov - sinai entropy quantifying the lack of predictability of a chaotic system . we now want to make these ideas more precise and clarify what is known and what is conjectured about our models . from a mathematical point of view , there are a large number of dynamical properties that a system may exhibit @xcite . those of interest to us are 1 . we need ergodicity for the time series analysis methods of secs . [ s : gp ] and [ s : isr ] ; that is , a single long trajectory is supposed to sample the dynamics of the whole phase space . we either know or conjecture that ergodicity holds in all of our models ( see below ) . 2 . decay of velocity correlations . we note that eqs . ( [ e : gk],[e : gkb ] ) below give the diffusion and burnett coefficients as integrals over velocity autocorrelation functions : the diffusion coefficient involves two - time correlations , while the burnett coefficient involves four - time correlations . for these coefficients to be defined the appropriate correlations must decay sufficiently quickly . numerical evidence for the existence or nonexistence of these coefficients for our models is given in sec . [ s : diff ] . 3 . a positive lyapunov exponent , corresponding to exponential separation of initially close trajectories , follows from the shape of the scatterers . the lorentz gas has a positive lyapunov exponent due to its convex scatterers , while the wind - tree models have all zero lyapunov exponents due to their ( piecewise ) flat scatterers . 4 . a positive kolmogorov - sinai entropy , which is our definition of chaos , and which the cohen - procaccia method ( see sec . [ s : gp ] ) is designed to compute . we either know or conjecture that the ks entropy is equal to the sum of the positive lyapunov exponents in our models ( pesin s theorem , see below ) . actually there is at most one positive lyapunov exponent in these systems . periodic orbit properties sensitive to chaoticity are described in sec . [ s : isr ] and used in secs . [ s : isr ] and [ s : esc ] . we now briefly discuss what is known about our models regarding these dynamical properties . the periodic lorentz gas has exponential decay of ( all two time ) correlations @xcite , which implies ergodicity and the convergence of the integral in ( [ e : gk ] ) below . for ( [ e : gkb ] ) see ref . the periodic wind - tree models are generically ergodic @xcite . pesin s theorem holds for all of these [ f , r , l]p models , see refs . @xcite for the lorentz gas and ref . @xcite for the wind - tree models . the infinite models are much more difficult to treat rigorously because the phase space is not compact . in order to make reasonable conjectures about the above properties , we use the solution of the diffusion equation ( [ e : gauss ] ) below , which is verified numerically in sec . [ s : diff ] for the models [ f , r , l]@xmath23 , to argue that the number of different scatterers hit by the particle in time @xmath34 is proportional to @xmath35 , as follows . in two dimensions , the probability density of the particle being at any point decays as @xmath36 . this means that in a time @xmath34 , the particle comes close ( say , within a radius @xmath37 ) to the point a number of times proportional to the integral of @xmath36 , that is @xmath38 . if the point corresponds to the location of a scatterer , we could say that the particle collides with each scatterer a number of times proportional to @xmath38 . in this case the total number of different scatterers hit by the particle in time @xmath34 might be expected to grow proportional to @xmath35 , which is what we find numerically in fig . [ f : stat ] . ( 300,320)(-70,0 ) ( 0,20 ) ( 155,10)@xmath34 ( -15,165)@xmath39 ( 40,285)l@xmath23 ( 40,275)r@xmath23 ( 40,265)f@xmath23 these observations have the following consequences for the chaotic properties of the infinite models . because the particle returns to each scatterer an infinite number of times , we expect that it passes arbitrarily close to every point in phase space , that is , the system is ergodic . the ks entropy gives the amount of information per unit time required to predict the trajectory knowing its infinite past . the unpredictability has two sources , the instability associated with the collisions in the lorentz gas ( that is , the positive lyapunov exponent ) , and the random positions of the scatterers with which the particle has not yet collided in all the models . the number of different scatterers hit by the particle grows as @xmath35 , so the rate decreases to zero as @xmath40 . we conclude from this argument that the random positions of the scatterers do not contribute to the ks entropy , although results at finite time might suggest otherwise , given that @xmath40 decays so slowly . in other words , pesin s theorem is satisfied for these infinite models also . ) below ) in two dimensions . in three dimensions , the density decays as @xmath41 which is integrable at infinity , so that the particle collides with each scatterer only a finite number of times and thus does not sample the whole phase space . in addition , there is a contribution to the ks entropy from the random positions of the scatterers , thus violating pesin s theorem . there is no contradiction here ; the proofs of pesin s theorem all demand that the phase space be compact . ] we have discussed a number of dynamical properties with regard to our models . the chaotic lorentz gas satisfies more of these properties than the nonchaotic wind - tree models . chaos is defined here by positive ks entropy and is clearly relevant fro diffusion , yet we show in subsequent sections that the periodic orbit properties also appear to be relevant to diffusion . the conclusion discusses an example of a model which is chaotic as it has positive ks entropy , but has periodic orbit properties and diffusive properties more similar to the nonchaotic wind - tree models . although we have a substantial number of relevant numerical results below , a full understanding of the dynamical properties ( ks entropy , periodic orbit properties or perhaps others ) most closely related to diffusion still eludes us . now we discuss diffusive properties . we begin with the diffusion equation , and then relate it to the mean square displacement of a particle . the diffusion equation @xmath42 for a probability density @xmath43 with a constant diffusion coefficient @xmath28 is linear . thus its general solution is an integral over green s functions given by the solution for an initial dirac delta distribution @xmath44 , that is , @xmath45\ ] ] is the conditional probability density that a particle initially at the point @xmath46 will be at the point @xmath47 a time @xmath34 later ; this is therefore a 2-time probability distribution ( or density ) function . the spatial dimension @xmath48 in our case . the diffusion equation is known to hold for the probability density of a particle undergoing deterministic dynamics in a number of systems including the lorentz gas @xcite . in the diffusion equation a macroscopic approximation is used , implying that @xmath43 is an average over space and time scales large compared to the characteristic microscopic lengths and times of the dynamics . the mean square displacement of the particle in the @xmath2-direction after a macroscopic time @xmath34 is obtained from ( [ e : gauss ] ) , @xmath49 where @xmath50 . this is the einstein relation for diffusion . note that @xmath51 is now an explicit function of time . similarly , the fourth and sixth cumulants are @xmath52 respectively . it is also possible to obtain dimensionless forms of the above expressions dividing them by the appropriate power of @xmath53 . in this form the fourth cumulant is usually called the kurtosis , and is a common measure of how close a probability density is to a gaussian distribution . straightforward manipulations applied to the mean square displacement lead to the green - kubo expression for the diffusion coefficient ( if it exists ) @xmath54 where @xmath55 is one ( say the @xmath2- ) component of the velocity and the subscript denotes the time . this shows that the existence of a diffusion coefficient is also related to the rate of decay of the velocity autocorrelation function . the integral diverges if the decay rate of the integrand is @xmath56 or slower , unless the integrand is oscillatory . an oscillatory integrand can not be ruled out , but is not typical for the lorentz gas as observed in numerical simulations . the diffusion equation is the lowest order macroscopic description of the diffusion process . the next approximation , in the direction of microscopic space and time scales , _ ie . _ when the probability density @xmath57 varies sufficiently rapidly , the right hand side of eq . ( [ e : diff ] ) may be augmented by terms such as @xmath58 , where @xmath59 is called the linear super - burnett coefficient @xcite . like the diffusion coefficient , it obeys an einstein relation @xmath60 this is consistent with the zero kurtosis found previously ( [ e : kurt ] ) : to find the kurtosis we divided by @xmath61 which is proportional to @xmath62 if the diffusion coefficient exists , while the burnett coefficient in ( [ e : burn ] ) gives a smaller term ( for large @xmath34 ) , proportional to @xmath34 . the burnett coefficient can also be written in the form of a green - kubo relation in terms of integrals over four - time velocity correlation functions , @xmath63 where @xmath64 as above and @xmath65 etc . a slow decay of the relevant correlation functions leads more easily to a divergence in this case than in eq . ( [ e : gk ] ) @xcite . this means that it is possible to have a well defined diffusion coefficient , but a divergent burnett coefficient , which occurs if the fourth cumulant increases faster than @xmath34 but slower than @xmath62 . we estimate the cumulants defined in eqs . ( [ e : msd]-[e : six ] ) numerically by choosing a large number ( typically @xmath66 ) of random initial conditions for the particle not inside a scatterer , for a single configuration of scatterers . we report only the cumulants in the @xmath2-direction , although we have checked that the @xmath3 distribution is similar . see figs . [ f : msd]-[f : six ] and table [ t : diff ] . ( 500,530)(-100,0 ) ( -100,30 ) ( 170,65 ) ( 155,20)@xmath34 ( -115,265)@xmath53 ( 0,442)@xmath34 ( 0,426)@xmath67 ( 0,410)@xmath62 ( 0,394)l@xmath23 ( 0,378)lp4 ( 0,362)r@xmath23 ( 0,346)rp4 ( 0,330)f@xmath23 ( 0,314)fp4 ( 300,320)(-70,0 ) ( 0,20 ) ( 165,10)@xmath34 ( -25,165)@xmath68 ( 60,261)@xmath34 ( 60,251)@xmath62 ( 60,241)@xmath69 ( 60,232)l@xmath23 ( 60,222)lp4 ( 60,213)r@xmath23 ( 60,203)rp4 ( 60,194)f@xmath23 ( 60,184)fp4 ( 300,320)(-70,0 ) ( 0,20 ) ( 165,10)@xmath34 ( -30,165)@xmath70 ( 60,261)@xmath34 ( 60,251)@xmath62 ( 60,241)@xmath71 ( 60,231)@xmath72 ( 60,222)l@xmath23 ( 60,212)lp4 ( 60,203)r@xmath23 ( 60,193)rp4 ( 60,184)f@xmath23 ( 60,174)fp4 .diffusive properties of our models , in terms of the diffusion coefficient @xmath28 , and the burnett coefficient @xmath59 if they exist . the existence of these coefficients is related to the rate of decay of the velocity autocorrelation functions of ( [ e : gk],[e : gkb ] ) . note that by this criterion , the periodic lorentz gas lp4 is more `` chaotic '' than the infinite lorentz model l@xmath23 , while the periodic fixed orientation wind - tree model fp4 is less `` chaotic '' than the infinite model f@xmath23 . only typical values of the coefficients for the periodic models are given , since they depend on the position of the scatterers . [ t : diff ] [ cols="^,^,^,^",options="header " , ] for five models , [ f , r , l]@xmath23 , [ r , l]p4 , the mean square displacement is found to be proportional to the time , and the fourth and sixth cumulants increase slower than @xmath62 and @xmath71 respectively at large times , indicating that the distribution approaches a gaussian . the exception to this rule is fp4 , for which the mean square displacement grows faster than the time , approximately as @xmath67 and the kurtosis is also nonzero . this superdiffusive behavior is not particularly stable in the sense that the mean square displacement for this model has larger fluctuations than the other models , and the exponent of approximately @xmath73 varies unpredictably between 1 and 2 with the exact positions of the scatterers and the size of the cell ( for example for fp6 or fp8 ) . we now look at how the chaotic and diffusive properties of our models are related in the light of our results , specifically of : ( a ) normal diffusion without chaos , ( b ) superdiffusion without chaos , ( c ) a possible relation between the diffusion coefficient and the periodic orbits , and ( d ) the burnett coefficient of the periodic lorentz gas . in no way does the data presented in figs . [ f : msd]-[f : six ] distinguish between chaotic and nonchaotic models . [ [ normal - diffusion ] ] normal diffusion + + + + + + + + + + + + + + + + our results show that it is possible to have a well defined diffusion coefficient and a gaussian distribution function without positive lyapunov exponents , that is , in the f@xmath23 , r@xmath23 and rp4 models . the rp4 model is particularly interesting in this regard , since the two most obvious sources of unpredictability : dispersive collisions and collisions with new randomly placed scatterers , are absent . recall that we argued ( in section [ s : chaos ] ) that these two processes might together determine the ks entropy , that is , the rate of loss of information about the particle s position and velocity . the rp4 model is thus the clearest example of a model in which there is normal diffusion , and zero ks entropy . an example with the same properties , using a periodic rhombus model has been studied in ref . @xcite , where a mean square displacement proportional to the time was observed , but the higher moments were not studied . [ [ superdiffusion ] ] superdiffusion + + + + + + + + + + + + + + the fp4 model , in contrast , has a mean square displacement growing faster than linearly with the time , and a non - gaussian distribution . this can be related to a slow decay of the velocity autocorrelation function ( see above ) , but a physical explanation of why this occurs in the fp4 model but not in the rp4 model is lacking . [ [ diffusion - and - periodic - orbits ] ] diffusion and periodic orbits + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is perhaps somewhat hazardous to deduce more from this data than the existence of normal diffusion ; however we would nevertheless like to point out a possible connection between the value of the diffusion coefficient ( given by the vertical displacement of the curves in the inset of fig . [ f : msd ] ) in the infinite models and the properties of periodic orbits discussed in section [ s : isr ] . we make no statement about the diffusion coefficient of the periodic models because this coefficient depends on the position of the scatterers . the idea is that the presence of periodic orbits in a model with randomly positioned scatterers might be expected to lower the diffusion coefficient , since the periodic orbits may keep the particle trapped in roughly the same place ; this takes time without contributing to the mean square displacement . thus we observe that the fixed orientation wind - tree model f@xmath23 has no periodic orbits and has the largest diffusion coefficient ; the random lorentz gas l@xmath23 has periodic orbits , but they are exponentially unstable , and has the next largest diffusion coefficient ; the random orientation wind - tree model r@xmath23 has periodic orbits with power law instability that can inhibit diffusion for a long time , and the diffusion coefficient is the smallest of the three . [ [ the - burnett - coefficient ] ] the burnett coefficient + + + + + + + + + + + + + + + + + + + + + + + the fourth cumulant grows linearly with time for the lp4 model , indicating a finite burnett coefficient , as expected , given its exponential decay of correlations @xcite . the other models , f@xmath23 , r@xmath23 , rp4 and l@xmath23 have a divergent burnett coefficient , which is not surprising since their decay of correlations is most likely as a power law . thus , from the more subtle properties of the two time ( displacement ) correlation function , we can deduce some properties relating to the rate of decay of higher order correlations , even special four - time ( velocity ) correlations , but not whether there is a positive lyapunov exponent , since l@xmath23 behaves similarly to r@xmath23 from this point of view . even in the intermediate region between the ballistic behavior at small times and the ( super)-diffusive behavior at large times in fig . [ f : msd ] , that is , where the distribution functions are measured at microscopic time scales , there is no clear distinction between the chaotic and nonchaotic models . we have now discussed at some length the connection between chaotic and diffusive properties of our models . while this has yielded valuable insight into how little chaoticity is needed for diffusion to occur , it has not provided us with an understanding of how chaotic and nonchaotic diffusion differ , let alone a method to distinguish chaotic and nonchaotic diffusive systems experimentally . the reason , given in ref . @xcite is that two time correlations ( for example our cumulants ) are insufficient to characterize chaos ; multi - time correlations ( or their underlying probability distributions ) are required . the remaining sections of this paper all use multi - time distributions of one form or another to investigate our models . we begin with a more detailed study of the grassberger - procaccia method than in ref . @xcite , where we found it to incorrectly classify the f@xmath23 model as chaotic . grassberger and procaccia @xcite gave the earliest practical methods for computing chaotic properties such as entropies and dimensions from experimental or computer generated time series . these are still the most popular methods in use , sometimes with minor variations . in ref . @xcite the conclusion of microscopic chaos using data from a brownian motion experiment was based on a slightly different method of cohen and procaccia @xcite . both methods are clearly reviewed in ref . @xcite . we consider these methods as applied to the calculation of the ks entropy @xmath74 , the non - zero value of which characterizes chaos . we recall the discussion of section [ s : chaos ] where we concluded that the wind - tree models all have zero ks entropy , but the lorentz gas has a ks entropy equal to its positive lyapunov exponent . the original grassberger - procaccia method @xcite computes a slightly different dynamical entropy @xmath75 ( defined using the square of the probability , @xmath76 rather than the conventional @xmath77 ) satisfying the inequality @xmath78 . thus a positive estimate @xmath79 implies that @xmath80 and hence chaoticity . we follow ref . @xcite in using an adaptation due to cohen and procaccia @xcite that allows @xmath74 to be estimated directly . we replace the experimental time series of the positions of the brownian particle in ref . @xcite by a numerical time series of any of our infinite or periodic diffusive models , containing the @xmath2 and @xmath3 positions of the particle at @xmath11 times uniformly spaced by a separation @xmath81 . unlike ref . @xcite , by removing data points . the new time step was denoted @xmath82 . we do not need to do this in our numerical work , and we do not use the notation @xmath82 for time steps to avoid confusion with the mean free time . ] our values of @xmath81 , equal to 1 and 0.01 time units are close to the dynamical time scale determined by the mean free time @xmath83 ( see sec . [ s : time ] ) . a single point on our numerical trajectory is denoted @xmath84 , where @xmath85 is an integer ranging from 0 to @xmath86 . for the fixed oriented square ( f ) models the trajectory contains all the phase space information except the velocity which can take only four values in this case . for the randomly oriented square ( r ) and the lorentz ( l ) models the velocity has one continuous degree of freedom . the brownian motion experiment uses only one component ( say , @xmath2 ) of the particle position , ignoring the huge number of other degrees of freedom contained in the fluid system . from a mathematical point of view , as long as the degrees of freedom are all coupled ( that is , all the degrees of freedom depend on each other directly or indirectly ) , this type of method usually converges to the correct entropy ( or dimension ) using as few as one measured variable , however the efficiency may then well be such that an unreasonably large number of data points is required for a reliable estimate of these quantities . a number ( typically @xmath87 to @xmath88 ) of ( in our case non - overlapping ) uniformly spaced subsequences of the long trajectory are denoted `` reference segments '' ( `` reference trajectories '' in ref . @xcite ) , having lengths @xmath89 ranging from 1 to 100 . these reference segments are compared with all subsequences of the trajectory having the same length . the comparison is made according to the position relative to the initial point of each subsequence , with a euclidean metric in real space and a maximum metric over the trajectory . thus , if @xmath85 and @xmath90 give the positions in the full trajectory of the initial points of the @xmath91 reference segments and @xmath92 total segments respectively , the distance between the segments beginning at @xmath85 and @xmath90 is defined as @xmath93 where @xmath94 etc . we are now in a position to count the number of recurrences , in what is called `` pattern probability '' in @xcite . for a given tolerance or spatial resolution @xmath37 we can define the probability of a certain pattern defined by the reference sequence beginning at @xmath85 as the proportion of all the subsequences of the full trajectory that are within a distance ( as defined above ) @xmath37 of the reference sequence : @xmath95\ ] ] and from this the `` pattern entropy '' @xmath96 which gives the information theoretic entropy of the dynamics with respect to the pattern ( of length @xmath89 and resolution @xmath37 ) probability distribution . the pattern entropy thus contains information about the @xmath89-time ( displacement ) distribution functions , as suggested by the discussion at the end of the previous section . when the pattern entropy increases linearly with time in the limit of large @xmath89 , it is possible to define an entropy per unit time , @xmath97\ ] ] in practice @xmath89 is finite , and @xmath74 is always bounded above by @xmath98 ( since it is an average of @xmath99 , where @xmath57 bounded below by @xmath100 ) and tends to saturate towards this value at long times . however , we can often find a large enough linear region in a plot of @xmath74 as a function of @xmath89 ( for fixed @xmath37 and @xmath81 ) to compute @xmath101 . in ref . @xcite @xmath101 was plotted as a function of @xmath37 with different curves corresponding to different @xmath81 . from dimensional arguments it follows that any diffusive process has a scale invariance leading to a dependence @xmath102 , which is in fact observed in such plots . assuming that there is sufficient data that all the limits ( of large @xmath14 , @xmath91 and @xmath89 ) can be approximated reliably , the ks entropy @xmath74 is equal to the maximum value of @xmath101 as @xmath37 and @xmath81 are varied . the measured value @xmath101 will be less than the true ks entropy @xmath74 if the values of @xmath37 and or @xmath81 are so large that not all of the information contained in the dynamics is represented in the time series . this effect does not explain why a nonchaotic system may appear to be chaotic , that is , why the measured value of @xmath101 is greater than @xmath103 , as in ref . @xcite : this must be due to a problem with the above mentioned limits , and we defer a discussion of this point until after we have presented our results . our results for each of the six models [ f , r , l][@xmath23,p4 ] are illustrated in fig . [ f : gpl ] with @xmath104 and in fig . [ f : gps ] with @xmath105 . the case @xmath104 samples the motion of the particle as it begins to diffuse among the scatterers . the models ( both chaotic and nonchaotic ) are indistinguishable , with the linear growth of the pattern entropy suggesting positive ks entropy , except that the superdiffusive model fp4 is showing signs that the ks entropy is zero , as a nonchaotic model should . ( 330,530)(-70,0 ) ( 0,360 ) ( 170,360 ) ( 0,190 ) ( 170,190 ) ( 0,20 ) ( 170,20 ) ( 85,10)@xmath89 ( -65,95)@xmath106 ( -10,480)f ( -10,310)r ( -10,140)l ( 85,515)@xmath23 ( 245,515)p4 ( 330,530)(-70,0 ) ( 0,360 ) ( 170,360 ) ( 0,190 ) ( 170,190 ) ( 0,20 ) ( 170,20 ) ( 85,10)@xmath89 ( -65,95)@xmath106 ( -10,480)f ( -10,310)r ( -10,140)l ( 85,515)@xmath23 ( 245,515)p4 for @xmath105 ( fig . [ f : gps ] ) , where mostly ballistic behavior occurs , the graphs of pattern entropy vs. number of time steps is no longer linear . the r and l models look much the same , while the f models have an unusual feature when the pattern entropy is equal to @xmath107 which is due to the four available wind directions , that is , only a fourth of the trajectories at a given @xmath108 point are close in phase space after a certain time . the pattern entropy is becoming linear at larger times in all models , suggesting positive ks entropy . we can draw the following conclusions about the use of grassberger - procaccia type methods for this class of problems : ( a ) measurements need to be made at microscopic time scales , ( b ) even at microscopic time scales , the method does not seem to distinguish between chaotic and nonchaotic dynamics , and ( c ) the reason seems to be that unfeasibly long time series are needed in order to do that . [ [ macroscopic - measurements ] ] macroscopic measurements + + + + + + + + + + + + + + + + + + + + + + + + it is clear that in an analysis as explained above , using large distance and time scales a diffusive system shows itself as almost independent of its microscopic dynamics ( whether chaotic , non - chaotic or stochastic ) , although we can not exclude the possibility that an unreasonably long trajectory might still give some evidence of weak correlations that survive for macroscopic times . in the brownian motion experiment , for example , there are of the order of @xmath109 collisions of the brownian particle with the surrounding solvent between each measurement of its position , so any correlations arising from , say , nonchaotic microscopic dynamics would be effectively averaged out between measurements . this is essential from the point of view of designing future experiments , but it is not the full story as it does not explain our results , which are obtained using short times . the brownian motion experiment could not investigate the motion at short time scales , but it is important to know whether data from an improved experiment might determine the microscopic chaoticity in principle . [ [ microscopic - measurements ] ] microscopic measurements + + + + + + + + + + + + + + + + + + + + + + + + our results do not suggest that the grassberger - procaccia type methods can distinguish between chaotic and nonchaotic diffusive models even at short times because the nonchaotic r models consistently give the same results as the chaotic l models . the only distinctions we have been able to make are between diffusion and superdiffusion for @xmath104 , and between the continuous velocity space of the r and l models and the four velocity directions of the f models . both of these distinctions can be made from a time series without using the intensive analysis of the grassberger - procaccia method . [ [ how - long - a - time - series - is - required ] ] how long a time series is required ? + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + our wind - tree models appear unpredictable ( in the sense of positive ks entropy ) because it takes the system some time to `` realize '' that it is nonchaotic , in other words , quite a long time series length @xmath14 is required . for example , the most pronounced nonchaotic diffusive model we have is rp4 . once the particle `` knows '' the positions and orientations of all the scatterers , the system is completely predictable , and hence a zero ks entropy is manifest . however it is necessary for the particle to enumerate the large number of ways of achieving this combinations . the eight collisions could occur in any order , giving an extra factor of @xmath110 . thus our estimate is @xmath111 . note that collisions may not be able to occur in any order ( thus lowering the estimate ) , but we have ignored situations where the particle collides with three or four sides of a scatterer , or with the same sides more than once ( thus greatly increasing the estimate ) . the trajectory length would need to be substantially larger than the true number of ways the particle can determine the scatterer configuration , in order to obtain reasonable statistics , so that nonchaotic recurrences dominate . ] with sufficient statistics that nonchaotic recurrences dominate the gp calculation , for which our trajectory length of @xmath112 is apparently insufficient . for the infinite models such as r@xmath23 the motion of the particle is never completely predictable , but the proportion of collisions that the particle encounters new scatterers decreases as @xmath40 according to the discussion of sec . [ s : chaos ] . the measurement of such a slow decrease is far beyond the capabilities of any feasible implementation of the grassberger - procaccia method . since the grassberger - procaccia approach does not seem to distinguish our chaotic and nonchaotic diffusive models , we turn to an alternative time series analysis method , with a view towards suggesting possible methods for analyzing future experiments that are able to sample the dynamics on microscopic time scales . such a method can not possibly enumerate all possible trajectory segments , as discussed above , but there are certain types of trajectory segments , namely those that are almost periodic , that stand out as predictable in a nonchaotic system . the method of the next section takes advantage of this property , and in effect distinguishes chaotic and nonchaotic systems by singling out this subset of all possible recurrences . we have seen in sec . [ s : gp ] that the variant of the grassberger - procaccia method used by gaspard et al can not distinguish between nonchaotic and chaotic models , for the reason that the time series ( either experimental or numerical ) is not long enough to provide a reasonable approximation to the infinite time limit implied by the method . we can attempt to circumvent this difficulty by looking for specific types of recurrences , as opposed to taking an average over all types of recurring trajectory segments . in particular we will show that a promising candidate for a useful specific recurrence is given by one that corresponds to an orbit that is almost periodic ( again , within a spatial tolerance @xmath37 ) . we can identify these almost periodic orbits as those that repeat ( within a distance @xmath37 ) their previous motion at equally spaced intervals of time , hence the name `` almost periodic recurrences '' ( apr ) . if we compare the apr approach with the grassberger - procaccia ( gp ) method , there are a number of similarities and differences . both methods are designed to deduce dynamical properties from time series data . gp aims to make quantitative statements about the ks entropy , whereas apr as yet only provides a qualitative statement about chaoticity as expressed in periodic orbit properties , which are not equivalent to the usual definition of chaos as positive ks entropy . both methods make use of the probability of recurrences . gp then averages over many probabilities , while apr singles out a few especially significant recurrences . this means , for example that in an intermittent system , where the dynamics switches irregularly from chaotic to regular behavior , gp measures only the average , chaotic , dynamics , while apr is sensitive to the lack of chaoticity in the regular regions of phase space . in other words , an intermittent system can have a positive lyapunov exponent , and also periodic orbits with nonchaotic properties . the apr method characterizes such a system as nonchaotic , which is an appropriate designation with regard to diffusive properties as we observe in sec . [ s : esc ] below , but which is inappropriate from the usual point of view , which is that chaos is defined as positive ks entropy . an example of such an intermittent system is given in the final discussion . we now discuss the properties of periodic orbits in chaotic and nonchaotic systems ( particularly our models ) , before describing the apr method in more detail and presenting the results . periodic orbits are of great importance in chaotic systems . although typical trajectories ( as defined by the liouville or natural measures ) are not periodic , there are many systems for which they are approximated by periodic orbits when the periodic orbits are dense in the phase space or on an attractor . this allows many properties to be computed from expansions involving periodic orbits @xcite . the properties of the periodic orbits are not directly related to the other dynamical properties discussed in sec . [ s : chaos ] , yet they do tend to differ between chaotic and nonchaotic systems , and so they constitute additional dynamical properties related to chaos . the properties of the periodic orbits in the lorentz gas are of most interest to us , and in some sense typical for chaotic systems , so we focus on these . in this and the following section we examine two relevant properties of periodic orbits , and sketch proofs of these properties . [ [ existence ] ] existence + + + + + + + + + there is an easy constructive proof of the existence of periodic orbits in the lorentz gas : choose two disks arbitrarily ; the line joining their centers gives a periodic orbit unless there is a disk interposing ; consider one of the original disks and the interposing disk , and repeat until no disk interposes . [ [ stability ] ] stability + + + + + + + + + the periodic orbits are exponentially unstable , that is , almost all trajectories beginning a small distance @xmath37 from the periodic orbit at time @xmath113 are at a distance @xmath114 at time @xmath34 , where @xmath115 is the maximum ( here the only positive ) lyapunov exponent , which depends on the periodic orbit @xmath116 . this exponential instability makes it very unlikely for a typical trajectory to remain near a given periodic orbit for long times . we study here the properties of periodic orbits of our wind - tree models , which are typical of nonchaotic systems , without attempting to classify all possible nonchaotic periodic behavior . for a complementary study of the periodic orbits in polygonal billiards , see ref . @xcite . not all wind - tree models have periodic orbits . the argument given above for the lorentz gas fails because an orbit connecting two square scatterers is periodic only if the scatterers have the same orientation ( of zero probability in the randomly oriented model ) and the particle moves perpendicular to the surface of the scatterers ( excluded by definition in the fixed orientation model ) . in general , the existence of periodic orbits depends on the locations and orientations of the scatterers . we will find that using generic locations of scatterers periodic orbits exist in the r@xmath23 model , but not in the f@xmath23 model . the periodic models ( such as fp4 and rp4 ) are harder to treat mathematically , and are not discussed here . [ [ f - model - nonexistence ] ] f model : nonexistence + + + + + + + + + + + + + + + + + + + + + the absence of periodic orbits for generic configurations of the f@xmath23 model follows directly from arguments in a proof of aarnes @xcite . in order for the particle to return to its starting point , an expression linear in the positions of the scatterers must vanish . the coefficients of the @xmath108 position of a scatterer in this expression are integers depending on which faces the particle hits , and are zero only if the particle hits opposite faces an equal number of times . a simple _ reductio ad absurdum _ argument shows then that there is a scatterer ( perhaps more than one ) , the furthest to the top and right ( largest value of @xmath117 ) , which has collisions on only one of its faces ( the lower left ) , and hence the coefficients corresponding to the position of this scatterer are nonzero . if the scatterers are randomly placed , no linear combination of the positions can vanish , so no periodic orbit can exist . of course , there are special configurations of the scatterers ( such as at the corners of a square aligned with the coordinate axes ) that allow periodic orbits . since there are therefore generically no periodic orbits in the f@xmath23 model , the remainder of this section refers to the r@xmath23 model . [ [ r - model - existence ] ] r model : existence + + + + + + + + + + + + + + + + + + the existence of periodic orbits for generic configurations of the r@xmath23 model follows from the observation that any set of configurations of a finite number of ( _ eg . _ 3 ) scatterers with nonzero ( lesbesgue ) measure includes configurations appearing somewhere in a generic infinite configuration . a period 3 orbit can always be found in an acute triangular billiard by minimizing the total path length , see fig . [ f : tri ] . if three scatterers outline an acute triangle and are positioned so that their faces contain the path of minimal length ( clearly of nonzero measure ) , a period 3 orbit exists . ( 280,280)(-50,0 ) ( 0,0 ) [ [ r - model - stability ] ] r model : stability + + + + + + + + + + + + + + + + + + finally we investigate the stability of periodic orbits in the r@xmath23 model . the combination of the linear dynamics of the free particle with length preserving reflections leads to a linear separation of nearby trajectories . that is , an initial separation of @xmath37 in the direction of the velocity leads to a separation @xmath118 in the position after time @xmath34 . another way of saying this is that the number of uniformly distributed trajectories remaining within a distance @xmath37 of the periodic orbit is @xmath119 for large @xmath34 . thus a particle in the r@xmath23 model is quite likely to spend a long time near a periodic orbit , in contrast to in the lorentz gas . now we return to our method of almost periodic recurrences for distinguishing chaotic properties of diffusive systems from a time series . this consists of counting the number of almost periodic sequences in the time series , thus hoping to exploit the different properties of periodic orbits in chaotic and nonchaotic systems as discussed in the previous two sections . as in the grassberger - procaccia method we begin with a time series containing @xmath11 positions of the particle @xmath84 spaced at a time interval @xmath104 , see sec . [ s : time ] . analogous to eq . ( [ e : dist ] ) we define a distance between two segments of length @xmath120 ( the period of an almost periodic orbit ) that begin at points @xmath85 and @xmath90 on the trajectory , @xmath121 using eq . ( [ e : dt ] ) we compute the number of initial points @xmath85 for which the orbit repeats within a tolerance @xmath37 , @xmath122\ ] ] where we use @xmath123 for the results presented in fig . [ f : isr ] . ( 330,470)(-50,0 ) ( 0,320 ) ( 150,320 ) ( 0,170 ) ( 150,170 ) ( 0,20 ) ( 150,20 ) ( 85,10)@xmath120 ( -15,95)@xmath124 ( -10,430)f ( -10,280)r ( -10,130)l ( 85,460)@xmath23 ( 235,460)p4 it is clear that the chaotic lorentz models , with an exponential decay of @xmath124 with @xmath120 can be easily distinguished from the nonchaotic f and r models which mostly have a slower and noisier decay , so we have achieved our objective of finding a time series analysis method that can distinguish between time series of chaotic and nonchaotic diffusive systems . it remains for us to explain the difference , and the plots obtained using the properties of the periodic orbits in these systems . for that , we note that there are two types of ( almost ) periodic orbits that might appear in the expression for @xmath124 : an orbit of length @xmath120 that has been ( almost ) repeated , and a shorter orbit of length @xmath125 that has appeared @xmath126 times , so that the total time is equal to @xmath127 . [ [ the - l - models ] ] the l models + + + + + + + + + + + + in the lorentz gas , the probability of remaining near a periodic orbit for a time @xmath120 is of order @xmath128 ( see sec . [ s : cpo ] ) , where @xmath115 is the maximum ( here the only ) positive lyapunov exponent , and depends on the periodic orbit @xmath116 . this is why both the infinite and periodic lorentz gases give an exponential form in fig . [ f : isr ] . because @xmath124 can be due to long orbits or repeats of short orbits , the exponential form depends on both ( 1 ) the exponential escape from short orbits with many repetitions and ( 2 ) the exponential instability of orbits with length , that is , that @xmath115 does not approach zero for long orbits . could become arbitrarily small . however , the probability of a long free path is also exponentially small , so the result is still exponential . ] [ [ the - r - models ] ] the r models + + + + + + + + + + + + in the randomly oriented wind - tree models , the probability of remaining near a periodic orbit for a time @xmath120 is of order @xmath129 ( see sec . [ s : npo ] ) . this form is not immediately apparent in fig . [ f : isr ] because orbits ( particularly those surviving for long times ) can be counted more than once due to contributions from an interval of different initial points @xmath85 . both the r@xmath23 model and the rp4 model continue beyond @xmath130 due to increasingly rare long orbits and many repeats of shorter orbits . for example , the almost periodic orbit giving rise to the conspicuous point with period @xmath131 in the rp4 model is repeated many times , also contributing to multiples of the period , double @xmath132114,115 ; triple @xmath132172,173 ; and so on up to @xmath1321033,1034 . it is clear that the dynamics is completely different to that of a chaotic system . [ [ the - f - models ] ] the f models + + + + + + + + + + + + in the fixed oriented wind - tree models , there are no periodic orbits ( see sec . [ s : npo ] ) . this means that any contribution to @xmath124 is due to orbits that are close to periodic . the infinite model f@xmath23 contains many arrangements of a few scatterers that come arbitrarily close to generating a true periodic orbit , so the plot is similar to that of the r models . the periodic model fp4 , in contrast , has only a few short orbits that are at all close to periodic , and none that can be repeated to give further contributions at higher values of @xmath120 , or these would be observed in the figure . what we have presented here barely scratches the surface of the almost periodic recurrences method and its variants . it would be quite easy to search for 3 or more occurrences of an ( almost ) periodic orbit , or to distinguish between a single repeat of a long orbit and many repeats of a short orbit . we have restricted the presentation here to a demonstration of the method as a means to distinguish time series of chaotic and nonchaotic diffusive systems , but we hope that it will be useful in more general contexts . the case of absorbing boundary conditions provides a rich context for illuminating the differences between chaotic and nonchaotic diffusion . this is because the diffusion equation with absorbing boundary conditions leads naturally to an exponential escape process , while the periodic orbits of the randomly oriented wind - tree model require a power law escape process ( see sec . [ s : npo ] ) , thus violating the diffusion equation . we consider here the probability that a single randomly placed particle will remain in an open system for a given time , or equivalently , the number of identical noninteracting particles remaining in the system after the same time . for chaotic dynamics the escape rate formalism of gaspard and nicolis @xcite uses the escape rate ( defined below ) to connect the diffusion coefficient to dynamical properties such as the ks entropy and the positive lyapunov exponents . for nonchaotic systems the escape process is qualitatively different depending on the properties of the periodic orbits as we have noted above , and the escape rate formalism must be generalized , if it is to make any sense at all . we now describe the escape rate formalism as it applies to chaotic systems , discuss the results for our models , and then attempt to generalize the formalism to include nonchaotic systems . a macroscopic description of escape on a square ( for simplicity ; other geometries are analogous ) is given by the diffusion equation ( [ e : diff ] ) for the particle density @xmath43 , together with the boundary condition @xmath133 along the lines @xmath134 , @xmath135 , @xmath136 and @xmath137 . the general solution is @xmath138 with the decay rate @xmath139 at long times , the solution is dominated by the slowest decaying mode , corresponding to the escape rate @xmath140 note that the escape of particles is exponential in ( [ e : diffsol ] ) . the escape rate formalism equates this macroscopic escape rate @xmath141 with the ( exponential ) escape rate of a ( microscopically ) chaotic system , which is related to its ks entropy @xmath74 and the sum of its positive lyapunov exponents @xmath142 ( here at most a single positive lyapunov exponent ) by @xcite @xmath143 the `` escape rate formula '' which generalizes pesin s formula ( sec . [ s : chaos ] ) to open systems . not only can the escape rate of chaotic systems be calculated from periodic orbit theory @xcite , but , as we will show below , there is an intimate connection between periodic orbits and the escape process in nonchaotic systems . we now present our numerical results , from which we learn how the above theory for chaotic systems is modified in the nonchaotic case . as described in sec . [ s : comp ] , the arrangement of scatterers used is the same as in the periodic case with @xmath13 ; the absorbing boundaries are one distance unit from the edge of the periodic cell , leading to an open system of size @xmath144 . as before we use fixed oriented squares ( f ) , randomly oriented squares ( r ) or circles ( l ) . we denote absorbing boundary conditions by a , so the full notation for these models is [ f , r , l]a18 . we place @xmath145 particles uniformly ( without overlapping the scatterers ) in the square of size @xmath144 and compute the number of particles remaining in the system as a function of time , see figs . [ f : esc ] , [ f : esc2 ] . ( 300,320)(-50,0 ) ( 0,20 ) ( 165,10)@xmath34 ( -15,165)@xmath146 ( 300,320)(-50,0 ) ( 0,20 ) ( 165,10)@xmath34 ( -15,165)@xmath146 the chaotic lorentz gas exhibits exponential decay as predicted by the diffusion equation for all times , but the nonchaotic wind - tree models deviate from exponential decay at late times . the results are well described by the empirical expressions @xmath147 for the l , r and f models respectively over the range of times considered . the initial exponential decay may be compared with eqs . ( [ e : diffsol]-[e : er ] ) to give an effective diffusion coefficient . we find @xmath148 , @xmath149 and @xmath150 , which are certainly consistent with the results of the mean square displacement , tab . [ t : diff ] in sec . [ s : diff ] , given that eqs . ( [ e : diffsol]-[e : er ] ) depend on the macroscopic diffusion equations , and so require a large system @xmath151 limit . the coefficient @xmath152 is obtained by matching eq . ( [ e : diffsol ] ) to the uniform probability density of the initial conditions ; this comes to @xmath153 with @xmath154 as the initial number of particles in the system . from this close agreement , we conclude that the early escape is well described by the diffusion equation with the same diffusion coefficient that appears in the mean square displacement of sec . [ s : diff ] ; there is no trace of non - diffusive behavior at short times . [ [ the - r - model ] ] the r model + + + + + + + + + + + we now discuss the long time behavior of the nonchaotic models . the randomly oriented wind - tree model has a @xmath36 decay in ( [ e : r ] ) due to its periodic orbits ; particles that are initially close to a periodic orbit lead to this power law as described in sec . [ s : npo ] . the coefficient gives an estimate of the density of periodic orbits ; @xmath155 is quite small compared to the number of particles , @xmath145 , so periodic orbits are relatively rare in a sense that is difficult to define precisely . in terms of the escape rate formalism , a power law decay corresponds to an escape rate @xmath156 which trivially satisfies ( [ e : erf ] ) , but yields no information about the diffusion coefficient @xmath28 , that is , ( [ e : er ] ) is not satisfied . [ [ the - f - model ] ] the f model + + + + + + + + + + + the fixed oriented wind - tree model shows a complete escape of all the particles in a finite time , as might be expected from the lack of periodic orbits . the dramatic transition from the initial exponential decay to a much slower ( at first ) linear decay is somewhat surprising . we interpret this as follows : while there are no exactly periodic orbits in the f model , it is possible for the particle to remain in an orbit that is nearly periodic for some time , allowing the particle to remain in the system longer than the exponential decay would predict . if there is a bundle of trajectories that survive for just 800 time units , then this would lead to a linear law because particles are initially evenly distributed along the bundle of trajectories . if there was , in addition , a large bundle of trajectories that survive for , say 500 time units , there would be a kink in the plot , with a sharp decrease in gradient at @xmath157 . the observation that the curve is close to linear ( see fig . [ f : esc2 ] ) thus implies that the lengths of these long living trajectories are strongly peaked around 800 time units . one such long lived orbit is depicted in fig . [ f : traj ] ; its length is seen to be due to two almost periodic orbits . the complete escape of the particles corresponds to an escape rate @xmath158 , which flagrantly violates both ( [ e : er ] ) and ( [ e : erf ] ) and yields no information about the diffusion coefficient @xmath28 . ( 350,350)(-50,0 ) ( 0,0 ) ( 55,150)a ( 65,270)b we conclude from these observations that , although the wind - tree models satisfy the diffusion equation when there are no boundaries , that is , the two time distribution function is of the correct gaussian form , such a macroscopic approximation is not valid when there are absorbing boundary conditions . this means we can not determine the diffusion coefficient in the same way as for the lorentz gas , eqs . ( [ e : diffsol]-[e : er ] ) , that is , from @xmath159 because the infinite time limit leads to zero or infinity as explained above . here @xmath160 is the probability of a particles remaining in a system of size @xmath9 for time @xmath34 ; this is obtained by observing many independent particles , and taking the limit of an infinite number of particles . we can , however , attempt to calculate the diffusion coefficient in the finite wind - tree models by taking the @xmath161 limit at the same time as the @xmath151 limit . this is because the diffusion equation is a good approximation even in the open case as long as the time is not too long , and the decay remains exponential . for example , assuming that the `` density of periodic orbits '' in the ra model is independent of @xmath9 ( which seems to be the case numerically ) , the transition time from exponential to power law decay , as determined by comparing the two terms in eq . ( [ e : r ] ) using eq . ( [ e : er ] ) is of order @xmath162 . this should be compared with the time scale for which the slowest decaying mode of the diffusion equation dominates , @xmath163 , see ( [ e : gmn ] ) . thus , for a given @xmath9 , there is a narrow range in time @xmath164 in which the decay is exponential , and in which a limit can be taken to obtain @xmath28 from the escape process . for example we can combine the @xmath9 and @xmath34 limits by setting @xmath165 and @xmath166 , @xmath167 we expect this relation to hold for all the models [ l , r , f]a . we emphasize that eq . ( [ e : narrow ] ) , if it is mathematically valid , still does not provide a practical method for computing the diffusion coefficient beyond the estimates we gave following eq . ( [ e : f ] ) . this is because the range of time scales @xmath164 grows so slowly with @xmath9 that extremely large system sizes are required for more precise estimates of @xmath28 . we can also turn the argument around , and suggest that the escape at long times of finite systems may be a good experimental technique for detecting ( or ruling out at some level ) nonchaotic microscopic dynamics . in any case , the range of validity of the diffusion equation as a macroscopic description of nonchaotic systems is restricted . we have explored the connections between microscopic chaos and diffusion . on a superficial level , the details of the microscopic dynamics seem to have little effect on the diffusion process : gaussian diffusion is observed in our rp4 model , which has scatterers with flat sides ( hence no exponential separation of nearby trajectories ) and is periodic ( hence there is no source of randomness from a disordered environment ) . in addition , the grassberger - procaccia method of time series analysis can not distinguish our chaotic and nonchaotic diffusive models because it would require impractically long data sets . on a deeper level , however , the subtle differences between chaotic and nonchaotic diffusion are quite apparent if you know where to look : our time series analysis method based on periodic orbits has no trouble distinguishing between the chaotic and nonchaotic models , and the long time behavior of escape from an open system is also determined by the properties of the chaotic versus nonchaotic periodic orbits . in the light of our results , we can make a few concrete suggestions for an experimental determination of the chaotic or nonchaotic properties in a diffusive system . firstly , as we remarked in ref . @xcite , it is necessary to make measurements on the distance and time scales relevant to the microscopic dynamics . in our models , this question is simplified by the fact that all microscopic time scales are of order one in our units , sec . [ s : time ] . once this is achieved , one could use an approach based on the method of almost periodic recurrences , which searches for periodic behavior ; alternatively the late time decay of particles from an open system can also elicit the chaotic or nonchaotic nature of the dynamics . this raises again the question of exactly what dynamical properties would be measured ; there are systems with the positive ks entropy of the lorentz gas as well as the power law unstable periodic orbits of the wind - tree models . we will now make some remarks about such systems , although we make no attempt to enumerate all degrees and classes of chaotic dynamics . our example in this discussion is a model containing both circular and randomly oriented square scatterers . the circular scatterers lead , as in the lorentz gas , to a positive lyapunov exponent , while the square scatterers lead , as in the randomly oriented wind - tree model , to power law unstable periodic orbits , using the same argument as in sec . [ s : npo ] . we have not simulated such a model numerically ; however the arguments used in the previous sections may be applied , leading to the following predictions : sec . [ s : conn ] : there will be a positive diffusion coefficient with a gaussian distribution function , as in both the l and r models . [ s : gp ] : the grassberger - procaccia method will describe the dynamics as chaotic , due to a positive lyapunov exponent and hence positive ks entropy , as in the l models . sec . [ s : isr ] : the apr method will describe the dynamics as nonchaotic , due to the power law unstable periodic orbits , as in the r models . sec . [ s : esc ] : such a system with absorbing boundaries will exhibit power law escape at long times , again due to the periodic orbits , as in the r model . the different conclusions reached by the gp and apr methods exemplify the fact that these methods are based on different dynamical properties ; for diffusion in an open system , it is clear that the mixed model is most similar to the r model , and hence that the apr method distinguishes the relevant dynamical property in this case . we conclude with some open questions and possibilities for further work . in terms of _ microscopic chaos _ , we have studied only a few models . we note that there are examples of lack of chaoticity we have not considered , such as the coexistence of chaotic and nonchaotic regions found in kam theory . in terms of _ diffusion _ , we note that in the light of sec . [ s : esc ] the diffusion equation is not a complete macroscopic description of nonchaotic diffusion . in terms of _ time series analysis _ , our apr method , while sufficient for our purposes here , could be much more developed and applied . with regard to _ periodic orbit theory _ , the methods developed for exploiting the importance of periodic orbits to compute properties of chaotic systems @xcite do not apply to nonchaotic systems , and yet we have seen that periodic orbits also play an important role in nonchaotic systems . we note that our infinite models , while sharing some of the properties of corresponding finite billiards , appear to fall outside the domain of current mathematics . for example , a recent study of infinite billiard systems @xcite is mostly restricted to cases with finite areas that are finitely connected ; our models obey neither of these conditions . it would be interesting to see if it is possible to rigorously determine the status of our models , particularly their ergodic properties , rate of decay of correlations and kolmogorov - sinai entropy within the framework of some mathematical theory of infinite nonchaotic systems . more generally , a mathematical understanding of statistical mechanics including the thermodynamic limit naturally leads to the study of infinite systems . finally , we remark that all the models we have considered with the exception of one ( fp4 ) are diffusive , limiting our investigation as to the presence of microscopic chaos or not to such systems . in addition , the precise role played by microscopic chaos as represented by the lyapunov exponents and `` macroscopic chaos '' , as embodied by the randomly placed scatterers for the existence of a diffusion process and the value of the diffusion coefficient , remains open . a similar but more complicated situation obtains when diffusion of momentum ( viscosity ) or energy ( heat conduction ) and other transport processes are considered . first of all we want to acknowledge that h. van beijeren suggested the wind - tree model as an important alternative to the lorentz gas for the study of microscopic chaoticity and diffusion , which led to the work of ref . @xcite and this paper . we thank him also , along with l. bunimovich , j. r. dorfman , p. gaspard and i. procaccia for stimulating and helpful discussions . this work was supported by the engineering research program of the office of basic energy sciences of the us department of energy under contract # de - fg02 - 88-er13847 . we give here a calculation of the mean free time for our models quoted in sec . [ s : time ] . our derivation is due to chernov @xcite , extended to the fixed oriented model , and with a technical caveat for the infinite models . refer to sec . [ s : comp ] for the definitions of @xmath168 , @xmath9 , @xmath14 and @xmath169 . the mean free time @xmath83 is equal to the mean free path , since the velocity is one . the mean free time is known exactly for billiard systems@xcite , which include the [ r , l]p models discussed here . we make a minor extension to allow the fp model ( which differs because we do not want to allow all velocity directions ) . we expect that the formula for the mean free time would still hold in the infinite models on physical grounds , but we can not justify this mathematically . briefly , the argument in ref . @xcite observes that the total volume of phase space @xmath170 can be computed in two different ways . one expression for the volume of phase space is as an integral over the area of the billiard , giving @xmath171 for the original wind - tree model with fixed orientations , and @xmath172 for random orientations or the lorentz gas . here , @xmath170 is the volume of phase space , including both position and velocity , while @xmath173 is the area accessible to the point particle . for our infinite billiards we think of a large but finite periodic system of length @xmath9 , and formally take the limit @xmath151 at the end of the calculation . since the area @xmath173 is equal to @xmath174 , we have @xmath175 here as @xmath22 . the prefactor @xmath176 comes from the four possible wind directions , and the @xmath177 from integrating over all possible directions . putting these expressions together we have @xmath178 for fixed oriented squares and @xmath179 for randomly oriented squares or circles . an alternative expression for the volume of phase space is an integral over the boundary of the billiard , where the contribution from each point is given by the free path / time @xmath82 . in this way we find @xmath180 for fixed oriented squares and @xmath181 for randomly oriented squares or circles . here @xmath57 is the length of the boundary , that is , the total perimeter of all the scatterers , @xmath83 is the _ mean _ free path / time , and the numerical prefactor is the integral over the component of velocity perpendicular to the boundary . for the wind - tree models we have @xmath182 where @xmath14 is the number of scatterers , so @xmath183 ; for the lorentz gas , we have @xmath184 . comparing the expressions for @xmath170 in both calculations , we find @xmath185 ( using our value of @xmath22 ) for the case of fixed oriented squares , @xmath186 for the case of random oriented squares , and @xmath187 for the lorentz gas . p. gaspard , m. e. briggs , m. k. francis , j. v. sengers , r. w. gammon , j. r. dorfman , and r. v. calabrese , nature * 394 * , 865 ( 1998 ) . c. p. dettmann , e. g. d. cohen and h. van beijeren , nature * 401 * , 875 ( 1999 ) . a. cohen and i. procaccia , phys . a * 31 * , 1872 ( 1985 ) . p. grassberger and i. procaccia , phys . rev . a * 28 * , 2591 ( 1983 ) . e. h. hauge and e. g. d. cohen , j. math . phys . * 10 * , 397 ( 1969 ) . h. van beijeren , rev . mod . phys . * 54 * , 195 ( 1982 ) . p. gaspard , _ chaos , scattering and statistical mechanics _ ( cambridge university press , cambridge , 1998 ) . w. w. wood and f. lado , j. comp . * 7 * , 528 ( 1971 ) . eckmann and d. ruelle , rev . mod . phys . * 57 * , 617 ( 1985 ) . young , ann . math . * 147 * , 585 ( 1998 ) . n. i. chernov and c. p. dettmann , physica a * 279 * , 37 ( 2000 ) . e. gutkin , j. stat . phys . * 83 * , 7 ( 1996 ) . m. wojtkowski , commun . phys . * 105 * , 391 ( 1986 ) . s. lepri , l. rondoni , and g. benettin , chao - dyn/9909004 . gaspard , adv . * xcix * , 369 ( 1997 ) . cvitanovi et al . d. biswas , pramana - j . phys . * 48 * , 487 ( 1997 ) . j. f. aarnes `` an ergodic problem for r - polygons '' dept . of mathematics , technical university of trondheim , norway ( unpublished,1974 ) . p. gaspard and g. nicolis , phys . lett . * 65 * , 1693 ( 1990 ) . p. gaspard and j. r. dorfman , phys . e * 52 * , 3525 ( 1995 ) . s. troubetzkoy , nonlinearity * 12 * , 293 ( 1999 ) . n. chernov , j. stat . phys . * 88 * , 1 ( 1997 ) .

graph models have been used in image analysis for a long time . the edited book @xcite gives an overview of methods in this field . however , approaches from quantitative graph theory such as graph indices have not played a significant role in these applications so far . this is to some extent surprising as it is not a far - fetched idea to model information contained in small patches of a textured image by graphs , and once this has been done , graph indices with their ability to extract in a quantitative form structural information from large collections of graphs lend themselves as a promising tool specifically for texture analysis . a first step in this direction has been made in @xcite where a set of texture descriptors was introduced that combines a construction of graphs from image patches with well - known graph indices . this set of texture descriptors was evaluated in @xcite in the context of a texture discrimination task . in @xcite , an example for texture - based image segmentation was presented based on this work . the present paper continues the work begun in @xcite and @xcite . its purpose is twofold . on one hand , it restates and slightly extends the experimental work from @xcite on texture segmentation , focussing on those descriptors that are based on entropy measures , which turned out particularly useful in the previous contributions . on the other hand , it undertakes a first attempt to analyse the graph - index based texture descriptors with regard to what kind of information they actually extract from a textured image . [ [ structure - of - the - paper . ] ] structure of the paper . + + + + + + + + + + + + + + + + + + + + + + + the remaining part of section [ sec - intro ] briefly outlines the fields of research that are combined in this work , namely quantitative graph theory , see section [ ssec - introqgt ] , graph models in image analysis with emphasis on the pixel graph and its edge - weighted variant , see section [ ssec - introgmia ] , and finally texture analysis , section [ ssec - introtex ] . in section [ sec - gi ] the construction of graph - entropy - based texture descriptors from @xcite is detailed . section [ sec - gac ] gives a brief account of the geodesic active contour method , a well - established approach for image segmentation that is based on the numerical solution of a partial differential equation . texture segmentation combining the graph - entropy - based texture descriptors with the geodesic active contour method is demonstrated on two synthetic examples that represent typical realistic texture segmentation tasks , and a real - world example in section [ sec - segex ] . some theoretical analysis is presented in section [ sec - tdana ] where ( one setup of ) graph - entropy - based texture descriptors is put into relation with fractal dimension measurements on a metric space derived from the pixel graph , and thus a connection is established between graph entropy methods and fractal - based texture descriptors . a short conclusion , section [ sec - conc ] , ends the paper . quantitative measures for graphs have been developed for almost sixty years in mathematical chemistry as a means to analyse molecular graphs @xcite . in the course of time , numerous _ graph indices _ have been derived based on edge connectivity , vertex degrees , distances , and information - theoretic concepts , see e.g. the work @xcite that classifies over @xmath1 descriptors from literature and subjects them to a large - scale statistical evaluation on several test data sets . recently , interesting new graph indices based on the so - called hosoya polynomial have been proposed @xcite . fields of application have diversified in the last decades to include e.g. biological and social networks , and other structures that can mathematically be modelled as graphs , see @xcite . efforts to apply statistical methods to graph indices across these areas have been bundled in the emerging field of _ quantitative graph theory _ @xcite . many contributions in this field group around the tasks of distinguishing and classifying graphs , and quantifying the differences between graphs . the first task focusses on the ability of indices to _ uniquely _ distinguish large sets of individual graphs , termed _ discrimination power _ @xcite . for the latter task , _ inexact graph matching , _ the _ graph edit distance _ @xcite or other measures quantifying the size of substructures that are shared or not shared between two graphs are of particular importance , see also @xcite . the concept of discrimination power has to be complemented for this purpose by the principle of _ smoothness _ of measures , see @xcite , which describes how similar the values of a measure are when it is applied to similar graphs . in @xcite , the quantitative measures of _ structure sensitivity _ and _ abruptness _ have been introduced in order to precisely analyse the discrimination power and smoothness of graph indices . these measures are based on the average and maximum , respectively , of the changes of graph index values when the underlying graph is modified by one elementary edit step of the graph edit distance . discrimination of graph structures by graph indices is also a crucial part of the texture analysis approach discussed here . thus , discrimination power and the related notions of high structure sensitivity and low abruptness matter also in the present context . however , unique identification of individual graphs is somewhat less important in texture analysis than when examining single graphs as in @xcite , as in texture analysis one is confronted with large collections of graphs associated with image pixels , and one is interested in separating these into a small number of classes representing regions . not only will each class contain numerous graphs , but also the spatial arrangement of the associated pixels is to be taken into account as an additional source of information , as segments are expected to be connected . as can be seen in @xcite , there are several ways in which image analysis can be linked to graph concepts . a large class of approaches is based on graphs in which the pixels of a digital image take the role of vertices , and the set of edges is based on neighbourhood relations , with @xmath2- and @xmath3-neighbourhoods as most popular choices in 2d , and similar constructions in 3d , see ( * ? ? ? * section 1.5.1 ) . to imprint actual image information on such a graph , one can furnish it with edge weights that are based on image contrasts . among others , the _ graph cut _ methods @xcite that have recently received much attention for applications such as image segmentation @xcite and correspondence problems @xcite make use of this concept . this setup is also central for the work presented here , see the more detailed account of the pixel graph and edge - weighted pixel graph of an image in section [ ssec - graphcons ] of the present paper . generalising the pixel - graph framework , the graph perspective allows to transfer image processing methods from the regular mesh underlying standard digital images to non - regular meshes that can be related to scanned surface data @xcite but arise also from standard images when considering non - local models @xcite that have recently received great attention in image enhancement . graph morphology , see e.g.@xcite , is one of these generalisations of image processing methods to non - regular meshes , but also variational and pde frameworks have been generalised in this way @xcite . we briefly mention that graphs can also be constructed , after suitable preprocessing , from vertices representing image regions , see ( * ? ? ? * section 1.5.2 ) , opening avenues to high - level semantic image interpretation by means of partition trees . comparison of hierarchies of semantically meaningful partitions can then be achieved e.g. using graph edit distance or related concepts @xcite . returning to the pixel - graph setup which we will also use in this work , see section [ ssec - graphcons ] , we point out a difference of our approach to those that represent the entire image in a single pixel graph . we focus here on subgraphs related to small image patches , thus generating large sets of graphs whose vertex sets , connectivity and/or edge weights encode local image information . to extract meaningful information from such collections , statistical methods such as entropy - based graph indices are particularly suited . in image analysis , the term _ texture _ refers to the small - scale structure of image regions , and as such it has been an object of intensive investigation since the beginnings of digital image analysis . for example , @xcite undertook approaches to define and analyse textures . [ [ complementarity - of - texture - and - shape . ] ] complementarity of texture and shape . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + real - world scenes often consist of collections of distinct objects which in the process of imaging are mapped to regions in an image delineated by more or less sharp boundaries . while the geometric description of region boundaries is afforded by the concept of _ shape _ and typically involves large - scale structures , texture represents the appearance of the individual objects , either their surfaces if images in the case of reflection imaging ( such as photography of opaque objects ) , or their interior if transmission - based imaging modalities ( such as transmission microscopy , x - ray , magnetic resonance ) are being considered . texture is then expressed in the distribution of intensities and their short - scale correlations within a region . a frequently used mathematical formulation of this distinction is the _ cartoon - texture model _ that underlies many works on image restoration and enhancement , see e.g. @xcite . in this approach , ( space - continuous ) images are described as the sum of two functions : a _ cartoon _ component from the space @xmath4 of functions of bounded variation , and a _ texture _ component from a suitable sobolev space . in a refined version of this decomposition @xcite , noise is modelled as a third component assigned to a different function space . note that also in image synthesis ( computer graphics ) the complementarity of shape and texture is used : here , textures are understood as intensity maps that are mapped on the surfaces of geometrically described objects . the exact frontier between shape and texture information in a scene or image , however , is model - dependent . the intensity variation of a surface is partly caused by geometric details of that surface . with a coarse - scale modelling of shape , small - scale variations are included in the texture description , whereas with a refined modelling of shape , some of these variations become part of the shape information . for example , in the texture samples shown in figure [ fi - e3](a ) and ( b ) , a geometric description with sufficiently fine granularity could capture individual leaves or blossoms as shapes , whereas the large - scale viewpoint treats the entire ensemble of leaves or blossoms as texture . [ [ texture - models . ] ] texture models . + + + + + + + + + + + + + + + capturing texture is virtually never possible on the basis of a single pixel . only the simplest of all textures , homogeneous intensity , can be described by a single intensity . for all other textures , intensities within neighbourhoods of suitable size ( that differs from texture to texture ) need to be considered to detect and classify textures . moreover , there is a large variety of structures that can be constitutive of textures , ranging from periodic patterns in which the arrangement of intensity values follows strict geometric rules , via near - periodic and quasi - periodic structures to irregular patterns where just statistical properties of intensities within a neighbourhood are characteristic of the texture . the texture samples in figure [ fi - e3](a ) and ( b ) are located rather in the middle of the scale where both geometric relations and statistics of the intensities are characteristic of the texture ; near - periodic stripe patterns as in the later examples , figures [ fi - e - stripes ] and [ fi - zebra ] , are more geometrically dominated . with emphasis on different categories of textures within this continuum , numerous geometric and statistic approaches have been made over the decades to describe textures . for example , frequency - based models @xcite emphasise the periodic or quasi - periodic aspect of textures . statistics on intensities such as @xcite mark the opposite end of the scale , whereas models based on statistics of image derivative quantities like gradients @xcite or structure tensor entries @xcite attempt to combine statistical with geometrical information . a concept that differs significantly from both approaches has been proposed in @xcite where textures are described generatively via grammars . also fractals @xcite have been proposed as a means to describe , distinguish and classify textures . remember that a fractal is a geometric object , in fact a topological space , for which it is possible to determine , at least locally , a minkowski dimension ( or , almost identical , hausdorff dimension ) which differs from its topological dimension . assume that the fractal is embedded in a surrounding euclidean space , and it is compact . then it can be covered by a finite number of boxes , or balls , of prescribed size . when the size of the boxes or balls is sent to zero , the number of them which is needed to cover the structure grows with some power of the inverse box or ball size . the minkowski dimension of the fractal is essentially the exponent in this power law . the minkowski dimension of a fractal is often non - integer ( which is the reason for its name ) , however , a more precise definition is that minkowski and topological dimensions differ , which also includes cases like the peano curve whose minkowski dimension is an integer ( @xmath5 ) but still different from the topological dimension ( @xmath6 ) . see also @xcite for different concepts of fractal dimension . textured images can be associated with fractals by considering the image manifold , i.e. the function graph if the image is modelled as a function over the image plane , which is naturally embedded in a product space of image domain and the range of intensity values . for example , a planar grey - value image @xmath7 has the image manifold @xmath8 . the dimension of this structure can be considered as a fractal - based texture descriptor . this approach has been stated in @xcite where fractal dimension was put into relation with image roughness and the roughness of physical structures depicted in the image . many works followed this approach , particularly in the 1980s and beginning 1990s when fractals were under particularly intensive investigation in theoretical and applied mathematics . in @xcite several of these approaches are reviewed . an attempt to analyse fractal dimension concepts for texture analysis is found in @xcite . the concept has also been transferred to the analysis of 1d signals , see @xcite . during the last two decades the interest in fractal methods has somewhat reduced but research in the field remains ongoing as can be seen from more recent publications , see e.g. @xcite for signal analysis , @xcite for image analysis with application in material science . with regard to our analysis in section [ sec - tdana ] that leads to a relation between graph methods and fractals , it is worth mentioning that already @xcite linked graph and fractal methods in image analysis , albeit not considering texture but shape description . [ [ texture - segmentation . ] ] texture segmentation . + + + + + + + + + + + + + + + + + + + + + the task of texture segmentation , i.e. decomposing an image into several segments based on texture differences , has been studied for more than forty years , see @xcite . a great variety of different approaches to the problem have been proposed since then . many of these combine generic segmentation approaches , that could also be implemented for merely intensity - based segmentation , with sets of quantitative texture descriptors that are used as inputs to the segmentation . for example , @xcite are based on active contour or active region models for segmentation , whereas @xcite is an example of a clustering - based method . nevertheless , texture segmentation continues to challenge researchers ; in particular , improvements on the side of texture descriptors are still desirable . note that the task of texture segmentation involves a conflict : on one hand , textures can not be detected on single - pixel level , necessitating the inclusion of neighbourhoods in texture descriptor computation . on the other hand , the intended output of a segmentation is a unique assignment of each pixel to a segment , which means to fix the segment boundaries at pixel resolution . to allow sufficiently precise location of boundaries , texture descriptors should therefore not use larger patches than necessary to distinguish the textures present in an image . the content of this paper is centred around a set of texture descriptors that have been introduced in @xcite based on graph representations of local image structure . this model seems to be the first that exploited graph models in discrimination of textures . note that even texture segmentation approaches in literature that use graph cuts for the segmentation task use non - graph - based texture descriptors to bring the texture information into the graph - cut formulation , see e.g. @xcite . our texture segmentation approach that was already shortly demonstrated in @xcite integrates instead graph - based texture descriptors into a non - graph - based segmentation framework , compare section [ sec - gac ] . throughout the history of texture processing , quantitative texture descriptors have played an important role . assigning a tuple of numerical values to a texture provides an interface to established image processing algorithms that were originally designed to act on intensities , and thereby to devise modular frameworks for image processing tasks that involve texture information . following this modular approach , we will attack the texture segmentation task by combining a set of texture descriptors with a well - established image segmentation algorithm . in this section we will introduce the texture descriptors whereas the following section will be devoted to describing the segmentation method . given the variety of different textures that exist in natural images , it can not be expected that one single texture descriptor will be suitable to discriminate arbitrary textures . instead , it will be sensible to come up with a set of descriptors that complement each other well in distinguishing different kinds of textures . to keep the set of descriptors at a manageable size , the individual descriptors should nevertheless be able to discriminate substantial classes of textures . on the other hand , it will be useful both for theoretical analysis and for practical computation if the set of descriptors is not entirely disparate but based on some powerful common concept . in @xcite a family of texture descriptors was introduced based on the application of several graph indices to graphs representing local image information . in combining six sets of graphs derived from an image , whose computation is based on common principles , with a number of different but related graph indices , this family of descriptors is indeed built on a common concepts . the descriptors were evaluated in @xcite in a simple texture discrimination task , and turned out to yield results competitive with haralick features @xcite , a well - established concept in texture analysis . in this comparison , graph indices based on entropy measures stood out by their texture discrimination rate . in the following , we recall the construction of texture descriptors from @xcite , focussing on a subset of the descriptors discussed there the first step is the construction of graphs from image patches . in the second step , graph indices are computed from these graphs . a discrete grey - scale image is given as an array of real intensity values sampled at the nodes of a regular grid . the nodes are points @xmath9 in the plane where @xmath10 , @xmath11 . the spatial mesh sizes @xmath12 and @xmath13 are often assumed to be @xmath6 in image processing , which we will do also here for simplicity . denoting the intensity values by @xmath14 and assuming that @xmath15 , @xmath16 , the image is then described as the array @xmath17 . the nodes of the grid , thus the pixels of the image , can naturally be considered as vertices of a graph in which neighbouring pixels are connected by edges . we will call this graph the _ pixel graph _ @xmath18 of the image . two common choices for what pixels are considered as neighbours are based on @xmath2-neighbourhoods , in which pixel @xmath19 has the two horizontal neighbours @xmath20 and the two vertical neighbours @xmath21 , or @xmath3-neighbourhoods , in which also the four diagonal neighbours @xmath22 are included in the neighbourhood . whereas the @xmath2-neighbourhood setting leads to a somewhat simpler pixel graph ( particularly , it is planar ) , whereas @xmath3-neighbourhoods are better suited to capture the geometry of the underlying ( euclidean ) plane . in this work , we will mostly use @xmath3-neighbourhoods . see 1.5 ) for more variants of graphs assigned to images . we upgrade the pixel graph to an _ edge - weighted pixel graph _ @xmath23 by defining edge weights @xmath24 for neighbouring pixels @xmath25 via @xmath26 i.e. an @xmath27 sum of the spatial distance of grid nodes @xmath28 ( where @xmath29 denotes the euclidean norm ) , and the contrast @xmath30 of their corresponding intensity values , weighted by a positive _ contrast scale _ @xmath31 . this construction can of course be generalised by replacing the euclidean norm in the image plane , and the @xmath27 sum by other norms . with various settings for these norms , it has been used e.g. in @xcite to construct larger spatially adaptive neighbourhoods in images , so - called _ morphological amoebas . _ see also @xcite for a more detailed description of the amoeba framework in a graph - based terminology . all graphs that will enter the texture descriptor construction are derived from the pixel graph or the edge - weighted pixel graph of the image . first , given a pixel @xmath32 and a radius @xmath33 , we define the euclidean patch graph @xmath34 as the subgraph of @xmath23 which includes all nodes @xmath35 with euclidean distance @xmath36 . in this graph , image information is encoded solely in the edge weights . second , we define the adaptive patch graph @xmath37 as the subgraph of @xmath23 which includes all nodes @xmath35 for which @xmath23 contains a path from @xmath32 to @xmath35 with total weight less or equal @xmath38 . in the terminology of @xcite , the node set of @xmath37 is a morphological amoeba of amoeba radius @xmath38 around @xmath32 , which we will denote by @xmath39 . note that the graph @xmath37 encodes image information not only in its edge weights , but also in its node set @xmath39 . one obvious way to compute @xmath39 is by dijkstra s shortest path algorithm @xcite with @xmath32 as starting point . a natural by - product of this algorithm , which is not used in amoeba - based image filtering as in @xcite etc . , is the dijkstra search tree , which we denote as @xmath40 . this is the third candidate graph for our texture description . image information is encoded in this graph in three ways : in the edge weights , the node set , and the connectivity of the tree . dropping the edge weights from @xmath40 , we obtain an unweighted tree @xmath41 which still encodes image information in its node set and connectivity . finally , a dijkstra search tree @xmath42 and its unweighted counterpart @xmath43 can be obtained by applying dijkstra s shortest path algorithm within the euclidean patch graph @xmath34 . whereas @xmath42 encodes image information in the edge weights and connectivity , @xmath43 does so only in the connectivity . applying these procedures to all pixels @xmath44 of a discrete image @xmath45 , we have therefore six collections of graphs which represent different combinations of three cues to local image information ( edge weights , node sets , connectivity ) and can therefore be expected to be suitable for texture discrimination . in the following we will drop the arguments @xmath32 , @xmath38 and use simply @xmath46 etc . to refer to the collections of graphs . in order to turn the collections of graphs into quantitative texture descriptors suitable for texture analysis tasks , the realm of graph indices developed in quantitative graph theory lends itself as a powerful tool . in @xcite , a selection of graph indices was considered for this purpose , including on one hand distance - based graph indices ( the wiener index @xcite , the harary index @xcite and the balaban index @xcite ) and on the other hand entropy - based indices ( bonchev - trinajsti indices @xcite and dehmer entropies @xcite ) . the so - obtained set of @xmath47 texture descriptors was evaluated in @xcite with respect to their discrimination power and diversity . using nine textures from a database @xcite , texture discrimination power was quantified based on simple statistics ( mean value and standard deviation ) of the descriptor values within single - texture patches , calibrating thresholds for certain and uncertain discrimination of textures within the set of textures . diversity of descriptors was measured based on the overlap in the sets of texture pairs discriminated by different descriptors . despite the somewhat ad - hoc character of the threshold calibration , the study gives valuable hints for the selection of powerful subsets of the @xmath47 texture descriptors . among the descriptors being analysed , particularly the entropy - based descriptors ranked medium to high regarding discrimination power for the sample set of textures . for the present work , we focus therefore on three entropy measures which we will recall in the following , namely the dehmer entropies @xmath48 and @xmath49 as well as bonchev and trinajsti s mean information on distances @xmath50 . the latter is restricted by its construction to unweighted graphs , and is therefore used with the unweighted dijkstra trees @xmath51 and @xmath52 . the dehmer entropies can be combined with all six graph collections . in @xcite , the dehmer entropies on the patch graphs @xmath46 and @xmath53 achieved the highest rates of certain discrimination of textures , and outperformed the haralick features included in the study . some of the other descriptors based on dehmer entropies as well as the bonchev - trinajsti information measures achieved middle ranks , combining medium rates of certain discrimination with uncertain discrimination of almost all other texture pairs ; thereby , they were still comparable to haralick features and distance - based graph indices . the measures considered here are based on shannon s entropy @xcite @xmath54 that measures the information content of a discrete probability measure @xmath55 , @xmath56 , @xmath57 . ( note that for @xmath58 one has to set in @xmath59 by limit . ) following @xcite , a discrete probability measure can be assigned to an arbitrary nonnegative - valued function @xmath60 , an _ information functional _ , via @xmath61 an entropy measure on an arbitrary information functional @xmath62 is then obtained by applying to . introduced in @xcite and further investigated in @xcite , the _ mean information on distances _ is the entropy measure resulting from an information functional on the path lengths in a graph . let a graph @xmath63 with @xmath64 vertices @xmath65 be given , and let @xmath66 denote the length of a shortest path from @xmath67 to @xmath68 in @xmath63 ( unweighted , i.e. each edge counting @xmath6 ) . let @xmath69 be the diameter of @xmath63 . for each @xmath70 , let @xmath71 the mean information on distances then is the entropy measure based on the information functional @xmath72 , i.e.@xmath73 let us shortly mention that @xcite also introduces the _ mean information on realised distances _ @xmath74 , which we will not further consider here . as an entropy measure , @xmath74 can be derived from the information functional @xmath66 on the set of all vertex pairs @xmath75 , @xmath76 , of @xmath63 . as pointed out in @xcite , @xmath74 can be generalised straightforwardly to edge - weighted graphs by measuring distances @xmath66 with edge weights , but a similar generalisation of @xmath77 would be degenerated because in generic cases all edge - weighted distances in a graph will be distinct , leading to @xmath78 for all realised values @xmath79 . therefore we will use the mean information on distances @xmath77 only with the unweighted graphs @xmath52 and @xmath51 . the two entropy measures @xmath80 and @xmath81 for unweighted graphs @xmath63 were introduced in @xcite . their high discriminative power for large sets of graphs was impressively demonstrated in @xcite . both measures rely on information functionals on the vertex set @xmath82 of @xmath63 whose construction involves spheres @xmath83 of varying radius @xmath79 around @xmath67 . note that the sphere @xmath83 in @xmath63 is the set of vertices @xmath68 with @xmath84 . for @xmath48 the information functional @xmath85 on vertices @xmath67 of an unweighted graph @xmath63 is defined as @xcite @xmath86 where @xmath87 is the cardinality of the @xmath79-sphere around @xmath68 , with positive parameters @xmath88 . ( note that @xcite used a general exponential with base @xmath89 . for the purpose of the present paper , however , this additional parameter is easily eliminated by multiplying the coefficients @xmath90 with @xmath91 . ) for @xmath49 the information functional @xmath92 relies on the quantities @xmath93 i.e. @xmath94 is the sum of distances from @xmath67 to all points in its @xmath79-sphere . with similar parameters @xmath88 as before , one defines then @xmath95 as pointed out in @xcite , both information functionals , and thus the resulting entropy measures @xmath48 , @xmath49 , can be adapted to edge - weighted graphs @xmath63 via @xmath96 where distances @xmath66 are now measured using the edge weights , and @xmath97\to\mathbb{r}^+_0 $ ] is a decreasing function interpolating a reverse partial sum series of the original @xmath98 coefficients . further following @xcite , we focus on the specific choice @xmath99 ( an instance of the _ exponential weighting scheme _ from @xcite ) and obtain accordingly @xmath100 with a positive constant @xmath101 , which yields @xmath102 with a positive constant @xmath101 , as the final form of the information functionals for our construction of texture descriptors . we use for our experiments a well - established segmentation method based on partial differential equations ( pde ) . introduced in @xcite , _ geodesic active contours _ ( gac ) perform a contrast - based segmentation of a ( grey - scale ) input image @xmath62 . of course , other contrast - based segmentation methods could be chosen , including clustering @xcite or graph - cut methods @xcite . advantages or disadvantages of these methods in connection with graph - entropy - based texture descriptors may be studied in future work . for the time being we focus on the texture descriptors themselves , thus it matters to use just one well - established standard method . from the input image @xmath62 , a gaussian - smoothed image @xmath103 is computed , where @xmath104 is a gaussian kernel of standard deviation @xmath105 . from @xmath106 , one computes an edge map @xmath107 with the help of a decreasing and bounded function @xmath108 with @xmath109 . a popular choice for @xmath110 is @xmath111 which has originally been introduced by perona and malik @xcite as a diffusivity function for nonlinear diffusion filtering of images . herein , @xmath112 is a contrast parameter that acts as a threshold distinguishing high gradients ( indicating probable edges ) from small ones . in addition to the input image , gac require an initial contour @xmath113 ( a regular closed curve ) specified e.g. by user input . this contour is embedded into a _ level set function _ @xmath114 , i.e. @xmath114 is a sufficiently smooth function in the image plane whose zero level - set ( set of all points @xmath115 in the image plane for which @xmath116 ) is the given contour . for example , @xmath114 can be introduced as a signed distance function : @xmath117 is zero if @xmath115 lies on @xmath113 ; it is minus the distance of @xmath115 to @xmath113 if @xmath115 lies in the region enclosed by @xmath113 , and plus the same distance if @xmath115 lies in the outer region . one takes then @xmath114 as initial condition at time @xmath118 for the parabolic pde @xmath119 for a time - dependent level - set function @xmath120 . at each time @xmath121 , an evolved contour can be extracted from @xmath122 as zero level - set . for suitable input images and initialisations and with appropriate parameters , the contours lock in at a steady state that provides a contrast - based segmentation . to understand equation one can compare it to the curvature motion equation @xmath123 that would evolve all level curves of @xmath124 by an inward movement proportional to their curvature . in , this inward movement of level curves is modulated by the edge map @xmath107 , which slows down the curve displacement at high - contrast locations , such that contours stick there . the name _ geodesic active contours _ is due to the fact that the contour evolution associated to can be understood as gradient descent for the curve length of the contour in an image - adaptive metric ( a riemannian metric whose metric tensor is @xmath107 times the unit matrix ) , thus yielding a final contour that is a geodesic with respect to this metric . in its pure form , geodesic active contours require the initial contour ( at least most of it ) to be placed outside the object to be segmented . in some situations , however , it is easier to specify an initial contour inside an object , particularly if the intensities within the object are fairly homogeneous but many spurious edges irritating the segmentation exist in the background region . moreover , despite being able to handle also topology changes such as a splitting from one to several level curves encircling distinct objects to some extent , it has limitations when the topology of the segmentation becomes too complex . as a remedy to both difficulties , one can modify by adding a _ force term _ @xmath125 to its right - hand side . adding a force term was proposed first in @xcite ( by the name of _ balloon force _ ) whereas the specific form of the force term weighted with @xmath110 was proposed in @xcite . depending on the sign of @xmath126 , this force exercises an inward ( @xmath127 ) or outward ( @xmath128 ) pressure on the contour , which ( i ) speeds up the curve evolution , ( ii ) supports the handling of complex segmentation topologies , and ( iii ) enables for @xmath128 also segmentation of objects from initial contours placed inside . the modified gac evolution with force term , @xmath129 will be our segmentation method when performing texture segmentation based on only one quantitative texture descriptor . in this case , the texture descriptor will be used as input image @xmath62 from which the edge map @xmath110 is computed . it is straightforward to extend the gac method , including its modified version with force term to multi - channel input images @xmath130 where each location @xmath115 in the image plane is assigned an @xmath131-tuple @xmath132 of intensities . a common case , with @xmath133 , are rgb colour images . in fact , equations and incur almost no change as even for multi - channel input images , one computes the evolution of a simple real - valued level - set function @xmath124 . what is changed is the computation of the edge map @xmath110 : instead of the gradient norm @xmath134 one uses the frobenius norm @xmath135 of the jacobian @xmath136 where @xmath137 is the gaussian - smoothed input image , @xmath138 with @xmath139 , yielding @xmath140 as edge map . equation with this edge map will be our segmentation method when performing texture segmentation with multiple texture descriptors . the input image @xmath130 will have the individual texture descriptors as channels . to weight the influence of texture descriptors , the channels may be multiplied by scalar factors . for numerical computation we rewrite pde as @xmath141 ( where we have omitted the argument of @xmath110 which is a fixed input function to the pde anyway ) . following established practice , we use then an explicit ( euler forward ) numerical scheme where the right - hand side is spatially discretised as follows . the first term , @xmath142 , is discretised using central differences . for the second term , @xmath143 , an upwind discretisation @xcite is used in which the upwind direction for @xmath124 is determined based on the central - difference approximations of @xmath144 . the third term , @xmath145 , is discretised with an upwind discretisation , too . here , the upwind direction depends on the components of @xmath146 and the sign of @xmath126 . although a detailed stability analysis for this widely used type of explicit scheme for the gac equation seems to be missing , the scheme works for time step sizes @xmath147 up to ca . @xmath148 ( for spatial mesh sizes of @xmath149 ) for @xmath150 , which needs to be reduced somewhat for non - zero @xmath126 . in our experiments in section [ sec - segex ] we use consistently @xmath151 . for the level - set function @xmath124 , we use the signed distance function of the initial contour as initialisation . since during the evolution the shape of the level - set function changes , creating steeper ascents in some regions but flattening slopes elsewhere , we re - initialise @xmath124 to the signed distance function of its current zero level set every @xmath152 iterations . in this section , we present experiments on two synthetic and one real - world test image that demonstrate that graph - entropy - based texture descriptors can be used for texture - based segmentation . an experiment similar to our second synthetic example was already presented in @xcite . 0.001 ( 1000 , 238 ) ( 127 , 0 ) pixels . * ( b ) * texture patch _ leaves _ , same size . * ( c ) * test image composed from ( a ) and ( b ) , @xmath153 pixels . both texture patches are converted to greyscale , downscaled and clipped from the _ vistex _ database @xcite . 1995 massachusetts institute of technology . developed by rosalind picard , chris graczyk , steve mann , josh wachman , len picard , and lee campbell at the media laboratory , mit , cambridge , massachusetts . under general permission for scholarly use . , title="fig : " ] ( 128 , 10 ) ( 381 , 0 ) pixels . * ( b ) * texture patch _ leaves _ , same size . * ( c ) * test image composed from ( a ) and ( b ) , @xmath153 pixels . both texture patches are converted to greyscale , downscaled and clipped from the _ vistex _ database @xcite . 1995 massachusetts institute of technology . developed by rosalind picard , chris graczyk , steve mann , josh wachman , len picard , and lee campbell at the media laboratory , mit , cambridge , massachusetts . under general permission for scholarly use . , title="fig : " ] ( 382 , 10 ) ( 635 , 0 ) pixels . * ( b ) * texture patch _ leaves _ , same size . * ( c ) * test image composed from ( a ) and ( b ) , @xmath153 pixels . both texture patches are converted to greyscale , downscaled and clipped from the _ vistex _ database @xcite . 1995 massachusetts institute of technology . developed by rosalind picard , chris graczyk , steve mann , josh wachman , len picard , and lee campbell at the media laboratory , mit , cambridge , massachusetts . under general permission for scholarly use . , title="fig : " ] ( 636 , 10 ) in our first example we use a synthetic image , shown in figure [ fi - e3](c ) , which is composed from two textures , see figure [ fi - e3](a ) an ( b ) , with a simple shape ( the letter ` e ' ) switching between the two . note that the two textures were also among the nine textures studied in @xcite for the texture discrimination task . with its use of real - world textures , this synthetic example mimicks a realistic segmentation task . its synthetic construction warrants at the same time a ground truth to compare segmentation results with . 0.001 ( 1000 , 492 ) ( 0 , 254)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 1 , 264 ) ( 254 , 254)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 255 , 264 ) ( 508 , 254)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 509 , 264 ) ( 762 , 254)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 763 , 264 ) ( 0 , 0)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 1 , 10 ) ( 254 , 0)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 255 , 10 ) ( 508 , 0)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 509 , 10 ) ( 762 , 0)(c ) ( values rescaled to @xmath154 $ ] ) . patch radii were fixed to @xmath155 , contrast scale to @xmath156 , weighting parameter to @xmath157 . * top row , left to right : * * ( a ) * @xmath158 . * ( b ) * @xmath159 . * ( c ) * @xmath160 . * ( d ) * @xmath161 . * bottom row , left to right : * * ( e ) * @xmath162 . * ( f ) * @xmath163 . * ( g ) * @xmath164 . * ( h ) * @xmath165 . , title="fig : " ] ( 763 , 10 ) figure [ fi - e3-gi ] shows the results of eight graph - entropy - based texture descriptors for the test image . in particular , the combination of the dehmer entropy @xmath48 with all six graph variants from section [ ssec - graphcons ] is shown as well as @xmath49 on the weighted dijkstra trees in non - adaptive and adaptive patches . patch radii were fixed to @xmath155 for both non - adaptive and adaptive patches , whereas the contrast scale was chosen as @xmath156 . these parameter settings have already been used in @xcite ; they are based on values that work across various test images in the context of morphological amoeba image filtering . further investigation of variation of these parameters is left for future work . visual inspection of figure [ fi - e3-gi ] indicates that for this specific textured image , the entropy measure @xmath48 separates the two textures well in combination in particular with the weighted dijkstra tree settings , in both adaptive and non - adaptive patches , see frames ( b ) and ( f ) . the other @xmath48 results in frames ( c ) , ( e ) and ( g ) show insufficient contrast along some parts of the contour of the letter ` e ' . the index @xmath166 in frame ( a ) , which was identified in @xcite as a descriptor with high texture discrimination power , does not distinguish these two textures clearly but creates massive over - segmentation within each of them . in a sense , this over - segmentation is the downside of the high texture discrimination power of the descriptor . note , however , that also other @xmath53-based descriptors tend to this kind of over - segmentation . regarding the @xmath49 index , figure [ fi - e3-gi ] ( d ) and ( h ) , there is a huge difference between the adaptive and non - adaptive patch setting . distinction of the two textures is much better when using non - adaptive patches . 0.001 ( 1000 , 238 ) ( 0 , 0)(c ) . the edge map is computed from the graph - entropy - based texture descriptor @xmath163 from figure [ fi - e3-gi](f ) using pre - smoothing with @xmath167 , perona - malik edge - stopping function with @xmath168 , with expansion force @xmath169 . * left to right : * * ( a ) * initial contour ( @xmath118 ) . * ( b ) * @xmath170 . * ( c ) * @xmath171 . * ( d ) * @xmath172 ( steady state ) . , title="fig : " ] ( 1 , 10 ) ( 254 , 0)(c ) . the edge map is computed from the graph - entropy - based texture descriptor @xmath163 from figure [ fi - e3-gi](f ) using pre - smoothing with @xmath167 , perona - malik edge - stopping function with @xmath168 , with expansion force @xmath169 . * left to right : * * ( a ) * initial contour ( @xmath118 ) . * ( b ) * @xmath170 . * ( c ) * @xmath171 . * ( d ) * @xmath172 ( steady state ) . , title="fig : " ] ( 255 , 10 ) ( 508 , 0)(c ) . the edge map is computed from the graph - entropy - based texture descriptor @xmath163 from figure [ fi - e3-gi](f ) using pre - smoothing with @xmath167 , perona - malik edge - stopping function with @xmath168 , with expansion force @xmath169 . * left to right : * * ( a ) * initial contour ( @xmath118 ) . * ( b ) * @xmath170 . * ( c ) * @xmath171 . * ( d ) * @xmath172 ( steady state ) . , title="fig : " ] ( 509 , 10 ) ( 762 , 0)(c ) . the edge map is computed from the graph - entropy - based texture descriptor @xmath163 from figure [ fi - e3-gi](f ) using pre - smoothing with @xmath167 , perona - malik edge - stopping function with @xmath168 , with expansion force @xmath169 . * left to right : * * ( a ) * initial contour ( @xmath118 ) . * ( b ) * @xmath170 . * ( c ) * @xmath171 . * ( d ) * @xmath172 ( steady state ) . , title="fig : " ] ( 763 , 10 ) finally , we show in figure [ fi - e3-gac ] geodesic active contour segmentation of the test image with the descriptor @xmath163 . we start from an initial contour inside the ` e ' shape , see figure [ fi - e3-gac](a ) , and use an expansion force ( @xmath169 ) to drive the contour evolution in an outward direction . frames ( b ) and ( c ) show two intermediary stages of the evolution , where it is evident that the contour starts to align with the boundary between the two textures . frame ( d ) shows the steady state reached after @xmath173 iterations ( @xmath172 ) . here , the overall shape of the letter ` e ' is reasonably approximated , with deviations coming from small - scale texture details . precision of the segmentation could be increased slightly by combining more than one texture descriptor . we do not follow this direction at this point . 0.001 ( 1000 , 492 ) ( 0 , 254 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 1 , 264 ) ( 254 , 254 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 255 , 264 ) ( 508 , 254 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 509 , 264 ) ( 762 , 254 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 763 , 264 ) ( 0 , 0 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 1 , 10 ) ( 254 , 0 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 255 , 10 ) ( 508 , 0 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 509 , 10 ) ( 762 , 0 ) pixels ) showing a stripe - textured shape in front of a noise background . * ( b d ) * texture descriptors based on graph entropies applied in adaptive patches , @xmath155 , @xmath156 , @xmath157 ; values rescaled to @xmath154 $ ] . * ( b ) * @xmath162 . * ( c ) * @xmath163 . * ( d ) * @xmath164 . * bottom row : * geodesic active contour segmentation of ( a ) using the texture descriptor ( @xmath48 on @xmath174 ) from ( b ) , same parameters as in figure [ fi - e3-gi ] except for @xmath175 . * left to right : * * ( e ) * initial contour ( @xmath118 ) . * ( f ) * @xmath176 . * ( g ) * @xmath177 . * ( h ) * @xmath178 ( steady state ) . , title="fig : " ] ( 763 , 10 ) in our second experiment , figure [ fi - e - stripes ] , we use again a synthetic test image where foreground and background segments are defined using the ` e ' letter shape such that again the desired segmentation result is known as a ground truth . also in this image we combine two realistic textures which can be seen as a simplified version of the foreground and background textures of the real - world test image , figure [ fi - zebra](a ) , used in the next section . this time , the foreground is filled with a stripe pattern whereas the background is noise with uniform distribution in the intensity range @xmath154 $ ] , see figure [ fi - e - stripes](a ) . in frames ( b)(d ) of figure [ fi - e - stripes ] we show the texture descriptors based on @xmath48 with the three graph settings in adaptive patches , using again @xmath155 and @xmath156 . the descriptor @xmath179 that was used for the segmentation in section [ ssec - segex-1 ] visually does not distinguish foreground from background satisfactorily here , whereas @xmath180 that provided no clear distinction of the two textures in section [ ssec - segex-1 ] clearly stands out here . this underlines the necessity of considering multiple descriptors which complement each other in distinguishing textures . our gac segmentation of the test image shown in frames ( e)(h ) is based on the texture descriptor @xmath180 and quickly converges to a fairly good approximation of the segment boundary . 0.001 ( 1000 , 758 ) ( 0 , 516 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 1 , 526 ) ( 339 , 516 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 340 , 526 ) ( 678 , 516 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 680 , 526 ) ( 0 , 258 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 1 , 268 ) ( 339 , 258 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 340 , 268 ) ( 678 , 258 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 680 , 268 ) ( 0 , 0 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 1 , 10 ) ( 339 , 0 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 340 , 10 ) ( 678 , 0 ) pixels ) , converted to greyscale . original image from ` http://en.wikipedia.org/wiki/file:grants_zebra.jpg ` , author ruby 1x2 , released to public domain . * ( b ) * texture descriptor @xmath158 , @xmath181 , @xmath156 , @xmath157 . * ( c ) * texture descriptor @xmath159 , parameters as in ( b ) . * middle row , left to right : * * ( d ) * texture descriptor @xmath161 , parameters as in ( b ) . * ( e ) * texture descriptor @xmath182 , parameters as in ( b ) . * ( f ) * initial contour for geodesic active contour segmentation ( @xmath118 ) . * bottom row , left to right : * * ( g ) * geodesic active contours with edge - stopping function computed from the texture descriptors @xmath161 shown in ( d ) and @xmath182 shown in ( e ) with @xmath183 , perona - malik edge - stopping function , @xmath184 , @xmath185 . * ( h ) * same as ( g ) but @xmath186 . * ( i ) * same as ( g ) but @xmath187 ( steady state ) . , title="fig : " ] ( 680 , 10 ) in our last experiment , figure [ fi - zebra ] , we consider a real - world image showing a zebra , see frame ( a ) . in a sense , this experiment resembles the synthetic case from section [ ssec - segex-2 ] because again a foreground dominated by a clear stripe pattern is to be distinguished from a background filled with small - scale detail . in frames ( b)(e ) four texture descriptors are shown . with regard to the higher resolution of the test image , the patch radius has been chosen slightly larger than in the previous examples , @xmath181 , whereas @xmath156 was retained . as can be seen in frame ( b ) , @xmath158 shows the same kind of over - segmentation behaviour as observed in section [ ssec - segex-1 ] , however , it also separates a large part of the zebra shape well from the background . the second descriptor , @xmath159 in frame ( c ) , appears unsuitable here because it does not yield sufficiently similar values within the black and white stripes to recognise these as a common texture . in contrast , @xmath161 and @xmath182 in figure [ fi - zebra](d ) and ( e ) , respectively , achieve this largely . our gac segmentation in frames ( f)(i ) uses a larger gaussian kernel for pre - smoothing than before , @xmath183 , to flatten out small - scale inhomogeneities in the texture descriptors , and combines the two descriptors from ( d ) and ( e ) . with these data , a large part of the zebra including the head and front part of the torso is segmented in the final steady state . not included are the rear part and the forelegs . note that in the foreleg part the stripes are much thinner than in the segmented region , apparently preventing the recognition of this texture as a continuation of the one from the head and front torso . in contrast , the rear part of the torso shown very thick stripes which under the patch size chosen decompose into separate ( homogeneous ) textures for black and white stripes , as is also visible in the texture descriptors ( d ) and ( e ) themselves . further investigation of parameter variations and inclusion of more texture descriptors might improve this preliminary result in the future . in this section , we undertake an attempt to analyse the texture descriptors based on the entropy measures @xmath48 and @xmath49 , focussing on the question what properties of textures are actually encoded in their information functionals @xmath85 and @xmath92 . part of this analysis is on a heuristic level at the present stage of research , and future work will have to be invested to add precision to these arguments . this applies to the limiting procedure in section [ ssec - graphfrac ] as well as to the concept of local fractal dimension arising in section [ ssec - fracdim ] . we believe , however , that even in its present shape , the analysis provided in the following gives valuable intuition about the principles underlying our texture descriptors . for the purpose of our analysis , we generalise the information functional @xmath85 from directly to edge - weighted graphs by replacing the series @xmath188 of cardinalities from with the monotone increasing function @xmath189 , @xmath190 that measures volumes of spheres with arbitrary radius . assuming the exponential weighting scheme and large @xmath191 this yields @xmath192 an analogous generalisation of is @xmath193 we assume now that the image is sampled successively on finer grids , with decreasing @xmath194 . note that the number of vertices of any region of the edge - weighted pixel graph , or any of the derived edge - weighted graphs introduced in section [ ssec - graphcons ] , grows in this process with @xmath195 . by using the volumes of spheres instead of the original cardinalities , provides a re - normalisation that compensates this effect in . thus , it is possible to consider the limit case @xmath196 . in this limit case , graphs turn into metric spaces representing the structure of a space - continuous image . additionally , these metric spaces are endowed with a volume measure which is the limit case of the discrete measures on graphs given by vertex counting . in simple cases , these metric spaces with volume measure can be manifolds . for example , for a homogeneous grey image without any contrast , the limit of the edge - weighted pixel graph is an approximation to a plane , i.e. a @xmath5-dimensional manifold . for an image with extreme contrasts in one direction , e.g. a stripe pattern , the edge - weighted pixel graphs will be path graphs , resulting in a metric space as limit which is essentially a @xmath6-dimensional manifold . finally , in the extreme case of a noise image in which neighbouring pixels have almost nowhere similar grey - values , the graph will practically decompose into numerous isolated connected components , corresponding to a discrete space of dimension @xmath197 . for more general textured images , the limit spaces will possess a more complicated topological structure . at the same time , it remains possible , of course , to measure volumes of spheres of different radii in these spaces . clearly , sphere volumes will increase with sphere radius . if they fulfil a power law , the ( possible non - integer ) exponent can immediately be interpreted as a dimension . the space itself is then interpreted as some type of fractal @xcite . the dimension concept underlying here is almost that of the minkowski dimension ( closely related to hausdorff dimension ) that is frequently used in fractal theory , with the difference that the volume measure here is inside the object being measured instead of in an embedding space . based on the above reasoning , values of the dimension will range between @xmath197 and @xmath5 . note that even in situations in which there is no global power law for the sphere volumes , and therefore no global dimension , power laws , possibly with varying exponents , will still be approximated for a given sphere centre in suitable ranges of the radius , thus allowing to define the fractal dimension as a quantity varying in space and resolution . this resembles the situation with most fractal concepts being applied to real - world data : the power laws that are required to hold for an ideal fractal across all scales will be found only for certain ranges of scales in reality . dijkstra trees , too , turn into @xmath6-dimensional manifolds in the case of sharp stripe images ; for other cases they will also yield fractals . fractal dimensions , wherever applicable , will be below those observed with the corresponding full edge - weighted pixel graphs , thus , the relevant range of dimensions is again bounded by @xmath197 from below and @xmath5 from above . one word of care must be said at this point . the fractal structures obtained here as limit cases of graphs for @xmath196 are not identical with the image manifolds whose fractal structures are studied as a means of texture analysis in @xcite and others . in fact , fractal dimensions of the latter , measured as minkowski dimensions by means of the embedding of the image manifold of a grey - scale image in three - dimensional euclidean space , range from @xmath5 to @xmath198 with increasing roughness of the image , whereas the dimensions measured in the present work go down from @xmath5 to @xmath197 for increasing image roughness . whereas it can be conjectured that these two fractal structures are related , future work will be needed to gain clarity about this relationship . based on the findings from the previous section , let us now assume that the limit @xmath196 from one of the graph structures results in a measured metric space @xmath199 of dimension @xmath200 $ ] , in which sphere volumes are given by the equation @xmath201 where @xmath202 is the volume of a unit sphere , @xmath203 denoting the gamma function . thus , we assume that @xmath204 interpolates the sphere volumes of euclidean spaces for integer @xmath205 . note that this assumption has indeed two parts . the first , , means that a volume measure on the metric space @xmath199 exists that behaves homogeneously with degree @xmath205 with regard to distances . in the manifold case ( integer @xmath205 ) , this is the case of vanishing curvature ; for general manifolds of integer dimension @xmath205 , would hold as an approximation for small radii . the second assumption , , corresponds to the euclideanness of the metric . for edge - weighted pixel graphs based on @xmath2- or @xmath3-neighbourhoods , the volume of unit spheres actually deviates from , even in the limit . however , with increasing neighbourhood size , is approximated better and better . most of the following analysis does not depend specifically on ; thus we will return to only later for numerical evaluation of information functionals . with we have @xmath206 where we have substituted @xmath207 . as a result , we obtain @xmath208 analogous considerations for @xmath92 from lead to @xmath209 as pointed out before , the metric structure of the fractal @xmath199 will in general be more complicated such that it does not possess a well - defined global dimension . however , such a dimension can be measured at each location and scale . the quantities @xmath85 and @xmath92 as stated in , can then be understood as functions of the local fractal dimension in a neighbourhood of vertex @xmath67 where the size of the neighbourhood the scale is controlled by the decay of the function @xmath210 in the integrands of and , respectively . as a result , we find that the information functionals @xmath85 and @xmath92 represent distribution over the input pixels of an image patch ( non - adaptive or adaptive ) in which the pixels are assigned different weights dependent on a local fractal dimension measure . the entropies @xmath48 and @xmath49 then measure the local homogeneity or inhomogeneity of this dimension distribution : for very homogeneous dimension values within a patch , the density resulting from each of the information functionals @xmath85 , @xmath92 will be fairly homogeneous , implying high entropy . the more the dimension values are spread out , the more will the density be dominated by a few pixels with high values of @xmath85 or @xmath92 , thus yielding low entropy . the precise dependency of the entropy on the dimension distribution will be slightly different for @xmath85 and @xmath92 and will also depend on the choice of @xmath211 . details of this dependency will be a topic of future work . 0.001 ( 1000 , 505 ) ( 5,500 ) ( 0,370 ) ( 190,260)@xmath205 ( 15,260)(a ) ( 333,500 ) ( 338,370 ) ( 525,260)@xmath205 ( 350,260)(b ) ( 661,500 ) ( 666,370 ) ( 853,260)@xmath205 ( 680,260)(c ) ( 5,240 ) ( 0,110 ) ( 190 , 0)@xmath205 ( 15 , 0)(d ) ( 333,240 ) ( 338,110 ) ( 525 , 0)@xmath205 ( 350 , 0)(e ) ( 661,240 ) ( 666,110 ) ( 853 , 0)@xmath205 ( 680 , 0)(f ) to give a basic intuition , we present in figure [ fi - fvfp ] graphs of @xmath212 and @xmath213 as functions of the dimension @xmath200 $ ] for selected values of @xmath211 . in computing these values , the specific choice for @xmath214 has been used . in the left column , frames ( a ) and ( b ) , we use @xmath157 as in our experiments in section [ sec - segex ] . here , both @xmath212 and @xmath213 increase drastically by almost @xmath215 from @xmath216 to @xmath217 and even by @xmath218 from @xmath217 to @xmath219 . in the resulting information functionals @xmath85 and @xmath92 , i.e. after applying @xmath220 to the functions shown in the figure , even pixels with only slightly higher values of the dimension strongly dominate the entire information density within the patch . for increasing @xmath211 , the rate of increment in @xmath85 and @xmath92 with @xmath205 becomes lower . for @xmath221 as shown in the second column , frames ( b ) and ( e ) , of figure [ fi - fvfp ] , the variation of @xmath212 and @xmath213 is already reduced to @xmath5 and @xmath2 , respectively , such that vertices across the entire dimension range @xmath222 $ ] will have a relevant influence on the information density . for even larger @xmath211 the dependency of @xmath85 and @xmath92 on @xmath205 becomes non - monotonic ( as shown in ( c ) for @xmath85 with @xmath223 ) and even monotonically decreasing ( for both @xmath85 and @xmath92 at @xmath224 ; not shown ) . it will therefore be interesting for further investigation to evaluate also the texture discrimination behaviour of the entropy measures for varying @xmath211 , as this may be a way to targeting the sensitivity of the measures specifically at certain dimension ranges . in this context , however , it becomes evident that the parameter @xmath211 plays two different roles at the same time . first , it steers the approximate radius of influence for the fractal dimension estimation . here , it is important that this radius of influence is smaller than the patch size underlying the graph construction , such that the cut - off of the graphs has no significant influence on the values of the information functional at the individual vertices . second , @xmath211 determines the shape and steepness of the function ( compare figure [ fi - fvfp ] ) that relates the local fractal dimension to the values of the information functionals . this makes it desirable to refine in future work the parametrisation of the exponential weighting scheme so that the two roles of @xmath211 are distributed to two parameters . in this paper , we have presented the framework of graph - index - based texture descriptors that has first been introduced in @xcite . particular emphasis was put on entropy - based graph indices that have proven in @xcite to afford medium to high sensitivity for texture differences . we have extended the work from @xcite in two directions . firstly , we have stated an approach to texture - based image segmentation in which the texture descriptor framework was integrated with geodesic active contours @xcite , a standard method for intensity - based image segmentation . this approach was already briefly introduced in @xcite and is demonstrated here by a larger set of experiments , including two synthetic and one real - world example . secondly , we have analysed one representative of the graph - entropy - based texture descriptors in order to gain insight about the image properties that this descriptor relies on . it turned out that it stands in close relation to measurements of fractal dimension of certain metric spaces that arise from the graphs in local image patches that underly our texture descriptors . although this type of fractal dimension measurement in images differs from existing applications of fractal theory in image ( and texture ) analysis , as the latter treat the image manifold as fractal object , results indicate that the two fractal approaches are related . our texture descriptor framework as a whole and also both novel contributions presented here require further research . to mention some topics , we start with parameter selection of the texture descriptors . in @xcite as well as in this paper , most parameters were fixed to specific values based on heuristics . a systematic investigation of the effect of variations of all these parameters is part of ongoing work . inclusion of novel graph descriptors proposed in the literature , e.g.@xcite , is a further option . the algorithms currently used for the computation of graph - entropy - based texture descriptors need computation times in the range of minutes already for small images . as the algorithms have not been designed for efficiency , there is much room for improvement which will also be considered in future work . both texture discrimination and texture segmentation have been demonstrated so far rather on a proof - of - concept level . extensive evaluation on larger sets of image data is ongoing . this is also necessary to gain more insight about the suitability of particular texture descriptors from our set for specific classes of textures . regarding the texture segmentation framework , the conceptual break between the graph - based set of texture descriptors and the partial differential equation for segmentation could be reduced by using e.g. a graph - cut segmentation method . it can be asked whether such a combination even allows for some synergy between the computation steps . this is not clear so far since the features used to weight graph edges are different : intensity contrasts in the texture descriptor phase ; texture descriptor differences in the graph - cut phase . in further course , the integration of graph - entropy - based texture descriptors into more complex segmentation frameworks will be a goal . unsupervised segmentation approaches are not capable to handle involved segmentation tasks ( like in medical diagnostics ) where highly accurate segmentation can only be achieved by including prior information on the shape and appearance of the objects to be segmented . state - of - the - art segmentation frameworks therefore combine the mechanisms of unsupervised segmentation approaches with model - based methods as introduced e.g. in @xcite . on the theoretical side , the analysis of the fractal limit of the descriptor construction will have to be refined and extended to include all six graph settings from section [ ssec - graphcons ] . relations between the fractal structures arising from the graph construction and the more image manifold more commonly treated in fractal - based image analysis will have to be analysed . generally , much more theoretical work deserves to be invested in understanding the connections and possible equivalences between the very disparate approaches to texture description that can be found in the literature . a graph - based approach like ours admits different directions of such comparisons . it can thus be speculated that it could play a pivotal role in understanding the relations between texture description methods , and create a unifying view on different methods that would also have implications for the understanding of texture itself . y. boykov , o. veksler , and r. zabih . markov random fields with efficient approximation . in _ proc . 1998 ieee computer society conference on computer vision and pattern recognition _ , pages 648655 , santa barbara , ca , june 1998 . a. buades , b. coll , and j. morel . a non - local algorithm for image denoising . in _ proc . 2005 ieee computer society conference on computer vision and pattern recognition _ , volume 2 , pages 6065 , san diego , ca , june 2005 . ieee computer society press . d. comaniciu and p. meer . robust analysis of feature spaces : color image segmentation . in _ proc . 1997 ieee computer society conference on computer vision and pattern recognition _ , pages 750755 , san juan , puerto rico , june 1997 . ieee computer society press . a. elmoataz , o. lezoray , v .- t . ta , and s. bougleux . partial difference equations on graphs for local and nonlocal image processing . in o. lezoray and l. grady , editors , _ image processing and analysis with graphs : theory and practice _ , chapter 7 , pages 174206 . crc press , boca raton , 2012 . m. ferrer and h. bunke . graph edit distance theory , algorithms , and applications . in o. lezoray and l. grady , editors , _ image processing and analysis with graphs : theory and practice _ , chapter 13 , pages 383422 . crc press , boca raton , 2012 . b. georgescu , i. shimshoni , and p. meer . mean shift based clustering in high dimensions : a texture classification example . in _ proc . 2003 ieee international conference on computer vision _ , volume 1 , pages 456463 , nice , oct . 2003 . h. ishikawa . graph cuts combinatorial optimization in vision . in o. lezoray and l. grady , editors , _ image processing and analysis with graphs : theory and practice _ , chapter 2 , pages 2563 . crc press , boca raton , 2012 . s. kichenassamy , a. kumar , p. olver , a. tannenbaum , and a. yezzi . gradient flows and geometric active contour models . in _ proc . fifth international conference on computer vision _ , pages 810815 , cambridge , ma , june 1995 . ieee computer society press . r. lerallut , . decencire , and f. meyer . image processing using morphological amoebas . in c. ronse , l. najman , and e. decencire , editors , _ mathematical morphology : 40 years on _ , volume 30 of _ computational imaging and vision _ , pages 1322 . springer , dordrecht , 2005 . o. lezoray and l. grady . graph theory concepts and definitions used in image processing and analysis . in o. lezoray and l. grady , editors , _ image processing and analysis with graphs : theory and practice _ , chapter 1 , pages 124 . crc press , boca raton , 2012 . l. lucchese and s. k. mitra . unsupervised segmentation of color images based on k - means clustering in the chromaticity plane . in _ proc . ieee workshop on content - based access of image and video libraries _ , pages 7478 , fort collins , co , 1999 . p. maragos . fractal aspects of speech signals : dimension and interpolation . in _ proc . 1991 international conference on acoustics , speech , and signal processing _ , volume 1 , pages 417420 , toronto , ontario , canada , may 1991 . l. najman and f. meyer . a short tour of mathematical morphology on edge and vertex weighted graphs . in o. lezoray and l. grady , editors , _ image processing and analysis with graphs : theory and practice _ , chapter 6 , pages 141173 . crc press , boca raton , 2012 . b. sandberg , t. chan , and l. vese . a level - set and gabor - based active contour algorithm for segmenting textured images . technical report cam-02 - 39 , department of mathematics , university of california at los angeles , ca , u.s.a . , july 2002 . m. welk . discrimination of image textures using graph indices . in m. dehmer and f. emmert - streib , editors , _ quantitative graph theory : mathematical foundations and applications _ , chapter 12 , pages 355386 . crc press , 2014 . m. welk and m. breu . morphological amoebas and partial differential equations . in p. w. hawkes , editor , _ advances in imaging and electron physics _ , volume 185 , pages 139212 . elsevier academic press , 2014 . l. zhu , w. ng , and s. han . classifying graphs using theoretical metrics : a study of feasibility . in j. xu , g. yu , s. zhou , and r. unland , editors , _ database systems for advanced applications _ , volume 6637 of _ lecture notes in computer science _ , pages 5364 . springer , berlin , 2011 .

labelled transition systems are a fundamental modelling formalism in many areas of computer science and one often needs to compare two or more such systems in applications . for example , in doing state compression prior to model checking one wants to ensure that the compressed system yields the same model checks as the uncompressed one . similarly , if one system is a specification and another one its implementation , then program correctness can be established by proving these systems to be equivalent . by the same token , if two systems are not equivalent , one may want to know to what degree this is so , e.g. in a risk analysis of a safety - critical system . this paper chooses bisimulation as the notion of equivalence of labelled transition systems . bisimulation is an established , sufficiently fine - grained notion of equivalence between labelled transition systems @xcite so any approximative notions , e.g. testing @xcite , have bisimulation as a well accepted point of reference . since quantitative aspects ought to be invariant under bisimulation , we stipulate that the quotient of all labelled transition systems with respect to bisimulation is the right conceptual space for reasoning about and comparing quantitative aspects of labelled transition systems . if two labelled transition systems are not bisimilar , one may require a quantitative measure of such differences and such a measure has many applications . we mention security protocols @xcite , where one system is the specification and the other is an implementation and where we may wish to quantify illicit information flow @xcite or the effort needed to expose implementation flaws ; modal specifications @xcite , where a specification captures a possibly infinite set of mutually non - bisimilar labelled transition systems ; and requirements engineering @xcite , where each system may be the modal specification of a particular viewpoint and _ consistency measures on modal specifications _ are sought . one principal aim of this paper is to unify several strands of established work in one integrated framework : metric semantics of processes la bakker & zucker @xcite ; use of hennessy - milner logic , domain theory and transition systems la abramsky @xcite ; means of under - specifying and refining processes la larsen & thomsen @xcite ; and representations of classical topological spaces as maximal - point spaces of domains la lawson @xcite . to that end , we use a domain @xmath0 , defined in @xcite and shown to be a universal model for finitely - branching modal transition systems and fully abstract for their refinement in loc . specifically , we discover that the metric induced by the lawson topology on @xmath0 is a generalization of the one in @xcite to modal transition systems ; that the subspace of maximal elements of @xmath0 is a stone space with respect to the lawson ( or scott ) topology ; and that this stone space is an isomorphic representation of the quotient of all labelled transition systems modulo bisimulation , so the topology and metric carry over to that quotient . since a stone space has a _ complete _ ultra - metric , our model has labelled transition systems that are not image - finite , allowing the modelling of continuous state spaces , but all labelled transition systems can be approximated by image - finite ones to any degree of precision . the compactness of this quotient space then makes it possible to study the topological structure of sets of implementations for modal transition systems , the second principal aim of this paper . in particular , our topological analysis shows that 3-valued model checking @xcite reasons about compact sets of labelled transition systems , namely the set of all 2-valued refinements of a given 3-valued system . we propose two measures , a pessimistic and an optimistic one , for how close any refining labelled transition systems of two such 3-valued systems can be . using compactness , we prove that the optimistic measure is zero iff the two 3-valued systems in question have a common refinement . our concepts and results are also _ robust under a change of representation _ , e.g. in moving from event - based to state - based systems or those that combine state and event information . it would be of interest to see whether results similar to the ones of this paper are obtainable for systems that explicitly represent time , probability ( e.g. as done in @xcite ) or other quantitative information . in section [ section : domain ] we review modal transition systems , their refinement , and a fully abstract domain model for these notions . section [ section : stone ] establishes the central result of this paper , showing that the maximal - points space of the fully abstract domain of section [ section : domain ] is a stone space and the quotient of all labelled transition systems with respect to bisimulation . in section [ section : applications ] we give three applications of the compactness of this maximal - points space : a compactness theorem for hennessy - milner logic on compact sets of implementations , an abstract interpretation of compact sets of implementations as scott - closed sets of modal transition systems , and a robust consistency measure for modal transition systems . section [ section : related ] states related work , and section [ section : conclude ] concludes . modal transition systems @xcite are defined like labelled transition systems , except that transitions come in two modes that specify whether such transitions must or may be implemented . a refinement relation between modal transition systems therefore associates to a modal transition system those refining labelled transition systems in which all implementation choices have been resolved . in this section we formalize these notions and present the domain of @xcite as a faithful mathematical model of the model - checking framework of modal transition systems . we define larsen & thomsen s modal transition systems @xcite , their refinement and other key concepts formally and present the domain @xmath0 which is a fully abstract model of such systems and their refinement @xcite . our results are shown within that domain . in this paper , let @xmath1 be a fixed finite set of events and @xmath2 the set of finite words over @xmath3 with @xmath4 denoting the word of length zero . the labelled transition systems considered here have events from @xmath3 only . the structural properties of our domain model require that we also define dams more general notion of mixed transition systems @xcite . a modal transition system @xmath5 has two transition relations @xmath6 on a set of states @xmath7 . the sets @xmath8 and @xmath9 specify _ contractual promises or expectations _ about the reactive capacity and incapacity of implementations , respectively . these guarantees are to be understood with respect to the refinement of states . we write `` _ a _ '' in @xmath10 to denote * a*sserted behavior and `` _ c _ '' in @xmath11 to denote * c*onsistent behavior and use these annotations in judgments @xmath12 and @xmath13 below with the same meaning . in figure [ fig : ref ] we see a contractual guarantee that any state refining @xmath14 can not have a transition labelled with @xmath15 to a state refining @xmath16 as the triple @xmath17 is not in @xmath18 . there is a contractual guarantee that any state refining @xmath19 has a @xmath8-transition labelled with @xmath15 to all states that refine @xmath14 or @xmath20 . 1 . * a _ mixed transition system _ @xcite is a triple @xmath21 such that , for every _ mode _ @xmath22 , the pair @xmath23 is a _ labelled transition system _ , i.e. @xmath24 . * if @xmath25 , then @xmath5 is a _ modal _ transition system @xcite . * we call @xmath5 _ image - finite _ iff for all @xmath26 , @xmath27 , and @xmath28 the set @xmath29 is finite . * a mixed transition system @xmath5 with a designated initial state @xmath30 is _ pointed _ , written @xmath31 . * we call elements of @xmath8 _ must - transitions _ and elements of @xmath32 _ may - transitions_. 2 . let @xmath33 be a mixed transition system . * a relation @xmath34 is a _ refinement within @xmath5 _ @xcite iff @xmath35 implies , for all @xmath27 , 1 . if @xmath36 , there exists some @xmath37 such that @xmath38 ; and 2 . if @xmath39 , there exists some @xmath40 such that @xmath38 . * we write @xmath41 or @xmath42 if there is some refinement @xmath43 with @xmath35 . in that case , @xmath44 _ refines _ ( _ is abstracted by _ ) @xmath45 . * states @xmath45 and @xmath44 are _ refinement - equivalent _ iff ( @xmath46 and @xmath47 ) . * let @xmath48 mean that @xmath49 refines @xmath30 in the mixed transition system that is the disjoint union of @xmath5 and @xmath50 ; @xmath31 and @xmath51 are refinement - equivalent iff @xmath30 and @xmath49 are refinement - equivalent in that union . * the _ implementations _ of @xmath31 are those pointed modal transition systems without may - transitions that refine @xmath31 . as the union @xmath52 of all refinements within @xmath5 is also a refinement within @xmath5 , @xmath52 is the greatest refinement relation within @xmath5 . please note that we use the relational inverse of the @xmath43 in @xcite , as done in @xcite , so our @xmath53 is written as @xmath54 in @xcite . larsen & thomsen s modal transition systems and their refinement @xcite are partial versions of labelled transition systems and bisimulation @xcite . a modal transition system represents those labelled transition systems that refine it , the implementations of @xmath5 . this representation is sound , for if a modal transition system @xmath5 refines a modal transition system @xmath50 , all labelled transition systems that refine @xmath5 also refine @xmath50 as @xmath55 is transitive . 1 . figures [ fig : ref ] and [ fig : reftwo ] depict modal transition systems , where dashed and solid lines depict may - transitions and must - transitions , respectively . the refinement @xmath43 identifies states with the same activity ; e.g. drinks with tomdrinks and bobdrinks etc . the mixed transition system on the left of figure [ fig : mixed ] is not a modal transition system but is refinement - equivalent to the modal transition system on the right of figure [ fig : mixed ] . # 1#2#3#4#5 @font ( 3471,2056)(0,-10 ) ( 957,1129)(2982,1129 ) ( 2862.000,1099.000)(2982.000,1129.000)(2862.000,1159.000)(2898.000,1129.000)(2862.000,1099.000 ) ( 597,1039)(1542,364 ) ( 1426.915,409.337)(1542.000,364.000)(1461.789,458.161)(1473.646,412.824)(1426.915,409.337 ) ( 3297,994)(2037,364 ) ( 2130.915,444.498)(2037.000,364.000)(2157.748,390.833)(2112.132,401.566)(2130.915,444.498 ) ( 3207,1309)(1722,1759)(822,1309 ) ( 915.915,1389.498)(822.000,1309.000)(942.748,1335.833)(897.132,1346.566)(915.915,1389.498 ) ( 462,1309)(12,1579)(372,2029)(597,1354 ) ( 530.592,1458.355)(597.000,1354.000)(587.513,1477.329)(570.437,1433.689)(530.592,1458.355 ) ( 1632,274)(597,274)(597,859 ) ( 627.000,739.000)(597.000,859.000)(567.000,739.000)(597.000,775.000)(627.000,739.000 ) ( 2172,274)(3297,274)(3297,814 ) ( 3327.000,694.000)(3297.000,814.000)(3267.000,694.000)(3297.000,730.000)(3327.000,694.000 ) ( 1677,184)(0,0)[lb ] ( 372,1084)(0,0)[lb ] ( 3072,1084)(0,0)[lb ] ( 1677,1219)(0,0)[lb ] ( 1092,769)(0,0)[lb ] ( 2217,769)(0,0)[lb ] ( 687,49)(0,0)[lb ] ( 2352,49)(0,0)[lb ] ( 2397,1669)(0,0)[lb ] ( 552,1804)(0,0)[lb ] # 1#2#3#4#5 @font ( 3855,4494)(0,-10 ) ( 777,912)(1722,1992 ) ( 1665.557,1881.936)(1722.000,1992.000)(1620.402,1921.446)(1666.686,1928.784)(1665.557,1881.936 ) ( 732,597)(372,237)(867,12)(867,507 ) ( 897.000,387.000)(867.000,507.000)(837.000,387.000)(867.000,423.000)(897.000,387.000 ) ( 3432,3612)(1992,4062)(867,3612 ) ( 967.275,3684.421)(867.000,3612.000)(989.559,3628.713)(944.992,3643.197)(967.275,3684.421 ) ( 867,3297)(1767,2262 ) ( 1665.620,2332.867)(1767.000,2262.000)(1710.897,2372.238)(1711.881,2325.387)(1665.620,2332.867 ) ( 1677,2127)(642,2127)(642,3207 ) ( 672.000,3087.000)(642.000,3207.000)(612.000,3087.000)(642.000,3123.000)(672.000,3087.000 ) ( 552,3612)(12,3972)(642,4467)(642,3792 ) ( 612.000,3912.000)(642.000,3792.000)(672.000,3912.000)(642.000,3876.000)(612.000,3912.000 ) ( 3432,3297)(2082,2262 ) ( 2158.980,2358.820)(2082.000,2262.000)(2195.486,2311.204)(2148.663,2313.108)(2158.980,2358.820 ) ( 2217,2127)(3567,2127)(3567,3207 ) ( 3597.000,3087.000)(3567.000,3207.000)(3537.000,3087.000)(3567.000,3123.000)(3597.000,3087.000 ) ( 1857,1947)(1857,1137)(1182,867 ) ( 1282.275,939.421)(1182.000,867.000)(1304.559,883.713)(1259.992,898.197)(1282.275,939.421 ) ( 1992,1947)(1992,1137)(3027,822 ) ( 2903.464,828.239)(3027.000,822.000)(2920.934,885.640)(2946.639,846.458)(2903.464,828.239 ) ( 1407,732)(2982,732 ) ( 2862.000,702.000)(2982.000,732.000)(2862.000,762.000)(2898.000,732.000)(2862.000,702.000 ) ( 1362,3432)(3072,3432 ) ( 2952.000,3402.000)(3072.000,3432.000)(2952.000,3462.000)(2988.000,3432.000)(2952.000,3402.000 ) ( 1767,2037)(0,0)[lb ] ( 417,3387)(0,0)[lb ] ( 417,687)(0,0)[lb ] ( 3117,3387)(0,0)[lb ] ( 3117,687)(0,0)[lb ] ( 777,4197)(0,0)[lb ] ( 2532,3972)(0,0)[lb ] ( 2577,2217)(0,0)[lb ] ( 867,2217)(0,0)[lb ] ( 1362,2847)(0,0)[lb ] ( 1767,3522)(0,0)[lb ] ( 642,1452)(0,0)[lb ] ( 1092,102)(0,0)[lb ] ( 1767,552)(0,0)[lb ] ( 2082,1182)(0,0)[lb ] ( 2307,2847)(0,0)[lb ] ( 1137,1182)(0,0)[lb ] we may identify modal transition systems @xmath56 with labelled transition systems @xmath57 and refinement between such modal transition systems with bisimulation @xcite and will freely move between these two representations of labelled transition systems and bisimulation subsequently . before we present the domain model for refinement of modal transition systems we use scott s interval domain @xcite as a motivating example that features most of the desirable properties of our domain model . figure [ fig : idom ] shows the interval domain and its ordering : @xmath58\leq [ r',s']$ ] iff ( @xmath59 and @xmath60 ) . in that case we say that @xmath61 $ ] refines @xmath58 $ ] . # 1#2#3#4#5 @font ( 5175,4860)(0,-10 ) ( 3015,4545 ) ( 3015,4545 ) ( 540,4545)(5040,4545)(2790,270)(540,4545 ) ( 2565,3420)(2340,4545 ) ( 2565,3420)(3690,4545 ) ( 2790,2475)(1215,4545 ) ( 2790,2475)(4455,4545 ) ( 2790,2475)(2792,2477)(2795,2481 ) ( 2801,2488)(2809,2498)(2820,2512 ) ( 2833,2528)(2847,2546)(2862,2566 ) ( 2876,2588)(2890,2610)(2902,2633 ) ( 2913,2656)(2921,2681)(2928,2707 ) ( 2931,2734)(2930,2762)(2925,2790 ) ( 2915,2816)(2903,2839)(2889,2858 ) ( 2875,2872)(2862,2882)(2850,2888 ) ( 2838,2892)(2827,2894)(2816,2895 ) ( 2805,2896)(2793,2897)(2780,2900 ) ( 2765,2905)(2747,2913)(2727,2923 ) ( 2705,2937)(2680,2953)(2655,2970 ) ( 2628,2989)(2604,3005)(2585,3017 ) ( 2569,3024)(2557,3028)(2548,3029 ) ( 2541,3028)(2535,3026)(2529,3024 ) ( 2524,3024)(2517,3026)(2510,3031 ) ( 2501,3042)(2492,3057)(2482,3079 ) ( 2475,3105)(2472,3131)(2471,3156 ) ( 2472,3178)(2474,3197)(2477,3213 ) ( 2481,3226)(2485,3236)(2489,3244 ) ( 2494,3251)(2498,3259)(2504,3268 ) ( 2509,3278)(2516,3292)(2523,3310 ) ( 2532,3331)(2542,3358)(2553,3388 ) ( 2565,3420)(2579,3456)(2593,3487 ) ( 2605,3512)(2616,3531)(2625,3544 ) ( 2634,3553)(2641,3558)(2648,3562 ) ( 2654,3566)(2661,3572)(2667,3580 ) ( 2675,3592)(2682,3609)(2689,3632 ) ( 2696,3659)(2700,3690)(2701,3724 ) ( 2699,3753)(2695,3774)(2690,3789 ) ( 2685,3798)(2680,3803)(2674,3806 ) ( 2668,3809)(2661,3814)(2653,3823 ) ( 2645,3837)(2634,3857)(2623,3884 ) ( 2610,3915)(2597,3946)(2586,3974 ) ( 2575,3996)(2567,4013)(2559,4026 ) ( 2552,4035)(2546,4043)(2540,4050 ) ( 2535,4058)(2530,4069)(2525,4082 ) ( 2521,4100)(2519,4120)(2520,4140 ) ( 2524,4155)(2530,4167)(2536,4176 ) ( 2542,4181)(2547,4184)(2551,4186 ) ( 2554,4186)(2558,4185)(2561,4184 ) ( 2566,4183)(2573,4182)(2582,4182 ) ( 2594,4183)(2611,4183)(2631,4184 ) ( 2655,4185)(2681,4184)(2706,4182 ) ( 2728,4179)(2746,4175)(2761,4171 ) ( 2774,4166)(2784,4160)(2794,4155 ) ( 2803,4150)(2813,4145)(2825,4140 ) ( 2840,4137)(2858,4135)(2879,4134 ) ( 2902,4136)(2925,4140)(2946,4147 ) ( 2963,4154)(2975,4162)(2984,4168 ) ( 2989,4174)(2992,4178)(2993,4182 ) ( 2993,4185)(2992,4189)(2992,4193 ) ( 2993,4200)(2995,4208)(2998,4220 ) ( 3003,4235)(3009,4254)(3015,4275 ) ( 3020,4301)(3024,4327)(3026,4351 ) ( 3027,4374)(3026,4395)(3025,4416 ) ( 3023,4435)(3021,4453)(3019,4469 ) ( 3018,4482)(3016,4492)(3015,4497)(3015,4500 ) ( 0,4545)(0,0)[lb ] ( 5175,4545)(0,0)[lb ] ( 2700,0)(0,0)[lb ] ( 990,4680)(0,0)[lb ] ( 2745,4635)(0,0)[lb ] ( 2070,4635)(0,0)[lb ] ( 3465,4635)(0,0)[lb ] ( 4320,4635)(0,0)[lb ] ( 2565,3195)(0,0)[lb ] ( 2700,2295)(0,0)[lb ] ( 3240,630)(0,0)[lb ] ( 3600,1215)(0,0)[lb ] the interval domain nicely illustrates some of the properties we expect our domain model @xmath0 to have . 1 . * refinement is complete for implementations : * real numbers @xmath62 $ ] represented as intervals @xmath63 $ ] are the `` implementations '' of intervals , so @xmath58 $ ] has all @xmath63 $ ] with @xmath64 $ ] as implementations . one can easily see that @xmath58 $ ] is refined by @xmath61 $ ] iff all implementations of @xmath61 $ ] are also implementations of @xmath58 $ ] . * universality : * the interval domain @xmath65 is universal for worst / best - case abstractions of subsets of @xmath66 $ ] . if we abstract @xmath67 $ ] by the interval @xmath68\in{{\mathbb i}}$ ] , any element of @xmath65 is the abstraction of at least one such @xmath69 . in fact , there is a galois connection @xmath70})}^{\rm op}\to { { \mathbb i}}$ ] and @xmath71})}^{\rm op}$ ] where @xmath72 $ ] is the monotone abstraction function , @xmath73 ) = [ r , s]$ ] is the monotone `` concretization '' function , and @xmath74 and @xmath75})}^{\rm op}}$ ] . * full abstraction : * the order on @xmath65 coincides with the refinement relation as the latter means reverse containment of implementations by item ( 1 ) above . 4 . * classical space as maximal - points space : * the set @xmath66 $ ] equipped with the compact euclidean topology is isomorphic as a topological space to the set of maximal elements of @xmath65 in the topology induced by the scott- or lawson - topology of @xmath65 . * denseness of computable structures : * intervals with rational endpoints approximate intervals to any degree of precision . * consistency measure : * the map @xmath76 defined by @xmath77,[r',s ' ] ) = [ \max(r , r'),\min(s , s')]$ ] , where @xmath78 $ ] is understood to be @xmath79 if @xmath80 , tells us whether its inputs are consistent with each other by checking whether its output is different from @xmath79 . non - overlapping intervals can not possibly approximate the same real number . the domain model @xmath0 for refinement of modal transition systems @xcite has similar properties which we discuss briefly here prior to their technical development in this paper . the completeness proof for implementations for refinement of modal transition systems does not depend on the compactness of @xmath81 , is non - trivial , and presented elsewhere @xcite . universality amounts to showing that every modal transition system has a refinement - equivalent embedding in the domain @xmath0 . full abstraction means that the order on @xmath0 equals the greatest refinement relation on @xmath0 interpreted as a modal transition system . the maximal - points space @xmath81 of @xmath0 gives us a precise model of labelled transition systems and their notion of `` nearness . '' this space turns out to be the quotient of labelled transition systems with respect to bisimulation such that the familiar metric based on tests expressed in hennessy - milner logic @xcite induces the topology on that space . finite - state labelled transition systems are shown to be dense in this space . finally , the compactness of this space is proved and a monotone consistency measure @xmath82 between two modal transition systems is then derived thereof . said compactness then renders a galois connection between compact sets of implementations and scott - closed sets of modal transition systems as shown in theorem [ theorem : gc ] below . apart from these similarities with @xmath65 , a key difference is that @xmath0 is algebraic and that the maximal - points space is therefore zero - dimensional . the reader familiar with domain theory @xcite may safely skip the next definition . 1 . * a _ topological space _ @xmath83 consists of a set @xmath69 and a family @xmath84 of subsets of @xmath69 such that @xmath85 and @xmath69 are in @xmath84 , and @xmath84 is closed under finite intersections and arbitrary unions . * elements @xmath86 are _ _ , complements @xmath87 with @xmath88 are _ @xmath84-closed _ , and sets that are @xmath84-open and @xmath84-closed are _ @xmath84-clopen_. 2 . * a subset @xmath89 of a partial order @xmath90 is _ directed _ iff ( for all @xmath91 there is some @xmath92 with @xmath93 ) . * a partial order @xmath90 is a _ iff all its directed subsets @xmath89 have a least upper bound @xmath94 . * we write @xmath95 for the set of _ upper bounds _ of @xmath89 . * we denote by @xmath96 the set of _ minimal upper bounds _ of @xmath89 . * an element @xmath97 is _ compact _ in a dcpo @xmath98 iff ( for all directed sets @xmath89 of @xmath98 with @xmath99 there is some @xmath100 with @xmath101 ) . we write @xmath102 for the set of compact elements of @xmath98 . * a dcpo @xmath98 is _ algebraic _ iff for all @xmath103 the set @xmath104 is directed with least upper bound @xmath105 . * for a finite subset @xmath106 of @xmath98 define , for all @xmath107 + @xmath108 * a _ bifinite domain _ , also known as an _ sfp_-domain , is an algebraic dcpo @xmath98 such that for every finite subset @xmath109 the set @xmath110 is finite , contained in @xmath102 , and @xmath111 where for any @xmath112 we write + @xmath113 * we call @xmath69 _ upper _ iff @xmath114 ; _ lower _ iff @xmath115 . 3 . for a bifinite domain @xmath98 , we define * the _ scott - topology _ @xmath116 to consist of all subsets @xmath117 of @xmath98 satisfying @xmath118 * the _ lawson - topology _ @xmath119 to consist of all subsets @xmath120 of @xmath98 such that @xmath121 implies the existence of some @xmath122 with @xmath123 ; and * the _ @xmath116-compact saturated _ subsets of @xmath98 to be the @xmath119-closed upper subsets of @xmath98 . the definitions of item ( 3 ) above are really characterizations @xcite . we use the _ initial _ solution @xmath0 of a domain equation , presented in @xcite and denoted by @xmath124 in loc . , as the domain whose set of maximal points we prove to be the stone space of pointed labelled transition systems modulo bisimulation . the items ( 2 ) and ( 3 ) of definition [ def : mtsd ] below are definition 8 and 9 of @xcite , respectively . [ def : mtsd ] 1 . the _ mixed powerdomain _ @xmath125}$ ] @xcite of a bifinite domain @xmath98 has as elements all pairs @xmath126 where @xmath127 is @xmath116-closed and @xmath117 is @xmath116-compact saturated such that @xmath127 and @xmath117 satisfy the mix condition + @xmath128 + the order on @xmath125}$ ] is defined by + @xmath129 2 . since @xmath125}$ ] is a bifinite domain whenever @xmath98 is bifinite and since the functors @xmath130 and @xmath131 are locally continuous @xcite , we can solve the domain equation + @xmath132}\ ] ] + over bifinite domains where @xmath133 denotes the product functor over all events in @xmath3 , and write @xmath0 for the _ initial solution _ of that equation . 3 . the domain @xmath0 may be _ interpreted as a pointed mixed transition system _ + @xmath134 + where the recursion @xmath135 of the equation ( [ equ : recursion ] ) for @xmath0 specifies that all elements @xmath136 in the set @xmath137 ( @xmath138 ) are exactly the @xmath139-successors ( @xmath140-successors ) of @xmath105 for @xmath141 in @xmath124 ( respectively ) . thus , the @xmath127 and @xmath117 in ( [ equ : mix ] ) model @xmath139- and @xmath140-transitions within @xmath124 , respectively . the order - theoretic mix condition ( [ equ : mix ] ) has an equivalent version for mixed transition systems . a mixed transition system @xmath33 satisfies the _ mix condition ( mc ) _ iff ( for all @xmath36 there is some @xmath142 such that @xmath143 ) . as shown in proposition 3 in @xcite , ( [ equ : mix ] ) ensures that @xmath124 satisfies the mix condition ( mc ) since the order on @xmath0 is a refinement within @xmath0 : for all @xmath144 there is some @xmath145 such that @xmath146 . [ example : mix ] figure [ fig : mixed ] demonstrates that mixed transition systems @xmath147 ) that satisfy the mix condition ( mc ) are refinement - equivalent to modal transition systems @xmath148 . therefore , such mixed transition systems are merely modal transition systems in disguise @xcite . # 1#2#3#4#5 @font ( 5488,2457)(0,-10 ) ( 1362,180 ) ( 462,1530 ) ( 2262,1530 ) ( 4512,180 ) ( 5412,1508 ) ( 3612,1530 ) ( 1317,225)(507,1440 ) ( 598.526,1356.795)(507.000,1440.000)(548.603,1323.513)(553.595,1370.108)(598.526,1356.795 ) ( 1473,287)(2238,1367 ) ( 2193.119,1251.737)(2238.000,1367.000)(2144.157,1286.418)(2189.447,1298.454)(2193.119,1251.737 ) ( 1362,360)(2127,1440 ) ( 2082.119,1324.737)(2127.000,1440.000)(2033.157,1359.418)(2078.447,1371.454)(2082.119,1324.737 ) ( 417,1575)(12,1980)(462,2430)(462,1755 ) ( 432.000,1875.000)(462.000,1755.000)(492.000,1875.000)(462.000,1839.000)(432.000,1875.000 ) ( 2172,1530)(597,1530 ) ( 717.000,1560.000)(597.000,1530.000)(717.000,1500.000)(681.000,1530.000)(717.000,1560.000 ) ( 4546,278)(5311,1358 ) ( 5266.119,1242.737)(5311.000,1358.000)(5217.157,1277.418)(5262.447,1289.454)(5266.119,1242.737 ) ( 5277,1530)(3702,1530 ) ( 3822.000,1560.000)(3702.000,1530.000)(3822.000,1500.000)(3786.000,1530.000)(3822.000,1560.000 ) ( 3522,1530)(3117,1935)(3567,2385)(3567,1710 ) ( 3537.000,1830.000)(3567.000,1710.000)(3597.000,1830.000)(3567.000,1794.000)(3537.000,1830.000 ) ( 192,1305)(0,0)[lb ] ( 2307,1305)(0,0)[lb ] ( 1182,1620)(0,0)[lb ] ( 552,2025)(0,0)[lb ] ( 1857,585)(0,0)[lb ] ( 1092,0)(0,0)[lb ] ( 687,630)(0,0)[lb ] ( 1452,900)(0,0)[lb ] ( 4152,1575)(0,0)[lb ] ( 4872,630)(0,0)[lb ] ( 3612,1980)(0,0)[lb ] ( 3297,1350)(0,0)[lb ] ( 4242,45)(0,0)[lb ] ( 5367,1260)(0,0)[lb ] by proposition 1 in @xcite and as seen in the previous example , the mix condition ( mc ) guarantees that the _ mixed _ transition system @xmath149 is refinement - equivalent to the _ modal _ transition system @xmath150 . therefore all reasoning that is invariant under refinement equivalence , as is the case in this paper , may be done with the latter modal transition system and we abuse notation to refer to that modal transition system as @xmath124 as well . the domain model @xmath0 is _ universal _ : there is an embedding @xmath151 from the class of image - finite pointed mixed transition system satisfying the mix - condition ( mc ) to elements of @xmath0 such that @xmath31 and @xmath152 are refinement - equivalent ( theorem 6.1 in @xcite ) . the domain model @xmath0 is _ fully abstract _ : for all @xmath153 , we have @xmath154 iff @xmath155 ( theorem 5 in @xcite ) . for sake of completeness , we sketch the construction of this embedding and needed aspects of the full abstraction proof in the next section . we show that the maximal elements of @xmath0 are precisely the representations of pointed labelled transition systems modulo bisimulation ; and that this quotient is a stone space and therefore determined by a complete ultra metric . we define the required notions from topology . [ def : ultra ] 1 . a topological space @xmath83 is 1 . _ compact _ iff for all @xmath156 with @xmath157 there is a finite subset @xmath158 with @xmath159 ; 2 . _ hausdorff _ iff for all @xmath160 in @xmath69 there are @xmath161 with @xmath162 , @xmath163 and @xmath164 ; 3 . _ zero - dimensional _ iff every @xmath84-open set is the union of @xmath84-clopens ; and 4 . a _ stone space _ iff it is zero - dimensional , compact , and hausdorff . 2 . a subset @xmath165 of @xmath83 is _ @xmath84-compact _ iff the topological space @xmath166 is compact . a subset @xmath89 of @xmath69 is _ dense _ in @xmath83 iff @xmath167 is non - empty for all non - empty @xmath168 . [ item : fourd ] an _ ultra - metric _ on @xmath69 is a function @xmath169 $ ] such that for all @xmath170 1 . @xmath171 iff @xmath172 ; 2 . @xmath173 ; and 3 . 5 . an ultra - metric @xmath169 $ ] _ determines a topology @xmath175 _ on @xmath69 whose elements are all those @xmath176 that are unions of sets of the form @xmath177 for @xmath178 and rational @xmath179 . a topological space @xmath83 is _ ultra - metrizable _ iff there is an ultra - metric @xmath169 $ ] such that @xmath180 . we denote by @xmath181 the set of _ maximal elements _ of @xmath0 . the set + @xmath182 + has a _ maximal - points space topology _ @xcite + @xmath183 8 . for @xmath184 , we write + @xmath185 since @xmath0 is a bifinite domain , the lawson condition @xcite holds for @xmath0 , namely that the topology @xmath186 is also induced by the @xmath187-topology : @xmath188 we remark that not all bifinite domains @xmath98 enjoy the property that @xmath189 is compact in the topology induced by @xmath116 or @xmath119 . we first record that @xmath190 is hausdorff and zero - dimensional . proposition [ prop : zero ] below holds for any algebraic domain satisfying the lawson condition @xcite . we state and prove that proposition for @xmath0 for sake of completeness . [ prop : zero ] the topological space @xmath191 is zero - dimensional and hausdorff . * every @xmath192 is the union of @xmath193-opens @xmath194 , @xmath195 , as @xmath0 is algebraic . but each @xmath194 is @xmath187-clopen as @xmath196 and @xmath194 is @xmath187-closed . from the lawson condition for @xmath0 , ( [ equ : lawsoncondition ] ) , we infer that @xmath197 is @xmath190-clopen and so @xmath186 is zero - dimensional as every @xmath198 is the union of such sets . * to show that @xmath186 is hausdorff , let @xmath199 . since @xmath0 is a partial order we may assume @xmath80 without loss of generality . since @xmath0 is algebraic , @xmath80 implies @xmath200 and @xmath201 for some @xmath195 . but @xmath197 is @xmath186-open and contains @xmath202 whereas @xmath203 is in @xmath204 which is @xmath190-open since @xmath197 is also @xmath190-closed . we use tools from temporal logic to develop a sufficient criterion for membership in @xmath205 . [ def : weak ] 1 . the set of formulas of _ hennessy - milner logic _ @xcite is generated by the grammar + @xmath206 + where @xmath141 ranges over the finite set of events @xmath3 . 2 . let @xmath207 be a pointed modal transition system . larsen s semantics , denoted by @xmath208 in @xcite for hennessy - milner logic in negation normal form , is depicted in figure [ fig : hml ] . 3 . we write @xmath209}}$ ] for @xmath210 and @xmath211 for @xmath212 subsequently for all @xmath27 and all @xmath213 and @xmath214 of hennessy - milner logic . @xmath215 for each @xmath28 we have @xmath216}}\phi & \mbox{iff } & \mbox{for all $ ( i,\alpha , i')\in{{r } ^{\it { \it \neg m}}}$ , $ ( n , i'){{\models } ^{\it { m}}}\phi$}\\ ( n , i){{\models } ^{\it { m}}}\phi\lor\psi & \mbox{iff } & \mbox{$((n , i){{\models } ^{\it { m}}}\phi$ or $ ( n , i){{\models } ^{\it { m}}}\psi)$\ , .}\end{array}\ ] ] please note that @xmath217}}\phi$ ] universally quantifies over transitions in the _ dual _ mode @xmath218 . [ example : two ] consider the modal transition system @xmath50 in figure [ fig : ref ] . 1 . [ item : one ] we have @xmath219 because of the @xmath18-transition @xmath220 . by the semantics of negation , this implies @xmath221 . we also infer @xmath222 as there is no state @xmath45 with @xmath223 . by the semantics of disjunction , these two judgments render @xmath224 . this judgment says that we ca nt determine that @xmath225 is asserted in state @xmath16 in @xmath50 . as that formula is a tautology over labelled transition systems we see that judgments @xmath226 under - approximate validity judgments `` all refinements of @xmath227 satisfy @xmath213 . '' as we show below , it turns out that the ability to capture these validity judgments for certain tautologies over labelled transition systems via @xmath12 is what characterizes modal transition systems that are refinement - equivalent to labelled transition systems . [ item : two ] we have @xmath228}}{{[{\rm talks}]}}({{\langle { \rm drinks}\rangle } } { { \it t\!t}}\lor\neg{{\langle { \rm drinks}\rangle } } { { \it t\!t}})$ ] as there is an @xmath18-path @xmath229 for the word @xmath230 and @xmath231 by item ( 1 ) . therefore , the check @xmath232}}{{[{\rm talks}]}}({{\langle { \rm drinks}\rangle } } { { \it t\!t}}\lor\neg{{\langle { \rm drinks}\rangle } } { { \it t\!t}})$ ] is unable to validate a tautology over labelled transition systems at state waits in @xmath50 . we sketch the definition of the embedding @xmath233 for an image - finite modal transition system @xmath31 such that @xmath31 and @xmath152 are refinement - equivalent @xcite . this construction follows ideas from algebraic semantics la nivat - courcelle - guessarian @xcite or la goguen - thatcher - wagner - wright @xcite in that we unfold pointed modal transition systems as finite trees for a fixed depth , adding a may - stub to all leaves of that tree for which there are still outgoing transitions in the pointed modal transition system . this unfolding is presented here via a simple process algebra . 1 . the grammar for the _ process algebra @xmath234 _ is + @xmath235 + where @xmath141 ranges over the finite set of events @xmath3 and no @xmath236 in @xmath237 is allowed to be @xmath238 or @xmath239 . 2 . for each @xmath240 let @xmath241 be as in figure [ fig : pad ] . 3 . for all @xmath240 , the structural operational semantics in figure [ fig : sos ] defines a pointed modal transition system @xmath242 _ { } } } , p)$ ] . l @xmath243 + + @xmath244 + + @xmath245 + + @xmath246 + + @xmath247 + + @xmath248 + + @xmath249 + + let @xmath240 be @xmath250 . then @xmath242_{}}},p)$ ] is refinement - equivalent to the image - finite pointed modal transition system in figure [ fig : app ] . we record that the denotational semantics of @xmath234 in @xmath0 matches the structural operational semantics . this proof is straightforward and amounts to showing that the saturations with @xmath251 and @xmath252 in @xmath0 do not break refinement equivalence as they always occur in the right direction . [ lemma : match ] for all @xmath240 , the modal transition system @xmath242_{}}},p)$ ] is refinement - equivalent to the mixed transition system @xmath253 . to define the embedding @xmath254 for an image - finite pointed modal transition system @xmath31 consider @xmath255 , unwind @xmath5 from @xmath30 as a tree @xmath256 $ ] such that all , and only , paths of length @xmath257 of @xmath5 are present . if a leaf of that tree has some @xmath18-successor in @xmath5 , create @xmath18-loops on that leaf for _ all _ events in @xmath3 ( a _ may - stub _ ) ; otherwise , leave it as is ( deadlock ) . by construction , this image - finite pointed modal transition system @xmath258,i)$ ] is the operational meaning @xmath259 _ { } } } , p_m)$ ] of a term @xmath260 so @xmath261 and lemma [ lemma : match ] imply that @xmath262 . thus @xmath263 is directed and we can set @xmath264 and note , shown in @xcite for bifinite domains without reference to a process algebra , that @xmath265 we may thus represent all @xmath195 in the form @xmath266 for some @xmath240 subsequently . ll & @xmath267 + + @xmath268 & @xmath269 + + @xmath270 & @xmath271 + + figure [ fig : app ] illustrates the construction of a finite approximation and depicts @xmath272,{\rm tomdrinks})$ ] for the pointed modal transition system @xmath273 of figure [ fig : reftwo ] . # 1#2#3#4#5 @font ( 3900,2634)(0,-10 ) ( 2037,225)(2037,1215 ) ( 2067.000,1095.000)(2037.000,1215.000)(2007.000,1095.000)(2037.000,1131.000)(2067.000,1095.000 ) ( 1722,225)(912,1215 ) ( 1011.207,1141.122)(912.000,1215.000)(964.770,1103.128)(965.192,1149.988)(1011.207,1141.122 ) ( 597,1575)(12,2070)(687,2340)(687,1755 ) ( 657.000,1875.000)(687.000,1755.000)(717.000,1875.000)(687.000,1839.000)(657.000,1875.000 ) ( 1992,1575)(1407,2070)(2082,2340)(2082,1755 ) ( 2052.000,1875.000)(2082.000,1755.000)(2112.000,1875.000)(2082.000,1839.000)(2052.000,1875.000 ) ( 2307,225)(3432,1215 ) ( 3361.733,1113.203)(3432.000,1215.000)(3322.096,1158.246)(3368.940,1159.507)(3361.733,1113.203 ) ( 1812,0)(0,0)[lb ] ( 462,1350)(0,0)[lb ] ( 1812,1350)(0,0)[lb ] ( 3162,1350)(0,0)[lb ] ( 822,495)(0,0)[lb ] ( 2307,720)(0,0)[lb ] ( 1407,990)(0,0)[lb ] ( 12,2475)(0,0)[lb ] ( 1632,2475)(0,0)[lb ] we define the characteristic formulas for terms @xmath236 of the process algebra @xmath234 , which will also be the characteristic formulas of the compact elements @xmath266 of @xmath0 . for each @xmath240 , we define the formula @xmath274 of hennessy - milner logic in figure [ fig : phip ] . l @xmath275 + + @xmath276 + + @xmath277}}\phi _ p\land\bigwedge _ { \beta\not=\alpha } \neg{{\langle { \beta}\rangle } } { { \it t\!t}}$ ] + + @xmath278}}\phi _ p\land \bigwedge _ { \beta\not=\alpha } \neg{{\langle { \beta}\rangle } } { { \it t\!t}}$ ] + + @xmath279 + + @xmath280 } } ( \bigvee _ { v\in \{\bot,{{\it t\!t}}\ } } \ { \phi _ { r'}\mid p+q\longrightarrow ^\alpha_v r'\})$ ] these formulas characterize their terms , for one can interchange refinement checks @xmath281 with model checks @xmath282 for all @xmath283 . [ lemma : psip ] for all @xmath283 we have @xmath284 we prove this by structural induction on @xmath240 . * we have @xmath285 iff ( there are no @xmath140-transitions out of @xmath105 ) iff @xmath286 . * we have @xmath287 for all @xmath283 , but also @xmath288 for all @xmath283 . * using induction on @xmath236 , we have @xmath289 iff ( there is a @xmath139-transition @xmath290 in @xmath0 with @xmath291 ; all @xmath140-transitions @xmath292 in @xmath0 satisfy @xmath293 ; and there are no @xmath140-transitions out of @xmath105 in @xmath0 for other events ) . this exactly captures @xmath294 . * by induction on @xmath236 , we have @xmath295 iff ( there are no @xmath140-transitions out of @xmath105 for events other than @xmath141 , and all @xmath140-transitions @xmath290 satisfy @xmath291 ) . but this captures @xmath296 . * let @xmath297 . then @xmath298 and induction express that all @xmath139-transitions out of @xmath299 to some @xmath300 can be answered by corresponding @xmath301 with @xmath302 ; whereas @xmath303 } } \bigl ( \bigvee \ { \psi _ { r'}\mid\exists v\in \{\bot,{{\it t\!t}}\}\colon p+q\longrightarrow ^\alpha_v r'\}\bigr ) $ ] states that all @xmath304 can be answered in @xmath305 by corresponding @xmath140-transitions to some @xmath300 such that @xmath306 by induction . so @xmath307 . this characterization is the key to proving that @xmath0 is fully abstract and that refinement is characterized by the semantics for hennessy - milner logic . [ cor : hjs04 ] 1 . the order on @xmath0 is the greatest refinement relation within @xmath124 . 2 . for all pointed modal transition systems @xmath31 and @xmath51 the following are equivalent : 1 . @xmath48 2 . for all @xmath213 of hennessy - milner logic , @xmath308 implies @xmath309 3 . for all @xmath213 of hennessy - milner logic , @xmath310 implies @xmath311 . 1 . that the order of @xmath0 is a refinement follows directly from the definition of @xmath124 . for the converse , we show `` @xmath312 implies that @xmath313 does not refine @xmath314 : '' first note that @xmath315 order - generates @xmath0 so @xmath312 implies @xmath316 and @xmath317 for some @xmath195 . then there is @xmath240 with @xmath318 so that , by lemma [ lemma : psip ] , for all @xmath319 : @xmath320 iff @xmath321 . thus , @xmath322 and @xmath323 imply that @xmath324 does not refine @xmath105 in @xmath124 . 2 . since @xmath325 and @xmath326 are dual with respect to negation , ( b ) and ( c ) are equivalent . the proof that ( a ) implies ( b ) is a straightforward structural induction on @xmath213 @xcite . that ( b ) implies ( a ) follows from item ( 1 ) , lemma [ lemma : psip ] , and the fact that @xmath0 is algebraic . we demonstrate that embeddings of pointed image - finite labelled transition systems are dense in @xmath327 , which we subsequently show to be the quotient space of all pointed labelled transition systems with respect to bisimulation . the denseness argument rests on the fact that embeddings of implementations are maximal elements of @xmath0 . [ lemma : aremax ] let @xmath283 be such that , for all @xmath213 of hennessy - milner logic , @xmath328 implies @xmath329 . then @xmath330 . consider such a @xmath105 and let @xmath154 in @xmath0 . since @xmath331 is a partial order and since @xmath0 is algebraic it suffices to show that @xmath332 . so let @xmath333 with @xmath334 . for @xmath335 of ( [ equ : phip ] ) , @xmath334 implies @xmath336 which implies @xmath337 by corollary [ cor : hjs04 ] as @xmath0 is fully abstract . but @xmath154 means @xmath338 as @xmath0 is fully abstract , and so @xmath339 by corollary [ cor : hjs04 ] as @xmath337 . by assumption on @xmath105 , this renders @xmath322 and so @xmath340 by ( [ equ : phip ] ) . [ prop : dense ] the set of all embeddings of pointed image - finite labelled transition systems is dense in @xmath341 . as any pointed image - finite labelled transition system @xmath342 is refinement - equivalent to @xmath343 @xcite , the embedding @xmath344 is in @xmath345 since it satisfies the assumptions of lemma [ lemma : aremax ] . let @xmath198 be non - empty , so @xmath346 for some @xmath192 and there is some @xmath347 with @xmath348 since @xmath349 is non - empty and @xmath0 is algebraic . let @xmath350 be obtained by replacing all @xmath79 in @xmath236 with @xmath239 and , for all @xmath351 , all prefixes @xmath352 with @xmath353 . then @xmath354 _ { } } } , q)$ ] refines @xmath242 _ { } } } , p)$ ] . since @xmath354_{}}},q)$ ] is a pointed _ labelled _ transition system and @xmath355 _ { } } } , r)$ ] is refinement - equivalent to @xmath356 for all @xmath357 by lemma [ lemma : match ] , we conclude @xmath358 by lemma [ lemma : aremax ] and @xmath359 is the embedding of a pointed image - finite labelled transition system . we show that @xmath191 is compact by proving , indirectly , that @xmath360 is @xmath187-closed . using results from @xcite one could show that @xmath360 is @xmath187-closed by finding a subset @xmath361 of @xmath315 that is a finitely branching tree and co - final in @xmath315 . given a candidate of such a @xmath361 , the property that is difficult to ascertain is that any two elements of @xmath361 that have an upper bound in @xmath315 are comparable . for example , consider the compact elements @xmath362 and @xmath363 , both of which have the compact element @xmath364 as an upper bound yet neither of them refines the other . faced with these difficulties , we therefore take a different route and realize @xmath360 as the set of those elements @xmath105 of @xmath0 that pass a set of judgments @xmath365 where @xmath366 are formulas of hennessy - milner logic . 1 . let @xmath367 , @xmath27 , and @xmath240 . then we define the hennessy - milner logic formula + @xmath368}}{{[{\delta _ 2}]}}\dots { { [ { \delta _ n}]}}({{\langle { \alpha}\rangle } } \phi _ p\lor \neg { { \langle { \alpha}\rangle } } \phi _ p)\ ] ] + with @xmath274 as in figure [ fig : phip ] . 2 . let @xmath369 be the set of all hennessy - milner logic formulas @xmath366 where @xmath370 , @xmath27 , and @xmath240 . 3 . for @xmath213 of hennessy - milner logic and all @xmath371 we define + @xmath372^{{\it m } } } } = \ { d\in { { \mathbb d}}\mid ( { { \mathcal { d}}},d){{\models } ^{\it { m}}}\phi \}\ , .\ ] ] 4 . let @xmath373^{{\it a}}}}$ ] . for each formula @xmath213 in @xmath369 , the test @xmath374 checks whether there is a certain @xmath140-reachable state from @xmath105 with a certain outgoing may - transition that can not be matched with a corresponding outgoing must - transition . accordingly , @xmath375 consists of those elements whose reachable states always find such a match . intuitively , those should be the elements that represent labelled transition systems . the formulas in items ( [ item : one ] ) and ( [ item : two ] ) of example [ example : two ] are in @xmath369 as @xmath376 is @xmath377 , @xmath378 , and @xmath379 . rather than proving directly that @xmath360 is @xmath187-closed , we first establish that @xmath375 is @xmath187-closed and then prove @xmath380 . whence maximal elements in @xmath0 are exactly those elements whose reachable may - transitions have matching must - transitions . as @xmath375 is the intersection of sets of the form @xmath381^{{\it a}}}}$ ] , we can show that the former is @xmath187-closed by proving that all latter sets are @xmath187-closed . we do this by structural induction on @xmath213 which requires a stronger induction hypothesis . [ lemma : denoteclosed ] for each @xmath213 of hennessy - milner logic , the sets @xmath381^{{\it a}}}}$ ] and @xmath381^{{\it c}}}}$ ] are @xmath187-clopen . in particular , @xmath375 is @xmath187-closed . we proceed with the first claim by structural induction on @xmath213 . this is evident for the clauses @xmath376 , negation , and conjunction since @xmath382^{{\it m } } } } = { { \mathbb d}}$ ] is @xmath187-clopen and clopens are closed under set complement ( @xmath383^{{\it a } } } } = { { \mathbb d}}\setminus { { [ \!\mid { \phi}\mid \!]^{{\it c}}}}$ ] and @xmath383^{{\it c } } } } = { { \mathbb d}}\setminus { { [ \!\mid { \phi}\mid \!]^{{\it a}}}}$ ] ) and finite intersections . we still require proofs for @xmath384 , where for each @xmath28 we have @xmath385^{{\it m } } } } = \{d\in { { \mathbb d}}\mid { { d } ^{\it { m}}}_\alpha \cap { { [ \!\mid { \phi}\mid \!]^{{\it m}}}}\not={\{\}}\}\ , .\ ] ] * let @xmath386 . by theorem 4.2 in @xcite , @xmath387^{{\it a}}}}$ ] is @xmath193-open for all @xmath214 of hennessy - milner logic , so @xmath388^{{\it a}}}}\in{{\sigma _ { { { \mathbb d}}}}}{\subseteq}{{\lambda _ { { { \mathbb d}}}}}$ ] and it suffices to show that @xmath388^{{\it a}}}}$ ] is @xmath187-closed , i.e. @xmath193-compact as an upper set . by induction , @xmath381^{{\it a}}}}$ ] is @xmath187-clopen ; it is also @xmath193-open so @xmath381^{{\it a } } } } = { { \uparrow \!\ ! _ { } \!{f_\phi}}}$ ] for a finite subset @xmath389 as @xmath0 is algebraic . by the definition of @xmath388^{{\it a}}}}$ ] , we have @xmath390^{{\it a}}}}$ ] iff @xmath391 iff @xmath392 ( as @xmath393 is a lower set ) . for each @xmath394 define @xmath395 by @xmath396 for all @xmath397 ; and @xmath398 . then @xmath399 is finite and @xmath400 . since @xmath401 for all @xmath402 , we get @xmath403^{{\it a}}}}$ ] as the latter set is upper . note that for each @xmath394 we have @xmath404 in @xmath0 iff @xmath405 . therefore , @xmath390^{{\it a}}}}$ ] implies @xmath406 . thus , @xmath388^{{\it a}}}}$ ] equals @xmath407 for the finite subset @xmath165 of @xmath315 . * let @xmath408 . from theorem 4.2 in @xcite we already know that @xmath388^{{\it c}}}}$ ] is @xmath193-closed and therefore @xmath187-closed . thus , it suffices to show that @xmath388^{{\it c}}}}$ ] is @xmath187-open . by induction , @xmath381^{{\it c}}}}$ ] is @xmath187-open and therefore @xmath409^{{\it c } } } } = { { [ \!\mid { \neg\phi}\mid \!]^{{\it a}}}}$ ] is @xmath187-closed ( and @xmath193-open ) , i.e. @xmath193-compact upper . since @xmath0 is algebraic , @xmath383^{{\it a } } } } = { { \uparrow \!\ ! _ { } \!{f_{\neg\phi}}}}$ ] for a finite subset @xmath410 of @xmath315 . thus , @xmath381^{{\it c } } } } = { { \mathbb d}}\setminus { { \uparrow \!\ ! _ { } \!{f_{\neg\phi}}}}$ ] . inspecting the definition of @xmath388^{{\it c}}}}$ ] , we infer @xmath390^{{\it c}}}}$ ] iff there is some @xmath411 such that @xmath412 . now let @xmath413^{{\it c}}}}$ ] . we claim that there are compact elements @xmath414 and @xmath415 with @xmath416^{{\it c}}}}$ ] , which concludes the proof since @xmath417 is @xmath187-open . choose any @xmath418 . as for @xmath419 , set @xmath420 for all @xmath397 ; and @xmath421 ; in particular , @xmath422 . note that @xmath423 in @xmath0 iff @xmath424 iff ( for some @xmath425 , @xmath426 ) . therefore , @xmath427^{{\it c}}}}$ ] . so @xmath375 is @xmath187-closed as the intersection of @xmath187-closed sets . in @xcite open sets are thought of as observable properties , so the denotations of hennessy - milner logic formulas in @xmath0 ( and in @xmath428 ) are closed under negation as observations . if we extend these denotations to the modal mu - calculus @xcite , we expect observable properties to correspond to sets in the borel algebra generated by @xmath193 . using the denseness of embeddings of image - finite labelled transition systems in @xmath428 , we can prove the inclusion @xmath429 . [ lemma : minc ] the set @xmath360 is contained in @xmath430 . let @xmath89 be the set of all embeddings @xmath344 of pointed image - finite labelled transition systems @xmath342 . then @xmath431 follows as * @xmath432 is refinement - equivalent to @xmath342 , * @xmath433 is valid over labelled transition systems for all @xmath213 of hennessy - milner logic , * @xmath434}}\phi$ ] is valid over labelled transition systems whenever @xmath213 is , and * @xmath325 is the standard semantics of hennessy - milner logic over labelled transition systems . by proposition [ prop : dense ] , @xmath89 is a dense subset of @xmath341 and so its superset @xmath435 is also dense in @xmath341 and is @xmath436-closed by the lawson condition for @xmath0 since @xmath437 is @xmath187-closed by lemma [ lemma : denoteclosed ] . but the only dense @xmath186-closed subset of @xmath438 is @xmath428 itself and so @xmath439 follows which implies @xmath440 . for a proof of the reverse inclusion @xmath441 we need to clarify the structure of elements in @xmath375 . [ lemma : aboutc ] let @xmath442 . then : 1 . all @xmath443 that are reachable from @xmath105 in the labelled transition system @xmath444 are in @xmath375 as well . 2 . for all @xmath27 we have @xmath445 . 3 . for all @xmath213 of hennessy - milner logic , @xmath446 implies @xmath329 . 1 . let @xmath136 be reachable from @xmath105 in @xmath447 and let @xmath448 be the word obtained by travelling from @xmath105 to @xmath136 on such a path . given @xmath449 , the concatenation @xmath450 is in @xmath451 and so @xmath452 . thus the path for @xmath453 above and @xmath454 ensure @xmath455 and so @xmath456 . 2 . let @xmath27 . since @xmath457 , it suffices to show @xmath458 . proof by contradiction : let @xmath459 . then @xmath460 and @xmath461 imply @xmath462 and so @xmath463 . as @xmath0 is algebraic and @xmath464 , there is some @xmath465 with @xmath466 and @xmath467 and so @xmath468 as @xmath137 is a lower set . but @xmath454 implies @xmath469 , as @xmath470 is @xmath471 , and so @xmath472 implies @xmath473 by the definition of @xmath388^{{\it m}}}}$ ] in ( [ equ : diauniv ] ) , contradicting @xmath474 . we use structural induction on @xmath213 . the cases for @xmath376 , negation , and conjunction are straightforward . let @xmath475 , so @xmath476 for some @xmath477 . by item ( 2 ) , there is some @xmath478 with @xmath479 . but then @xmath476 and @xmath479 imply @xmath480 by corollary [ cor : hjs04 ] . since @xmath481 is reachable from @xmath105 in @xmath447 it is in @xmath375 by item ( 1 ) . thus , we can apply induction on @xmath482 and get @xmath483 . since @xmath484 , this renders @xmath485 . we have now all the machinery at our disposal for stating and proving our main results in the next two theorems . [ theorem : one ] the set @xmath360 equals @xmath375 . in particular , @xmath205 is @xmath187-closed and @xmath191 is a stone space in which the set of embeddings of pointed image - finite labelled transition systems is dense . from item ( 3 ) of lemma [ lemma : aboutc ] and lemma [ lemma : aremax ] we infer @xmath486 . lemma [ lemma : minc ] then renders @xmath380 . by lemma [ lemma : denoteclosed ] , this means that @xmath360 is @xmath187-closed . by propositions [ prop : zero ] and [ prop : dense ] , it suffices to show that @xmath487 is compact . let @xmath488 for @xmath489 . by the definition of @xmath436 , each @xmath490 is of the form @xmath491 for some @xmath492 . since @xmath0 is a bifinite domain , @xmath493 is compact @xcite . since @xmath360 is @xmath187-closed it is @xmath187-compact as a @xmath187-closed subset of the compact space @xmath493 . from @xmath494 and @xmath196 we infer that @xmath495 . the @xmath187-compactness of @xmath81 therefore implies the existence of a finite set @xmath158 with @xmath496 . but then @xmath497 follows . theorem [ theorem : one ] is of interest in its own right since @xmath189 is not @xmath119-closed for bifinite domains @xmath98 in general . but we also have to demonstrate that @xmath428 is the desired quotient space of labelled transition systems modulo bisimulation . given a topological space @xmath83 let @xmath498}$ ] be the _ poset of all @xmath84-compact subsets of @xmath69 _ , ordered by reverse inclusion : @xmath499 iff @xmath500 . [ theorem : two ] 1 . the embedding @xmath151 for pointed image - finite modal transition systems given in @xcite extends to pointed modal transition systems such that labelled transition systems are embedded into @xmath81 . conversely , for any @xmath501 the pointed mixed transition system @xmath314 is refinement - equivalent to a labelled transition system . ( it does nt `` type check '' to ask whether @xmath502 is _ bisimilar _ to a labelled transition system ; but @xmath503 and @xmath504 are merely saturation artifacts of the model . ) we have the isomorphism + @xmath505}\ ] ] + of sets where @xmath506 models the @xmath141-successors of @xmath203 as the @xmath186-compact set @xmath507 , for each @xmath27 . 1 . whenever a state @xmath45 has infinitely many states @xmath508 as @xmath141-successors for @xmath18 , choose a finite subset @xmath106 of @xmath509 , retain transitions @xmath510 and their must / may status for all @xmath511 , discard all @xmath510 with @xmath512 , and create a _ may - stub _ @xmath513 ( @xmath514 ) and a may - transition @xmath515 . doing this for all events while , at the same time , unfolding @xmath31 as a tree ensures that all approximations are image - finite with limit @xmath254 such that @xmath516 is refinement - equivalent to @xmath31 . in particular , @xmath517 by lemma [ lemma : aremax ] whenever @xmath31 is a labelled transition system . 2 . let @xmath330 and @xmath27 . the set @xmath518 is in @xmath430 , which equals @xmath360 , and @xmath445 by lemma [ lemma : aboutc ] and theorem [ theorem : one ] . combining this with ( [ equ : mix ] ) , we infer @xmath519 . but since @xmath375 is closed under states reachable in @xmath444 , we may assume this representation for all elements @xmath324 reachable from @xmath105 in @xmath444 . therefore , @xmath502 is refinement - equivalent to the modal transition system with no may - transitions that replaces @xmath520 with @xmath521 for all @xmath27 and all @xmath324 reachable from @xmath105 in @xmath444 . the isomorphism follows from the equation for @xmath0 and lemmas 34.5 and 25 of @xcite ; the latter is stated for @xmath522-domains @xmath98 , which are bifinite , but its proof only requires that @xmath189 is @xmath119-closed . an immediate consequence of these two main theorems is that sets of implementations of modal transition systems are compact in the quotient space modulo bisimulation . [ cor : cl ] for each pointed modal transition system @xmath523 , its set of implementations is compact in the quotient space of labelled transition systems modulo bisimulation . the set of implementations of @xmath523 in @xmath428 is @xmath524 which is @xmath187-closed as the intersection of two @xmath187-closed sets and so it is @xmath190-compact . we now discuss some of the consequences of the compactness of @xmath190 : a compactness theorem for hennessy - milner logic on compact sets of implementations , an abstract interpretation of compact sets of implementations as scott - closed sets of modal transition systems , and a robust consistency measure for modal transition systems . compactness of @xmath525 , stated in terms of hennessy - milner logic , is familiar from first - order logic but here secured without appeal to a complete proof system . such semantic techniques for proving compactness are not new , we mention model - theoretic techniques based on ultra - filters . a compactness theorem for hennessy - milner logic alone already follows from its standard encoding in first - order logic . however , we prove a compactness result that goes beyond hennessy - milner logic as it applies to compact sets of labelled transition systems , in particular to the set of common implementations of finitely - many pointed modal transition systems . for a single such system , @xmath526 , we then regain the familiar compactness theorem for hennessy - milner logic . our result is stronger than this familiar theorem as the sets of implementations of pointed modal transition systems are not expressible through hennessy - milner logic . in theorem [ theorem : mc](2 ) below we see that these sets are expressible in hennessy - milner logic extended with greatest fixed points for finite - state modal transition systems . [ coro : one ] 1 . let @xmath527 be a set of formulas of hennessy - milner logic and @xmath165 a @xmath190-compact set such that for all finite subsets @xmath528 of @xmath527 there is some @xmath529 that satisfies @xmath530 . then there is some @xmath531 that satisfies all formulas of @xmath527 . 2 . in particular , let @xmath527 be a set of formulas of hennessy - milner logic and @xmath532 a finite set of pointed modal transition systems such that for all finite subsets @xmath528 of @xmath527 there is a pointed labelled transition system that refines all @xmath533 and satisfies @xmath530 . then there is a pointed labelled transition system that refines all @xmath533 and satisfies all formulas of @xmath527 . by corollary [ cor : cl ] it suffices to prove item ( 1 ) . by duality of consistency ( i.e. satisfiability ) and validity , it suffices to prove the dual statement of item ( 1 ) : assume that every @xmath534 satisfies as least one @xmath535 and show that there is a finite set @xmath536 such that @xmath537 is valid over the set @xmath165 . by this assumption , we have @xmath538 where @xmath539^{{\it a}}}}\cap \max ( { { \mathbb d}})\mid \phi\in \gamma\}$ ] is a subset of @xmath190 as all @xmath381^{{\it a}}}}$ ] are in @xmath193 by theorem 4.2 in @xcite . as @xmath165 is @xmath190-compact , there is a finite @xmath158 with @xmath540 , i.e. @xmath541^{{\it a } } } } = { { [ \!\mid { \bigvee \delta}\mid \!]^{{\it a}}}}$ ] for a finite set @xmath536 . thus all @xmath534 satisfy @xmath542 . figure [ fig : cr ] depicts schematically the set of common implementations of two pointed modal transition systems @xmath314 and @xmath543 , the intersection of the implementations of @xmath105 and @xmath324 . this is a compact subset of @xmath428 and so we get a compactness theorem for hennessy - milner logic on that set . cousot & cousot s abstract interpretation framework @xcite approximates concrete objects and their transformations by abstract objects and transformations such that reasoning on abstract objects is sound for their concretizations . in a simple setting , one has given a set @xmath165 of concrete objects ( e.g. computer programs ) and a partial order @xmath544 of abstract objects , a monotone abstraction function @xmath545 , and a monotone concretization function @xmath546 . the value @xmath547 should represent the best approximation of @xmath548 within the partial order @xmath544 and @xmath549 should represent the set of those concrete objects that are abstracted by @xmath550 . one can encode these intuitions by making @xmath141 and @xmath551 a galois connection @xcite , a notion we define below . let @xmath165 be the set of natural numbers and @xmath552 where @xmath553 is the top element and @xmath349 and @xmath554 are incomparable . define @xmath555 to be @xmath349 if all elements of @xmath69 are odd ; @xmath554 if all elements of @xmath69 are even ; and @xmath553 otherwise . then @xmath556 and @xmath557 etc . define @xmath558 , @xmath559 , and @xmath560 . then @xmath561 says that @xmath349 is the least element that soundly represents the set @xmath562 . the equation @xmath563 shows that the abstract value of @xmath562 has a larger set of concrete objects . we want to apply this framework in our setting . from the compactness of @xmath190 corollary [ cor : cl ] infers that the set @xmath564 is @xmath190-compact for all pointed modal transition systems @xmath523 . said @xmath190-compact set comprises all the implementations of @xmath31 . conversely , a @xmath190-compact set @xmath165 of labelled transition system can be approximated by any pointed modal transition system @xmath523 satisfying @xmath565 . ideally , one wants an _ optimal _ such @xmath523 , one for which the difference @xmath566 is minimal . of course , this optimality is ensured for any @xmath165 of the form @xmath564 . the next example shows that there is no optimal @xmath523 in general . consider two pointed modal transition trees @xmath567 and @xmath568 that have a common refinement but do not refine each other . in general , there will be more than one minimal upper bound of the set @xmath569 in @xmath0 so there can not be a @xmath283 such that @xmath570 equals the @xmath190-compact set @xmath571 . the fact that modal transition systems can not be such optimal abstractions of @xmath190-compact sets seems to be related to the incompleteness of modal transition systems for abstraction - based model checking @xcite since @xmath0 is not bounded complete . but there is a galois connection between @xmath190-compact subsets of @xmath428 and @xmath193-closed subsets of @xmath0 . for a @xmath190-compact set @xmath165 its set of concretizations is the scott - closed set of all @xmath523 for which @xmath565 . conversely , a scott - closed subset @xmath127 of pointed modal transition systems is abstracted as the set of those pointed labelled transition systems that implement all elements of @xmath127 . 1 . let @xmath572 } = \{l\mid l\ \mbox{${{\sigma _ { { { \mathbb d}}}}}$-closed}\}$ ] be the set of @xmath193-closed subsets of @xmath0 , ordered by set inclusion : @xmath127 is less than or equal to @xmath573 iff @xmath574 . 2 . let @xmath575 and @xmath576 be complete lattices . a _ galois connection _ @xcite is a pair of monotone maps @xmath577 and @xmath578 such that for all @xmath579 we have @xmath580 and for all @xmath581 we have @xmath582 . in that case , @xmath141 is the _ upper adjoint _ of @xmath551 . [ theorem : gc ] the maps @xmath583}\to { { \mathcal l } [ { { { \mathbb d}}}]}$ ] and @xmath584}\to { { \mathcal c}[{{{\mathbb x}}},{\tau_{{\mathbb x}}}]}$ ] defined by @xmath585 form a galois connection , where @xmath141 is the upper adjoint of @xmath551 . * the map @xmath551 is well defined . first @xmath154 implies @xmath586 and so @xmath587 is a lower set . second let @xmath588 be directed in @xmath589 . then @xmath590 and the latter equals @xmath591 , so @xmath587 is @xmath193-closed . * the map @xmath141 is well defined . for if @xmath127 is empty , then @xmath592 is @xmath190-compact ; and if @xmath127 is non - empty , @xmath593 is the intersection of @xmath187-closed elements and so @xmath187-closed whence @xmath190-compact . * the map @xmath551 is monotone . let @xmath499 , i.e.@xmath500 . then @xmath594 means @xmath595 and so @xmath596 follows . therefore @xmath597 and so @xmath598 . * the map @xmath141 is monotone . let @xmath574 . then @xmath599 and so @xmath600 . * to see @xmath601}}$ ] let @xmath602}$ ] . then @xmath603 clearly contains @xmath127 . * to see @xmath604}}$ ] let @xmath605}$ ] . then @xmath606 obviously contains @xmath165 . theorem [ theorem : gc ] remains to be valid if we reverse the orders on the domains @xmath607}$ ] and @xmath572}$ ] and swap the names @xmath141 and @xmath608 throughout the theorem and its proof . in that case , a @xmath609-compact set @xmath165 is abstracted by a set @xmath127 of pointed modal transition systems and any such @xmath127 has a set of pointed labelled transition systems as concretizations . this view is perhaps more natural . we explicitly state the metrics @xmath610 for pointed modal transition systems and @xmath611 for pointed labelled transition systems . the latter is then used to define a consistency measure on modal transition systems as an alternative to the metric @xmath610 . fix an enumeration @xmath612 of @xmath234 and set @xmath613 then the topology determined by @xmath614 and @xmath615 is @xmath187 and @xmath186 , respectively . for practical purposes we wish to enumerate @xmath240 in increasing modal depth of @xmath274 in ( [ equ : phip ] ) , corresponding to the iterative unfolding of the functional for bisimulation @xcite . in that case , @xmath611 is essentially the metric in @xcite . these metrics are standard and well understood but result in consistency measures if lifted to compact sets of implementations . we define the _ consistency measure _ @xmath616\colon { { \mathbb d}}\times{{\mathbb d}}\to{{\mathbb i}}$ ] by @xmath617 and use this as an alternative to the metric @xmath610 for comparing the pointed modal transition systems @xmath314 and @xmath543 . note that @xmath618 and @xmath619 are optimistic and pessimistic measures ( respectively ) from the point of view of an implementor . figure [ fig : cr ] shows a scenario where two pointed modal transition systems @xmath314 and @xmath543 have a common refinement , and so @xmath620 . # 1#2#3#4#5 @font ( 6222,3783)(0,-10 ) ( 3240,3240 ) ( 3240,3240 ) ( 1620,3240 ) ( 1620,3240 ) ( 5040,3240 ) ( 5040,3240 ) ( 810,3240)(6210,3240)(3510,90)(810,3240 ) ( 1575,3240)(2610,2115)(3780,3240 ) ( 2565,3240)(4500,2475)(5040,3240 ) ( 1575,3465)(3735,3465 ) ( 3735,3465)(3735,3375 ) ( 1575,3465)(1575,3375 ) ( 4500,2475)(4501,2477)(4503,2481 ) ( 4507,2489)(4513,2501)(4521,2515 ) ( 4530,2533)(4540,2553)(4551,2575 ) ( 4562,2597)(4572,2621)(4583,2644 ) ( 4594,2668)(4603,2692)(4613,2717 ) ( 4621,2742)(4629,2766)(4635,2790 ) ( 4638,2807)(4641,2822)(4642,2835 ) ( 4642,2844)(4642,2851)(4641,2857 ) ( 4640,2860)(4639,2861)(4638,2862 ) ( 4637,2862)(4635,2861)(4633,2860 ) ( 4631,2860)(4629,2859)(4627,2859 ) ( 4624,2860)(4620,2861)(4616,2864 ) ( 4611,2867)(4605,2871)(4598,2876 ) ( 4590,2880)(4571,2887)(4555,2891 ) ( 4544,2892)(4536,2892)(4530,2891 ) ( 4524,2890)(4515,2888)(4501,2886 ) ( 4481,2884)(4455,2880)(4429,2875 ) ( 4409,2869)(4396,2864)(4389,2859 ) ( 4384,2854)(4379,2849)(4372,2844 ) ( 4360,2840)(4342,2837)(4320,2835 ) ( 4305,2836)(4292,2837)(4282,2839 ) ( 4275,2840)(4270,2841)(4266,2842 ) ( 4264,2843)(4261,2843)(4259,2845 ) ( 4255,2848)(4250,2852)(4244,2859 ) ( 4237,2868)(4230,2880)(4223,2899 ) ( 4221,2916)(4223,2929)(4228,2939 ) ( 4234,2948)(4240,2956)(4244,2966 ) ( 4245,2979)(4241,2996)(4230,3015 ) ( 4218,3029)(4206,3039)(4196,3045 ) ( 4189,3049)(4185,3051)(4181,3053 ) ( 4177,3053)(4171,3054)(4161,3055 ) ( 4145,3057)(4123,3059)(4095,3060 ) ( 4069,3059)(4046,3057)(4026,3054 ) ( 4011,3051)(4001,3048)(3993,3045 ) ( 3986,3041)(3980,3038)(3972,3034 ) ( 3960,3031)(3945,3027)(3924,3023 ) ( 3899,3019)(3870,3015)(3838,3012 ) ( 3813,3009)(3796,3006)(3786,3003 ) ( 3782,3000)(3780,2996)(3778,2993 ) ( 3774,2992)(3764,2992)(3747,2995 ) ( 3722,3003)(3690,3015)(3665,3027 ) ( 3642,3041)(3622,3054)(3607,3066 ) ( 3595,3077)(3586,3087)(3579,3096 ) ( 3574,3105)(3568,3114)(3562,3123 ) ( 3553,3133)(3542,3144)(3527,3156 ) ( 3509,3169)(3488,3183)(3465,3195 ) ( 3441,3206)(3419,3213)(3402,3217 ) ( 3387,3218)(3375,3218)(3364,3215 ) ( 3355,3212)(3348,3208)(3341,3204 ) ( 3336,3200)(3333,3198)(3331,3196)(3330,3195 ) ( 2610,2115)(2608,2117)(2604,2121 ) ( 2598,2127)(2589,2137)(2578,2149 ) ( 2566,2164)(2554,2180)(2542,2197 ) ( 2530,2215)(2521,2233)(2514,2253 ) ( 2509,2273)(2508,2295)(2512,2317 ) ( 2520,2340)(2534,2361)(2550,2380 ) ( 2567,2393)(2584,2403)(2600,2410 ) ( 2616,2413)(2630,2414)(2644,2415 ) ( 2658,2415)(2673,2417)(2689,2419 ) ( 2706,2424)(2726,2433)(2747,2444 ) ( 2769,2459)(2790,2475)(2806,2490 ) ( 2819,2503)(2829,2514)(2837,2523 ) ( 2843,2529)(2847,2533)(2850,2536 ) ( 2852,2538)(2854,2539)(2855,2540 ) ( 2855,2542)(2856,2544)(2856,2549 ) ( 2855,2556)(2853,2566)(2849,2578 ) ( 2843,2593)(2835,2610)(2822,2630 ) ( 2807,2648)(2791,2664)(2775,2677 ) ( 2760,2687)(2746,2695)(2732,2702 ) ( 2719,2707)(2705,2713)(2692,2719 ) ( 2678,2727)(2663,2736)(2649,2747 ) ( 2634,2761)(2621,2775)(2610,2790 ) ( 2603,2809)(2608,2820)(2622,2823 ) ( 2641,2821)(2663,2816)(2684,2811 ) ( 2702,2809)(2714,2811)(2714,2820 ) ( 2700,2835)(2685,2845)(2667,2856 ) ( 2648,2865)(2627,2874)(2607,2881 ) ( 2586,2888)(2566,2893)(2547,2898 ) ( 2527,2903)(2508,2907)(2489,2912 ) ( 2470,2917)(2452,2924)(2434,2931 ) ( 2417,2940)(2402,2949)(2391,2960 ) ( 2385,2970)(2386,2980)(2394,2989 ) ( 2407,2994)(2423,2998)(2439,2999 ) ( 2457,2999)(2475,2998)(2494,2996 ) ( 2513,2995)(2532,2993)(2552,2993 ) ( 2573,2994)(2594,2996)(2616,3001 ) ( 2636,3007)(2655,3015)(2669,3024 ) ( 2678,3032)(2681,3038)(2678,3043 ) ( 2670,3047)(2660,3050)(2647,3051 ) ( 2633,3052)(2619,3054)(2607,3056 ) ( 2598,3059)(2594,3064)(2596,3071 ) ( 2606,3080)(2626,3092)(2655,3105 ) ( 2681,3115)(2711,3126)(2742,3136 ) ( 2774,3145)(2807,3154)(2840,3163 ) ( 2874,3171)(2908,3179)(2942,3187 ) ( 2976,3195)(3009,3202)(3042,3209 ) ( 3073,3215)(3102,3221)(3127,3227 ) ( 3149,3231)(3167,3234)(3179,3237 ) ( 3188,3239)(3193,3240)(3195,3240 ) ( 855,1170)(0,0)[lb ] ( 3735,0)(0,0)[lb ] ( 4365,2250)(0,0)[lb ] ( 0,810)(0,0)[lb ] ( 2520,1890)(0,0)[lb ] ( 3285,3060)(0,0)[lb ] ( 1350,3015)(0,0)[lb ] ( 1710,3600)(0,0)[lb ] ( 4635,3060)(0,0)[lb ] ( 4230,585)(0,0)[lb ] since @xmath621 is @xmath186-compact for all @xmath622 by corollary [ cor : cl ] , @xmath623 and @xmath624 are the metric analogue of symmetric @xmath625 and @xmath626 lifts of relations from elements to subsets , here of @xmath627 to @xmath186-compact subsets , respectively . the standard metric @xmath628 between compact subsets @xmath570 and @xmath629 , the hausdorff distance , is the symmetric @xmath630-lift of @xmath611 to @xmath186-compact subsets and so @xmath631 such consistency measures are of particular interest if @xmath105 and @xmath324 represent different viewpoints @xcite of the same system such that the degree of consistency between these descriptions needs to be explored . we prove that @xmath618 is a robust measure in that its kernel consists of those pairs of pointed modal transition systems that have a common refinement . [ theorem : mc ] 1 . for all @xmath153 , we have @xmath620 iff @xmath314 and @xmath313 have a common refinement . 2 . deciding whether two finite - state modal transition systems have a common refinement is reducible to checking the satisfiability of a modal mu - calculus formula with greatest fixed points only . 1 . we use theorems [ theorem : one ] and [ theorem : two ] repeatedly . if @xmath314 and @xmath313 have a common refinement , there is some @xmath632 and so @xmath633 as @xmath634 . conversely , let @xmath620 . then for each @xmath635 there are @xmath636 and @xmath637 with @xmath638 . since @xmath341 is compact , there is a convergent subsequence @xmath639 of @xmath640 with limit @xmath641 and so @xmath642 as the latter is @xmath186-closed . since @xmath643 for each @xmath644 , this implies @xmath645 and so @xmath641 is in all @xmath186-closed sets that contain @xmath646 . therefore , @xmath641 is in @xmath629 and so @xmath647 is a common refinement of @xmath314 and @xmath313 . if @xmath31 and @xmath51 are finite - state , we show that there are formulas @xmath648 and @xmath649 of the modal mu - calculus with greatest fixed points only such that the modal mu - calculus formula @xmath650 is satisfiable over labelled transition systems iff @xmath31 and @xmath51 have a common refinement . larsen & thomsen implicitly define these formulas in the system of recursive equations ( 3 ) of @xcite where , for each state @xmath45 in @xmath651 , + @xmath652 } } ( \bigvee _ { ( s,\alpha , s')\in{{r } ^{\it { c } } } } x_{(m , s')}))\ ] ] + as a greatest fixed point . if @xmath45 has finitely many reachable states in @xmath5 , then @xmath653 is expressible in the modal mu - calculus , using a `` calling context '' on the set of states @xmath44 that are @xmath18-reachable from @xmath45 and static scoping of the greatest fixed - point operators @xmath654 . now for all pointed labelled transition systems @xmath342 we have @xmath655 iff @xmath656 where we can use the proof of ( 3 ) in @xcite which works in our setting as conjunctions and disjunctions need not be finite . let @xmath5 be the modal transition system from figure [ fig : ref ] . we write @xmath657 as a formula of the modal mu - calculus with greatest fixed points only . let @xmath658 } } z_{\rm dr } \land { { [ { \rm talks } ] } } x^{\rm dr}_{(m,{\rm ta})}\land { { [ { \rm orders } ] } } x^{\rm dr}_{(m,{\rm wa})}\\ x^{\rm dr}_{(m,{\rm ta } ) } & = & \nu z_{\rm ta}. { { [ { \rm drinks } ] } } z_{\rm dr } \land { { [ { \rm orders } ] } } x^{\rm dr\ , ta}_{(m,{\rm wa})}\nonumber\\ x^{\rm dr}_{(m,{\rm wa } ) } & = & \nu z_{\rm wa}. { { \langle { \rm newpint}\rangle } } z_{\rm dr } \land { { \langle { \rm newpint}\rangle } } x^{\rm dr}_{(m,{\rm ta})}\land { { [ { \rm newpint}]}}(z_{\rm dr}\lor x^{\rm dr\ , wa}_{(m,{\rm ta})})\nonumber\\ x^{\rm dr\ , wa}_{(m,{\rm ta } ) } & = & \nu z_{\rm ta}. { { [ { \rm drinks } ] } } z_{\rm dr } \land { { [ { \rm orders } ] } } z_{\rm wa}\nonumber\\ x^{\rm dr\ , ta}_{(m,{\rm wa } ) } & = & \nu z_{\rm wa}. { { \langle { \rm newpint}\rangle } } z_{\rm dr } \land { { \langle { \rm newpint}\rangle } } z_{\rm ta}\land { { [ { \rm newpint}]}}(z_{\rm dr}\lor z_{\rm ta})\nonumber\end{aligned}\ ] ] where the superscripts in @xmath653 record the `` calling context '' of the recursions . so @xmath623 measures the _ degree of inconsistency _ of @xmath502 and @xmath313 , a lower bound on the difference between their implementations , @xmath624 is an upper bound on such a difference , and none of them is a metric : from item ( [ item : fourd ] ) of definition [ def : ultra ] , @xmath618 satisfies only ( b ) and @xmath619 satisfies only ( b ) and ( c ) . the reducibility of common refinement checks to satisfiability checks in the modal mu - calculus yields exptime as a weak upper bound on its complexity . since the formulas are defined in terms of greatest fixed points only , one can indeed show a stronger result : the decision problem of common refinements is in ptime @xcite . our results also apply to 3-valued model checking frameworks in which system observables are state propositions or a combination of state propositions and events . this is so since godefroid & jagadeesan s translation between modal transition systems ( events only ) , partial kripke structures @xcite ( state propositions only ) , and kripke modal transition systems @xcite ( events and state propositions ) and their translations of the respective temporal logic formulas is shown to preserve and reflect refinement and the meaning of model checks @xcite . bakker & zucker use domain equations and metric completions for a metric and denotational treatment of concurrency in @xcite . lawson proposes the notion of a maximal - point space to represent classical topological spaces as maximal points of a domain in the topology induced by the domain s lawson- and scott - topology @xcite . abramsky @xcite provides a fully abstract domain of synchronization trees for _ partial _ bisimulation between labelled transition systems that have a divergence predicate . the domain equation of loc . cit . uses a sum construction on the convex powerdomain . maximal points are not part of that paper s agenda and are therefore not discussed therein . labelled transition systems with a divergence predicate and partial bisimulation are recognized as certain modal transition systems and their refinement in @xcite . mislove et al . present a fully abstract domain model , which combines the probabilistic power domain with a convex variant of the plotkin powerdomain , for finite - state processes with non - deterministic and probabilistic choice @xcite . alessi et al . @xcite introduce a category of @xmath659-domains with a compositional maximal - points space functor to stone spaces . they show that all bifinite domains @xmath98 for which @xmath189 is a stone space are scott - continuous retracts of @xmath659-domains . in particular , @xmath0 is such a retract by theorem [ theorem : one ] . we suspect that @xmath0 is not an @xmath659-domain since @xmath660}$ ] is not an @xmath522-domain for the @xmath522-domain @xmath661 @xcite , although @xmath660}$ ] is the second iteration of the domain equation ( [ equ : recursion ] ) for @xmath0 when @xmath662 . the paper @xcite presents the domain @xmath0 and its modal transition system @xmath124 , both denoted as @xmath124 in loc.cit . , and proves full abstraction and a characterization of @xmath0 s compact elements in terms of formulas of hennessy - milner logic . in @xcite it is shown that the co - inductive refinement of modal transition systems has an extensional description : a pointed modal transition system @xmath31 refines a pointed modal transition system @xmath51 if , and only if , the set of implementations of @xmath31 is a subset of the implementations of @xmath51 . dams & namjoshi @xcite show that finite - state modal transition systems are incomplete as abstractions of infinite - state modal transition systems for modal mu - calculus checking . they propose focused transition systems as a generalization of modal transition systems , show completeness for this class of models , and define a game semantics for refinement of focused transition systems and a game semantics for model checks of alternating tree automata on focused transition systems . it is straightforward to write down a domain equation for focused transition systems but a programme of maximal - points spaces wo nt directly render pointed kripke structures since , as noted in @xcite , focused transition systems can have maximal refinements that have inconsistent constraints on propositions at states . in @xcite consistency , satisfiability , and validity problems are studied for collectively model checking a set of views endowed with labelled transitions , hybrid constraints on states , and atomic propositions . a ptime algorithm for deciding whether a set of views has a common refinement ( consistency ) it is proved that deciding whether a common refinement satisfies a formula of the hybrid mu - calculus @xcite ( satisfiability ) , and its dual ( validity ) , are exptime - complete . two generically generated `` summary '' views are defined that constitute informative and consistent common refinements and abstractions of a set of views ( respectively ) . di pierro et al . @xcite develop a quantitative notion of process equivalence as the basis for an approximative version of non - interference and precise quantifications of information leakage . they present two semantics - based analyzes for approximative non - interference where one soundly abstracts the other . desharnais et al . @xcite show that each continuous - state labelled markov process has a sequence of finite acyclic labelled markov processes as abstractions which is precise for a probabilistic modal logic ; an equivalence between the category of markov processes and simulation morphisms and a recursively defined domain , viewed as a category , is given . desharnais et al . @xcite define a pseudo metric between labelled concurrent markov chains where zero distance means weak bisimilarity . the metric is characterized in a real - valued modal logic and shown to allow for compositional quantitative reasoning . we presented the fully abstract and universal domain model @xmath0 for pointed modal transition systems and refinement of @xcite . using techniques from concurrency theory and topology , we demonstrated that @xmath0 is the right fully abstract and universal model for labelled transition systems and bisimulation since the quotient space of all pointed labelled transition systems with respect to bisimulation , @xmath191 , is obtained as the maximal - points space of @xmath0 . we furthermore revealed the fine - structure of @xmath428 , notably we proved that its topology @xmath186 inherited from the scott- and lawson - topology of @xmath0 is compact , zero - dimensional , and hausdorff ( a stone space ) . in particular , @xmath186 is determined by a computationally meaningful , complete ultra - metric @xmath615 for which image - finite labelled transition systems approximate labelled transition systems to any degree of precision . modulo refinement , @xmath663 is image - finite for all @xmath195 , so this denseness also applies to modal transition systems for the lawson - topology and its metric @xmath610 . thus our results unify denotational , operational , and metric semantics of labelled and modal transition systems . we finally derived consequences of this compact representation : a compactness theorem for hennessy - milner logic on compact sets of implementations , an abstract interpretation of compact sets of implementations as scott - closed sets of modal transition systems , and a robust consistency measure for modal transition systems . radha jagadeesan suggested working with the mixed powerdomain in @xcite . glenn bruns , alessandra di pierro , patrice godefroid , dimitar guelev , chris hankin , radha jagadeesan , achim jung , ralph kopperman , david schmidt , and herbert wiklicky are thanked for helpful comments and discussions . this paper is an extended journal version of @xcite and reflects the thorough and thoughtful comments made by the anonymous referees of the lics 2004 conference and the journal logical methods in computer science . f. alessi , p. baldan , and f. honsell . partializing stone spaces using sfp domains . in m. bidoit and m. dauchet , editors , _ tapsoft97 conference proceedings _ , volume 1214 of _ lecture notes in computer science _ , pages 478489 , lille , france , 14 - 18 april 1997 . springer verlag . g. bruns and p. model checking partial state spaces with @xmath664-valued temporal logics . in _ proc . of the 11th international conference on computer aided verification _ , volume 1633 of _ lecture notes in computer science _ , pages 274287 . springer verlag , july 1999 . g. bruns and p. godefroid . generalized model checking : reasoning about partial state spaces . in _ proc . of the 11th international conference on concurrency theory _ , volume 1877 of _ lecture notes in computer science _ , pages 168182 . springer verlag , august 2000 . p. cousot and r. cousot . abstract interpretation : a unified lattice model for static analysis of programs . in _ proc . of the 4th acm symposium on principles of programming languages _ , pages 238252 , los angeles , california . acm press , 1977 . d. dams and k. namjoshi . the existence of finite abstractions for branching time model checking . in _ proc . of the nineteenth annual ieee symposium on logic in computer science _ , pages 335344 , 13 - 17 july , turku , finland . ieee computer society press , 2004 . j. desharnais , v. gupta , r. jagadeesan , and p. panangaden . approximating labeled markov processes . in _ proc . of the 15th annual ieee symposium on logic in computer science _ , pages 95106 , santa barbara , california , 26 - 29 june 2000 . ieee computer society press . j. desharnais , v. gupta , r. jagadeesan , and p. panangaden . the metric analogue of weak bisimulation for probabilistic processes . in _ proc . of the 17th annual ieee symposium on logic in computer science _ , pages 413422 , copenhagen , denmark , july 2002 . ieee computer society . godefroid , m. huth , and r. jagadeesan . abstraction - based model checking using modal transition systems . in _ proc . of the 12th international conference on theory and practice of concurrency _ , volume 2154 of _ lecture notes in computer science _ , pages 426440 . springer verlag , august 2001 . p. godefroid and r. jagadeesan . automatic abstraction using generalized model checking . in e. brinksma and k. g. larsen , editors , _ proc . of the 14th international conference on computer aided verification _ , volume 2404 of _ lecture notes in computer science _ , pages 137150 , copenhagen , denmark , july 2002 . springer verlag . p. godefroid and r. jagadeesan . on the expressiveness of 3-valued models . in l. d. zuck , p. c. attie , a. cortesi , and s. mukhopadhyay , editors , _ proc . of the 4th international conference on verification , model checking and abstract interpretation _ , volume 2575 of _ lncs _ , pages 206222 , new york , january 2003 . springer verlag . m. huth , r. jagadeesan , and d. a. schmidt . modal transition systems : a foundation for three - valued program analysis . in d. sands , editor , _ proc . of esop2001 _ , pages 155169 . springer verlag , april 2001 . m. huth . beyond image - finiteness : labelled transition systems as a stone space . in : _ proc . of the nineteenth annual ieee symposium on logic in computer science _ , pages 222231 , 13 - 17 july , turku , finland , ieee computer society press , 2004 . k. g. larsen . modal specifications . in j. sifakis , editor , _ automatic verification methods for finite state systems _ , number 407 in lecture notes in computer science , pages 232246 . springer verlag , june 1214 1989 . international workshop , grenoble , france . u. sattler and m. vardi . the hybrid @xmath667-calculus . in r. gor ' e , a. leitsch , and t. nipkov _ proc . of the first international joint conference on automated reasoning_. lecture notes in computer science 2083 , pages 7691 , siena , italy , 18 - 23 june , springer verlag , 2001 . i. sommerville , p. sawyer , and s. viller . viewpoints for requirements elicitation : a practical approach . in _ proc . of the 1998 international conference on requirements engineering _ , colorado springs , colorado , april 6 - 10 1998 . ieee computer society press .

the concept of nonlinear self - localization is of importance for many physical phenomena , and has appeared in a number of different contexts since the pioneering work by landau @xcite on the polaron problem in the 1930s . in recent years , much attention has been devoted to studies of spatially localized and time - periodic vibrational modes in anharmonic lattices ( see e.g. @xcite , @xcite for recent reviews ) . the general existence of such modes , which have been termed _ discrete breathers _ , or _ intrinsic localized modes _ , as robust solutions to nonlinear ( and in general non - integrable ) lattice - equations was suggested in 1988 by takeno _ et al _ @xcite . later , their existence was rigorously proven under rather general conditions by mackay and aubry @xcite by considering the limit of uncoupled oscillators ( the so called _ anticontinuous _ or _ anti - integrable _ limit ) . by means of the implicit function theorem , they showed that the trivial solution of a single - site localized vibration at the uncoupled limit could be continued into a localized breather solution for non - zero coupling between the oscillators , provided that the individual oscillators are anharmonic , and that no multiples of the breather frequency resonate with the bands of linear excitations ( phonons ) . as was demonstrated first in @xcite , the ideas of the rigorous proof can be turned into an efficient numerical scheme to calculate breather solutions to any desired accuracy . since the discrete breathers appear under very general conditions in anharmonic lattices and provide efficient means of energy localization , they have been proposed as candidates to explain experimentally observed localization of energy in many different physical areas , e.g. dna dynamics @xcite . although , from a fundamental and mathematical viewpoint , the existence theorems for discrete breathers provide an important cornerstone for understanding the dynamics of anharmonic lattices , it is probably of even larger physical importance to understand the behaviour of a system close to an exact breather solution . by linearizing the lattice - equations around the exact solution , one can obtain an approximate description of the dynamics of weakly perturbed breathers , and in particular the linear stability properties determining whether small perturbations will grow exponentially or not . it was shown in @xcite,@xcite that the simplest , single - site , breathers are generally linearly stable close to the uncoupled limit , and numerical investigations using standard floquet analysis ( see e.g. @xcite ) have shown that linearly stable breathers typically exist also for rather large values of the inter - site coupling . however , when considering time - scales large compared to the breather period , the mere linear stability of a breather does no longer guarantee the eternal existence of the breather in the presence of small perturbations , and there are still many questions remaining concerning the different mechanisms by which breathers may grow or decay , or possibly finally be destroyed . if the breathers have a finite life - time , the determination of this life - time is of large importance for understanding the role of breathers in real systems . it is the purpose of this paper to investigate in more detail some mechanisms for breather growth and decay in a simple model system , the discrete nonlinear schrdinger ( dnls ) equation . the dnls equation is generic in the sense that it describes slowly ( in time ) varying modulational waves in discrete systems in a rotating - wave approximation ( see e.g. @xcite ) ; however , due to its extra symmetry properties ( see sec . [ sec : perturbation ] ) it exhibits some nongeneric features among discrete systems , e.g. exact quasiperiodic breathers @xcite . the single - site breathers of the dnls equation are stationary states which are linearly stable for all inter - site coupling , and which reduce to the nls soliton in the continuum limit ( see e.g. @xcite ) . an important application appears in nonlinear optics , where the single - site dnls - breather describes a discrete spatial soliton in an array of weakly coupled waveguides @xcite ; recent experimental observations @xcite confirm the successful use of the dnls - model in this context . some recent numerical investigations @xcite have shown that dnls - breathers can be spontaneously created from noisy backgrounds , in a similar manner as was previously observed for klein - gordon @xcite and fpu @xcite lattices . typically , this spontaneous energy localization was observed to occur in two steps . in the first step , a large number of small breathers are created as a result of the modulational instability @xcite of travelling plane waves occurring for certain wave number regimes . the second step proceeds by inelastic collisions between the breathers , in which systematically the big breathers grow at expense of the smaller ones . thus , the outcome will be a small number of large breathers , together with some remaining background of small - amplitude ( phonon ) oscillations . however , it can generally not be concluded from the numerical simulations that this is the true final state of the system , and actually long - time simulations for fpu - chains @xcite revealed also a third step , in which the interaction with the phonon oscillations leads to the final destruction of the breather and equipartition of energy . thus , to elucidate the nature of the final states for typical initial conditions in anharmonic chains , it is necessary to obtain a better understanding of the mechanisms for interactions between breathers and small - amplitude perturbations . in this paper , we take the following approach . as initial state , we consider an exact single - site breather solution , and add a small perturbation corresponding to a time - periodic eigensolution to the equations of motion linearized around the breather . these solutions , which can be either localized or extended in space , constitute a complete set in which an arbitrary initial perturbation can be expanded . the localized solutions correspond to internal modes of the breather @xcite , while the excitation of an extended solution corresponds to a standing - wave ( i.e. , non - propagating ) phonon interacting with the breather . in sec . [ sec : perturbation ] we describe the model and outline the perturbational approach which forms the analytical backbone for the interpretation of the numerical results presented in secs . [ sec : internal ] and [ sec : phonons ] . [ sec : internal ] discusses the long - time consequences of the interaction between the breather and its internal modes , while sec . [ sec : phonons ] concerns the interaction between the breather and small - amplitude standing - wave phonons of different wave vectors . we will find that in both cases , a simple argument based on the conservation laws can be used to obtain a sufficient condition for breather growth . in sec . [ sec : instabilities ] we discuss another type of mechanism for breather growth and destruction which becomes appreciable when the amplitude of the standing wave is non - negligible ( and consequently the perturbational approach can be expected to fail ) , and which has its origin in the recently discovered oscillatory instabilities of the nonlinear standing - wave phonon themselves @xcite . finally , we make some concluding remarks in sec . [ sec : concluding ] . concerning the numerical simulations of the dynamics presented in this paper , unless otherwise stated they always apply for a system of infinite size ( finite size systems are considered only in sec . [ sec : instabilities ] ) . the simulations have been performed either by using very large system sizes or by appending damping regions of various sizes to the boundaries ; in all cases we have carefully checked that the boundary conditions have no essential influence on our results . we consider the following form of the dnls hamiltonian with canonical conjugated variables @xmath0 : @xmath1 this yields the dnls equation @xmath2 which , in addition to the hamiltonian ( [ dnlsham ] ) , also conserves the total _ excitation norm _ ( or _ power _ in nonlinear optics applications ) , @xmath3 the conservation laws for the norm and hamiltonian are , through noether s theorem , related to the invariance of the dnls equation ( or , more precisely , of its corresponding action integral ) under infinitesimal transformations in phase ( @xmath4 ) and time ( @xmath5 ) , respectively . defining the norm density @xmath6 and hamiltonian density @xmath7 as in eqs . ( [ norm ] ) resp . ( [ dnlsham ] ) , the conservation laws can be expressed in terms of continuity equations as @xmath8 @xmath9 with the _ ( norm ) current density _ @xmath10\ ] ] and the _ hamiltonian flux density _ @xmath11 , % \ ] ] respectively . these conservation laws are discrete analogs to those existing for the continuous nls equations with general nonlinearities ( see e.g. @xcite ) , however there is no discrete counterpart to the momentum conservation law since the discrete equation has no continuous translational symmetry in space . furthermore , we note that the transformation @xmath12 in ( [ dnls ] ) is equivalent to @xmath13 , and thus we will for the rest of this paper only consider @xmath14 without loss of generality . the single - site dnls - breather is a stationary - state solution to eq . ( [ dnls ] ) of the form @xmath15 where the time - independent shape @xmath16 depends on the frequency @xmath17 and is spatially localized with a single maximum at a lattice site . the breather exists for all @xmath18 ; the limit @xmath19 ( or @xmath20 ) corresponding to the anticontinuous limit , where @xmath16 is localized at a single lattice - site , while the limit @xmath21 corresponds to the continuous limit , where @xmath16 approaches the nls soliton . the single - site breather is a ground state solution to eq . ( [ dnls ] ) in the sense that it minimizes the hamiltonian ( [ dnlsham ] ) for a fixed value of the norm ( [ norm ] ) , i.e. , @xmath22 , where the frequency @xmath17 appears as the lagrange multiplier ( see e.g. @xcite ) . the norm for the single - site breather , @xmath23 , is known to be a monotonously increasing function of @xmath17 , while the hamiltonian , @xmath24 , is negative and monotonously decreasing ( see e.g. @xcite ) . from the minimization condition , these functions will be related as @xmath25 to describe the dynamics close to the breather ( [ stationary ] ) , we introduce the following perturbation expansion : @xmath26 where @xmath27 is the initial perturbation and @xmath28 . thus , as for the usual stability analysis of stationary states ( see e.g. @xcite ) , the perturbation is applied in a frame rotating with the breather frequency @xmath17 . substituting into eq . ( [ dnls ] ) and identifying coefficients for consecutive powers of the small parameter @xmath29 yields an infinite set of equations , which from 0th to 4th order read : @xmath30 where the operator @xmath31 ( which is linear over the field of real numbers ) is defined from the first equality in eq . ( [ 1st ] ) . the 0th order equation ( [ 0th ] ) gives the breather shape @xmath16 ( which for the single - site breather can be assumed real and positive without loss of generality ) , while the 1st order equation ( [ 1st ] ) is the linearization of the dnls equation around the breather . to obtain the solutions to the linearized equations ( [ 1st ] ) , we proceed in a similar way as is usually done for continuous generalized nls models ( see e.g.@xcite ) and introduce a substitution of the form @xmath32 so that @xmath33 substituting ( [ subst ] ) into eq . ( [ 1st ] ) and assuming @xmath34 real yields @xmath35 where the operators @xmath36 and @xmath37 are hermitian . thus , we can obtain the eigenfrequencies @xmath38 and the corresponding eigenvectors @xmath39 from matrix diagonalization , @xmath40 to make connection to earlier work @xcite , we remark that the vector @xmath41 is an eigenvector of the floquet matrix with eigenvalue @xmath42 , where the time - period @xmath43 here is arbitrary since the operators @xmath36 and @xmath37 are time - independent . ( the symplectic floquet matrix is @xmath44 , where @xmath45 is obtained from @xmath46 by changing @xmath37 into @xmath47 . ) thus , ( [ subst ] ) is the linear combination of two complex conjugated floquet eigensolutions which makes @xmath48 and @xmath49 real . for the single - site breather , all eigenvalues @xmath38 of @xmath46 are always real , implying the linear stability of the breather for all parameter values @xmath50 @xcite . accordingly , we can also choose the eigenvectors @xmath39 of @xmath46 to be real and normalized , in which case the phase of the amplitude @xmath51 describes the symmetry properties of the solution ( [ subst ] ) under time reversal : choosing @xmath51 real yields a time - symmetric solution , @xmath52 , while choosing @xmath51 purely imaginary yields a time - antisymmetric solution , @xmath53 . for an infinite system , the spectrum of the ( non - hermitian ) matrix @xmath54 can generally be divided into a continuous ( phonon ) part , corresponding to extended eigenvectors , and a point spectrum corresponding to localized eigenvectors . the phonon spectrum for any localized solution @xmath16 is easily obtained from the limit @xmath55 , since the condition @xmath56 implies that the operators @xmath57 and @xmath58 become identical and eqs . ( [ w_n ] ) - ( [ u_n ] ) reduce into two uncoupled equations for the linear combinations @xmath59 resp . assuming @xmath61 and @xmath62 , respectively , yields the dispersion relations @xmath63 from eqs . ( [ w_n ] ) - ( [ u_n ] ) . thus , the continuous spectrum of the matrix @xmath54 consists of two branches , symmetrically located around @xmath64 , and since @xmath65 for the single - site breather these two branches never overlap . note also that two eigenvectors with eigenvalues @xmath66 correspond to the same solution to eq . ( [ 1st ] ) ( changing the sign of @xmath38 in eq . ( [ subst ] ) is equivalent to changing @xmath67 , @xmath68 , @xmath69 ) , and therefore it is enough to consider e.g. @xmath70 , in which case @xmath71 always vanishes exponentially as @xmath72 . when @xmath73 is not too large , the linear spectrum around the single - site breather contains also two pairs of nonzero isolated eigenvalues @xmath38 , which correspond to the two internal modes of the breather @xcite . one of these modes is a spatially symmetric , breathing , mode , while the other is a spatially antisymmetric , translational or pinning mode . numerically , it has been found that the breathing mode exists for @xmath74 , while the pinning mode exists for @xmath75 . the numerically calculated internal mode frequencies @xmath38 as a function of breather frequency @xmath17 are shown in fig . note that as @xmath76 , the breathing mode frequency approaches the lower edge of the phonon band ( but always stays outside the band @xcite ) , while the pinning mode frequency approaches zero ( but always stays nonzero ) . this is consistent with the fact that the soliton solution of the continuous nls equation has no breathing mode ( due to its exact integrability ) , and has a translational mode with zero frequency due to the translational symmetry of the nls equation . to obtain a complete set of solutions to eq . ( [ 1st ] ) in which an arbitrary initial perturbation @xmath77 can be expanded , we must include also the zero - frequency solutions , which generally can be written as a superposition of two fundamental modes . one of these modes ( phase mode @xcite ) is the solution @xmath78 to the homogeneous equation ( [ w_n ] ) , @xmath79 . the corresponding perturbation @xmath80 describes a rotation of the overall phase of the breather . the second mode ( growth mode @xcite ) is obtained by solving the inhomogeneous equation @xmath81 , which has the solution @xmath82 . the corresponding solution to eq . ( [ 1st ] ) is @xmath83 , and corresponds to a time - linear growth of the perturbation representing a small change in the breather frequency . although the set of eigensolutions ( [ subst ] ) together with the two zero - frequency modes forms a basis for the space of solutions to eq . ( [ 1st ] ) ( there are no bifurcations , which could result in additional marginal modes @xcite with time - linear growth at degenerate eigenvalues ) , this basis is in general not orthogonal using the ordinary scalar product , and typically there is a considerable overlap between the solution corresponding to the internal breathing mode and the zero - frequency modes . however , in analogy with e.g. @xcite , we can define a pseudo - scalar product between any two vectors @xmath84 and @xmath85 by @xmath86 this product is formally not a true scalar product , since for the general case the product of a vector with itself as defined by ( [ scalar ] ) is not necessarily positive . however , when @xmath39 is a real eigenvector of @xmath54 we have @xmath87 from ( [ w_n ] ) , and the operator @xmath57 is positive definite for all @xmath88 @xcite . with this product , it follows from ( [ w_n ] ) - ( [ u_n ] ) that all eigensolutions with different ( real ) eigenfrequencies @xmath38 are orthogonal in the sense that @xmath89 and the only nonzero product involving the zero - frequency modes is the cross - product between the phase mode and the growth mode @xcite : @xmath90 where @xmath23 is the norm ( [ norm ] ) of the breather with frequency @xmath17 . we also remark that , since the product ( [ scalar ] ) multiplied by a factor @xmath91 is just the symplectic product between the two vectors @xmath92 and @xmath93 , the sign of the product of an eigenvector with itself can be interpreted as the negative of the krein signature of the corresponding pair of floquet eigenvalues @xcite . we now consider as initial state a single - site breather perturbed in the direction of a single eigenmode ( [ subst ] ) ( localized or extended ) of the linearized equations ( [ 1st ] ) , and wish to describe qualitatively the long - time effects of this perturbation using the expansion ( [ perturbation ] ) . in general , taking into account terms up to order @xmath94 in this expansion yields a solution to the dnls equation which is correct to @xmath95 , i.e. , for long but finite time - scales for small initial perturbations ( note that the expansion parameter @xmath29 plays the same role as the mode amplitude @xmath51 . ) . as is wellknown however , this kind of expansion in general diverges due to resonances between solutions to the homogeneous equation ( [ 1st ] ) and the inhomogeneous terms appearing in the right - hand sides of eqs . ( [ 2nd])-([4th ] ) and the corresponding higher - order equations . a resonance with a solution belonging to the continuous spectrum results in a bounded but nonlocalized solution corresponding to outgoing radiation , while a resonance with an eigenfunction belonging to the discrete spectrum gives a spatially localized response which diverges linearly in time . however , up to any finite order these divergences can be systematically removed by allowing a slow time dependence of the independent variables , which in our case are taken to be the mode amplitude @xmath51 and the breather frequency @xmath17 . this procedure adds additional terms to the equations , which can be tuned so that the divergent parts of the response disappear . in other words , these two quantities are used as collective variables which , together with the outgoing radiation fields , are expected to describe the main features of the asymptotic dynamics if the initial perturbation is sufficiently small . the second order correction is given by the inhomogeneous eq . ( [ 2nd ] ) , which with the substitution ( [ subst ] ) becomes ( choosing @xmath34 , @xmath96 , and @xmath97 real without loss of generality ) : @xmath98 .\end{aligned}\ ] ] thus , the right - hand side contains one static part and one part involving the frequencies @xmath99 . it acts as a periodic force with frequencies 0 and @xmath100 , and since all terms contain the factor @xmath34 this force is localized at the breather region . the response to this force will remain bounded and localized unless the corresponding homogeneous equation ( [ 1st ] ) has a solution with frequency 0 resp . @xmath101 which is non - orthogonal to the corresponding part of the right - hand side in ( [ 2nd2 ] ) . as will be shown in sec . [ subsec : higher - order ] , a non - zero overlap between the static part of ( [ 2nd2 ] ) and the zero - frequency solutions of ( [ 1st ] ) yields a ( time - independent ) shift of the breather frequency . moreover , if @xmath102 , so that @xmath100 is inside the phonon band of the homogeneous equation , a resonance will generally occur , resulting in radiation with frequency @xmath103 emitted from the breather region . the strength of the radiation field is determined by the ( generally nonzero ) overlap between the @xmath101-part of ( [ 2nd2 ] ) and the corresponding homogeneous solution ( see sec . [ subsec : higher - order ] ) . in a similar way , we obtain that the right - hand - side of the third order equation ( [ 3rd ] ) contains the frequencies @xmath38 and @xmath104 , the fourth order equation ( [ 4th ] ) contains the frequencies 0 , @xmath105 , and @xmath106 , and in general the _ p_th order equation contains as its highest harmonic the frequency @xmath107 . accordingly , we conclude that if @xmath108 so that @xmath107 belongs to the phonon band , the perturbed breather will radiate to _ p_th order . the consequences of this radiation for the breather itself will be discussed in secs . [ sec : internal ] and [ sec : phonons ] for the cases of localized and extended perturbations @xmath109 , respectively . with the initial perturbation @xmath77 of the single - site breather corresponding to a spatially localized eigenmode of the linearized equations ( [ 1st ] ) , it is clear from the discussion in sec . [ subsec : strategy ] that higher order radiation always will be created , since for any internal mode frequency @xmath38 there is always an integer @xmath94 such that @xmath107 belongs to the phonon band and ( [ pradiation ] ) is fulfilled . moreover , from the numerical results presented in fig . [ fig1 ] we find that the spatially symmetric breathing mode always radiates to second order , since ( [ pradiation ] ) always is fulfilled for @xmath110 , while the antisymmetric pinning mode radiates to second order only when @xmath111 . thus , due to this radiation from the breather the total norm contained in any finite region around the breather(i.e . , the total norm of breather + internal mode ) will always decrease with time . however , the main concern here is the long - time effect of the internal - mode excitation on the breather itself , and thus we must investigate whether there will also be some transfer of energy between the breather and its internal mode . we will first ( sec . [ subsec : numerics ] ) show results from direct numerical integration of eq . ( [ dnls ] ) ; then we will give two alternative approaches to the analytical interpretation of these results based on the higher - order equations ( [ 2nd])-([4th ] ) ( sec . [ subsec : higher - order ] ) resp . the conservation laws ( [ normcons])-([hamcons ] ) ( sec . [ subsec : conservation ] ) . in fig . [ fig2 ] , we show a typical example on the long - time evolution of a breather when the initial perturbation is taken in the direction of its internal breathing mode . as is seen from fig . [ fig2 ] ( a ) , the amplitude of the breathing mode decays slowly with time as a consequence of the losses due to generation of second order radiation , and a careful study of its envelope @xmath112 indicates that it decays as @xmath113 where @xmath114 is a constant . this is consistent with a similar result obtained for the continuum nls equation with generalized ( non - cubic ) nonlinearity @xcite ; the analytical motivation for this result ( which is analogous to that of the continuum model given in ref . @xcite ) is given in the following subsections . however , the main result of this section is illustrated in fig . [ fig2 ] ( b ) and ( c ) . fig . [ fig2 ] ( b ) is obtained by calculating the time - average of the central - site intensity as @xmath115 where @xmath116 is a set of closely spaced time - instants . it is clear that the interaction between the breather and its internal mode asymptotically leads to an _ increase _ of the average peak intensity , i.e. , to _ breather growth_. the same phenomenon is illustrated also in fig . [ fig2 ] ( c ) , where we have plotted the difference between the instantaneous breather frequency calculated at time @xmath117 , @xmath118 , and the frequency of the unperturbed breather @xmath119 . from this figure , we can also conclude that there are two different mechanisms causing the shift of breather frequency . firstly , there is an initial ( almost instantaneous ) rather large frequency shift , which can be interpreted as an adaption of the initially perturbed breather to the breather which is closest to the initial condition . as is shown below in sec . [ subsec : higher - order ] , this time - independent frequency shift , which is observed to be always positive for the breathing mode , is a consequence of the overlap between the static part of the right - hand side of the second - order equation ( [ 2nd2 ] ) and the zero - frequency modes . secondly , there is the slow , continuous increase of the breather frequency which corresponds to the slow increase of @xmath120 in fig . [ fig2 ] ( b ) , indicating a continuous transfer of norm from the internal mode to the breather . it is described by the static part of the right - hand side of the fourth - order equation ( [ 4th ] ) ( see sec . [ subsec : higher - order ] ) . when the initial perturbation of the breather is taken in the direction of its internal pinning mode we observe , just as for the breathing mode , that the breather - internal mode interaction asymptotically _ always leads to breather growth_. an example is shown in fig . [ fig3 ] , where the parameter values have been chosen so that the lowest harmonic that enters the phonon band is @xmath121 ( @xmath122 ) , we observe two qualitative differences compared to the case with breathing mode excitation . firstly , since in this case the first phonon resonance occurs only in the third order equation ( [ 3rd ] ) , the decay of the internal mode amplitude will be slower , and a good fit is obtained by @xmath123 . this agrees with the general result when @xmath124 is the lowest harmonic that enters the phonon band obtained for the continuum generalized nls equation in ref . @xcite ; the derivation of the corresponding result for the discrete case ( see eq . ( [ ageneral ] ) ) is given in sec . [ subsec : conservation ] . as a consequence of the slower decay of the internal mode amplitude , the breather growth will also be slower when @xmath125 , as can be seen from fig . [ fig3 ] ( c ) by comparing the time - scales with those of fig . secondly , the initial shift of the breather frequency for a pinning mode excitation is always much smaller than for the breathing mode excitation ( also when @xmath100 is in the phonon band ) , and is when @xmath126 also observed to be negative . the explanation for this is given in sec . [ subsec : higher - order ] . however , it is important to stress that also in the cases where the initial frequency shift is negative , we find that the continuous breather growth always will give an asymptotic frequency shift which is positive . here , we will analyze the higher - order equations ( [ 2nd])-([4th ] ) by making use of the strategy of systematically removing the appearing divergent parts as outlined in sec . [ subsec : strategy ] ( in analogy with the treatment of the continuous nls - type equations in e.g. refs . first , we show how the dominating contribution to the time - independent frequency shift observed in the numerical simulations above can be calculated from the static part of the right - hand side of the second - order equation ( [ 2nd ] ) . this frequency shift can be explicitly taken into account by replacing @xmath17 in eq . ( [ perturbation ] ) with @xmath127 , where @xmath119 is the unperturbed breather frequency and @xmath128 the second - order shift to be determined . this implies that the additional term @xmath129 will be added to the right - hand side of eq . ( [ 2nd2 ] ) . writing the response to the static part of ( [ 2nd2 ] ) as @xmath130 with real @xmath131 and @xmath132 then yields @xmath133 with @xmath46 as defined by eq . ( [ matrix ] ) . if the expansion of the right - hand side of ( [ mstatic ] ) in the complete set of vectors consisting of the eigenvectors of @xmath46 ( including the phase mode ) and the growth mode contains some component on either of the two zero - frequency modes , the response @xmath134 will not remain bounded but diverge linearly with time . thus , in order to remove this divergency , the frequency shift @xmath128 must be chosen so that both these components are identically zero . the component on the growth mode is trivially zero , while the component on the phase mode is obtained by applying the pseudo - scalar product ( [ scalar ] ) with the vector corresponding to the growth mode and using ( [ overlap ] ) . demanding this component to be zero yields @xmath135 this is typically positive for the breathing mode , since the dominating contribution to the sum in ( [ static ] ) comes from the central site @xmath136 , and @xmath137 is always positive . for the spatially antisymmetric pinning mode , there is no contribution to this sum from the central site , since @xmath138 and @xmath139 are zero . the change from a positive to a negative frequency shift when increasing @xmath17 in this case is related to a qualitative change of the nature of the growth mode @xmath140 : for @xmath141 we find that @xmath142 for all @xmath143 , so that all terms in the sum in eq . ( [ static ] ) are negative , while @xmath144 becomes positive also for sites in the neighborhood of @xmath136 for smaller @xmath17 . for the rest of the analysis in this subsection , we assume for calculational simplicity that the internal mode frequency is such that @xmath105 is inside the phonon band ( and thus it is not applicable for the pinning mode excitation when @xmath145 ) . then , the non - static part of the right - hand side of the second order equation ( [ 2nd2 ] ) will generally give rise to a non - localized response , which can be written in the form @xmath146 . this response corresponds to the radiation field going out from the breather region , and since the right - hand side of ( [ 2nd2 ] ) is spatially localized and symmetric , this field should asymptotically correspond to two identical linear waves propagating to the left ( right ) for @xmath147 ( @xmath148 ) . thus , the boundary conditions can be written as @xmath149 with @xmath150 according to ( [ adisp ] ) . defining for general @xmath151 the matrix @xmath152 ( cf . ( [ matrix ] ) ) as @xmath153 the functions @xmath154 and @xmath155 are seen from ( [ 2nd2 ] ) to be determined by @xmath156 since for general @xmath151 every eigenvector of @xmath157 with eigenvalue @xmath158 is also an eigenvector of @xmath46 with eigenvalue @xmath159 , the right - hand side can be expanded on the basis of eigenvectors of @xmath46 ( including the zero - frequency modes ) . the strength of the radiation field is then given by the expansion coefficient for the ( continuous spectrum ) eigenvector of @xmath46 with eigenvalue @xmath100 , since this corresponds to the eigenvalue @xmath160 of @xmath161 , and thus a spatially non - bounded response in ( [ m2omegap ] ) . using the orthogonality relation ( [ orthogonal ] ) , this coefficient is simply the overlap between the right - hand side of ( [ m2omegap ] ) and the eigenvector of @xmath46 with eigenvalue @xmath100 calculated with the pseudo - scalar product ( [ scalar ] ) , which is generally nonzero . next , we show how the dominating contribution to the decay of the internal mode amplitude as given by eq . ( [ decay ] ) is obtained from the condition that @xmath162 in the third - order equation ( [ 3rd ] ) should remain bounded . to this end , we assume a slow time - dependence of the internal mode amplitude of the form @xmath163 , and consider the response to the terms with frequency @xmath38 in the right - hand side of eq . ( [ 3rd ] ) corrected by the additional terms @xmath164 appearing as a consequence of including the time - dependence of @xmath51 and the second - order frequency shift @xmath128 from eq . ( [ static ] ) in the perturbation expansion ( [ perturbation ] ) . writing the response to this part as @xmath165 yields @xmath166 a bounded response for @xmath167 exists only if the vector on the right - hand side of ( [ momegap ] ) has no component in the direction of the internal mode eigenvector @xmath39 , since this is the eigenvector corresponding to the eigenvalue zero of @xmath168 . using the orthogonality relation ( [ orthogonal ] ) , this component is obtained by application of the product ( [ scalar ] ) with the vector @xmath39 , and the condition that this component must be zero determines the time - evolution of @xmath51 . considering only the absolute value @xmath169 , the resulting equation has the form @xmath170 is given by @xmath171 } { \sum_n u_n w_n } = \frac{8 c \omega_p |r_2|^2 \sin q_2 } { \sum_n w_n { \cal l}_0 w_n } > 0 , \ ] ] where the second equality is obtained using eqs . ( [ w_n ] ) , ( [ bc ] ) and ( [ m2omegap ] ) , and the positivity of @xmath172 follows from the fact that , as mentioned in sec . [ subsec : linear ] , the operator @xmath57 is positive definite for all @xmath173 @xcite . finally , we show how the continuous increase of the breather frequency appears from the divergent response to the static part of the right - hand side of the fourth - order equation ( [ 4th ] ) . in its unmodified form , this part is given by @xmath174 now , it is clear from ( [ static ] ) and ( [ decay ] ) that the time - dependence of @xmath51 will induce a time - dependence of the second - order frequency shift @xmath128 of the form @xmath175 , so that we can express the total breather frequency up to order @xmath176 as @xmath177 , where a 4th order correction @xmath178 has also been included . then , we must also take into account the time - dependence of the breather shape @xmath34 by writing @xmath179 . as a consequence , the term @xmath180 will be added to the expression ( [ static4 ] ) for @xmath181 in the right - hand side of ( [ 4th ] ) , and writing the response to this total force as @xmath182 with real @xmath183 and @xmath184 yields @xmath185 the response @xmath186 will be bounded in time only if the right - hand side of ( [ m4 ] ) has no component either on the growth mode or on the phase mode , which gives two conditions for the determination of @xmath187 and @xmath178 . using ( [ scalar])-([overlap ] ) , we obtain by demanding the expansion coefficient for the growth mode to be zero : @xmath188 \right . \nonumber\\ \left . + \phi_n \left[2 u_n w_n { \rm im}\left ( u_n^{(2)}\right)+\left(3w_n^2-u_n^2\right ) { \rm im}\left ( w_n^{(2)}\right ) \right ] \right\ } .\end{aligned}\ ] ] ( similarly , @xmath178 is obtained by demanding the component on the phase mode to be zero . ) thus , the dominating contribution to the frequency growth should be of order @xmath189 , so that with the approximate time - dependence ( [ decay ] ) of the internal - mode amplitude we obtain qualitatively @xmath190 which is in good agreement with the numerically observed time - dependence of the frequency shift as shown in fig . [ fig2 ] ( c ) . however , the positivity of @xmath187 is not easily seen from the expression ( [ lambda2dot ] ) , and therefore we will in the following subsection derive an alternative expression from which the positivity follows immediately , using the conservation laws for the norm resp . hamiltonian . we consider first the conservation law ( [ normcons ] ) for the total norm ( [ norm ] ) contained in any large but finite region around the breather . averaging over a time - interval @xmath191 $ ] and using eqs . ( [ perturbation ] ) , ( [ subst ] ) , we can write the time - averaged norm to second order in @xmath192 ( putting @xmath193 ) as @xmath194 . \label{normav}\ ] ] ( note that there will be no contribution at order 2 from the static second - order correction @xmath131 , since the renormalization of the breather frequency @xmath17 according to ( [ static ] ) yields @xmath195 . ) in the case when @xmath105 belongs to the phonon band , we obtain the following balance equation , which is correct up to order @xmath196 : @xmath197 where we have used eqs . ( [ normcons ] ) , ( [ jndef ] ) and ( [ bc ] ) , and @xmath198 is as defined below eq . ( [ bc ] ) . similarly , we can use the conservation law ( [ hamcons ] ) for the hamiltonian ( [ dnlsham ] ) together with the general expression for the hamiltonian flux density ( [ jhdef ] ) for a small - amplitude plane wave @xmath199 , @xmath200 to write the balance equation for the total time - averaged hamiltonian in the same region for the case of second order radiation : @xmath201 which is also correct to order @xmath196 . the lowest order contribution to the derivative @xmath202 can be obtained using the first equality in the equation of motion ( [ dnls ] ) and its complex conjugate as follows : @xmath203 which is always negative for an internal mode excitation since @xmath204 . thus , we can combine the two balance equations ( [ dndt ] ) and ( [ dhdt ] ) , and using eqs . ( [ nh ] ) and ( [ dhda2 ] ) we obtain the expression ( [ decay ] ) for @xmath112 with @xmath172 as in ( [ gamma ] ) , together with the following expression for the frequency growth rate from which its positivity is immediately seen : @xmath205 this approach also has the advantage that it is easily generalized to the case where the lowest harmonic that enters the phonon band is @xmath107 with @xmath206 , i.e. , for the pinning mode excitation when @xmath207 . then , we can write the boundary conditions at the infinities to lowest order in @xmath51 as @xmath208}$ ] , @xmath209 , where @xmath210 and @xmath211 according to ( [ adisp ] ) . consequently , we can proceed exactly as above , writing down the balance equations for the norm and hamiltonian to order @xmath212 just by modifying the right - hand sides of eqs . ( [ dndt ] ) resp . ( [ dhdt ] ) by replacing @xmath196 with @xmath212 , @xmath213 with @xmath214 , @xmath198 with @xmath215 , and @xmath105 with @xmath107 . combining the balance equations yields the following general expression for the time - dependence of the internal mode amplitude , @xmath216^ { 1/(2p-2)}}\;,\ ; \gamma_p = \frac{4pc|r_p|^2 \sin q_p } { \sum_n u_n w_n } > 0 , \label{ageneral}\ ] ] which is the analog to the expression obtained with similar arguments in @xcite for the continuum nls models . and most importantly , we obtain a general expression for the breather frequency growth rate which is positive for all @xmath94 : @xmath217 thus , integrating eq . ( [ lambdadotgeneral ] ) using the time - dependence ( [ ageneral ] ) of the internal - mode amplitude , we obtain that the dominating contribution to the frequency growth generally can be written qualitatively as @xmath218 let us finally point out that the approach used in this section provides a simple physical interpretation of why the interaction between the breather and its internal mode in addition to generate radiation also should lead to breather growth . since the expression ( [ hamflowpw ] ) for the hamiltonian flux density of a small - amplitude plane wave is positive when @xmath219 and @xmath220 have the same sign , the hamiltonian energy for a plane wave always propagates in the same direction as the wave itself . thus , the second ( or higher ) order radiation emitted from the breather region will always carry away a _ positive _ amount of the hamiltonian energy , or , equivalently , negative hamiltonian energy will flow into the breather region . moreover , from eq . ( [ dhda2 ] ) it is clear that the contribution to the hamiltonian from the internal mode always is negative and a monotonously _ decreasing _ function of its amplitude , and thus the decay of the internal mode would cause an _ increase _ of the hamiltonian in the breather region . consequently , since the hamiltonian of the pure breather is a monotonously decreasing function of the breather frequency , the breather should grow so that the total hamiltonian of ( breather+internal mode ) decreases . a similar mechanism was recently found to cause soliton growth in the parametrically driven continuum nls equation in the regime of oscillatory instability @xcite , and this type of argument has also been used to explain the quasi - collapse of a broad excitation to a narrow localized state in the two - dimensional dnls equation @xcite . we choose , as in the previous section , the initial perturbation @xmath77 to be an eigensolution of the linearized equations ( [ 1st ] ) , but now we consider the case of a spatially extended perturbation . choosing without loss of generality a solution of the form ( [ subst ] ) with positive frequency @xmath38 yields the asymptotic behaviour @xmath221 where the wave vector @xmath220 @xmath222 is determined by the dispersion relation ( [ adisp ] ) , and @xmath223 is the phase shift across the breather . thus , the excitation of an extended eigenmode corresponds to an interaction between the breather and a non - propagating , standing - wave phonon with small amplitude @xmath51 . as was mentioned in the introduction , these standing - wave phonons are generally unstable , but since the instabilities become exponentially weak in the small - amplitude limit they are expected to have very little effect on the breather for the perturbation sizes and time - scales considered in this section . we will return to the effects on the breather of these instabilities in sec . [ sec : instabilities ] , where larger perturbations are considered . in contrast to the case of excitation of a localized internal mode discussed in sec . [ sec : internal ] , where higher order radiation always was emitted from the breather region , the condition ( [ pradiation ] ) yields that for the standing - wave perturbation the breather will radiate to higher order only if @xmath224 , so that both @xmath38 and @xmath225 are inside the phonon band . in terms of the phonon wave vector @xmath220 , this means that there is a critical value @xmath226 , @xmath227 such that for @xmath228 second order radiation will be emitted from the breather region , while for @xmath229 ( @xmath230 ) _ all _ multiples of @xmath38 are outside the phonon band , and no higher order radiation is emitted . as will be shown below , these two regions yield qualitatively different scenarios for the long - time evolution of the perturbed breather . we also note that for @xmath231 we have @xmath232 , so that for the highly localized , high - frequency breathers _ no phonons can generate higher - order radiation _ ( note also that there are no internal modes in this regime ) . furthermore , we always have @xmath233 , so that the regime of higher - order radiation generation is a subset of the regime @xmath234 where modulational instability for _ travelling _ plane waves occurs @xcite . let us first discuss the case @xmath235 . a typical example of the long - time evolution of a breather interacting with a small - amplitude standing - wave phonon with @xmath235 is illustrated by fig . [ fig4 ] . as is seen from fig . [ fig4 ] ( a ) , the amplitude of the oscillations remains essentially constant in time , but a closer inspection reveals that the average value of @xmath236 asymptotically _ increases _ with an apparently constant rate ( see inset in fig . [ fig4 ] ( b ) ) . similarly , fig . [ fig4 ] ( b ) shows that also the total norm contained in any finite region around the breather asymptotically increases linearly with time . we find that these results are generic for all cases when the phonon wave vector @xmath235 ( the spatial symmetry of the phonon is not important for the asymptotic behaviour ) , and thus we conclude that in this regime , the breather can pump energy from the phonon ( which is infinite for an infinite system ) , and thereby grow . in the same spirit as for the internal mode excitation in sec . [ subsec : conservation ] , we can give a simple argument based on the conservation laws to motivate why the generation of second order radiation should lead to breather growth . to this end , we assume that the initial standing - wave phonon is infinitely extended , and that far away from the breather a stationary regime will be reached corresponding to the following boundary conditions @xmath237 e ^{i\lambda t } \ ; ; \ ; n \rightarrow \pm \infty . % \label{scattering}\ ] ] thus , we have taken into account the second - order radiation with frequency @xmath100 generated at the breather region but neglected possible higher - order radiation ; moreover the resonance at the original phonon frequency @xmath38 in the third - order equation ( [ 3rd ] ) has been taken into account by allowing the incoming and outgoing complex amplitudes @xmath51 and @xmath238 to be different . we can then , in analogy with eqs . ( [ dndt ] ) and ( [ dhdt ] ) , write the balance equations for the total norm and hamiltonian contained in a region around the breather averaged over a time - interval @xmath191 $ ] as @xmath239 , \label{dndtph}\ ] ] and @xmath240 , \label{dhdtph}\end{aligned}\ ] ] respectively . here , we have used the facts that the time - average of the norm current density ( [ jndef ] ) and the hamiltonian flux density ( [ jhdef ] ) are additive quantities for small - amplitude plane waves , and that in the stationary regime , the mode amplitudes @xmath51 , @xmath238 and @xmath213 are time - independent . combining eqs . ( [ dndtph ] ) and ( [ dhdtph ] ) and using ( [ nh ] ) , we obtain that the breather frequency grows with a constant rate given by @xmath241 the physical interpretation of this result is , similarly as for the case of internal mode excitation , that the generation of higher - order radiation results in a net flow of negative hamiltonian energy into the breather region , which is absorbed by the breather by increasing its frequency and maximum amplitude . this process is similar to the one observed for the two - channel phonon scattering on breathers in a klein - gordon model with a morse potential in ref . @xcite ; however in the latter case the second outgoing wave resulted from a resonance in the linearized equations and were therefore of the same order of magnitude as the incoming wave , and moreover the outcome in this case was breather decay since the energy of a klein - gordon breather is an increasing function of its amplitude . considering now the regime @xmath242 where all multiples of @xmath38 are outside the phonon band , the most important conclusion from our extensive numerical investigations is that we _ never observe breather growth_. instead , we sometimes ( but not always ) observe a very slow decrease of @xmath120 , and an increase of the fluctuations around this mean - value . this behaviour is illustrated by fig . [ fig5 ] ( a ) and ( b ) . a possible interpretation of these results is that , since the higher harmonics which are created by the breather - phonon interaction can not propagate , they stay trapped around the breather . thus , this could lead to a transfer of energy from the pure breather , which acquires more and more internal frequencies and becomes a chaotic breather @xcite . another possible interpretation is that the increase of the oscillation amplitude is connected with the oscillatory instabilities of the standing waves ; as we will show in the next section these instabilities provide a mechanism for breather decay . however in some cases , illustrated by fig . [ fig5 ] ( c ) and ( d ) , the oscillation amplitude as well as its average value apparently approaches a constant limit value . we have at present no explanation for this behaviour ( as will be discussed in the next section , there exist exact phonobreather solutions which could be candidates for such a final state , but they are unstable ) ; it is possible that the time ranges that we were able to study with sufficient numerical accuracy in these cases simply were too short to observe the scenario described by fig . [ fig5 ] ( a ) and ( b ) . to conclude this section , we repeat our main result that breather growth is observed if and only if @xmath235 , where @xmath226 is given by ( [ qc ] ) . the fact that @xmath232 for @xmath243 thus implies the existence of an upper limit beyond which the breather can not grow with the type of perturbations considered here . we would also like to relate our results to recent numerical simulations of breathers interacting with _ propagating _ phonons in klein - gordon @xcite and fpu @xcite lattices . for the klein - gordon lattice with a ( soft ) morse on - site potential , phonons with small wave vector @xmath220 were observed to yield breather growth , while phonons with large @xmath220 caused breather decay . for a fpu lattice with hard anharmonicity , the opposite situation was observed , i.e. , small-@xmath220 phonons caused breather decay and large-@xmath220 phonons breather growth . the fact that the situation for the hard fpu lattice was opposite to that of the soft klein - gordon lattice could be expected , since in the former case the modulational instability occurs for large @xmath220 , whereas soft klein - gordon and dnls lattices with @xmath14 are modulationally unstable for small @xmath220 . however , we stress that the relation between plane - wave modulational instability and breather growth is nontrivial , and at least for the case considered in this paper the critical value @xmath226 for breather growth from interaction with standing - wave phonons differs from the critical value @xmath244 for modulational instability of travelling waves . in this section , with the aim at describing the interaction between a breather and a standing - wave phonon with non - negligible amplitude , we will take a slightly different point of view than in the preceding sections . instead of choosing as initial condition an exact breather solution and adding a perturbation corresponding to an eigenmode of the linearized equations , we will here consider initial conditions which are exact _ phonobreather _ @xcite ( or _ nanopteron _ ) solutions . by definition , a phonobreather consists of a spatially localized breather on top of a spatially extended tail which is a nonlinear , standing - wave phonon . ( there are also solutions where the tail is a propagating wave @xcite , but we will not discuss them further here . ) phonobreathers exist generically for nonlinear lattice - equations ( see e.g. @xcite ) , but their existence normally requires an integer relationship between the breather and phonon frequencies . however , for the dnls equation phonobreathers exist for _ any _ ( rational or irrational ) relation between the two frequencies @xcite , as a consequence of the additional invariance of the equation under global phase transformations . since the phonobreathers are exact solutions consisting of a breather part and a standing - wave part , one could expect them to be attractors for the initial conditions considered in sec . [ sec : phonons ] . however , as was shown recently @xcite , generically for soft klein - gordon and dnls models with @xmath14 _ all _ phonobreathers with phonon wave vector @xmath245 will be _ linearly unstable _ if , for fixed phonon amplitude @xmath51 , the linear coupling @xmath246 is larger than some threshold value @xmath247 ( i.e. , away from the anticontinuous limit ) . ( for lattices with hard potentials and dnls with @xmath248 , the stable phonobreather has @xmath249 . ) these instabilities are caused by an oscillatory instability of the standing - wave phonon itself , which can be understood by considering the construction of a nonlinear standing wave with wave vector @xmath220 close to @xmath250 at the anticontinuous limit @xmath251 by introducing a periodic array of discommensurations or defects in the nonlinear phonon with wave vector @xmath250 and amplitude @xmath51 , @xmath252 , which is linearly stable for all @xmath51 and @xmath14 @xcite . in the anticontinuous limit @xmath251 , each defect consists when @xmath253 of one extra site with @xmath254 added to the @xmath250-phonon , which consequently suffers an additional phase shift of @xmath250 across each defect . for @xmath234 , each defect consists of several consecutive zero - amplitude sites with associated phase shifts ( a general method for generating the anticontinuous coding sequence for standing - wave phonons from a circle map is described in @xcite ) ; in this case it is also useful to consider the periodic repetition of sites with @xmath255 as defects of the zero - amplitude state . the limit case of one isolated zero - amplitude defect , which is a discrete counterpart of the dark - soliton solution of the continuum nls equation , was investigated in @xcite . the linear stability analysis of this mode showed that , although it is stable close to the anticontinuous limit , it suffers a bifurcation for @xmath256 where two pairs of eigenvalues of the eigenvalue problem ( [ matrix ] ) go out in the complex plane . the resulting oscillatory instability occurs due to a resonance between a mode localized around the defect ( the defect pinning mode ) and linear radiation modes . it was shown that for finite systems , the mode recovers its stability above some upper critical value of @xmath257 ( since the wavelength of the resonating linear modes becomes larger than the system size ) ; however this critical value increases with system size so that in the limit of an infinite system , the instability persists for all @xmath258 but with a growth - rate that decreases in an exponential - like fashion when approaching the continuum limit @xmath259 . this instability was shown to result in the defect becoming mobile ( in nls terms , the stationary black soliton with zero minimum intensity transforms into a moving grey soliton with non - zero minimum intensity ) and radiation being emitted . in terms of the phase dynamics , this describes a moving , slowly spreading phase kink . the instability scenario for the standing - wave phonons is basically the same as for the isolated defect , with the essential difference that the localized pinning modes associated with the individual , periodically repeated defects now will form a continuous defect band. in general , the krein signature @xcite of this defect band is opposite to that of the bands associated with the non - zero amplitude sites @xcite ( for @xmath234 there are generally several defect bands , but they will have the same krein signature ) , and as a consequence resonances between the bands will occur if the linear coupling @xmath246 is large enough , giving rise to similar oscillatory instabilities as described above ( details are given in @xcite ) . we remark that earlier analysis @xcite of standing waves in nonlinear lattices , based on a quasi - continuum approximation , did not reveal these instabilities since their origin is the discrete nature of the lattice . let us now return to the main objective of this section , namely to study the effect of the oscillatory standing - wave instabilities on the phonobreathers . we find that the families of phonobreathers which are stable close to the anticontinuous limit and whose tails approach harmonic standing waves in the small - amplitude limit can be constructed from anticontinuous standing - wave solutions at @xmath251 , placing the breather at a zero - amplitude site of the phonon and adjusting it so that the resulting solution is either symmetric or antisymmetric around the breather site . denoting the anticontinuous breather amplitude by @xmath260 ( the phase of the breather site is unimportant when @xmath261 @xcite ) , this yields the following possibilities : \(i ) for @xmath262 the antisymmetric anticontinuous solution ( here @xmath263 ) @xmath264 with the asymptotic behaviour @xmath265 in the continuum limit @xmath259 . ( note that this solution is antisymmetric only at @xmath251 . ) \(ii ) from ( [ i ] ) we can construct a symmetric solution for @xmath262 by introducing an additional phase shift of @xmath250 at one side of the breather site , giving for @xmath263 @xmath266 with the asymptotic behaviour @xmath267 in the continuum limit . \(iii ) for @xmath268 , @xmath269 even , the number of consecutive zero - amplitude sites is odd and the antisymmetric anticontinuous solution is for e.g. @xmath270 @xmath271 behaving as @xmath272 in the continuum limit . \(iv ) from ( [ iii ] ) the symmetric solution for @xmath268 , @xmath269 even , is constructed by a phase shift as above , giving for @xmath270 @xmath273 with the asymptotic behaviour @xmath274 in the continuum limit . \(v ) for @xmath268 , @xmath269 odd , the number of consecutive zero - amplitude sites is even and we must add an extra site to obtain the antisymmetric anticontinuous solution , which for e.g. @xmath275 becomes @xmath276 behaving as @xmath277 in the continuum limit . ( a solution with similar properties is obtained by instead removing one zero - amplitude site ; however its symmetric counterpart is always unstable . ) \(vi ) the symmetric counterpart of ( [ v ] ) for @xmath268 , @xmath269 odd , is constructed by a phase shift as above , giving for @xmath275 @xmath278 with the asymptotic behaviour @xmath279 in the continuum limit . a typical example on the time - evolution for an initially very weakly perturbed phonobreather with phonon wave vector @xmath262 and phonon amplitude small but non - negligible compared to the breather amplitude is illustrated in fig . ( the example in the figure belongs to type ( i ) , but similar dynamics is observed also for the spatially symmetric states of type ( ii ) . ) we can clearly distinguish two different steps leading to the final breather destruction . the first step is the linear oscillatory instability described above , which leads to the generation of new internal frequencies of the breather , and to the movement of the defect sites in a similar way as for the case of an isolated defect . in the second step , the moving defects start interacting , and a close inspection of fig . [ fig6 ] ( a ) shows that neighboring defects tend to merge and create regions of accumulated phase fluctuations travelling around in the lattice . these will interact with the breather , and apparently cause its decay . when the breather has decayed sufficiently to have an excitable pinning mode , it will start to move in the lattice but with rapidly decreasing amplitude , and it will finally be destroyed . we have at present no complete understanding for the mechanism by which the interaction of the breather with the moving phase kinks cause its decay , but we remark that a similar scenario was observed when adding to the dnls equation an external , parametric white noise term @xcite . in that case , the white - noise approximation allowed a qualitative understanding of the breather decay as a consequence of phase fluctuations by using a collective coordinate approach . however , to observe this scenario for breather destruction it is necessary ( at least for a finite system ) that the phonon amplitude is not too small compared to the breather amplitude . if we increase the breather amplitude in fig . [ fig6 ] ( or decrease the phonon amplitude ) sufficiently , we find that although the oscillatory instability develops , the fluctuations created in the second step will be too weak to cause the breather to decay , and it will live seemingly forever as a chaotic phonobreather. the absence of decay for small perturbations can be viewed as a consequence of the fact that the single - site dnls breather is nonlinearly ( lyapunov ) stable for norm - conserving perturbations , in the sense that @xmath280 remains arbitrarily close to the breather for all times if the initial perturbation is small enough @xcite . thus , it is clear that for finite systems , the breather can not be destroyed unless the phonon amplitude exceeds some critical value , while nothing can be said about the infinite system since any infinitely extended phonon obviously has an infinite norm . with @xmath234 , the first step resulting from the oscillatory instability occurs in a similar way as for @xmath262 : the breather acquires new frequencies and the small - amplitude sites of the phonon start moving . by instead making the interpretation that the sites with non - zero amplitude at the anticontinuous limit are defects in the zero - amplitude state , their movement can be seen as a consequence of the repulsive interaction between spatially separated , small - amplitude breathers with opposite phases observed e.g. in @xcite . we also observe , similarly as for @xmath262 , the merging of neighboring defects , but in this case their interaction with the breather will not lead to breather decay , but rather to breather growth if the original amplitude of the breather is not too large . a typical example is illustrated in fig . it can be seen by a careful inspection of fig . [ fig7 ] ( a ) how the merging of small - amplitude sites results in localized humps of larger amplitude reminiscent of small - amplitude moving breathers travelling around in the lattice . the interaction of these humps with the original breather leads to growth of the latter in a similar way as observed in @xcite . however , this growth stops when the breather amplitude has reached a critical value which is close to ( but apparently smaller than ) that corresponding to the limit value @xmath281 for small - amplitude perturbations found in sec . [ sec : phonons ] ( the latter corresponds to @xmath282 ) . the final state appears also here to be a chaotic phonobreather ; we have followed the time evolution of this kind of state for times up to @xmath283 without seeing any signs of decay . also , if the initial breather frequency is chosen above the critical value @xmath281 , we typically do not observe breather growth ; instead the mean value of the chaotic amplitude oscillations resulting from the oscillatory instability remains close to the initial amplitude . investigating the interaction between discrete nonlinear schrdinger breathers and small perturbations , we have found firstly that exciting an internal mode of the breather always leads to a slow energy transfer to the breather , i.e. , to breather growth . furthermore , we found that a dnls - breather can pump energy from a small - amplitude standing - wave phonon , provided that the phonon wave vector is smaller than the critical value @xmath226 given by eq . ( [ qc ] ) . in both cases , the mechanism for breather growth involves the higher - order generation of radiating modes . since this mechanism disappears at the threshold value @xmath281 of the breather frequency , it is impossible for a breather to grow beyond this value with the type of small - amplitude perturbations considered here . to analyze the interaction between breathers and standing - wave phonons of small but non - negligible amplitude , we considered the long - time evolution of weakly perturbed exact phonobreather solutions . the instabilities of these , originating in oscillatory instabilities of the nonlinear phonons , where shown to lead to propagating inhomogeneities whose interaction with the breather provided a mechanism for breather decay and destruction ( when the phonon wave vector @xmath262 ) or growth ( @xmath284 and @xmath285 ) . as was mentioned already in the introduction , the existence of the two conserved quantities ( [ dnlsham ] ) and ( [ norm ] ) makes the dnls equation non - generic among nonlinear lattice equations , and it is therefore necessary to investigate to what extent the results obtained in this paper apply also for klein - gordon and fpu lattices . we plan to address these questions in a forthcoming publication , but let us stress already here that the perturbational approach used here for the dnls equation needs to be modified to account for the fact that generically , the dynamics of the breather involves also higher harmonics of its fundamental frequency . moreover , the approach used in secs . [ subsec : conservation ] and [ sec : phonons ] based on the conservation laws can not be directly applied in the absence of a second conserved quantity . however , in view of the wide applicability of the dnls equation ( and in particular its appearance as a limit case of general lattice equations as mentioned in the introduction ) , we believe that the mechanisms for breather growth and destruction described in this paper are essential ingredients also for the corresponding processes in general lattice models . we thank yu . s. kivshar for giving us an early preprint of ref . @xcite , i. v. barashenkov for directing our attention to ref . @xcite , and a. m. morgante for discussions on phonobreathers and standing - wave instabilities . m. j. acknowledges a marie curie research training grant from the european community . a preliminary version of these results was presented at the conference nonlinearity 99 ( heraklion , may 10 - 14 , 1999 ) . t. dauxois , m. peyrard , and c. r. willis , physica d * 57 * , 267 ( 1992 ) ; t. dauxois , m. peyrard , and a. r. bishop , phys . rev . e * 47 * , 684 ( 1993 ) ; m. peyrard , in _ nonlinear cooperative phenomena in biological systems _ , edited by l. matsson ( world scientific , singapore , 1998 ) , p. 276 , and references therein . a. m. morgante , oral communication at the conference _ nonlinearity 99 _ , heraklion , 10 - 14 may 1999 , organized by t. geisel and g. p. tsironis ; a. m. morgante , g. kopidakis , m. johansson , and s. aubry , in preparation . b. gaididei , s. f. mingaleev , p. l. christiansen , and k. . rasmussen , phys . e * 55 * , 6141 ( 1997 ) ; k. . rasmussen , p. l. christiansen , m. johansson , yu . b. gaididei , and s. f. mingaleev , physica d * 113 * , 134 ( 1998 ) . , @xmath286 , and @xmath287 . ( a ) shows the time - evolution of the central - site intensity @xmath236 , ( b ) shows its time - average @xmath120 calculated using eq . ( [ average ] ) , while ( c ) shows the instantaneous shift of breather frequency @xmath288 . the solid line in the main figure in ( c ) is a fit using eq . ( [ shift ] ) with @xmath289 , @xmath290 , @xmath291 , and @xmath292.,title="fig:",height=332 ] , @xmath286 , and @xmath287 . ( a ) shows the time - evolution of the central - site intensity @xmath236 , ( b ) shows its time - average @xmath120 calculated using eq . ( [ average ] ) , while ( c ) shows the instantaneous shift of breather frequency @xmath288 . the solid line in the main figure in ( c ) is a fit using eq . ( [ shift ] ) with @xmath289 , @xmath290 , @xmath291 , and @xmath292.,title="fig:",height=332 ] , @xmath286 , and @xmath287 . ( a ) shows the time - evolution of the central - site intensity @xmath236 , ( b ) shows its time - average @xmath120 calculated using eq . ( [ average ] ) , while ( c ) shows the instantaneous shift of breather frequency @xmath288 . the solid line in the main figure in ( c ) is a fit using eq . ( [ shift ] ) with @xmath289 , @xmath290 , @xmath291 , and @xmath292.,title="fig:",height=332 ] , @xmath293 , and @xmath287 . ( a ) shows the time - evolution of @xmath294 , where @xmath136 is the central site of the breather , ( b ) ( main figure ) shows the instantaneous shift of breather frequency @xmath288 , and inset in ( b ) shows the time - average @xmath120.,title="fig:",height=521 ] , @xmath293 , and @xmath287 . ( a ) shows the time - evolution of @xmath294 , where @xmath136 is the central site of the breather , ( b ) ( main figure ) shows the instantaneous shift of breather frequency @xmath288 , and inset in ( b ) shows the time - average @xmath120.,title="fig:",height=521 ] . parameters are @xmath295 , @xmath296 ( @xmath297 ) , @xmath298 , and @xmath287 . ( a ) shows the time - evolution of the central - site intensity @xmath236 , ( b ) ( main figure ) shows the total norm contained in a region of 120 sites around the breather , and inset in ( b ) shows the time - average @xmath299.,title="fig:",height=498 ] . parameters are @xmath295 , @xmath296 ( @xmath297 ) , @xmath298 , and @xmath287 . ( a ) shows the time - evolution of the central - site intensity @xmath236 , ( b ) ( main figure ) shows the total norm contained in a region of 120 sites around the breather , and inset in ( b ) shows the time - average @xmath299.,title="fig:",height=498 ] . parameters are @xmath295 , @xmath287 , and in ( a ) , ( b ) @xmath300 ( @xmath301 ) and @xmath302 , resp . in ( c ) , ( d ) @xmath303 ( @xmath304 ) and @xmath298 . ( a ) , ( c ) show the time - evolution of the central - site intensity @xmath236 , while ( b ) , ( d ) show its time - average.,title="fig:",height=302 ] . parameters are @xmath295 , @xmath287 , and in ( a ) , ( b ) @xmath300 ( @xmath301 ) and @xmath302 , resp . in ( c ) , ( d ) @xmath303 ( @xmath304 ) and @xmath298 . ( a ) , ( c ) show the time - evolution of the central - site intensity @xmath236 , while ( b ) , ( d ) show its time - average.,title="fig:",height=302 ] . parameters are @xmath295 , @xmath287 , and in ( a ) , ( b ) @xmath300 ( @xmath301 ) and @xmath302 , resp . in ( c ) , ( d ) @xmath303 ( @xmath304 ) and @xmath298 . ( a ) , ( c ) show the time - evolution of the central - site intensity @xmath236 , while ( b ) , ( d ) show its time - average.,title="fig:",height=302 ] . parameters are @xmath295 , @xmath287 , and in ( a ) , ( b ) @xmath300 ( @xmath301 ) and @xmath302 , resp . in ( c ) , ( d ) @xmath303 ( @xmath304 ) and @xmath298 . ( a ) , ( c ) show the time - evolution of the central - site intensity @xmath236 , while ( b ) , ( d ) show its time - average.,title="fig:",height=302 ] ) with @xmath263 for a system of 120 sites ( periodic boundary conditions ) , perturbed only by the numerical truncation errors . the breather frequency is @xmath305 , phonon frequency @xmath306 ( phonon amplitude @xmath307 ) , and @xmath287 . ( a ) shows @xmath308 , while ( b ) ( main figure ) shows the intensity of the breather central site @xmath236 . inset in ( b ) shows the inverse participation number @xmath309 , which gives a qualitative measure of the degree of localization.,title="fig:",height=476 ] ) with @xmath263 for a system of 120 sites ( periodic boundary conditions ) , perturbed only by the numerical truncation errors . the breather frequency is @xmath305 , phonon frequency @xmath306 ( phonon amplitude @xmath307 ) , and @xmath287 . ( a ) shows @xmath308 , while ( b ) ( main figure ) shows the intensity of the breather central site @xmath236 . inset in ( b ) shows the inverse participation number @xmath309 , which gives a qualitative measure of the degree of localization.,title="fig:",height=476 ] ) with @xmath270 for a system of 120 sites ( periodic boundary conditions ) , perturbed only by the numerical truncation errors . the breather frequency is @xmath310 , phonon frequency @xmath311 ( phonon amplitude @xmath312 ) , and @xmath287 . ( a ) shows @xmath308 , while ( b ) shows the intensity of the breather central site @xmath236.,title="fig:",height=519 ] ) with @xmath270 for a system of 120 sites ( periodic boundary conditions ) , perturbed only by the numerical truncation errors . the breather frequency is @xmath310 , phonon frequency @xmath311 ( phonon amplitude @xmath312 ) , and @xmath287 . ( a ) shows @xmath308 , while ( b ) shows the intensity of the breather central site @xmath236.,title="fig:",height=519 ]

after the effect of spin transfer torque ( stt ) in thin magnetic films had been predicted @xcite and then experimentally proven @xcite , it was generally believed that current - controlled spin valve devices would replace soon the memory cells operated by external magnetic field . such a technological progress , if realized , would certainly offer higher data storage density and faster manipulation with the information stored on a hard drive memory . however , it became clear soon that some important issues must be solved before devices based on spin torque could be used in practice . the most important is the reduction of current density needed for magnetic excitation ( switching ) in thin films , as well as enhancement of switching efficiency and thermal stability . some progress has been made by using more complex spin valve structures and/or various subtle switching schemes based on optimized current and field pulses @xcite . a significant enhancement of thermal stability can be achieved by replacing a simple free layer ( single homogeneous layer ) with a system of two magnetic films separated by a thin nonmagnetic spacer , known as composite free layer ( cfl ) . the spacer layer is usually thin so there is a strong rkky exchange coupling between magnetic layers @xcite . in practice , antiferromagnetic configuration is preferred as it reduces the overall magnetic moment of the cfl structure and makes the system less vulnerable to external magnetic fields and thermal agitation . when the antiferromagnetically coupled magnetic layers are identical , we call the structure synthetic antiferromagnet ( syaf ) . if they are different , then the cfl has uncompensated magnetic moment and such a system will be referred to as synthetic ferrimagnet ( syf ) . current and/or field induced dynamics of cfls is currently a subject of both experimental and theoretical investigations @xcite . switching scheme of syaf by magnetic field pulses has been proposed in ref . [ ] , and then the possibility of current - induced switching of syaf was demonstrated experimentally @xcite . in turn , the possibility of critical current reduction has been shown for a cfl with ferromagnetically coupled magnetic layers @xcite . however , the reduction of critical current in the case of antiferromagnetically coupled cfls still remains an open problem . in a recent numerical study on switching a syaf free layer @xcite it has been shown that the corresponding critical current in most cases is higher than the current required for switching of a simple free layer , and only in a narrow range of relevant parameters ( exchange coupling , layer thickness , etc . ) the critical current is reduced . hence , proper understanding of current - induced dynamics of cfls is highly required . we also note , that cfl can be used as a polarizer , too . indeed , it has been shown recently @xcite that syaf used as a reference layer ( with magnetic moment fixed fixed to adjacent antiferromangentic layer due to exchange anisotropy ) might be excited due to dynamical coupling @xcite with a simple sensing layer @xcite . the main objective of this paper is to study current - induced dynamics of a cfl with antiferomagnetic rkky coupling in metallic spin valve pillars . we consider a system @xmath0/@xmath1/@xmath2/@xmath3/@xmath4/@xmath5 , shown in fig . [ fig : scheme ] , where @xmath0 is an antiferromagnetic layer ( used to bias magnetization of the reference magnetic layer @xmath1 ) , @xmath3 and @xmath5 are two magnetic layers , while @xmath2 and @xmath4 are non - magnetic spacers . the part @xmath3/@xmath4/@xmath5 constitutes the cfl structure with antiferromagnetic interlayer exchange coupling . we examine current - induced dynamics of both syaf and syf free layers . these two structures differ only in the thickness of @xmath3 layer , while rkky coupling and other pillar parameters remain the same . we assume that spin - dependent electron transport is diffusive in nature , and employ the model described in refs . [ ] and [ ] . an important advantage of this model is the fact that it enables calculating spin current components and spin accumulation consistently in all magnetic and nonmagnetic layers , as well as current - induced torques exerted on all magnetic components . the torques acting at the internal interfaces of cfl introduce additional dynamical coupling between the corresponding magnetic layers . consequently , the magnetic dynamics of cfl has been modelled by two coupled macrospins and described in terms of landau - lifshitz - gilbert ( llg ) equation . in addition , we derive some analytical expressions for critical currents from the stability conditions of linearized llg equation in the static points @xcite , and discuss results in the context of numerical simulations . the paper is organized as follows . in section 2 we describe the assumed models for spin dynamics and stt calculations . in section 3 we analyze stt acting on cfl and present results from numerical simulations on current - induced switching and magnetic dynamics . some additional information on stt calculation can be found in the appendix . critical currents are derived and discussed in section 4 . finally , summary and general conclusions are in section 5 . scheme of the spin valve pillar structure with a composite free layer . ] in the macrospin approximation , magnetization dynamics of the cfl is described by two coupled landau - lifshitz - gilbert ( llg ) equations , @xmath6 for @xmath7 , where @xmath8 stands for a unit vector along the net spin moment of the @xmath9-th layer , whereas @xmath10 and @xmath11 are the effective field and current - induced torque , respectively , both acting on @xmath8 . the damping parameter @xmath12 and the saturation magnetization @xmath13 are assumed the same for both magnetic components of the cfl . furthermore , @xmath14 is the gyromagnetic ratio , @xmath15 is the vacuum permeability , and @xmath16 stands for thickness of the @xmath17 layer . the effective magnetic field for the @xmath17 layer is @xmath18 where @xmath19 and @xmath20 . in the latter equation , @xmath21 is the external magnetic field applied along the easy axis in the layers plane ( and oriented opposite to the axis @xmath22 ) , @xmath23 is the uniaxial anisotropy field ( the same for both magnetic layers ) , and @xmath24 is the self - demagnetization field of the @xmath17 layer with @xmath25 being the corresponding demagnetization tensor . similarly , @xmath26 describes the magnetostatic influence of the layers @xmath1 and @xmath27 on the layer @xmath17 , respectively . here , @xmath28 is the saturation magnetization of the layer @xmath29 , which might be generally different from @xmath13 . components of the tensors @xmath25 , @xmath30 and @xmath31 used in our simulations have been determined by the numerical method introduced by newell _ et al . _ this method was originally developed for magnetic systems with non - uniform magnetization . to implement it into a macrospin model we considered discretized magnetic layers with uniform magnetizations , calculated tensors in each cell of the layer , and then averaged them along the whole layer . since these tensors are diagonal , the demagnetization and magnetostatic fields can be expressed as @xmath32 , and @xmath33 , with @xmath34 , @xmath35 , and @xmath36 denoting the components of the vector @xmath8 ( @xmath37 ) in the coordinate system shown in fig.1 . finally , @xmath38 stands for the rkky exchange field acting on @xmath8 , which is related to the rkky coupling constant as @xmath39 @xcite . to include thermal effects we add to the effective field ( [ eq : heff ] ) a stochastic thermal field @xmath40 . for both spins its components obey the rules for gaussian random processes @xmath41 and @xmath42 , where @xmath43 and @xmath44 . here , @xmath45 is the noise amplitude , which is related to the effective temperature , @xmath46 , as @xmath47 where @xmath48 is the boltzmann constant , and @xmath49 is the volume of @xmath17 layer . in general , the current - induced torques acting on @xmath50 and @xmath51 can be expressed as a sum of their in - plane and out - of - plane components @xmath52 and @xmath53 , respectively . in a cfl structure , the layer @xmath3 is influenced by stt induced by the polarizer @xmath1 , as well as by stt due to the layer @xmath5 . in turn , the layer @xmath5 is influenced by the torques from the layer @xmath3 . hence we can write [ eqs : stt ] @xmath54\ , , \\ { { \bm \tau}_{1 \perp } } & = i\bhs_1 \times \left ( b_1^{(0 ) } \bhs_0 + b_1^{(2 ) } \bhs_2 \right)\ , , \\ { { \bm \tau}_{2 \parallel } } & = ia_2^{(1)}\ , \bhs_2 \times \left ( \bhs_2 \times \bhs_1 \right)\ , , \\ { { \bm \tau}_{2 \perp } } & = ib_2^{(1)}\ , \bhs_2 \times \bhs_1\ , , \end{aligned}\ ] ] where @xmath55 is the charge current density , which is positive when electrons flow from the layer @xmath5 towards @xmath1 ( see fig . [ fig : scheme ] ) , while the parameters @xmath56 and @xmath57 ( @xmath58 ) are independent of current @xmath55 , but generally depend on magnetic configuration . we write the current density in the spin space as @xmath59 , where @xmath60 is the particle current density ( @xmath61 ) , @xmath62 is the spin current density ( in the units of @xmath63 ) , @xmath64 is the vector of pauli matrices , and @xmath65 is a @xmath66 unit matrix . in frame of the diffusive transport model @xcite , the parameters @xmath56 and @xmath57 are given by the formulas @xmath67 where the angles @xmath68 and @xmath69 are given by @xmath70 and @xmath71 . here , @xmath72 and @xmath73 are transversal ( to @xmath50 ) components of spin current in the layer @xmath2 ( taken at the @xmath2/@xmath3 interface ) written in the local coordinate system of @xmath50 . thus , @xmath72 is parallel to the vector @xmath74 ( lying in the plane defined by @xmath75 and @xmath50 ) and @xmath73 is aligned with @xmath76 ( perpendicular to the plane defined by @xmath75 and @xmath50 ) . note , that the @xmath22-component of spin current is aligned along @xmath50 and does not contribute to stt @xcite . similarly , we define the torque amplitudes acting inside the cfl . here , @xmath77 and @xmath78 are the components of spin current in the layer @xmath4 ( taken at the n2/f1 interface ) , transversal to @xmath50 and lying respectively in the plane and perpendicularly to the plane defined by @xmath50 and @xmath51 . analogically , @xmath79 and @xmath80 are spin current components in @xmath4 transversal to @xmath51 . the relevant spin current transformations are given in appendix [ sec : transformations ] . note , that the spin current components depend linearly on @xmath55 , so the parameters @xmath56 and @xmath57 are independent of the current density @xmath55 . in this section we present results on our numerical simulations of current - induced dynamics for two metallic pillar structures including cfl with antiferromagnetic interlayer exchange coupling . as described in the introduction , the considered pillars have the general structure @xmath0/@xmath1/@xmath2/@xmath3/@xmath4/@xmath5 ( see fig.(1 ) ) . more specifically , we consider spin valves cu - irmn(10)/py(8)/cu(8)/co(@xmath81)/ru(1)/co(@xmath82 ) - cu , where the numbers in brackets stand for the layer thicknesses in nanometers . the layer py(8 ) is the permalloy polarizing layer with its magnetization fixed due to exchange coupling to irmn . in turn , co(@xmath81)/ru(1)/co(@xmath82 ) is the cfl ( @xmath3/@xmath4/@xmath5 structure ) with antiferromagnetic rkky exchange coupling _ via _ the thin ruthenium layer . the coupling constant between co layers has been assumed as @xmath83 , which is close to experimentally observed values @xcite . here , we shall analyze two different geometries of cfl . the first one is a syaf structure with @xmath84 , while the second one is a synthetic ferrimagnet ( syf ) with @xmath85 . simulations have been based on numerical integration of the two coupled llg equations ( [ eq : llg ] ) with simultaneous calculations of stt , see eq.([eqs : stt ] ) . we have assumed typical values of the relevant parameters , i.e. , the damping parameter has been set to @xmath86 , while the uniaxial anisotropy field @xmath87 in both magnetic layers of the cfl . in turn , saturation magnetization of cobalt has been assumed as @xmath88 , and for permalloy @xmath89 . the demagnetization field and magnetostatic interaction of magnetic layers have been calculated for layers of elliptical cross - section , with the major and minor axes equal to @xmath90 and @xmath91 , respectively . for both structures under consideration we have analyzed the current - induced dynamics as a function of current density and external magnetic field . the results have been presented in the form of diagrams displaying time - averaged values of the pillar resistance . numerical integration of eq.([eq : llg ] ) has been performed using corrector predictor heun scheme , and the results have been verified for integration steps in the range from @xmath92 up to @xmath93 . the stt components acting on cfl spins have been calculated at each integration step from the spin currents , which have been numerically calculated from the appropriate boundary conditions @xcite . similarly , resistance of the studied pillars has been calculated from spin accumulation in frame of the model used also for the stt description ( for details see also ref . ) . let us analyze first the angular dependence of stt components in the structures under consideration . although the thicknesses of magnetic layers in the studied syaf and syf structures are different , the angular dependence of stt components as well as their amplitudes are very similar . thus , the analysis of stt in syaf applies also qualitatively to the studied syf free layer . ( color online ) angular dependence of the cartesian components of stt , in the units of @xmath94 , acting at n / f interfaces , when magnetization of the syaf structure is rotated rigidly with both @xmath50 and @xmath51 remaining in the corresponding layer planes . here , @xmath95 is an angle between @xmath50 and @xmath96 . @xmath51 is tilted away from the antiparallel configuration with @xmath50 by an angle of @xmath97.,width=302 ] first , we analyze stt components in the case when syaf is rotated as a rigid structure , i.e. the antiparallel configuration of @xmath50 and @xmath51 is maintained . to have a nonzero torque between @xmath3 and @xmath5 layers , @xmath51 has been tilted away from the antiparallel configuration by an angle of @xmath97 . figure [ fig : stt1 ] shows all three cartesian components ( see fig.1 for definition of the coordinate system ) of stt acting at n / f interfaces as a function of the angle @xmath95 between @xmath96 and @xmath50 . while the @xmath98 and @xmath22-components are in the plane of the layers ( the spins of cfl are rotated in the layer plane ) , the component @xmath99 is normal to the layer plane . however , @xmath100 remains negligible at all interfaces of the cfl . the stt acting at @xmath2/@xmath3 reveals a standard ( non - wavy @xcite ) angular dependence , and vanishes when @xmath50 is collinear with @xmath75 . its amplitude is comparable to stt in standard spin valves with a simple free layer . the stt at @xmath3/@xmath4 and @xmath4/@xmath5 interfaces also depends on the angle @xmath95 . however , they are about two order of magnitude smaller , which is a consequence of a small angle ( @xmath97 ) assumed between @xmath50 and @xmath51 . as will be shown in the following , cfl is usually not switched as a rigid structure , but generally forms a configuration which deviates from the antiparallel one . figure [ fig : stt2 ] shows how the stt components at the @xmath3/@xmath4 and @xmath4/@xmath5 interfaces vary when @xmath51 is rotated from @xmath96 by an angle @xmath101 , while @xmath50 remains fixed and is parallel to @xmath102 . in such a case , the torque acting at @xmath2/@xmath3 interface remains zero , as @xmath50 stays collinear to @xmath75 . as before , the out - of - plane components are also negligible in comparison to the in - plane ones . the in - plane components of stt reveal standard angular dependence at both interfaces . the amplitude of stt at the internal interfaces of cfl is comparable to that acting at the @xmath2/@xmath3 interface in the case of noncollinear configuration of @xmath75 and @xmath50 , when @xmath51 is fixed in the direction antiparallel to @xmath75 . ( color online ) angular dependence of the cartesian stt components in the units of @xmath103 acting on f1/n2 and n2/f1 interfaces when @xmath104 and @xmath51 is rotated from @xmath96 in the layer s plane by an angle @xmath101.,width=302 ] when @xmath50 is noncollinear to @xmath75 , the spin accumulation in @xmath2 layer increases and consequently the amplitude of stt at @xmath3/@xmath4 and @xmath4/@xmath5 decreases . in turn , when @xmath50 is antiparallel to @xmath75 , the stt inside the cfl structure is reduced by more than a factor of 2 . nonetheless , the stt acting at the internal interfaces of the studied cfl layers might have a significant effect on their current - induced dynamics and switching process , provided the magnetic configuration of cfl might deviate remarkably from its initial antiparallel configuration . first , we examine dynamics of the syaf free layer . from symmetry we have @xmath105 , and we have set @xmath106 , which corresponds to @xmath107 . we have performed a number of independent numerical simulations modelling syaf dynamics induced by constant current and constant in - plane external magnetic field . the latter is assumed to be smaller than the critical field for transition to spin - flop phase of syaf . accordingly , each simulation started from an initial state close to @xmath108 . to have a non - zero initial stt for @xmath50 , both spins of the syaf have been tilted by @xmath97 in the layer plane so that they remained collinear . from the results of numerical simulations we have constructed a map of time - averaged resistance , shown in fig . [ fig : saf_dynam1](a ) . the resistance has been averaged in the time interval of @xmath109 following initial @xmath110 equilibration time of the dynamics . the diagram shows only that part of the resistance , which depends on magnetic configuration , and hence varies with cfl dynamics @xcite . the constant part of resistance , due to bulk and interfacial resistances of the studied structure , has been calculated to be as large as @xmath111 . ( color online ) ( a ) averaged resistance of cu - irmn(10)/py(8)/cu(8)/co(2)/cu(1)/co(2)-cu spin valve pillar with a saf free layer as a function of current density and applied magnetic field . examples of switching processes at @xmath112 and @xmath113 ( b e ) and @xmath114 ( f i ) . figures ( b ) and ( f ) show dynamics of @xmath22-components of both spin moments , ( c ) and ( g ) present the overal magnetization of the free layer , ( d ) and ( h ) show the corresponding variation of pillar resistance , ( e ) and ( i ) show the spin trajectories of @xmath50 ( red solid line ) and @xmath51 ( black dashed line ) in the time interval from @xmath115 to @xmath116 , where switching takes place . , width=302 ] for the assumed initial configuration , magnetic dynamics has been observed only for negative current density . when the current is small , no dynamics is observed since the spin motion is damped into the closest collinear state ( @xmath108 , marked as @xmath117 ) of high resistance . after exceeding a certain threshold value of current density , there is a drop in the averaged resistance , which indicates current induced dynamics of the syaf free layer . figures [ fig : saf_dynam1](b ) and ( f ) show that this drop is associated with switching of the whole syaf structure into an oposite state ( @xmath118 , marked as @xmath119 ) from fig.4(a ) follows that the threshold current for dynamics onset markedly depends on the applied field and reaches maximum at a certain value of @xmath21 , @xmath120 . furthermore , it appears that mechanisms of the switching process for @xmath121 and @xmath122 are qualitatively different . to distinguish these two mechanisms , we present in figs . [ fig : saf_dynam1](b i ) basic characteristics of switching , calculated for @xmath123 and for fields @xmath113 , which is below @xmath124 [ figs . [ fig : saf_dynam1](b e ) ] , and @xmath125 , which lies above @xmath124 [ figs . [ fig : saf_dynam1](f i ) ] . figs . [ fig : saf_dynam1](b ) and ( f ) present time evolution of the z - components of both spins . to better understand the syaf dynamics , in figs . [ fig : saf_dynam1](c ) and ( g ) we plotted the amplitude of overal syaf magnetization , defined as @xmath126 . this parameter vanishes for antiparallel alignment of both spins of cfl , but becomes nonzero when the configuration deviates from the antiparallel one . magnetization of syaf is also a measure of the cfl coupling to external magnetic field . futhermore , figs . [ fig : saf_dynam1](d ) and ( h ) show the corresponding time variation of the resistance , @xmath127 , which might be directly extracted from experimental measurements as well . in addition , in figs . [ fig : saf_dynam1](e ) and ( i ) we show the trajectories of @xmath50 and @xmath51 in the real space taken from the time interval from @xmath128 to @xmath129 . in addition , from figs . [ fig : saf_dynam1](a ) it has been found that the point where the threshold current reaches its maximum is located at @xmath130 , which indicates its relation to magnetostatic interaction of f2 and fixed polarizer . this also has been confirmed by analogical simulations disregarding the magnetostatic coupling between magnetic layers , which resulted in similar diagram , but with @xmath131 ( not shown ) . this fact significantly facilitates understanding the mechanism of syaf switching . the initial configuration assumed above was @xmath132 with @xmath102 ( @xmath133 ) . when the magnitude of current density is large enough and @xmath134 , orientation of @xmath50 becomes unstable and @xmath50 starts to precess with small angle around @xmath135 . initial precession of @xmath50 induces precession of @xmath51 mainly via the rkky coupling . generally , response to the exchange field is slower than current - induced dynamics . therefore , a difference in precession phase of @xmath51 and @xmath50 appears , and configuration of syaf deviates from the initial antiparallel one . this in turn enhances the stt acting on f2 , which tends to switch @xmath51 . its amplitude , however , is small in comparison to the strong rkky coupling . further scenario of the dynamics depends then on the external magnetic field . when @xmath121 [ figs . [ fig : saf_dynam1](b e ) ] the zeeman energy of @xmath51 has a maximum in the initial state and external magnetic field tends to switch @xmath51 to the opposite orientation . competition between the torques acting on syaf results in out - of - plane precessions of both spins . after several precessions @xmath50 reaches the opposite static state , which is stable due to stt . in turn , @xmath51 is only slightly affected by stt , and its dynamics is damped in the external magnetic and rkky exchange fields . in contrast , when @xmath122 [ figs . [ fig : saf_dynam1](f i ) ] , zeeman energy of f2 has a local minimum in the initial state , which stabilizes @xmath51 . therefore , in a certain range of current density , syaf does not switch but remains in self - sustained coherent in - plane precessions ( red area in the upper part of fig . [ fig : saf_dynam1](a ) ) . for a sufficient current density , the syaf structure becomes destabilized and the precessional angle increases until the spins pass the @xmath136-plane . consequently , the precessional angle decreases and spins of the syaf are stabilized in the opposite state ( @xmath137 ) . moreover , as shown in fig . [ fig : saf_dynam1](c ) , the switching process for @xmath121 is connected with high distortion of syaf configuration , where @xmath138 in a certain point reaches its maximum value ( corresponding to parallel orientation of both spins ) . contrary , the @xmath138 remains small for @xmath122 [ figs . [ fig : saf_dynam1](g ) ] , and the effective magnetic moment of the free layer stays smaller than magnetic moment of a single layer . this might play an important role in applications of spin - torque devices based on cfls . the two switching mechanisms described above dominate the current - induced dynamics when the current density is close to the dynamics threshold . for higher current densities , the nonlinearities in syaf dynamics become more pronounced , which results in bistable behavior of the dynamics , especially for @xmath121 and @xmath139 . in that region , the number of out - of - plane precessions before syaf switching increases with the current density . however , their precessional angle increases in time and consequently @xmath50 might reach an out - of - plane static point slightly tilted away from the @xmath140 direction while @xmath141 remains in the layers plane . the out - of - plane static states ( marked as @xmath142 ) have small resistance and appear as dark red spots in the diagram shown in fig . [ fig : saf_dynam1](a ) . in addition , from the analysis of the dispersion of pillar resistance ( not shown ) one finds that except of a narrow region close to the dynamics threshold with persistent in - plane precessions , no significant steady - state dynamics of syaf element appears . as will be shown below , such a behavior might be observed when cfl becomes asymmetric ( syf free layer ) . let us study now spin valve with syf as a free layer , assuming @xmath143 and @xmath144 . accordingly , @xmath145 remains @xmath146 while @xmath147 is reduced to @xmath148 . as in the case of syaf , from the averaged time - dependent part of the pillar resistance we have constructed a diagram presenting current - induced dynamics , see fig . [ fig : sf_dynam1](a ) . the static part of resistance is now @xmath149 . ( color online ) ( a ) averaged resistance of cu - irmn(10)/py(8)/cu(8)/co(4)/cu(1)/co(2)-cu spin valve pillar with a syf free layer , presented as a function of current density and applied magnetic field . examples of current - induced dynamics for @xmath150 and @xmath151 ( b e ) and @xmath113 ( f i ) . panels ( b ) and ( f ) show dynamics of @xmath22-components of @xmath50 and @xmath51 , ( c ) and ( g ) present the overal magnetization of the free layer , ( d ) and ( h ) show the corresponding variation of pillar resistance , ( e ) and ( i ) show spin trajectories of @xmath50 ( red solid line ) and @xmath51 ( black dashed line ) taken from a time interval as large as @xmath152 after @xmath153 of initial equilibration.,width=302 ] the diagram has some features similar to those studied in the previous subsection . however , the maximum critical current is shifted towards negative values of @xmath21 , even if magnetostatic interaction between magnetic layers is neglected . this asymmetry is caused by the difference in exchange and demagnetization fields acting on layers @xmath3 and @xmath5 . moreover , this difference leads to more complex dynamics of the cfl s spins than that in the case of syaf . generally , there are several dynamic regimes to be distinguished . the first one is the region of switching from @xmath117 configuration to the opposite one , @xmath119 , which is located at largest values of @xmath21 in the diagram . mechanism of the switching is similar to that of syaf shown in figs . [ fig : saf_dynam1](f i ) , where cfl changes its configuration just _ via _ in - plane precessional states with a small value of @xmath138 ( weak distortion of the antiparallel alignment of @xmath50 and @xmath51 ) . furthermore , the darker area above @xmath154 indicates one of the possible self - sustained dynamic regimes of syf , i.e. the in - plane precessions ( ipp ) ; see figs . [ fig : sf_dynam1](b e ) . this precessional regime starts directly after the syf switching , and @xmath50 and @xmath51 precess around @xmath96 and @xmath135 , respectively . due to different effective fields in @xmath3 and @xmath5 , and energy gains due to stt , the spins precess with different precessional angles [ fig . [ fig : sf_dynam1](e ) ] and consequently different frequencies . because of the strong interlayer coupling and spin transfer between the layers , amplitudes of their precessions are periodically modulated in time . this modulation appears also in the time dependence of pillar resistance . conversely , below @xmath154 the dynamic is dominated by large angle out - of - plane precessions ( opp ) of both spins , as shown in figs . [ fig : sf_dynam1](f i ) . this dynamic state is connected with a strong distortion of the antiparallel cfl configuration , i.e. large value of @xmath138 , and large variation of the resistance . from fig . [ fig : sf_dynam1](i ) one can see that trajectories of @xmath50 and @xmath51 are rather complicated including both ipp and opp regimes with dominant opp . from the analysis of current - induced dynamics we found that self - sustained dynamics in structures with syf free layer is much richer than that in systems with syaf free layer [ see figs . [ fig : sf_dynam1](b i ) ] . therefore , in this section we restrict ourselves to dynamic regimes of the syf free layer only . more specifically , we shall examine the power spectral density ( psd ) as a function of current density and external magnetic field . in the simulation we started from @xmath155 and changed current density in steps @xmath156 at a fixed applied field . as before , to protect the syf dynamics from collapsing into collinear static state , we assumed small thermal fluctuations corresponding to @xmath157 . at each step we simulated the dynamics of coupled cfl s spins and calculated psd . as in ref . , we assumed that the input current is split between a load with resistance @xmath158 and pillar with resistance @xmath159 . hence , voltage on the pillar has been calculated as @xmath160 $ ] , where we assumed @xmath161 , and @xmath162 is the cross - section of the pillar ( ellipsoid with the major and minor axes equal to @xmath163 and @xmath164 nm , respectively ) . then , at a given @xmath55 we calculated voltage in the frequency domain , @xmath165 , using fast fourier transformation over the period @xmath166 following the equilibration time of @xmath167 . the power spectral density has been defined as @xmath168 , where @xmath169 . power spectral density calculated for the spin valve with syf free layer at @xmath157 and @xmath113 ( a ) and @xmath170 ( b ) . ( c ) and ( d ) show steady time evolution of spins @xmath22-components in a time window of @xmath171 after the equilibration at @xmath113 and @xmath172 . ( e ) and ( f ) show steady time evolution of the time - dependent part of spin valve resistance in a time window of @xmath173 after the equilibration at @xmath113 and @xmath174 ( both @xmath50 and @xmath51 precess out of layer plane ) and @xmath175 ( @xmath51 performs out - of - plane precessions while @xmath50 precesses in the layer s plane ) , respectively . panels ( g ) and ( h ) depict trajectories of @xmath50 ( red solid line ) and @xmath51 ( black dashed line ) corresponding to resistance oscillations ( e ) and ( f ) , respectivelly . ] figures [ fig : psd](a ) and ( b ) show psd calculated at @xmath113 and @xmath170 , respectively . the former case corresponds to that part of the diagram in fig . [ fig : sf_dynam1](a ) , which includes opp modes , while in the latter case we observed ipp only . let us analyze first the situation in fig . [ fig : psd](a ) . when current passes through the corresponding threshold value , both spins start precessing in the layers plane , similarly as shown in fig . [ fig : sf_dynam1](b ) . apart from the main peak in the psd at @xmath176 , two additional minor peaks close to @xmath177 are visible . we attribute them to the oscillations of precessional amplitudes of both spins . with increasing amplitude of the current density , the precessional angles of both spins increase and their precessional frequencies slightly decrease . moreover , with increasing current the frequencies of the minor peaks become closer and closer , until they finally coincide . at this point the psd becomes widely distributed along the whole range of observed frequencies , which is an evidence of noisy variation of the resistance . an example of spin dynamics in this region is shown in figs . [ fig : psd](c ) and ( d ) which have been taken in a time window as large as @xmath171 after the equilibration period for @xmath113 and @xmath178 [ within the broad feature of psd in fig . [ fig : psd](a ) ] . firstly , the figures show that @xmath51 starts to perform out - of - plane precessions as a result of the competition between stt and rkky coupling . secondly , one can note thermally activated random transitions of @xmath50 between opp and ipp modes . these random transitions modify opp precessions of @xmath51 as well . simultaneous dynamics of both spins causes chaotic variation of spin valve resistance and broadens the psd . the quasi - chaotic feature of the spin dynamics in this range of current densities can be seen also on the spin trajectories , which cover almost the whole sphere ( not shown ) . further increase in current density leads to stabilization of the opp mode of @xmath50 . hence spin valve resistance becomes more periodic [ see fig . [ fig : psd](e ) ] and psd reveals a narrow peak again . since both spins perform rather complicated dynamics including ipp but dominated by opp regime [ see fig . [ fig : psd](g ) ] , we observe a blue - shift in psd with current , which is connected with a decrease in the precessional angles . however , at a certain value of @xmath55 we notice an abrupt drop in the peak s frequency . at this current density the stt acting on the left interface of layer @xmath3 starts to dominate the dynamics of @xmath50 and enables only small angle ipps along the @xmath75 direction , which modifies the trajectory of @xmath51 . @xmath51 still remains in the opp regime [ see fig . [ fig : psd](h ) ] and hence the blue - shift with current appears . the fact that ipp of @xmath50 still influence the dynamics of the whole syf is also shown in fig . [ fig : psd](f ) , which presents the dynamic part of the spin valve resistance at @xmath179 and @xmath113 . as a result of ipps of @xmath50 , amplitude of the resistance varies periodically . in addition , comparison of figs . [ fig : psd](e ) and ( f ) shows that the simultaneous opps of both spins lead to stronger variation of the resistance than in the case when the layers are in the ipp state . contrary , at @xmath180 one observes only ipp modes of both spins similar to those shown in fig . [ fig : sf_dynam1](e ) . the in - plane precessional angle increases with current density and hence the peak frequency in psd decreases and becomes broader . in real systems , however , one might expect the peaks narrower than those obtained in the macrospin simulations , as observed in standard spin valves with a simple free layer @xcite . first , we derive some approximate expressions for critical current density needed to induce dynamics of cfl , derived from linearized llg equation . in metallic structures , the out - of - plane torque components are generally much smaller than the in - plane ones , and therefore will be omitted in the analytical considerations of this section ( @xmath181 ) . the coupled llg equations in spherical coordinates can be then written as @xmath182 where @xmath183 is a 4-dimensional column vector which describes spin orientation in both layers constituting the cfl , and @xmath184 stands for the torque components , @xmath185 and @xmath186 , with @xmath187 and @xmath188 denoting unit vectors in local spherical coordinates associated with @xmath8 . in turn , the @xmath189 matrix @xmath190 takes the form @xmath191 static points of the cfl dynamics have to satisfy @xmath192 and @xmath193 for both @xmath194 and @xmath195 . these conditions are obeyed in all collinear configurations , i.e. @xmath196 . additional four trivial static points can be found in the out - of - plane configurations with @xmath197 and @xmath198 . following ref . we linearize eq . ( [ eq : llg_spherical ] ) by expanding @xmath199 into a series around the static points , which leads to @xmath200 where @xmath201 is a jacobian matrix of @xmath202 components . the matrix product @xmath203 defines here the dynamic matrix @xmath204 . this matrix allows one to study stability of cfl s spins in their static points . if @xmath205 is negative , the static point is stable , otherwise it is unstable . hence , the condition for critical current is @xcite @xmath206 . to obtain threshold current for dynamics onset of individual spins in the cfl , we assume first that one of the spins is fixed in its initial position and investigate stability of the second spin . the dynamic matrix @xmath207 becomes then reduced to a @xmath208 matrix . considering initial position of saf with @xmath108 ( i.e. @xmath209 and @xmath210 ) , marked as @xmath117 , and polarizer @xmath211 , the stability condition leads to the following critical currents @xmath212 and @xmath213 for @xmath50 and @xmath51 , respectively : @xmath214\ , , \label{eq : critcurr1_du}\ ] ] with @xmath215 , and @xmath216\ , , \label{eq : critcurr2_du}\ ] ] with @xmath217 . in both above expressions @xmath218 , @xmath219 , and @xmath220 are taken in the considered static point , while the demagnetization field for the @xmath9-th layer is given by @xmath221 analogically , one can derive similar formulas for critical currents in the opposite @xmath222 magnetic configuration of the cfl . now we relax the assumption that one of the spins is fixed , and consider both spins of the cfl as free . then , we calculate the trace of the whole @xmath189 matrix , which leads to the following expression for critical current destabilizing the whole cfl : @xmath223\ , , \end{split}\ ] ] where @xmath224 . since the spins of cfl are antiparallel in the considered static point , @xmath225 is independent of external magnetic field . the above equation describes the critical current at which the cfl is destabilized as a rigid structure ( unaffected by external magnetic field along the @xmath22-axis ) . numerical calculations presented below show that critical current is usually smaller than that given by eq.(12 ) . apparently , as shown by numerical simulations , there is a phase shift in initial precessions of @xmath50 and @xmath51 . such a phase shift slightly perturbs initial antiparallel configuration and might reduce the critical current for the dynamics onset . similar formula also holds for the opposite configuration ( @xmath119 ) , where the critical current is given by @xmath226\ , , \end{split}\ ] ] with @xmath227 . now we discuss the theoretical results on critical currents in the context of those following from numerical simulations . let us consider first the critical currents for individual spins of the syaf free layer , assuming that the second spin remains stable in its initial position , eqs . ( [ eq : critcurr1_du ] ) and ( [ eq : critcurr2_du ] ) . the corresponding results obtained from the formula derived above are presented in table [ tab : crit ] , where we have omitted a weak dependence on @xmath21 . for the studied structure with syaf free layer , @xmath212 is negative while @xmath213 is positive . from our analysis follows , that the current density at which dynamics appears in the simulations ( fig . [ fig : saf_dynam1](a ) ) is higher than that given by @xmath212 . however , we checked numerically that the critical value @xmath212 agrees with the critical current obtained from simulations when assuming @xmath50 as free and fixing @xmath51 along @xmath96 . .[tab : crit ] critical current densities in the units of @xmath228 , calculated according to equations ( [ eq : critcurr1_du ] ) , ( [ eq : critcurr2_du ] ) , and ( [ eq : iccfl ] ) for both syaf and syf free layers . [ cols="<,^,^,^,^ " , ] following the above discussion of the cfl dynamics , one can understand the shift of the threshold current as follows . initially , when the current density exceeds @xmath212 , @xmath50 becomes destabilized . then , @xmath51 responses to the initial dynamics of @xmath50 with similar coherent precession . however , @xmath51 should still be stable in its initial position at this current density and common precessions of the two coupled spin moments damps the initial dynamics . accordingly , syaf ends up in the closest static state ( @xmath117 ) . however , as the current density increases , the initial precessions of @xmath50 become more pronounced , which in turn means that the initial antiparallel configuration becomes distorted and @xmath51 becomes destabilized . this results in coupled dynamics of both spins and finally leads to switching of the syaf structure . on the other hand , we have also calculated the critical current for the whole syaf structure according to eq . ( [ eq : iccfl ] ) , and for the given structure we got @xmath229 shown in fig . [ fig : saf_dynam1](a ) by the dashed vertical line ( see also table i ) . equation ( [ eq : iccfl ] ) describes stability of the whole cfl , and since the interlayer coupling is strong , @xmath229 corresponds to the current density at which both spins become destabilized simultaneously preserving their antiparallel orientation . as can be seen in fig . [ fig : saf_dynam1](a ) , this is the maximum threshold current density for current - induced dynamics . because the rigid structure consisting of two antiparallel spins is not influenced by an external homogeneous magnetic field , there is no dependence of @xmath229 on @xmath21 . nevertheless , from our numerical simulation follows that the threshold current for the syaf dynamics , @xmath230 , obeys the condition @xmath231 , provided that @xmath232 or ( as in our case ) @xmath213 has different sign . when the syaf is in the @xmath119 configuration , the spin accumulation and spin current are different from those in the @xmath117 configuration ( at the same voltage ) . this in turn leads to different spin torques , which is the reason why the critical currents destabilizing @xmath119 state are different from those for @xmath117 , as shown in table [ tab : crit ] . from the critical currents one can expect relatively symmetric hysteresis with applied current in structures with syaf . in contrast , @xmath233 for the syf is negative , similarly as @xmath229 , but it is significantly larger , which indicates lack of hysteresis . to compare switching of the syaf and syf free layers from the @xmath117 to @xmath119 configurations with the opposite one ( @xmath119 to @xmath117 ) , we have simulated dynamics of the corresponding cfls assuming @xmath234 and varying current density . the simulations have been performed in the quasistatic regime , i.e. , for each value of current density the spin dynamics was first equilibrated for @xmath110 and then averaged values of spin components and pillar resistance were calculated from the data taken for the next @xmath109 of dynamics . in order to prevent the system from collapsing into a static state with zero torque , we have included a thermal stochastic field corresponding to @xmath157 [ see eq . ( [ eq : dth ] ) ] . starting from @xmath155 and going first towards negative currents we have constructed the current dependence of the averaged resistance and related @xmath22-components of both spins , as shown in fig . [ fig : hyster ] . hysteresis loops of the resistance for the studied pillars with syaf ( a ) and syf ( d ) free layers . panel ( b ) and ( c ) depict spin dynamics of @xmath50 and @xmath51 in syaf , respectivelly , corresponding to resistance loop ( a ) . panels ( e ) and ( f ) show dynamics of @xmath50 and @xmath51 in syf , respectivelly , corresponding to resistance loop ( d ) . the initial point of each hysteresis loop is marked with a dot . the arrows indicate direction of the current change . figures ( g ) and ( h ) correspond to the upper parts of ( a ) and ( d ) , in which however the effects due to magnetostatic field of the reference layer to the cfl spins have been omitted . ] for both syaf [ figs . [ fig : hyster](a c ) ] and syf [ figs . [ fig : hyster](d f ) ] free layers , one can see relatively symmetric hysteresis with the current density . in both cases direct switching from @xmath117 to @xmath119 state occurs at a current density comparable to @xmath229 . in contrast , in the case of syf free layer , the second transition ( @xmath119 to @xmath117 ) appears at a current density which is very different from that predicted by the linearized llg model . moreover , in both cases switching from @xmath119 to @xmath117 state does not appear directly , but through some precessional states . more precisely , as the positive current density increases , both spins start precessing in the layers plane prior to switching . the in - plane precessions are connected with a significant drop in the resistance and with a reduction of the @xmath235-components . the range of ipp regime is particularly large in the case of syf . from the analysis of spins trajectories one may conclude that the angle of ipps increases with increasing current density , and after exceeding a certain threshold angle cfl switches to the @xmath117 configuration . the other factor giving rise to the the difference in switching from @xmath119 to @xmath117 and from @xmath117 to @xmath119 follows from the fact that the magnetostatic interaction of the cfl s layers with the polarizer is different in the @xmath117 and @xmath119 states . to prove this we have constructed analogical hysteresis loops for syaf and syf free layers disregarding magnetostatic interaction with the @xmath1 layer ; see figs . [ fig : hyster](g ) and ( h ) . for both syaf and syf free layers we observe now large decrease in @xmath127 for both switchings . this implies that both switchings are realized _ via _ in - plane precessions , in contrast to the case when @xmath1 influences the cfl dynamics via the corresponding magnetostatic field . while the hysteresis loop for syaf remains symmetric , the one for syf becomes highly asymmetric . the asymmetry of syf loop is due to a significant asymmetry of stt in @xmath119 and @xmath117 states , which was previously shaded by the magnetostatic coupling with the layer @xmath1 . we have studied current - induced dynamics of syaf and syf composite free layers . by means of numerical simulations we identified variety of dynamical regimes . the most significant difference between dynamics of syaf and syf free layers concerns the evidence of self - sustained dynamics of both cfl spins . while in the case of syaf only coupled in - plane precessions in a narrow window of external parameters ( @xmath21 and @xmath55 ) are observed , syf free layer reveals more complex and richer dynamics , with the possibility of coupled out - of - plane precessions which might be interesting from the application point of view . furthermore , as shown by numerical simulations , both syaf and syf are switchable back and forth without the need of external magnetic field . for syaf element two possible ways of switching have been identified . since they lead to different switching times , their identification might be crucial for optimization of switching in real devices with syaf free layers . however , one has to note that the diagrams shown in figs . [ fig : saf_dynam1 ] and [ fig : sf_dynam1 ] may be changed when magnetization in cfl becomes non - homogeneous . a disadvantage of the studied structures is their relatively low efficiency of switching , i.e. high amplitude of critical current and long switching time . in order to show more sophisticated ways of tuning the cfl devices , we have analyzed critical currents derived from the linearized llg equation . the formula ( [ eq : iccfl ] ) has been identified as the maximum value of critical current at which dynamics of the cfl structure should be observed . this formula reveals some basic dependence of critical current on spin valve parameters , and therefore might be useful as an initial tool for its tuning . however , in some cases non - linear effects in cfl dynamics might completely change the process of cfl switching , as shown by the presented numerical simulations . but the effects of non - linear dynamics go beyond the simple approach of linearized llg equation , and their study requires more sophisticated nontrivial methods and/or numerical simulations . this work was supported by polish ministry of science and higher education as a research project in years 2010 - 2011 , and partly by eu through the marie curie training network spinswitch ( mrtn - ct-2006 - 035327 ) . the authors thank m. gmitra for helpful discussions . one of us ( pb ) also thanks l. lpez daz , e. jaromirska , u. ebels , and d. gusakova for valuable suggestions . torque acting on the left interface of @xmath3 is calculated from @xmath99 and @xmath98 components of @xmath236 , where @xmath237 is spin current vector in @xmath2 layer ( written in global frame ; shown in fig . [ fig : scheme ] ) , and @xmath238 , where @xmath239 is the matrix of rotation by angle @xmath12 along the axis @xmath240 in the counterclockwise direction when looking towards origin of the coordinate system . hence @xmath241 components can be written as where @xmath243 are spherical coordinates of @xmath50 in the global frame . similarly , we define torques amplitudes on the left interface of @xmath5 from the components of transformed spin current vector @xmath244 . in this case , however , @xmath245 is not written in the global frame , but in the local coordination system coordinate system connected with @xmath50 . to rotate local coordinate system of @xmath50 to local coordinate system of @xmath51 we need to know spherical angles @xmath246 and @xmath247 of vector @xmath51 in the local coordinate system of @xmath50 . this might be done by transforming first @xmath51 vector to local coordinate system of @xmath50 as @xmath248 and calculate its angles @xmath246 and @xmath247 . then we can calculate components of @xmath249 similarly as for the left interface equation in @xmath4 , which is adjacent non - magnetic interface from the right - hand side of @xmath3 , are written in local coordinate system of @xmath50 . to apply the definition of @xmath251 and @xmath252 we need to rotate the local coordinate system so , that its @xmath98-axis will lie in the layer given by vectors @xmath50 and @xmath51 . this might be done by single rotation of local coordinate system around its @xmath22-axis by angle @xmath253 , @xmath254 , where

binary motion is now well known to generate a spiral pattern in the circumstellar envelope of a mass losing star @xcite . however , it is not generally recognized that the binary - induced pattern , vertically extended from the orbital plane , exhibits a ring - like pattern with an inclined viewing angle . at an inclination close to edge - on , in particular , the binary - induced spiral - shell appears as eccentric arcs in the plane of the sky ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? even at a low inclination ( close to face - on ) , the pitch angle of the spiral pattern decreases with radius from the center of binary orbital motion @xcite , such that the outer part of the spiral approximates a concentric ring structure . this misidentification may occur , especially , for the outer ring - like patterns of ( pre-)planetary nebulae ( ppn / pn ) whose inner parts are contaminated by the bright , fast - moving lobes ( e.g. , ngc 7027 @xcite , egg nebula @xcite ) . because of the incompleteness of the observed ring - like patterns , whether they arise from spherical shells due to the pulsation of single stars or spiral - shell patterns due to binary orbital motion is inconclusive . in this paper , we focus on the interpretation that the incomplete ring - like patterns observed in the past may be caused by the motion of ( unknown ) binary stars . linking a binary scenario with the ring - like patterns observed on the circumstellar envelopes of asymptotic giant branch ( agb ) stars and their remnant material in the outer parts of ppn / pn has an interesting implication for the late phases of stellar evolution . since the bipolar ( often multi - polar ) structures , including jets and tori , observed in the ppn / pn ( e.g. , * ? ? ? * ; * ? ? ? * ) likely originate from the central binary systems @xcite , a fundamental issue remains concerning the origin of the ring - like patterns in the outer parts of the ppn / pn . such a ring - like pattern is commonly assumed as the consequence of the periodic mass loss of a _ single _ star in the previous agb phase , due to stellar pulsation ( e.g. , * ? ? ? * ; * ? ? ? * ) or thermal pulsation ( e.g. , * ? ? ? * ) on the spherically symmetric wind . for example , the archetypal agb star , irc+10216 , reveals the circumstellar pattern in the optical and infrared images @xcite , which is characterized by an interval corresponding to @xmath12501700 years . interestingly , @xcite noted that this pattern interval of irc+10216 does not agree with either the stellar pulsation ( 649 days , * ? ? ? * ) or the thermal pulsation ( 600033000 years , * ? ? ? similar discrepancy has also been found in several ppn / pn @xcite . a binary - induced spiral - shell model , in which the pattern interval time is defined by the binary orbital period without a strong limitation on the scale , may possibly serve as an alternative explanation for the ring - like ( but not perfectly symmetric ) patterns formed in the agb phase . in the binary model , the observed spiral - shell pattern of a circumstellar envelope can constrain the binary mass and orbital properties . for the carbon - rich agb star , afgl 3068 , displaying a well - defined spiral circumstellar pattern in dust scattered light , @xcite estimated the orbital period simply from the arm interval and wind expansion speed , ignoring the inclination . for another carbon star , r sculptoris , exhibiting a spiral pattern in addition to a thin thermal pulse shell , @xcite characterized the thermal pulsation with the help of their modeled binary properties for the spiral pattern features . similarly , the circumstellar patterns in close binary systems may be used in determining the relative importance and observational consequence of the accretion disks surrounding the companion star , as theoretically investigated ( e.g. , * ? ? ? * ; * ? ? ? furthermore , @xcite found that the bow shock around a disk can lead to complicated time - dependent post - shock structures for orbital separations less than about 20 au . as such , the features of a spiral - shell pattern provide indirect evidence for the presence of a binary companion to the agb star . in cases where the companion is obscured by the circumstellar envelope , the detection of spiral features may be the only diagnostic probe for , otherwise hidden binary systems . in a previous paper @xcite , we took account of the effect of inclination angle , for the first time , in modeling an observed circumstellar pattern within the binary framework . it was shown that the elongated spiral shape in the plane of the sky , seen in dust scattered light , can be used to constrain key binary quantities ( i.e. , inclination of the orbital plane , orbital period , companion mass , and mass ratio ) . however , this analysis left a degeneracy in such model parameters since the apparent shape in the plane of the sky reflects both the binary motion and projection effect due to the inclination of the orbital plane . in this paper , we show that the degeneracy can be lifted by using gas kinematics from high resolution observations , which provide three - dimensional information . to illustrate the modeling method , we apply the model to the object , cit 6 , by revisiting its incomplete ring - like pattern observed in molecular line emission with the spiral - shell due to the motion of ( hypothesized ) central binary stars . in section [ sec : cit ] , we introduce the known properties of cit 6 and the existing molecular line observation with the very large array ( vla ) , which are used to constrain the model parameters . in section [ sec : geo ] , we formulate a simple analytic model based on the oblate spheroid geometry of an archimedes spiral - shell with a binary orbital inclination @xmath2 . this simple analytic model can be easily applied to other sources in order to ascertain whether a spiral - shell pattern exists . in section [ sec : deg ] , a parameter space analysis is carried out following the method of @xcite based on the elongated shape of the pattern in the plane of the sky . in section [ sec : hyd ] , the results of hydrodynamic radiative transfer simulations are presented , from which we use molecular line kinematics to resolve the degeneracy in the parameter space analysis . we find that the simulated spiral - shell model can reproduce the observational data in the channel maps and position - velocity ( p - v ) diagrams , providing a better fit than obtained by using a spherically symmetric shell model . in section [ sec : dis ] , we discuss the advantage and limitation of the current model . finally in section [ sec : sum ] , we summarize our findings and suggest future applications . cit 6 ( also known as rw lmi , irc+30219 , iras 10131 + 3049 , afgl 1403 ) is an extreme carbon star and believed to be in the transition from the agb to the post - agb phase . @xcite found the presence of a nascent bipolar nebula , providing evidence that the evolutionary phase of cit 6 lies just pass the tip of agb . its effective temperature of 2800k @xcite and the bolometric luminosity of @xmath3 @xcite also correspond to the values at the tip of agb phase of the evolutionary track for a star with the initial mass of 23 , according to the single stellar evolution package of @xcite assuming a solar metallicity . the distance to cit 6 is estimated to be @xmath4pc based on the pulsation period ( @xmath5 yr ) using a period - luminosity relation ( * ? ? ? * and references therein ) . the importance of cit 6 stems from the existence of partial rings of molecular line emission based on vla interferometric observational data obtained by ( * ? ? ? * see also the red - scaled image in figure[fig : chm ] ) at high angular resolution ( @xmath6 ) and high spectral resolution ( @xmath7 ) . @xcite modeled this line emission pattern with four spherical shells , using the shell radii , expansion velocities , and central velocities as free parameters for the individual shells . their model fitting results in different central velocities between the modeled shells , which may imply the departure from spherical symmetry of the circumstellar pattern about the central agb star . @xcite remarked on the existence of the asymmetric features such as the spiral shape ( relatively well traced in the velocity channel of @xmath8 ) , stronger emission observed in the redshifted side , and lack of emission in the west and northwest near the central channels . @xcite showed that the optical depth effect can not reproduce the asymmetric line profile of cit 6 for a physical structure assumed to be spherically symmetric . their three - dimensional morphokinematic model instead suggested several incomplete shells for the structure of cit 6 . here we suggest an alternative interpretation for the asymmetric shape of the circumstellar pattern as an indication for a central binary star system . some evidence for binarity is indeed found in the hubble space telescope ( hst ) images with two compact cores separated by @xmath9 at the position angle ( pa , measured from north to east ) of 10 @xcite . the northern lobe varies in its flux and polarization with the known pulsation period of the carbon star ( @xmath5 yr ; @xcite ) , while the bluer southern lobe is stable and has a polarization direction @xmath10 different from the northern red component @xcite . these characteristics of the southern lobe , differing from those of the carbon star and the northern lobe , suggest that the blue southern lobe is illuminated by a companion star in the vicinity @xcite . the existence of a companion star is supported by the consistence of the polarization direction of h@xmath11 emission @xcite to the blue component . from the observed flux at 4500 , @xcite inferred that the companion is likely later than spectral type g0 ( @xmath12 ) if it is a main - sequence star . the orbital motion of a mass losing star in a binary system creates a spiral - shell pattern in its circumstellar envelope ( e.g. , * ? ? ? we consider the overall morphology of the spiral - shell pattern as an oblate spheroid , following @xcite . in the coordinates having the vertical axis @xmath13 aligned with the orbital axis , the oblate spheroid can be described by ( see figure[fig : geo ] ) @xmath14 where the axial ratio @xmath15 is defined by the ratio of pattern propagation speeds @xmath16 in the orbital plane and along the polar axis . here , the radial propagation speed of the spiral pattern in the orbital plane is affected by the orbital motion of the mass losing star with the orbital speed of @xmath17 to be @xmath18 @xcite , while the pattern propagation speed along the orbital axis is the average wind speed @xmath19 corresponding to the intrinsic wind speed . we define new coordinates @xmath20 rotated by an inclination @xmath2 , in which the observer is at infinity in the direction of the @xmath21-axis . a rotational transformation about the @xmath22-axis by the angle @xmath2 is defined by @xmath23 , @xmath24 , and @xmath25 . substitution of these into equation ( [ eqn : obl ] ) yields @xmath26 where @xmath27 and @xmath28 refer to the tangential coordinate @xmath29 and the azimuthal angle in the @xmath22@xmath30 plane , respectively , defining @xmath31 and @xmath32 . a factor @xmath33 is defined as @xmath34 , thus @xmath35 alternative forms of equation ( [ eqn : inc ] ) are @xmath36 or @xmath37 because the oblate spheroid geometry originates from the latitudinal variation of the wind velocity , these coordinates @xmath27 and @xmath21 can be linearly transformed to the tangential and line - of - sight velocities of the wind . thus , the tangential component is @xmath38 representing the pattern propagation speed in the inclined plane within the @xmath39-channel . the second term in the right - hand - side of equation ( [ eqn : v_t ] ) shows the inclination effect , and the last term is related to the channel velocity representing the height of the relevant material from the midplane . from equation ( [ eqn : v_t ] ) , the line - of - sight velocity can be written as @xmath40 at a given @xmath41 . the maximum line - of - sight velocity is @xmath42 at the orbital center ( i.e. , at the center of mass in the circular orbit case ) . a circumstellar spiral pattern due to the orbital motion of the agb star ( orbital radius @xmath43 ; orbital velocity @xmath17 ) for a given @xmath21 is defined by a differential equation in the polar coordinates @xmath44 as , @xmath45 where the tangential velocity @xmath41 represents the pattern propagation velocity in the layer within the @xmath39-channel . the velocity @xmath41 is given by equation ( [ eqn : v_t ] ) as a function of @xmath28 with given wind velocity , orbital velocity , and inclination angle . as the velocity of an agb wind is nearly constant beyond the wind acceleration region ( typically less than several times the stellar radius ; @xcite ) , the integrated form of equation ( [ eqn : dif ] ) well approximates to @xmath46 for the pattern located at a large radius @xmath27 . here , @xmath47 is the distance to the source . in order to obtain approximate constraints on the binary properties , we explore the possible parameter space for the system of interest , following the method described in @xcite using an overall oblate spheroidal morphology of the pattern . four observed quantities are required , corresponding to the projected axial ratio @xmath48 , the arm pattern spacing @xmath49 along the major axis , the projected binary separation @xmath50 , and the angular position of the stars relative to the major axis of the elliptical pattern shape @xmath51 . here , @xmath52 refers to a position angle measured from the north to the east direction and hereafter @xmath28 is defined as the angle relative to the line of nodes of the binary orbit ( @xmath53 ; the intersection of the orbital plane with the plane of the sky ) measured in the counterclockwise direction . refer to @xcite for the details of this method using an overall oblate spheroid shape . we first identify the line of nodes of the binary orbit as the major axis of the elongated spiral pattern of the hc@xmath0n(43 ) line emission in figure[fig : chm ] . because of the insufficient sensitivity of observation , it is challenging to precisely determine the angular direction of the major axis only by ellipse fitting . thus , we also use the fact that the p - v diagram is most symmetric along the line of nodes and most asymmetric in the perpendicular direction ( middle and right panels of figure[fig : pvd ] , respectively ; see also section [ sec : inc ] ) . from the resulting ellipse fit with the major axis at @xmath54 , we measure the axis ratio @xmath48 of 1.15 and the pattern spacing @xmath49 of @xmath55 along the major axis . we assume the two compact components from the optical images are members of a binary ( or their surrounding gaseous envelopes ) following @xcite , which provides a projected separation of @xmath56 and a position angle of the binary at @xmath57 . the line of nodes nearly coincides with the alignment of the two stars , implying that @xmath50 is approximately equivalent to the actual binary separation . in addition to these four observed quantities , the distance of 400pc is adopted ( see section [ sec : cit ] ) , thus constraining the binary properties in a parameter space spanning the average wind velocity @xmath19 versus the orbital velocity @xmath17 of the mass losing star ( figure[fig : deg ] ) . the gray - colored area in figure[fig : deg ] shows the possible sets of binary parameters . the first constraint for the gray area is the mass of the evolved carbon star ( @xmath58 , black solid line ) greater than the mean mass of white dwarfs @xcite . in particular , the stellar luminosity of @xmath3 for cit 6 suggests the core mass of @xmath59 with 2% accuracy based on @xcite , providing the lower limit on the current mass , @xmath58 , of the agb star . the other constraint stems from the maximum line - of - sight velocity of the pattern @xmath60 corresponding to the half width of the hc@xmath0n(43 ) line , measured as 18 with an uncertainty of @xmath61 considering the spectral resolution . in previous papers @xcite , the lower limit of the maximum line - of - sight velocity was loosely constrained by the condition @xmath62 , which is satisfied at any inclination . we further calculate the maximum line - of - sight velocity taking account of the inclination effect . from equation ( [ eqn : vrm ] ) , the measured line - of - sight velocity range @xmath63 is marked in figure[fig : deg ] as a set of yellow dotted lines . the plausibility of a particular model parameter set can be easily checked by the simple analytic model in section [ sec : geo ] . among the possible models in figure[fig : deg ] , the parameter set marked by a plus symbol is applied to the analytic model . the model consists of an estimate of the agb star mass of @xmath64 , companion mass @xmath65 , binary separation @xmath66au , inclination @xmath67 , and average wind velocity @xmath68 with the distance @xmath69pc . the result is overlaid in each velocity channel map of figure[fig : chm ] by a gray - colored solid line , showing a reasonable match with the outer edge of the observed pattern ( red - scaled ) , particularly for the middle - row panels displaying the near - central velocity channels . furthermore , this inclined spiral can explain the wiggle feature of the observed pattern in the polar coordinates , clearly seen in figure[fig : pol ] . without considering the inclination effect ( i.e. , assuming face - on ; @xmath70 ) , the pattern propagation speed @xmath41 in each @xmath39-velocity channel is constant in equation ( [ eqn : v_t ] ) , indicating that the pattern should appear as a straight line in the polar coordinates with the slope determined by equation ( [ eqn : pol ] ) . for comparison , a concentric spherical shell pattern forms horizontal lines in the @xmath28-@xmath27 coordinates in figure[fig : pol ] . therefore , the wiggle feature , which is also found in other sources ( e.g. , r sculptoris , see figure 2a in @xcite ) , is a natural outcome of the inclined spiral model , and is distinguishable from the face - on view of the spiral model for a binary system and the spherical shell model for a single star . the region delineated by the gray - colored area in the parameter space ( figure[fig : deg ] ) , based on the elongated spiral shape projected on the plane of the sky , can be sharpened by taking advantage of three - dimensional information provided by the molecular line kinematics . we explore the parameter space in the vicinity of the @xmath71 line with hydrodynamic simulations followed by radiative transfer calculations for the hc@xmath0n molecular line . the parameters for the model which best matches the aspects of the vla observation @xcite , described in sections [ sec : inc ] and [ sec : res ] , are summarized in table[tab : par ] ( and by a plus sign in figure[fig : deg ] ) . we note that these parameters are not independent of each other due to their constraints imposed by the parameter space analysis . the basic equations of hydrodynamics are integrated using the flash3 code @xcite based on a piecewise parabolic method @xcite . we construct a centrally concentrated , static grid utilizing the block - structured adaptive mesh refinement paramesh package and adjusting the refinement level based on the distance from the stars . the maximum refinement level of eight with the grids of @xmath72 per block is used for the simulation domain of 8400au@xmath738400au@xmath734200au ( @xmath74 at 400pc , corresponding to the image size of the vla observation in consideration ) in three - dimensional cartesian coordinates , assuming the mirror symmetry about the orbital @xmath22@xmath30 plane coordinates defined in section [ sec : geo ] and figure[fig : geo ] . the prime notation is omitted from now on . ] . the spatial gridding corresponds to a resolution of 1au ( @xmath75 ) in the central region , assigning a sufficient number of grid cells at the wind generating radius @xmath76 within which physical conditions are reset every simulational timestep , in order to assure the ejection of gas from the mass losing star in a spherical shape . the gas temperature over the simulation domain is scaled by the temperature @xmath77 at the wind generating radius @xmath76 from the mass losing star . we set @xmath78 , and @xmath77 is calculated by the temperature distribution @xmath79 of an adiabatic gaseous medium with a @xmath80 density distribution . the observed effective temperature @xmath81 is set at the assumed stellar photospheric radius @xmath82 or 2.59au . an adiabatic equation of state with the adiabatic index @xmath83 of 1.4 is used . the gravitational potentials of the stars are treated as plummer potentials ( see , e.g. , * ? ? ? * ) using the pointmass implementation of the flash3 code with modifications to include the gravitational softening radii and the motion of the two stars . the gravitational softening radius of the mass losing star does not change the result as it is smaller than the wind generating radius . the gravitational softening radius of the companion is set to be comparable to the hoyle - lyttleton accretion radius . an additional simulation with the companion s softening radius reduced by a factor of 2 is compared with the model presented in this paper , in order to verify its insignificant contribution to the overall shape of the arm pattern . the accretion disk of the companion is expected to have a negligible size if it is gauged by the impact parameter ( 0.004 au with the large separation of our binary system ) in @xcite . resolving the accretion disk for a wide binary system is computationally very expensive and beyond the scope of this paper . the flash3 results are used as inputs to the radiative transfer calculation for hc@xmath0n lines with the sparx code . sparx , standing for simulation package for astrophysical radiative transfer , is a software package for calculating molecular excitation and radiative transfer of ( dust ) continuum and molecular lines at millimeter / submillimeter wavelengths . it adopts the accelerated monte carlo ( amc ) algorithm developed by @xcite , and finds full non - local thermodynamical equilibrium ( non - lte ) solutions of molecular level population and radiation fields iteratively and consistently . instead of assuming that the molecular excitation is simply in equilibrium corresponding to the local temperature , non - lte demands only statistical equilibrium of the level population through collisional and radiative excitation and de - excitation processes , which can be solved with the detailed balance equation . additional ( de-)excitation mechanisms , such as infrared radiative pumping , have not been incorporated in the package . as a post - processing tool , the sparx requires inputs of molecular gas density , temperature , velocity as well as the linewidth distribution , which can be provided by the results of a hydrodynamic simulation . knowledge of the molecular abundance distribution and parameters such as dipole moments , line frequencies , and collisional rates are also required by the sparx . the final sparx output includes synthetic images which can be used for generating spectra and p - v diagrams , as input for a telescope filter for simulated observations , or for other scientific analysis . details of the package such as the code implementation and its verification with benchmark cases will be presented in liu et al . ( 2013 , in preparation ) . for the sparx calculation , the hc@xmath0n molecular parameters are obtained from the leiden atomic and molecular database ( lamda ) @xcite . the hc@xmath0n abundance of @xmath84 is adopted throughout the envelope @xcite , while a spherical central region of 2 was carved out with hc@xmath0n abundance set to zero in order to mimic the central hole seen in the observed emission of hc@xmath0n ( also found in its parent molecule cn ; @xcite ) . for computational efficiency , we use the spatial and spectral resolutions of @xmath85 and 0.4 in the sparx calculation , which are sufficient to simulate the vla observation of @xcite summarized in section[sec : cit ] . to further compare the sparx output with the vla observation , we perform a convolution with a gaussian beam of @xmath86 and the rebinning to 1 over the synthetic images . the inferred binary parameters are sensitive to the orbital inclination as illustrated in figure[fig : deg ] . hence , determination of the inclination can lift the degeneracy that remain in the parameter space analysis based only on the pattern shape in the plane of the sky . with the binary parameters in table[tab : par ] , figure[fig : inc ] compares the images for the non - zero velocity channels and the p - v diagrams along the major ( i.e. , @xmath22-axis ; line of nodes ) and minor ( i.e. , @xmath30-axis ; perpendicular to the line of nodes ) axes for five different inclinations of 0 ( face - on ) , 30 , 50 , 70 , and 90 ( edge - on ) . the channel images ( top row ) illustrate the change of pattern shape from a spiral in the face - on view to ellipses in the edge - on view . in particular , the spiral pattern is split along @xmath87-axis in the image for @xmath88 and attached to the adjacent component in the image for @xmath89 , making a new ring . the p - v diagrams of figure[fig : inc ] ( middle and bottom rows ) reveal notable features . the middle row show that , regardless of inclination , the p - v diagram along the line of nodes ( i.e. , @xmath22-axis ) appears relatively symmetric about the central velocity because of the mirror symmetry of the overall oblate spheroid geometry about the orbital plane . in contrast , the p - v diagram along the minor axis highly depends on the inclination angle ( bottom row ) . one important implication is that the pa dependence of the p - v diagram can be used to determine the line of nodes of the binary orbit ( as in section [ sec : deg ] ) . moreover , the comparison between the p - v diagrams along major and minor axes helps to infer the inclination angle of the system from the observed data cube . for example , the pa dependence of the p - v diagram is minimized in the face - on case . the p - v diagram along the minor axis ( bottom row ) provides further information on the orbital inclination . at a mid - range inclination ( see e.g. @xmath90 ) , it reveals an asymmetry in intensity between the redshifted and blueshifted components , the degree of asymmetry of which differs with position , and thus the overall shape of p - v diagram is tilted . hence , the degree of tilt and spectral asymmetry is dependent on the inclination . on the other hand , the maximum line - of - sight velocity ( i.e. , @xmath39 at @xmath91 ) increases with inclination as predicted in equation ( [ eqn : vrm ] ) . this increase of maximum velocity makes the outline of p - v diagram at the high end velocity at each @xmath30 position ( @xmath92 ) shallower in the face - on case ( @xmath93 ) and becoming steeper as the inclination increases . all these features of the p - v diagrams can be used to constrain the inclination angle of the binary system . in the framework of the binary model , the companion creates a gravitationally - induced density wake in a flatter spiral shape , overlapped with the spiral - shell pattern due to the orbital motion of the agb star @xcite . the overlapped region appears as knots in the plane of the sky at the pa depending on the inclination ( see figure 10 of * ? ? ? * ) because the knots represent the companion s wake relatively thin in the vertical direction from the orbital plane @xcite . in figure[fig : inc ] , the top panels for a non - zero velocity channel show such knots at high inclinations ( @xmath94 in this model ) at the pa varying with the inclination ( marked by arrows ) . in the p - v diagram along the minor axis ( bottom row ) , knots appearing with low inclinations ( @xmath95 ) also shift the positions depending on the inclination . on the other hand , in the p - v diagram along the line of nodes of orbits ( middle row ) all knotty structures are present at the zero velocity , confirming the existence of the companion s wake on the orbital plane . in short , the shape of p - v diagram , the degree of spectral asymmetry , the position of knots both in individual channels and in individual p - v diagrams provide important clues in identifying the best model . it is also worth noting that the velocity dispersion of individual arm component is dominated by the velocity variation due to the binary orbital motion ( @xmath96 ) . for the binary system in table[tab : par ] with an inclination angle of 60 , the above features reasonably match the vla observation . figure[fig : chm ] shows the modeled hc@xmath0n(43 ) line emission with the parameters in table [ tab : par ] in green color , overlaid with the observational data displayed in red color . a good correlation exhibits between the observed and modeled data in the individual velocity channels , albeit the outermost arm in the observation is barely detected at the @xmath97 level , where the noise level @xmath98mjy / beam refers the root - mean - square of the line - free channels in the vla observation . the observed channel images ( red ) range between 1 and 5mjy / beam . to present the pattern shape beyond the sensitivity limit of the observation , the modeled channel images ( green ) are displayed from 0.1 to 2mjy / beam . all arc patches in the observed line emission are well located on the modeled spiral pattern . at the central channels ( middle panels of figure[fig : chm ] ) the model reveals elongated spiral patterns , while at higher velocity channels ( top and bottom panels ) ring - like patterns are reproduced . here , we emphasize that the ring - like features appear in the spiral - shell models with considerably large inclination angles ( see * ? ? ? it suggests that ring - like patterns observed in the agb circumstellar envelopes and in the outer parts of ppn / pn do not necessarily imply an underlying spherically symmetric nature . the inclination angle of the presented model is 60 . the binary model in figure[fig : chm ] shows the breakup of rings and their unequal spacing between adjacent arms . this appears in the intermediate range of inclination angles , reflecting the transition from a spiral shape to a ring shape , as explained in section [ sec : inc ] . in the @xmath8 channel , for example , the ring - like pattern is incomplete revealed by the gap in the east at @xmath99 ( along the east side of minor axis ; see also the first panel of figure[fig : pvd ] ) . the ring radius of the northern part of the gap is larger than the radius of the southern part , in which the model and observation match sensibly . the companion s gravitational wake appears as knots in the modeled channel images located in specific pas ( marked by arrows in figure[fig : chm ] ) as determined by the inclination . in the zero - velocity channel ( middle panel of figure [ fig : chm ] ) representing the midplane , the knots are arranged along the line of nodes of the orbits ( i.e. , @xmath100 and 190 ) . the knots shift to the east in the redshifted velocity channels and to the west in the blueshifted channels . the direction of shift of the knots with the channel velocity is determined by the orientation of the companion s orbit . the model knots eventually converge on the east and west ends of the ring pattern at the maximum velocity channels , which results in the thickening and brightening of the pattern at such locations . the pas of knots in the modeled channel images are consistent with bright emission spots in the corresponding vla images . for example , the observed image ( in figure[fig : chm ] ) exhibits thickening and brightening features in the ring pattern on the east and west ends at the @xmath101 and @xmath102 channels , respectively . another example is found at the southwest in the @xmath103 and @xmath8 channels , where the detected emission in the outermost ring ( red ) corresponds to the position of bright knotty structure in the model ( green ) . at the locations of these knots , the arm ( in red and green images ) is found considerably inside the analytic archimedes spiral model without considering the companion s gravitational wake ( gray solid line ) . this decrease of arm radii is explained by ( * ? ? ? * figure 9 and text ) , which showed that a shock front ( appearing as knots in the channel images ) at the location of the companion s wake reduces the speed of the downstream flow and , therefore , the pattern propagation speed . we have shown that a numerical hydrodynamic radiative transfer model for cit 6 demonstrates that the observed incomplete ring patches can be naturally understood within the framework of a single spiral - shell pattern induced by a binary motion . in contrast , the previous spherical shell model @xcite and incomplete shell model @xcite required the use of individual shell parameters without providing physical understanding for their variations . moreover , the binary model reveals that the knotty structures result from the overlapping between a spiral - shell shaped pattern due to the orbital motion of the mass losing star and a more flattened spiral pattern onto the orbital plane due to the companion s motion . in this picture , the knots are aligned in a specific pa , which systematically changes with the velocity channel . this implies that such regularly - aligned knots are a natural consequence of the model . the total flux integrated over the full simulation domain is related to the mass loss rate and the hc@xmath0n abundance for a given temperature distribution . we have adopted the mass loss rate of @xmath104 and relative molecular abundance of @xmath84 from @xcite . unlike @xcite , who assumed the local thermal equilibrium with a single temperature , the temperature distribution in our model is determined by an adiabatic gas equation of state . the resulting total flux is smaller than the vla observation by a factor of 2 . this implies that a complete modeling of these parameters requires studies involving a further fine - tuning of the input parameters as well as the introduction of molecular chemistry . such studies will also necessitate incorporating observations of various molecular species , transitions , and angular resolutions . however , these parameters merely scale the flux level , but do not change the shape of the spiral pattern which is the focus of this paper . the fact that the observed redshifted emission component is stronger than the blueshifted component is , however , not reproduced in our spiral model , albeit a small magnitude of spectral asymmetry is found depending on the inclination . this implies that our model is incomplete and that the asymmetry along the line - of - sight does not originate solely from the binary motion . the observed spectral asymmetry may agree with a presence of a bow shock due to the interaction between the circumstellar and interstellar media if the space motion away from us is significant ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? however , the vla data does not show a shift of the local standard of rest ( lsr ) velocity from the central velocity of the spiral - shell pattern . other possible effects which may enhance the asymmetry include physical and chemical inhomogenities in the surrounding matter or in the dust formation phase at the onset of the stellar wind ( e.g. , * ? ? ? * ; * ? ? ? * ) if propagated to large scales . the missing emission in the west side of the central channels is also beyond the scope of this paper . this observed feature can not be explained simply by a spiral - shell model nor by a spherical shell model . this spatial asymmetry is also found in a different transition of hc@xmath0n and in its parent molecule cn @xcite . we speculate that such lack of emission could possibly be caused by a chemical effect reducing the molecular abundance , rather than being affected by the gas density . compared to the other model sets in the parameter space determined by the projected pattern shape ( figure[fig : deg ] ) , for instance at @xmath105 , the presented model shows a better match with the observation for the pattern features explained in section [ sec : inc][sec : res ] in both channel maps and p - v diagrams . however , the difference of our fiducial model from the other comparison model sets are not decisive for the given uncertainties in the model parameters . in particular , observational uncertainties due to the limited sensitivity in determining the elongated shape and the arm spacing , as well as the determination of the source distance , stellar luminosity , photospheric temperature , and binary separation enter . follow - up observations with a higher sensitivity for a chemically stable molecule such as co would address some of these uncertainties . we have developed a method to investigate the incomplete ring - like feature of a spiral - shell pattern in an outflowing circumstellar envelope , due to the orbital motion of central binary stars with an inclination with respect to the orbital plane , providing an alternative explanation to models invoking periodic mass loss events of a single star . as a first step in this method , a parameter space study is performed as in @xcite and improved upon by taking account of the inclination angle for the constraint on the wind expansion velocity . the results of such an analysis significantly narrow down the parameter space that can provide a fit to the observed pattern shape in the plane of the sky . this reduction in the parameter space is important for the second step where a modest number of hydrodynamic and radiative transfer investigations can be carried out . the pattern shape , degree of asymmetry , position of knots in individual velocity channels and p - v diagrams are used via detailed modeling to constrain binary properties . we have shown that gas kinematics is important in constraining the binary properties , which would otherwise remain degenerate under the constraints from the parameter space analysis based only on the shape of the circumstellar spiral - shell pattern in the plane of the sky . we have also demonstrated that a simple analytic analysis for the elongated spiral model based on an archimedes spiral - shell pattern viewed as a function of inclination angle for comparison to individual velocity channels is useful for determining the applicability of the model to observational data . finally , observed circumstellar patterns often reveal wiggles with varying slope in the polar coordinates that can be adequately described by the elongated spiral model , not by a face - on archimedes spiral nor by spherical shells . we have used the agb star , cit 6 , as the first example to illustrate a method which can be used to help to constrain the binary properties given the existence of high resolution and sensitivity vla data for the hc@xmath0n(43 ) molecular line . this first success of applying the binary - induced spiral - shell model to an observed incomplete ring - like pattern opens the possibility of connecting the ring - like patterns commonly found in the agb circumstellar envelopes and in the outer parts of ppn / pn and pointing to the conceivable presence of central binary systems . we acknowledge a detailed comments from an anonymous referee , and thank dr . mark claussen for providing the calibrated fits image of the vla data . this research is supported by the theoretical institute for advanced research in astrophysics ( tiara ) in the academia sinica institute of astronomy and astrophysics ( asiaa ) . the computations presented here have been performed through the asiaa / tiara computing resource . the sparx code development is supported by the coordinated / coherent hydrodynamic and astrophysical research , modeling and synthesis ( charms ) project in asiaa . the flash code is developed by the doe - supported asc / alliance center for astrophysical thermonuclear flashes at the university of chicago . balick , b. , & frank , a. 2002 , , 40 , 439 binney , j. , & tremaine , s. 2008 , galactic dynamics ( princeton , nj : princeton univ . press ) bond , h. 1996 , hst press release stsci - pr96 - 05 colella , p. , & woodward , p. r. 1984 , journal of computational physics , 54 , 174 chau , w. , zhang , y. , nakashima , j .- i . , deguchi , s. , & kwok , s. 2012 , , 760 , 66 claussen , m. j. , sjouwerman , l. o. , rupen , m. p. , et al . 2011 , , 739 , l5 cohen , m. 1979 , , 186 , 837 cohen , m. , & hitchon , k. 1996 , , 111 , 962 decin , l. , royer , p. , cox , n. l. j. , et al . 2011 , , 534 , a1 de marco , o. 2009 , , 121 , 316 dinh - v .- trung , & lim , j. 2009 , , 701 , 292 edgar , r. g. , nordhaus , j. , blackman , e. g. , & frank , a. 2008 , , 675 , l101 fryxell , b , . olson , k. , ricker , p. , et al . 2000 , , 131 , 273 he , j. h. 2007 , , 467 , 1081 hfner , s. 2007 , why galaxies care about agb stars : their importance as actors and probes , 378 , 145 hogerheijde , m. r. , & van der tak , f. f. s. 2000 , , 362 , 697 huggins , p. j. 2007 , , 663 , 342 huarte - espinosa , m. , carroll - nellenback , j. , nordhaus , j. , frank , a. , & blackman , e. g. 2013 , , 1448 hurley , j. r. , pols , o. r. , & tout , c. a. 2000 , , 315 , 543 kim , h. 2011 , , 739 , 102 kim , h. , & taam , r. e. 2012a , , 744 , 136 kim , h. , & taam , r. e. 2012b , , 759 , 59 kim , h. , & taam , r. e. 2012c , , 759 , l22 leo , i. c. , de laverny , p. , mkarnia , d. , de medeiros , j. r. , & vandame , b. 2006 , , 455 , 187 ladjal , d. , barlow , m. j. , groenewegen , m. a. t. , et al . 2010 , , 518 , l141 le bertre , t. 1992 , , 94 , 377 lindqvist , m. , schier , f. l. , lucas , r. , & olofsson , h. 2000 , , 361 , 1036 loup , c. , forveille , t. , omont , a. , & paul , j. f. 1993 , , 99 , 291 madej , j. , naleyty , m. , & althaus , l. g. 2004 , , 419 , l5 maercker , m. , mohamed , s. , vlemmings , w. h. t. , et al . 2012 , , 490 , 232 mastrodemos , n. , & morris , m. 1999 , , 523 , 357 mattsson , l. , hfner , s. , & herwig , f. 2007 , , 470 , 339 mauron , n. , & huggins , p. j. 2006 , , 452 , 257 mayer , a. , jorissen , a. , kerschbaum , f. , et al . 2011 , , 531 , l4 monnier , j. d. , tuthill , p. g. , & danchi , w. c. 2000 , , 545 , 957 morris , m. 1987 , , 99 , 1115 nordhaus , j. , & blackman , e. g. 2006 , , 370 , 2004 paczyski , b. 1971 , , 21 , 271 sahai , r. , morris , m. r. , & villar , g. g. 2011 , , 141 , 134 sahai , r. , trauger , j. t. , watson , a. m. , et al . 1998 , , 493 , 301 schmidt , g. d. , hines , d. c. , & swift , s. 2002 , , 576 , 429 schier , f. l. , van der tak , f. f. s. , van dishoeck , e. f. , & black , j. h. 2005 , , 432 , 369 soker , n. 1994 , , 270 , 774 soker , n. 2000 , , 312 , 217 su , k. y. l. 2004 , asymmetrical planetary nebulae iii : winds , structure and the thunderbird , 313 , 247 theuns , t. , & jorissen , a. 1993 , , 265 , 946 trammell , s. r. , dinerstein , h. l. , & goodrich , r. w. 1994 , , 108 , 984 trammell , s. r. , & goodrich , r. w. 1997 , aas , 191 , 47.15 wareing , c. j. , zijlstra , a. a. , speck , a. k. , et al . 2006 , , 372 , l63 wilkin , f. p. 1996 , , 459 , l31 willson , l. a. , & hill , s. j. 1979 , , 228 , 854 woitke , p. , & niccolini , g. 2005 , , 433 , 1101 wood , p. r. 1979 , , 227 , 220 zhang , y. , kwok , s. , & dinh - v - trung 2009 , , 691 , 1660

the classic model for galaxy formation within the cold dark matter ( cdm ) paradigm assumes that gas within galaxy halos initially reaches the temperature of the virialized halo . during the collapse of the halo , the process of violent relaxation of the dark matter shocks the gas component to the virial temperature . any subsequently accreted gas is also shocked to the virial temperature as it enters the virial radius @xcite . this assumption has been routinely adopted by many analytical works ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? given a density profile , a cooling radius can be calculated , inside of which gas can radiate away its energy while conserving its specific angular momentum to form a centrifugally supported disk that grows from the inside out ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . in general , these semi - analytic models ( sams ) have been successful in reproducing many of the present day properties of galaxies . some studies have called this virial temperature assumption into question , however . recent theoretical work has found that , in fact , the majority of baryons may never reach the virial temperature as they accrete onto galaxies @xcite . * ; * ? ? ? * hereafter k05 and k08 , respectively ) examined how gas gets into galaxies using a n - body + smoothed particle hydrodynamics ( sph ) simulation of a cosmological volume . they found that galaxies with halo masses below a few 10@xmath2 m@xmath3 are dominated by `` cold '' gas accretion . this cold gas is never shock heated to the virial temperature of the galaxy , and is thought to be accreted predominantly along filaments in the cosmic web ( e.g. , * ? ? ? further support for this new paradigm has come from using a similarly large cosmological volume simulated with an adaptive mesh refinement ( amr ) code . for galaxy halos to shock heat infalling gas , the cooling rate of gas behind the shock must be slower than the compression rate of the infalling gas @xcite . this requirement leads to two regimes in which gas is able to avoid shock heating , and instead cools onto the galaxy on a free fall timescale instead . first , galaxies below a certain critical mass are not massive enough to support a stable shock because their dynamical time is shorter than the cooling time of the halo gas . opt08 find that the halo mass where cold and shocked accretion contribute equally is in good agreement with the theoretical analysis of , with a critical mass of 10@xmath4 m@xmath3 . this critical mass is roughly independent of redshift . second , even when a shock is present , cold gas accretion can occur along filaments that penetrate deep inside the hot halo . the ability of filaments to feed galaxy growth has been noted for some time @xcite . k05 find that their cold flow gas accretion is largely along filaments . these filaments are efficient at funneling gas to the galaxy , and can bring material to the galaxy from a much larger radius than predicted in the standard spherical accretion model . the mass contributed by filaments is a strong function of redshift ( db06 , opt08 ) . note that in the first case above , when a galaxy is in the low mass regime incapable of supporting a stable shock , cold gas accretion can occur both in a quasi - spherical fashion and along filaments , while in the second , high mass regime in which a hot halo exists , cold gas accretion occurs primarily along preferred directions in filaments . lccccccccr h579 & 3.4@xmath510@xmath6 & 85 & 40 & 5.1@xmath510@xmath7 & 0.03 & @xmath85 & 0.17 & @xmath09.5@xmath510@xmath9 + dwf1 & 1.4@xmath510@xmath2 & 135 & 70 & 1.3@xmath510@xmath9 & 0.01 & 1.5 & 0.15 & @xmath05.3@xmath510@xmath10 + mw1 & 1.1@xmath510@xmath11 & 270 & 130 & 5.1@xmath510@xmath9 & 0.05 & 2.5 & 0.3 & @xmath04.8@xmath510@xmath10 + gal1 & 3.3@xmath510@xmath11 & 385 & 190 & 1.1@xmath510@xmath10 & 0.02 & 2.8 & 0.3 & @xmath03.7@xmath510@xmath10 + h277 & 7.1@xmath510@xmath2 & 230 & 115 & 3.8@xmath510@xmath9 & 0.03 & 2.5 & 0.35 & @xmath02.3@xmath510@xmath10 + mw1.lo & 1.1@xmath510@xmath11 & 265 & 130 & 5.0@xmath510@xmath9 & 0.07 & 2.5 & 0.6 & @xmath05.8@xmath510@xmath9 + mw1.ad & 1.1@xmath510@xmath11 & 270 & 130 & 5.1@xmath510@xmath9 & 0.07 & 2.5 & 0.6 & @xmath03.0@xmath510@xmath9 + galaxies above the critical mass for shocking appear to be dominated by hot gas accretion as in the standard model , but they may have been fed by cold flow filaments even after a shock develops , particularly at high @xmath12 , and they are also built in part by the mergers of smaller galaxies that were dominated by cold flows . thus , cold flow gas accretion is an important component of galaxy formation that has remained largely ignored until recently . previous work on cold flow gas accretion in simulations have addressed the issue using large cosmological volumes , furnishing a statistical description of how galaxies get their gas . this provides a useful starting point for modifying semi - analytic models of galaxy formation . however , with large volumes and many simulated galaxies , numerical resolution for any individual galaxy must be sacrificed . hence , these simulations were unable to resolve the internal structure of galaxies and investigate the impact of cold flows on the formation of galactic _ disks_. opt08 follow gas as it enters the virial radius of galaxies , and examined the temperature history of that gas as it flows toward the central galaxy , down to 0.1 r@xmath13 . however , limited resolution prevented them , or k05 or k08 , from being able to resolve disks in their galaxies . thus , the impact of cold flows to disk assembly , and in particular the growth of the stellar disk component , has yet to be addressed . rather than sample a cosmological volume at low resolution , in this paper we investigate gas accretion in several very high resolution simulations , in order to study the impact of cold gas accretion on the assembly of stellar disks . each of these galaxies has at least a million particles within the virial radius by redshift zero . this high resolution provides two advantages over previous simulation work on this subject . first , it allows us to resolve the internal structure of our galaxies and their disks . second , we overcome artificial broadening of shocks in sph so that we are able to identify entropy increases associated with accretion shocks , allowing us to explicitly search for shocks rather than simply using a temperature threshold as in previous works ( k05 , k08 , opt08 ) . we are thus capable of examining the temperature and entropy history of accreted gas that eventually settles to the disk and forms stars . these simulated galaxies are run within a @xmath14cdm context and span two orders of magnitude in galaxy halo mass , allowing us to study the role of cold flows in galaxies from 0.01 l@xmath1 to l@xmath1 . this paper is outlined as follows : in section [ sims ] we briefly discuss the simulations used in this study . in section [ gasaccr ] we describe our method to identify shocked and unshocked gas accretion to galaxies , and find that the fraction of accreted gas that is shocked is a strong function of galaxy mass , in agreement with previous works . in section [ diskgrowth ] we extend the results of section [ gasaccr ] to investigate the role that unshocked , cold gas accretion has in building the stellar disk of galaxies , as a function of galaxy mass . we briefly compare our results on disk growth to analytic models of disk galaxy formation . in section [ issues ] we discuss the various effects of our numerical scheme on the results , and conclude that our results are robust ( those readers interested strictly in the science results may wish to skip this section ) . finally , in section [ conclusions ] we discuss the implications of our results for existing analytic models of disk galaxy formation , and conclude . the simulations used in this study are the culmination of an effort to create realistic disk galaxies . as discussed in @xcite , at @xmath12 = 0 disk galaxies in our simulations overcome the angular momentum catastrophe and resolve the internal structure of disks ( see also * ? ? ? * which shows that galaxies in this paper lie on the tully - fisher relation ) . this crucial improvement allows us to undertake the current investigation , and is made possible both by dramatically increased numerical resolution and by the inclusion of a simple but physically motivated recipe to describe star formation ( sf ) and the effects of subsequent supernova ( sn ) feedback ( full details in * ? ? ? * ) . our adopted sn feedback and cosmic uv background @xcite drastically reduce the number of galaxy satellites containing a significant stellar population , avoiding the well known `` substructure problem '' @xcite . our sn feedback is also effective at regulating sf , preventing gas from artificially cooling too early and forming stars , and hence we produce smaller bulges and halos in subsequent mergers @xcite . this regulation by sn also reproduces the observed stellar mass - metallicity relationship for galaxies , both at the present time and at high redshift @xcite . at @xmath12 = 3 , these simulations reproduce the metallicities , incidence rate , and column density distributions of observed damped lyman alpha absorbers ( dlas * ? ? ? four simulated disk `` field '' galaxies were chosen to cover two orders of magnitude in halo virial mass , from 3.4@xmath510@xmath6 m@xmath3 to 3.3@xmath510@xmath11 m@xmath3 . the three most massive galaxies have been discussed in detail in ( * ? ? ? * ; * ? ? ? * i.e. , dwf1 , mw1 , gal1 ) , but they have been rerun at eight times the mass resolution and twice the spatial resolution for this study . all galaxies are simulated using gasoline @xcite in a flat , @xmath14-dominated cosmology . the three most massive galaxies were run using a wmap year 1 cosmology , while a wmap year 3 cosmology was adopted for the newest , lowest mass galaxy ( hereafter h579 ) . the three least massive galaxies were originally drawn from uniform volumes ranging from 25 mpc to 50 mpc , while gal1 was selected from a 100 mpc box . these galaxies were then resimulated at higher resolution using the volume renormalization technique @xcite , which allows us to resolve fine structure while capturing the effect of large scale torques . we adopt the same sn and sf efficiency parameters as in @xcite and a kroupa initial mass function @xcite . characteristics of each galaxy at @xmath12 = 0 are given in table 1 . galaxies and their parent halos were identified using ahf @xcite . ahf determines the virial radius , r@xmath13 , for each halo at each output time step using the overdensity criterion for a flat universe following @xcite . ahf identifies all halos that exist in the high resolution region of our simulations at each output time step , and the individual particles that belong to them . we can thus track a single gas particle through the entire simulation , as it is accreted to a galaxy , as that galaxy later merges with another , and as the gas particle forms stars . ahf allows us to identify the main galaxy at each output time step from @xmath12 = 0 to higher @xmath12 , and the particles that belong to it . in this paper , we identify the progenitor halos of the main galaxy all the way back to @xmath12 = 6 . thus , as can be seen in table 1 , particles are traced back to a higher @xmath12 than the last major merger for these galaxies . for example , our mw1 galaxy undergoes a last major merger at @xmath152.5 , and is easily identified at lower redshifts . at higher @xmath12 , we follow the most massive of the two progenitors to study the galaxy s history . @xcite and @xcite showed that the mass function in their simulations converged at a minimum of 64 dark matter ( dm ) particles per halo . thus , we require a minimum of 64 dm particles for halo identification in ahf . halos with fewer particles are considered as `` smooth '' gas accretion instead , but in practice these halos are `` dark '' and have no bound baryons due to heating from our uv background . the uv background turns on at @xmath12 = 6 and by @xmath12 = 4 has a significant effect on the lowest mass halos , heating their gas so that only halos with @xmath0100 or more particles are of sufficient mass to retain bound gas particles ( with velocities below the escape velocity of the halo ) . therefore , unresolved halos with fewer than 64 dm particles should contribute little or no bound gas , and have no significant impact on the gas accretion study in this paper . resolution effects are explored further in section [ uv ] . our highest ( lowest ) resolution runs have particle masses of 4.5@xmath510@xmath7 m@xmath16 ( 2.9@xmath510@xmath10 m@xmath16 ) , 1.6@xmath510@xmath7 m@xmath16 ( 4.3@xmath510@xmath9 m@xmath16 ) , and 4.7@xmath510@xmath17 m@xmath16 ( 1.3@xmath510@xmath9 m@xmath16 ) for dark matter , gas , and stars , respectively , and a force resolution of 0.15 - 0.3 kpc . these galaxies were simulated to have a similar dynamical range and similar number of particles within the virial radius . for a more detailed discussion of the simulations , see @xcite . we examine the temperature , entropy , and accretion histories for gas particles accreted to the main galaxy in these four simulations . for all cases , each gas particle s properties are determined at @xmath12 = 6 , @xmath12 = 5 , and @xmath12 = 4 , after which the values are sampled approximately every 320 myr until @xmath12 = 0 . the gas in these simulations is allowed to cool via atomic cooling , adopting the cooling function for a primordial ( metal free ) gas . we show in section [ mc ] that adding metal line cooling to the simulations has no effect on the results presented in this paper . major mergers have been defined as having a @xmath03:1 mass ratio , or lower . the galaxies in this sample tend to have relatively quiescent merging histories , with the last major merger redshifts being greater than 1.5 . this similarity is by design , in order to compare gas accretion histories of galaxies with similar pasts . in a future paper , we will explore gas accretion and disk assembly in a set of l@xmath1 galaxies ( @xmath010@xmath11 at @xmath12 = 0 ) that span a broader range in space in spin parameters and in last major merger redshifts . in addition to the four main galaxies studied here , a few other simulations have been run to investigate numerical convergence . mw1 is explored at half of the dynamical resolution ( mw1.lo , the same resolution simulation discussed at length in * ? ? ? this lower resolution run is also studied with cooling turned off ( the adiabatic case , mw1.ad ) and with metal line cooling added ( mw1.mc ) . an additional l@xmath1 galaxy , h277 , with m@xmath13 = 7.1@xmath510@xmath2 m@xmath3 , is used to investigate the effects of output time step resolution on our temperature and entropy criteria . before investigating the role of shocked versus unshocked gas in our simulations , we first want to identify the gas that has been accreted smoothly to our galaxy halos , as opposed to gas that has been accreted from other galaxy halos . it is this smoothly accreted gas that will be available to potentially shock as in the classic model . we first find all gas particles that have ever been identified as part of the main galaxy , back to @xmath12 = 6 . ahf allows us to identify those gas particles which have been part of other galaxies before merging with the main galaxy . those gas particles that are already in the main halo progenitor at @xmath12 = 6 ( i.e. , the earliest step at which gas particles are traced ) are labeled as `` early '' accretion . we adopt an extreme definition and label as `` clumpy '' accretion any particles that have at any output step belonged to a galaxy halo other than the main galaxy we re considering . the majority of this gas will be accreted in merger events ( note that in this definition , even those particles that come from the smaller progenitor in a last major merger get labeled as clumpy ; it includes both the hot and cold gas accreted in this merger ) , though a small fraction of this clumpy gas may also include gas that was tidally stripped or blown out of a galaxy halo and later accreted to the main galaxy we re considering . the remaining gas that does not meet our `` clumpy '' criterion is labeled as `` smoothly '' accreted gas . [ fig1 ] shows the fraction of gas accreted since @xmath12 = 6 that has been identified as `` clumpy '' ( green , at top of bars ) or `` smooth '' ( the blue and red bars , combined ; this separation of smooth is discussed in detail in section [ decomp ] ) , in order of increasing galaxy mass for our four main galaxies . the total halo mass ( dm+stars+gas ) of each of the four galaxies is listed below their respective bar . the key feature of fig . [ fig1 ] is the dominance of smoothly accreted gas in the formation of these galaxies , despite our extreme definition for `` clumpy '' accretion . the first two columns of table [ accr ] list the ratios of `` smooth '' and `` clumpy '' gas accretion to total gas accreted to the galaxy since @xmath12 = 6 . in all cases , @xmath075% or more of the gas accreted to the galaxy since @xmath12 = 6 has never belonged to another galaxy halo ( the adiabatic case is discussed in section [ adiabatic ] ) . because we resolve halos down to the mass limit where the uv background suppresses galaxy formation , the major numerical uncertainty in our smooth gas accretion fractions is due to our finite output time interval ( 320 myr ) . due to output time step resolution , our fractions in table [ accr ] for smoothly accreted gas will be an upper limit . the issue of output step resolution is explored below ( section [ timeres ] ) and it is shown that , for a milky way type galaxy , increasing time resolution will identify a higher fraction of `` clumpy '' material , but not by more than 10% , so smoothly accreted gas still dominates the formation of this galaxy . the dominance of smooth gas accretion in the baryonic growth of galaxies agrees with previous simulation results @xcite . @xcite looked at this question , in particular . their mass resolution allowed them to fully examine gas accretion in galaxies with masses greater than 5.4@xmath510@xmath6 m@xmath3 in halo mass , and they found that galaxies above this mass predominantly gain mass via smooth accretion rather than in mergers . in fact , they find that mergers account for no more than 25% of mass accretion at redshifts greater than 1 , and no more than 35% of growth at @xmath12 @xmath18 0.5 . k05 and k08 also concluded that , in a global sense , the growth of galaxies is dominated by smooth gas accretion rather than mergers . lcccc h579 & 0.96 & 0.04 & 0.00 & 1.00 + dwf1 & 0.87 & 0.13 & 0.10 & 0.90 + mw1 & 0.73 & 0.27 & 0.65 & 0.35 + gal1 & 0.75 & 0.25 & 0.79 & 0.21 + h277 & 0.73 & 0.27 & 0.44 & 0.56 + mw1.lo & 0.78 & 0.22 & 0.62 & 0.38 + mw1.ad & 0.49 & 0.50 & 0.84 & 0.16 + fig . [ fig1 ] shows that , in general , there is a trend in the amount of `` clumpy '' accretion as a function of galaxy mass . this figure suggests that as galaxy masses get smaller , the amount of material accreted from other galaxy halos decreases . this is at odds with the merger trees of pure dark matter halos of different masses , which are statistically indistinguishable ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? however , fig . [ fig1 ] represents only accreted gas , and in practice there is a decrease in the baryon fraction of halos with decreasing halo mass . the reionization of the universe acts to unbind baryons from halos with v@xmath19 below @xmath030 km / s , and the fraction of unbound baryons increases as halo mass decreases further . our uv background mimics this effect of reionization . gas loss due to sne feedback also becomes increasingly effective at these lower masses . the combined effect is for halos to become increasingly dark matter dominated toward lower halo mass . as smaller halos merge with even smaller , more dark matter dominated halos , the amount of baryonic material accreted from other halos decreases with smaller halo masses , as seen in fig . [ fig1 ] . [ fig3 ] shows the accretion rate of gas to the virial radius as a function of time for each of the four main galaxies . lines denoting smoothly accreted gas and `` clumpy '' accretion gas are shown in purple ( solid ) and red ( dashed ) , respectively . major mergers are conspicuous for all but the lowest mass galaxy , h579 , for which the last major merger occurred at a redshift greater than 5 . the obvious result here , again , is that smooth accretion of gas that has never belonged to another galaxy halo dominates the gas accretion history of all of these disk galaxies . [ fig3 ] demonstrates that this is true at all redshifts , and not just for the cumulative case that is shown in fig . [ fig1 ] . having now identified two subsets of gas accretion to the virial radius , `` smooth '' or `` clumpy , '' we now focus on the gas that has been smoothly accreted . in the classic analytic picture of disk galaxy formation , this smoothly accreted gas will be shocked as it enters the virial radius of the galaxy . db06 showed , instead , that there is a critical mass below which galaxies are unable to support a stable shock at the virial radius . if a galaxy is below this mass , then gas may not shock until it reaches the disk , where the density contrast is high . in massive galaxies , even after a stable shock develops and propagates out through the galaxy s halo , filaments of cold gas may penetrate inside the shock radius . these filaments may shock when they reach a radius closer to the disk . smoothly accreted gas is followed until it reaches a radius of 30 kpc from the galaxy center ( or r@xmath13 at high @xmath12 , whichever is smaller ) . we stop following gas particles at 30 kpc to avoid contamination from feedback effects that occur within our galaxy disks that also lead to a strong increase in entropy and temperature , mimicking the shocks we wish to identify . similarly , we exclude `` clumpy '' gas from consideration in the remaining analysis , as feedback from merger induced star formation may mimic shocks . by definition , a particle undergoes an entropy increase when it encounters a shock . previous work that searched for shocked gas within simulations used a temperature criterion rather than an entropy criterion . our high resolution simulations allow us instead to search for the entropy increase associated with shocks . we searched all smoothly accreted gas for an entropy jump that corresponded to a mach number of 3 or more . however , we found that not all of this shocked gas was associated with an increase in temperature to near the virial temperature of the halo ( we discuss in section [ entropy ] that this is unlikely to be a resolution effect ) . for the purposes of this work , we are most interested in comparing to semi - analytic models of galaxy formation , which adopt the assumption that gas starts at the virial temperature of the halo before cooling to the disk . thus , our adopted definitions below include a temperature criterion . at each time step , we determine the densities and temperatures for each gas particle and compare to the properties of the main halo at that step . in the limit of a strong shock , it follows ( see also db06 ) from the rankine - hugoniot shock jump conditions for a singular isothermal sphere : @xmath20 and @xmath21 where @xmath22 is the gas density prior to encountering the shock , @xmath23 is the post - shock density , and t@xmath13 is the virial temperature of the galaxy halo that the gas encounters . we require that a gas particle must reach 3/8 t@xmath13 to be labeled as `` shocked . '' in practice , however , due to the heating of the intergalactic medium ( igm ) by our uv background , 3/8 t@xmath13 of low mass halos is below the temperature floor of the igm . alternatively , this means that the entropy of these halos is below the entropy floor of the igm ( discussed further in the next section ) , where we adopt the entropy definition of s = log@xmath24(t@xmath25/@xmath26 ) . thus , for a particle to be labeled as `` shocked , '' it must reach a minimum entropy threshold in addition to a temperature ( 3/8 t@xmath13 ) threshold . this entropy criterion requires @xmath27 . \label{eq:3}\ ] ] for each smoothly accreted gas particle at each time step we ask what minimum change in entropy , @xmath28s , is required for the particle to go from its initial entropy , s@xmath29 at step _ t _ , to s@xmath30 by the next time step , _ t_+@xmath31 t ( where , again , @xmath31 t is @xmath0320myr at most redshifts ) . if the entropy at the next time step is in fact @xmath32 s@xmath30 , and the temperature of the particle at that next time step is @xmath32 3/8 @xmath33 , we count that particle as having undergone a shock . in summary , to identify shocked gas we require that a gas particle undergoes a strong increase in both entropy and temperature , using two simple rules : @xmath28s @xmath32 s@xmath30-s@xmath29 , and t@xmath30 @xmath32 3/8 @xmath33 . gas particles which meet these criteria are labeled as `` shocked , '' and the remaining smoothly accreted gas particles are then identified as `` unshocked . '' if filaments undergo either strong shocks or a series of weak shocks within the halo , our definitions will identify both of these processes . finally , we restrict our label of `` shock '' to those particles that shock at the time they enter the main galaxy s virial radius , or afterward . we note that many of our particles do in fact undergo entropy jumps which meet our definition of shocks at earlier times . it is common for these earlier shocks to occur while the gas particle is still at distances of several r@xmath13 from the main galaxy , and at times many gyr before they are ultimately accreted . as we presume these shocks have nothing to do with galaxy accretion ( our focus here ) , we ignore them ( and note that these shocks are likely to occur when the uv background turns on , or in collapse in filaments ) . [ fig1 ] shows the results of our search . the smoothly accreted gas has been divided into `` shocked '' ( red ) and `` unshocked '' ( blue ) , based on the temperature and entropy criteria described above . immediately obvious is that there is a strong trend for the amount of shocked gas within a galaxy to increase with virial mass . the last two columns of table [ accr ] quantify the fraction of shocked and unshocked gas mass to the total _ smoothly accreted _ gas mass . when considering only the fraction relative to smoothly accreted gas ( rather than all gas ) , the trend with mass in the table is even stronger than originally appears in fig . [ fig1 ] . @xcite , k05 , bd06 , k08 , and opt08 found that there is a minimum halo mass required to support a stable shock . similarly , a minimum halo mass to shock is built into our definition , which can be deduced from eqn . [ eq:3 ] , as follows . gas particles in the intergalactic medium will be at the mean baryonic density for that redshift , @xmath34 . however , their temperature will be set by the ionizing uv background radiation . thus , these particles have some entropy , s@xmath35 . if particles with s@xmath35 are to shock and reach s@xmath30 as they enter the galaxy s virial radius , their densities must jump to 4@xmath34 , and their temperature must be 3/8 t@xmath13 . for low halo masses , though , with low t@xmath13 , s@xmath30 is less than s@xmath35 . if this is the case , our shock definitions would lead to the false conclusion in which every accreted gas particle appears to be shocked . effectively , the sound speed in the infalling gas is comparable to the infall velocity so there is no opportunity for a supersonic - subsonic transition . clearly , we must require s@xmath30 @xmath8 s@xmath35 for gas to be physically shocked . this requirement then leads to a minimum halo mass that is capable of shocking , as a function of redshift . for our lowest mass galaxy , h579 , s@xmath30 never becomes greater than s@xmath35 . hence , all of the smoothly accreted gas in h579 is identified as unshocked gas . dwf1 does not shock gas until it reaches a halo mass of 1.1@xmath510@xmath2 m@xmath3 at @xmath12 = 0.6 . this late shock development is seen in detail in fig . [ rate2 ] , which is the same as fig . [ fig3 ] , but now with smoothly accreted gas divided into shocked and unshocked gas . meanwhile , our two most massive galaxies ( mw1 and gal1 ) reach large enough masses to shock very early on . for mw1 , this happens at @xmath12 = 4 when the galaxy is 8.9@xmath510@xmath6 m@xmath3 . gal1 is 1.6@xmath510@xmath2 m@xmath3 at @xmath12 = 6 ( as far back as we trace gas ) , at which time it is already capable of supporting shocks . although these four galaxies do not provide a statistical sample , the mass at which their shocks exist are in good agreement with previous results . for dwf1 and mw1 , the masses listed in the previous paragraph are the masses at which a shock is capable of developing , and thus when a hot halo might develop . gal1 is already capable of shocking at @xmath12 = 6 , so this mass is an upper limit to when a stable shock develops . opt08 show that shocking near the disk radius begins to occur when dark matter halos reach masses of a few@xmath510@xmath6 m@xmath3 to 10@xmath2 m@xmath3(see their figure 4 ) , in good agreement with the masses found for our galaxies . we note that the shock definition of these authors differs from ours , but find below ( section [ temp ] ) that our two definitions yield similar results . these four galaxies were chosen to have their last major mergers at @xmath36 1.5 . if we had instead examined galaxies with late major mergers , they may not cross the shock threshold until low redshifts , and have even higher fractions of unshocked gas accretion . our galaxies here assemble the majority of their mass early , and then quietly accrete gas afterward , meaning that they effectively cross the mass threshold for stable shocks early . thus , by adopting galaxies with similar merger histories , we guarantee that we are comparing galaxies across a range in mass , but with other properties being similar . because of their range in final mass , each galaxy crosses the shock threshold at differing times . galaxies that cross this threshold sooner , i.e. , more massive galaxies , therefore have a longer period of time to accrete shocked gas . this extended time capable of shocking is partly responsible for the well defined trend of accreted shock fraction with increasing mass , in that these galaxies have similar histories and thus display this trend sharply . however , larger volume cosmological simulations that are capable of statistically sampling a large number of galaxies also show this trend ( k05 , k08 , opt08 ) . fig [ fig7 ] shows histograms of the maximum temperature attained by all the gas that is smoothly accreted by our four galaxies . the temperature shown here is the maximum temperature the particle reaches before we stop tracing it ( i.e. , until the particle reaches 30 kpc of the galaxy s center ) . the solid line shows all of the smoothly accreted gas , while the dotted lines show the component of that smooth accretion that has remain unshocked , and the dashed lines show the shocked component . in general , the shocked / unshocked gas is well divided into hot / cold as well . h579 has no shocked gas , but for our two most massive galaxies , mw1 and gal1 , the two components are well separated in temperature . this separation is not altogether surprising , as part of our requirement for being identified as shocked is that the particle reach a minimum temperature ( 3/8 t@xmath13 ) as well as have a sudden entropy increase . the exception is dwf1 . while the unshocked gas extends to much lower temperatures than the hot gas , it overlaps the shocked gas . this result is due to the requirement mentioned above that the shocked entropy of the galaxy must be above the ambient , intergalactic entropy ( s@xmath35 , set by the mean baryon density and the uv temperature floor ) . dwf1 does not cross this threshold until @xmath150.6 . consequently , there exists gas that reaches temperatures above 3/8 t@xmath13 of dwf1 at higher redshifts , and is not included as shocked . it is difficult to say immediately whether this gas _ should _ be counted as shocked gas . note that even the two most massive galaxies , which are capable of supporting a shock over basically their entire traced histories , include a portion of gas that reached high temperatures ( above a few 10@xmath9 k ) yet is labeled as unshocked because it does not undergo a strong entropy increase . previous works on this subject have attempted to identify shocked and unshocked gas by using only a temperature cut ( e.g. , k05 and opt08 used a cut of 2.5@xmath510@xmath9 k ) . in that case , our hot , unshocked gas would have been included as shocked gas . we will investigate this alternative definition to search for hot and cold flows later in the discussion ( section [ temp ] ) . despite the small inclusion of hot gas in the unshocked definition , we find that our definitions do an excellent job of separating gas spatially , into gas accreted along preferred directions in filaments ( unshocked ) and gas accreted from a more spherical distribution ( shocked ) . this distinction was also found by k05 and opt08 , using their temperature definition . [ highz ] demonstrates this for our mw1 galaxy . at @xmath12 = 4 and 3 , the left hand column shows in green the gas particles that will eventually be ( or have already been ) accreted to the galaxy but remain unshocked , while the right hand column shows in green those gas particles that will be accreted ( or already have been ) to the galaxy and shocked . the underlying colors are simply a gas density map , for reference ( black is least dense , white is densest structure ) . each frame is @xmath01100 comoving kpc on a side . shown as a faint green box is the virial radius of our mw1 galaxy at each time . at @xmath12 = 4 , the galaxy has only just crossed the entropy threshold to allow shocked gas , so that it is only accreting unshocked gas at this step . however , by @xmath12 = 3 the galaxy is capable of accreting shocked gas , but is still accreting a substantial amount of unshocked gas in filaments penetrating deep inside of the virial radius . this demonstrates that the two modes of accretion can occur simultaneously at high @xmath12 . the immediately obvious feature in fig . [ highz ] is the clear distinction in spatial distribution at these high redshifts , before most of the gas has been accreted . the unshocked gas is collapsing into filaments that are feeding the galaxy s growth . because these filaments are capable of penetrating deep inside the virial radius , this cold gas is being delivered much closer to the galaxy disk ( see also section [ adiabatic ] , and hence appears to occupy a smaller spatial scale than the shocked gas . the gas that will later be accreted and shocked , on the other hand , has a much more uniform or spherical distribution . the shocked gas develops into a hot halo , much further from the disk . now that we have identified how gas gets to the galaxy , and whether it shocks or remains unshocked as it approaches the disk , we wish to take it one step further and discover what role the various gas accretion mechanisms play in building the galaxy stellar disk . to identify disk particles , the galaxies are first aligned so that the disk angular momentum vector lies along the _ z_-axis . j@xmath37/j@xmath19 is calculated for each star particle in the galaxy , where j@xmath37 is the angular momentum in the _ x - y _ plane , and j@xmath19 is the momentum that a particle would have in a circular orbit with the same orbital energy ( see also * ? ? ? disk stars are identified as those having near circular orbits , so that j@xmath37/j@xmath19 @xmath8 0.8 . for a full discussion on our disk identification , see brook _ et al . _ ( 2008 , in prep ) . [ fig8 ] shows the fraction of stellar disk mass at redshift zero that has formed from gas of various sources , for each of our four main galaxies . again , the total halo mass of the four galaxies is listed below their respective bar , for reference . the total of the blue and red bars shows the fraction of stellar mass in the disk that formed from smoothly accreted gas , with red indicating the fraction that shocked as it entered the galaxy , and blue indicating the fraction that remained unshocked . the green bar shows the fraction of the stellar disk mass that formed from `` clumpy '' accretion , i.e. , gas that at some point belonged to another galaxy halo . finally , the black bar shows the fraction of stellar disk mass that was accreted directly _ as stars _ from other galaxies , while the white shows the fraction of stellar disk mass that formed from gas that was already in the galaxy at @xmath12 = 6 . [ fig1 ] showed that smoothly accreted gas dominates over clumpy accretion for all four of our galaxies . [ fig8 ] shows that this smooth gas goes on to form stars , so that the majority of the disk is composed of stars formed from smoothly accreted gas rather than from gas accreted in mergers or _ stars _ accreted in mergers . table [ disk ] is similar to table [ accr ] but for the stellar disk at redshift zero . unlike table [ accr ] , in which the fraction of smooth and clumpy gas totaled to 1.0 , in table [ disk ] the fractions of stars formed from smooth , clumpy , and `` early '' gas now sum to 1.0 , where `` early '' gas is that which was already in the galaxy at @xmath12 = 6 , i.e. , as far back as we trace the history of our gas particles . however , the fractions of stars that formed from shocked or unshocked gas ( the last two columns in fig . [ fig8 ] ) again sum to 1.0 ( representing the total of stars formed from smoothly accreted gas ) . it can now be seen by comparing tables [ disk ] and [ accr ] that the amount of shocked gas that makes it to the disk to form stars is lower than the total amount of shocked gas that is accreted to the galaxy . in fact , table [ disk ] shows that for all but our most massive galaxy , the stellar disk mass is dominated by stars formed from unshocked , smoothly accreted gas . this domination of unshocked gas in building the stellar disk is even true for mw1 , despite the fact that the majority of smoothly accreted gas to mw1 gets shocked . lcccc h579 & 0.96 & 0.03 & 0.00 & 1.00 + dwf1 & 0.75 & 0.14 & 0.06 & 0.94 + mw1 & 0.64 & 0.30 & 0.24 & 0.76 + gal1 & 0.63 & 0.29 & 0.69 & 0.31 + h277 & 0.67 & 0.23 & 0.32 & 0.68 + mw1.lo & 0.70 & 0.28 & 0.28 & 0.72 + the importance of unshocked gas in the growth of the stellar disk is shown more explicitly in fig . [ diskgrow ] , which shows the star formation histories of those stars that are in the galaxy disk at redshift zero , for each of the four main galaxies ( note , not the sfh of the entire galaxy ) . here we can see that the disk sfr is , in fact , dominated at basically _ all times _ by stars forming from smoothly accreted , unshocked gas , for all but the most massive galaxy , gal1 . even in gal1 , unshocked gas dominated the growth of the stellar disk at an early period ( to @xmath151 ) ; only later does shocked gas begin to dominate . it is interesting to compare fig . [ diskgrow ] to fig . [ rate2 ] , which shows the gas accretion rate onto the virial radius with time . note that in no case is the gas accretion rate exactly balanced by the star formation rate , though again we emphasize that we are only including disk sf here ( and excluding , for example , the sfr of the bulge ) . our two most massive galaxies are particularly interesting . [ rate2 ] shows that their gas accretion histories are generally dominated by smoothly accreted , _ shocked _ gas . however , the star formation history of the mw1 disk is instead dominated by smoothly accreted , _ unshocked _ gas . star formation from gas that has been shocked initially increases steadily with time , until about 5 gyr , after which the disk sfr due to this shocked component still remains relatively low , and eventually declines . gal1 is a dramatic example of this . shocked gas dominates the gas accretion history of the galaxy at essentially all times in the traced history . yet again the initial star formation of the disk is instead dominated by unshocked gas accretion , while the star formation from shocked gas slowly increases with time , before peaking at about 9 gyr and again declining . this delay in star formation for the shocked gas is not surprising . as this gas reaches higher temperatures ( and entropies ) , it has much longer cooling times , slowing the rate at which it can cool to the disk and form stars . as the galaxy grows in time and reaches even higher virial temperatures , the radius at which shocking occurs increases ( discussed further in section [ adiabatic ] ) . this increase in shock radius will potentially make later accretion to the disk even more difficult , due to even longer cooling times . we have quantified the different delay times ( the time between accretion to the virial radius and star formation in the disk ) of the two types of smooth gas accretion . we find those stars that form from shocked and unshocked gas , and measure the average time between a gas particle s accretion to the virial radius and the time when it begins to form stars . this delay is shorter for the unshocked gas . the typical delay time for unshocked gas in our mw1 and gal1 galaxies is 1 - 2 gyr . for shocked gas , the delay times in mw1 are typically 3 - 4 gyr , while for the hotter gal1 the times are longer , 4 - 6 gyr . note that these delays are not intended to be representative of the cooling times of the particles . the delay between accretion and star formation includes both the cooling time of the gas particle onto the disk , the natural time the gas might spend in the disk before producing stars , and subsequent delays in sf due to feedback processes in the disk , which vary as a function of galaxy mass @xcite . in the two lowest mass galaxies , the delay time for unshocked gas is typically longer than for our two highest mass galaxies ( 3 - 4 gyr rather than 1 - 2 gyr ) due to the increasing importance of feedback effects at lower masses . the main conclusion , however , is that we expect longer cooling times for shocked gas compared to clumpy or unshocked gas . thus , clumpy and unshocked gas will be able to make it to the disk to form stars at a faster rate than the shocked gas . in the next two sections we discuss the impact of cold flow ( unshocked ) gas accretion on disk formation . [ gal1 ] shows the rest frame @xmath38 band surface brightness for gal1 at @xmath12 = 1 . this image was created using sunrise , a monte carlo radiative transfer code @xcite . the growth of the disk in gal1 is dominated by unshocked gas accretion prior to @xmath12 = 1 , which clearly results in an observable disk by this redshift . in a future paper we will investigate the evolution of disk sizes since @xmath12 = 1 . [ fig5 ] shows sunrise images of the @xmath39 band surface brightness of stars at @xmath12 = 0 that form from the various accretion modes in our mw1 simulation . the left panel shows stars formed from unshocked , smoothly accreted gas ; the middle panel shows stars that formed from shocked , smoothly accreted gas ; the right panel shows stars that formed _ in the galaxy _ from gas that had been accreted from other halos ( i.e. , clumpy gas accretion ) . these panels show that the stars formed from the clumpy and unshocked , cold gas are found in both the stellar disk and bulge component , while the stars that are able to form from shocked gas are instead predominantly in a disk . this disk formed from shocked gas is also thinner than the disk created by the unshocked or clumpy gas . the scale height of the shocked disk is @xmath025% smaller than the scale height of the unshocked disk . the morphology indicated by fig . [ fig5 ] can be explained in terms of the accretion times of the various gas components , and the subsequent cooling rate of that gas . the earliest disk stars are predominantly formed by unshocked and `` clumpy '' gas , which dominate the accretion rate to the galaxy at the earliest times . this gas can rapidly cool and form stars , but at these early times the galaxy is still undergoing a series of mergers ( major and minor ) that will both heat the forming disk ( leading to a larger scale height for the stars created from unshocked and clumpy gas ) and lead to bulge formation . the shocked gas , on the other hand , is both the dominant mode of accretion at later times and takes substantially longer to cool to the disk after accretion . the shocked gas will therefore settle into the disk after the chaotic merger period . this late settling allows the shocked gas to cool to a thin disk before forming stars ( see also * ? ? ? hence , the scale height of the stellar component formed from this shocked gas is smaller than the unshocked case . [ sb ] dissects these sunrise images further , showing the radially averaged surface brightness profiles for each of these three components of gas accretion . as expected from fig . [ diskgrow ] , the stars that form from unshocked gas dominate the surface brightness ( shown as blue diamonds ; clumpy as green + s ; shocked as red x s ; total as black triangles ) . these profiles are fit well by a sersic bulge component plus exponential disk . the scale lengths , @xmath40 , for each type of accretion are listed in the upper right corner of fig . the stars formed from shocked gas have a longer disk scale length ( 4.1 kpc ) than those formed from unshocked ( 3.5 kpc ) or clumpy gas ( 3.3 kpc ) . although the picture is confused slightly due to the longer cooling times of the shocked gas , this scale length difference may not be surprising given the differing accretion times for these two components . although there is still a low level of unshocked gas accretion at the virial radius all the way to redshift zero , the majority of the unshocked gas is accreted prior to @xmath12 = 2 when the mean angular momentum is expected to be lower , and when chaotic merging will heat the disk . after this time , shocked gas dominates the gas accretion to the halo , when the mean angular momentum is expected to be higher , and will therefore form a disk with a longer scale length . the angular momentum content of this gas will be investigated in detail in a future paper . an important point to draw from the previous two sections is that , particularly as seen in our two most massive galaxies , a significant amount of star formation is possible in disks prior to @xmath12 = 1 ( when the universe was 6 gyr old ) . as we have shown above , this early star formation is predominantly due to the contribution from unshocked gas accretion . of course , the disks continue to grow since @xmath12 = 1 , but an existing disk is in place already by that time . this is in accord with observations that find that large galaxy disks must be assembled prior to @xmath12 = 1 @xcite , or even by @xmath12 = 2 @xcite . theoretical models of disk formation suggest instead that large disks must have formed since @xmath12 = 1 and hence undergone a large change in scale length since then , @xcite . even the most up - to - date models still suggest a discrepancy with observational results @xcite , with observed disks being larger than predicted at @xmath12 = 1 . we will address disk sizes in a future paper , as it also requires a detailed examination of the angular momentum content of the accreted gas . we focus here on the age of the stellar disk . it is particularly interesting to examine the stellar disk growth for our two most massive galaxies , which have been dominated by shocked gas accretion over most of the history we trace here . the unshocked gas accreted to these galaxies never reaches a temperature above 3/8 t@xmath13 . sams adopt a model in which this gas would first have been heated to the virial temperature of the halo as it entered the virial radius , thus delaying its ability to cool quickly to the disk . as demonstrated in fig . [ fig7 ] , the temperature difference between the shocked and unshocked gas can be a factor of 10 . this temperature discrepancy may lead to significant differences between the sfrs found here and in sams . in order to quantify the difference in sfrs between sams and our simulations , we estimate the effect on the disk sfh if the cold , unshocked gas had instead first been heated to t@xmath13 . the resulting change in the disk star formation rate is not straightforward to assess . even our cold flow gas does not immediately form stars when it reaches the disk . rather , there is a delay in star formation due to feedback effects in the disk . as found by @xcite , this delay is a function of galaxy mass , being longer in low mass galaxies , so that low mass galaxies form stars inefficiently compared to higher mass galaxies . as discussed in section [ early ] , the time between accretion to the virial radius and formation of disk stars is much longer for shocked gas than for unshocked gas . we find the shocked and unshocked gas particles accreted between each output time step interval , and the mean difference in formation time for stars that subsequently form from that gas . to quantify the affect of heating our unshocked gas to high temperatures , we adopt this average difference in formation time between shocked and unshocked star particles , and add this delay to the formation times of the star particles born from unshocked gas . although not precise , this method captures the relevant physics , including cooling times and suppression of star formation by feedback , and will provide a meaningful estimate of the effects of assuming that all disk gas has been shock heated . the result of adding this delay is shown in fig . [ cooltheory ] , for mw1 and gal1 . the top panels show the actual disk sfrs from our simulations as a solid line , exactly as in fig . [ diskgrow ] , but now showing the total disk sfr from unshocked , shocked , and clumpy accreted gas together . the dashed line shows what the resulting sfr would be if the unshocked gas had been heated to the same temperatures as the shocked gas . the middle shows the resulting cumulative mass of the stellar disk as a function of time , again as actually seen in the simulations ( solid line ) and with the false delay added to the unshocked gas ( dashed line ) . the bottom panels show the difference between the two disk masses in the middle panel ( the delayed disk mass as a fraction of the actual disk mass from the simulation ) . [ cooltheory ] demonstrates that if the cold gas had reached virial temperatures that the sfr in the disk would have been significantly delayed . rather than being able to build the stellar disk within the first few gyr , the disk does not begin to grow substantially until @xmath151 the bottom panel shows that the disk is 50% less massive until @xmath12 = 1 in the case of strong heating than it actually is in the simulation , where the unshocked gas does not approach the virial temperature . we discuss in this section the impact of key numerical effects on our results . the effect likely to have the largest impact on our results is the interval between our outputs . given our finite step interval of @xmath0320 myr , it is possible that a particle with a short cooling time may radiate away its energy quickly enough that we miss an increase in temperature and entropy associated with a shock event ( making our shocked fractions a lower limit ) . this omission is more likely to occur for those particles shocking near the disk ( at @xmath0 30 kpc ) where the densities are much higher ( and hence cooling times shorter ) than at the virial radius . to investigate the effect of our step size on our results , we have run a new simulation with output steps at 1/12 of the time resolution presented above . this new simulation , h277 , is a milky - way like galaxy with a rather quiet merger history and a total halo mass of 7.1@xmath510@xmath2 m@xmath3 ( see table [ simsum ] ) at @xmath12 = 0 . for the same step interval as was used above , the gas accretion results for this galaxy agree well with the trends shown for the previous four galaxies . from table [ accr ] it can be seen that the fraction of shock accreted gas lies nicely between dwf1 and gal1 , as might be expected for this mass of galaxy . a change in time resolution has an effect at two different stages of our analysis . first , as just mentioned , we may miss particles with short cooling times that should have been counted by our `` shocked '' definition . second , `` clumpy '' gas is identified as belonging at any output time step to a halo other than the main galaxy halo . we are likely to miss gas particles that belong to another galaxy for less than @xmath0320 myr ( the time between output steps ) , e.g. , because they are subsequently lost from the galaxy due to sne . the result of missing this clumpy accretion will be to increase the number of smoothly accreted particles instead . these two stages of our analysis are not independent of one another , since the sample of shocked gas is drawn only from those gas particles that are identified as smooth accretion . we wish to quantify them separately . we first test how the number of particles identified as `` clumpy '' changes as a function of time resolution . we investigate the change at 12x the original step interval used above ( i.e. , fine sampling corresponds to @xmath027 myr resolution per output ) . when using our standard step interval , the smooth gas makes up 73% of all the gas accreted by galaxy h277 over the history of the trace . by increasing the output interval resolution by a factor of 12 , the gas identified as smoothly accreted declines , so that it composes 66% of all the gas accretion over the galaxy s history , for a total decline of 7% . thus , by definition , the fraction of gas accreted as `` clumpy '' over the galaxy s history must increase by 7% . this result means that our absolute numbers stated above for the fraction of gas that is smoothly accreted to each galaxy is an upper limit . an examination of the accretion history of this galaxy shows smooth accretion with characteristic spikes indicative of mergers . the amount of gas in these spikes identified as `` smooth '' decreases in our fine time step limit , but the spikes still remain , suggesting we are still missing some gas in mergers . by generously estimating the mass of gas contributing to these accretion spikes , we find that we are missing only an additional 1% of clumpy gas accretion . thus , we agree with previous studies that find that the majority of gas mass is accreted smoothly rather than in mergers . next , we use the `` smooth '' particles identified in the standard output resolution ( 320 myr outputs , where smooth gas accounts for 73% of all the gas accreted over the history of the galaxy ) to isolate the effect of output interval on the fraction of particles identified as shocked . at finer step intervals , the fraction of smooth gas that is shocked increases by 2.3% . this increase is quite small , and occurs primarily at @xmath12 @xmath8 2 , likely to due to the fact that we do not adequately sample cooling times at high redshift . to verify this , we calculated the instantaneous cooling rates for particles in an annulus between 30 - 40 kpc ( as close to the disk as we trace them ) and inside 10 kpc of r@xmath13 at each of our `` standard '' analysis intervals . for h277 , the virial radius does not reach 30 kpc until @xmath12 = 2.65 , and the cooling times of the particles within r@xmath13 are lower than our standard step resolution of 320 myr until this redshift . after this redshift , however , the galaxy grows quickly and the cooling times are longer than our output intervals , so that we are capable of easily identifying shocked gas . thus , it is not surprising that the addition of shocked gas when increasing time resolution comes at high redshifts . we have done a similar cooling time analysis for our four main galaxies . in the case of dwf1 , mw1 , and gal1 , we do not fully resolve cooling times within the halos until redshifts 3 , 3 , and 4.5 , respectively . however , because dwf1 is not massive enough to support a stable shock until @xmath12 @xmath00.6 , our time step undersampling at high @xmath12 has no impact on our resulting fraction of shocked gas . for mw1 and gal1 , we can expect about a 5% change , as we found in h277 , and perhaps less since we resolve their cooling times to higher redshift than in the h277 case . as an interesting note , our lowest mass galaxy , h579 , always has cooling times longer than our standard output step interval ( though it is never massive enough to shock , so this is not important for our results ) . it may seem odd at first that this galaxy has such long cooling times , but it is a result of the lower densities in this halo . our uv background is capable of ionizing much of the hydrogen gas in the halo of this galaxy , causing it to be inefficient at cooling . we verified that turning off our uv background lowers the instantaneous cooling times substantially , by almost an order of magnitude @xcite . by definition , a shock is a phenomenon that leads to a sharp discontinuity in the physical properties of a gas . due to the nature of smoothed particle hydrodynamic codes , these properties for any given gas particle will be a smoothed estimate based on the properties of its nearest neighbors . thus , there is some concern that sph will do a poor job of resolving shocks . gasoline adopts the standard sph artificial viscosity formulation @xcite , with a correction to detect and reduce viscous transport in shearing regions @xcite . @xcite investigated gasoline s performance in this area with both the standard @xcite shock tube test , and a spherical adiabatic collapse test @xcite . it was found that gasoline does an excellent job of matching the analytic solutions to within a few particle separations , and matches the post - shock entropy values very well . it can be possible for low numerical resolution to artificially broaden shocks ( due to entropy generated by the artificial viscosity used in sph ) , allowing gas to radiate energy while passing through a shock rather than heating and undergoing an entropy jump @xcite . this broadening is unlikely to be an issue at the high resolutions used here @xcite , but we verified numerical convergence by using a lower resolution version of our mw1 run , mw1.lo , and performing the same analysis . this run has 1/8 of the mass resolution and 1/2 the force resolution of mw1 ( and is the same galaxy discussed in detail in * ? ? ? table [ accr ] demonstrates that the shocked fraction in this run is in excellent agreement with mw1 . since our shock definition includes a temperature criterion , this implies that we are not missing high temperature gas , and thus that we have a sufficient numerical resolution to identify entropy jumps and are not missing heating in shocks . hence , our cold flow gas accretion is not an artifact of numerical resolution . previous groups who have attempted to investigate the role of cold flows in galaxy gas accretion have used a different criterion to identify shocked and unshocked gas , adopting a strict temperature cutoff . we examine in this section how our results differ if we use the same temperature definition as other authors . k05 and opt08 have noted that the temperature distribution of accreted particles can be bimodal , with a minimum at 2.5@xmath510@xmath9 k. they use this minimum to separate gas into cold accretion and hot accretion modes . we also see this minimum in the temperature distribution , as shown in fig . [ mwhist ] for our mw1 galaxy . this figure shows the maximum temperature of gas particles that have been accreted to the galaxy between successive redshift intervals . up until @xmath12 = 0.5 , there is a strong bimodality . the dashed vertical line shows the separation at 2.5@xmath510@xmath9 k used by k05 , k08 , and opt08 . our accreted gas particles clearly demonstrate this same minimum . these two groups have previously noted that this minimum is a function of the cooling curve , and has therefore has little direct relation to the actual modes of hot or cold accretion . however , it is a convenient point at which to separate the gas into hot or cold accretion modes . to assess how our results are affected by these definitions , we repeat the above analysis to identify shocked gas in our smoothly accreted gas component , using the temperature criterion of k05 and opt08 . the same set of smoothly accreted particles is used , so that the only change between the two definitions will be in the resulting shocked or unshocked fraction ( or , more appropriately , hot and cold fraction ) . the difference resulting from these two separate definitions is remarkably small . the temperature criterion leads to slightly more hot gas being accreted to the galaxy than shocked gas , but the largest difference is only 7% , in the case of gal1 . the remaining galaxies have differences of 3% or less . h579 never reaches the temperature cutoff identified by k05 , and thus both studies agree than it has never accreted shocked / hot gas . these tests lead us to conclude that the two different definitions yield very similar results . this agreement may be partly due to the fact that we also impose some sort of temperature criterion ( requiring that our gas not only undergo an entropy increase , but reach 3/8 t@xmath13 as well ) . however , this result was not guaranteed . we emphasize that the temperature criterion alone is based on the physics of the cooling curve , and does not reflect the physics of the gas accretion shocks . hence , we choose to add an entropy criterion to our definition , as it is unique to each galaxy halo and the physics of the various accretion modes . @xcite found that the mass function of our galaxies is resolved down to a minimum of 64 dm particles . for the runs used here , that corresponds to total halo masses of 8.1@xmath510@xmath10 m@xmath3 , 6.1@xmath510@xmath10 m@xmath3 , 4.9@xmath510@xmath41 m@xmath3 , and 1.9@xmath510@xmath42 m@xmath3 for h579 , dwf1 , mw1 , and gal1 , respectively . these masses are well below the limit where the uv field will suppress galaxy formation @xcite , and thus we avoid issues due to sub - resolution merging . that is , there should be no unresolved halos that are being counted incorrectly as `` smooth '' accretion . this assertion was verified by running the same analysis on a lower resolution version of our mw1 galaxy , mw1.lo , with a mass resolution of 1.9@xmath510@xmath42 m@xmath3 at 64 dm particles . this mass is just below the limit that the uv field should suppress galaxy formation . indeed , the smooth gas fraction agrees to within a few percent with mw1 ( see table [ accr ] ) . mw1.lo does have a five percent higher fraction of smooth accretion , but given fluctuations in the simulations , this is excellent agreement . the dominant effect of the uv background in our simulations is to suppress galaxy formation at low masses , and hence lower the amount of clumpy accretion . gas which would have otherwise been in small halos is now free to be smoothly accreted to our galaxies . this was confirmed by running mw1.lo with the uv background off ( and cooling and sf turned off as well ) . the results for this galaxy , mw1.ad , are shown in table [ accr ] . it is clear from the first two columns that the amount of `` clumpy '' accretion material increases from about 20% to about 50% . this confirms that our uv background is partly responsible for reducing the amount of gas accreted from other halos , and hence increasing the amount of gas available for smooth accretion . the more interesting effect of the uv background is on the amount of shocked gas accretion to the galaxy . our uv background ( in combination with the mean baryonic density of the universe as a function of redshift ) sets a temperature floor in the igm , and hence an entropy floor for igm gas . for a galaxy to be capable of shocking accreted gas , the virial entropy of the galaxy halo must be above this background entropy by a certain factor . hence , based on our definitions , the uv background sets a minimum halo mass that will be able to shock , as function of time . @xcite and db06 showed that the shock radius in the case of adiabatic accretion ( i.e. , no cooling ) is well matched to the virial radius in a spherically symmetric case . they also demonstrated that when cooling is turned on , the same galaxy is not initially massive enough to support a stable shock , so that accreted gas shocks at the disk of the assembling galaxy . only after a critical mass is reached can the galaxy support a shock . we have again used mw1.ad with the uv background , cooling , and sf turned off ( i.e. , the adiabatic case ) to test these results . as noted above , the amount of smooth accretion drops for mw1.ad , as much of the gas is able to remain bound to its dark matter halo . however , from table [ accr ] ( last two columns ) it is clear that the amount of smooth gas accretion that is shocked also increases in this case , increasing to 84% . this behavior is expected , and is similar to the standard spherical model in which all gas gets heated to the virial temperature of the galaxy as it is accreted . the fact that there is unshocked gas at all is due to the fact that this simulation is not spherically symmetric , and filaments of cold gas still exist that are capable of penetrating well inside of the halo without shocking . [ shockcomp ] shows the radii at which particles are shocked inside of r@xmath13 , as a function of time , for the mw1.lo and adiabatic , mw1.ad , runs . contours are used for clarity , with black representing the least dense concentration of particles , and white peaks the most dense . the solid lines show the virial radii of the galaxies as a function of time . the main difference between the two runs is at high redshift . in the adiabatic case , particles are able to shock inside the virial radius even at high @xmath12 , while in the mw1.lo run there is no shocking at the highest redshifts . shocks begin to occur near the disk at redshift 4 , and shocked particles fill the space between the disk and the virial radius at lower redshifts . the radius with the highest concentration of shocked particles moves outward with time in both cases . metal line cooling has recently been added to gasoline ( shen et al . , in prep . ) . mw1.lo was rerun with metal cooling on , to find if the new cooling has any effect on the results here . however , the smoothly accreted gas ( be it shocked or unshocked ) has nearly primordial metallicities , and the resulting fractions of shocked and unshocked accreted gas are identical to the results for mw1.lo with only h and he cooling . likewise , the fraction of disk stars formed from shocked and unshocked gas is also identical to mw1.lo . we have tracked the history of gas accreted to four high resolution disk galaxies that span two orders of magnitude in galaxy mass . for all of these galaxies , early disk growth can be predominantly attributed to cold gas , and up to milky way masses the sf of the stellar disk is dominated by unshocked , cold flow gas accretion at essentially all times . this dominance of cold gas in the formation of disk stars can exist even when the the majority of gas accreting to the virial radius is being shocked and reaching high temperatures . at galaxy masses above l@xmath1 , an early phase of disk growth prior to @xmath12 = 1 can be attributed to the presence of cold gas accretion . we have examined gas temperature histories in relation to the virial temperature of the halo that they are accreted to . while the shocked component is capable of reaching temperatures near the virial temperature of the halo , the unshocked component of gas accretion remains below 3/8 of the virial temperature at all times . the important role of unshocked , cold flow gas in the growth of galaxies is consistent with previous studies that adopt an explicit temperature cutoff , determined by the cooling curve , to identify cold flow gas . we confirm that the ability of smoothly accreted gas to shock is a strong function of mass . however , our focus here has been to study the subsequent role of this cold gas accretion in the growth of stellar disks . because the unshocked gas does not reach the same high temperatures as shocked gas , it is capable of cooling to the disk faster and forming stars . it is important to note that this early cooling of unshocked gas to the disk is not a manifestation of the historic `` overcooling '' problem . this can be seen by examining the results from our stellar mass - metallicity relationship , which matches the observed relation at both low and high redshifts @xcite . as discussed by both @xcite and @xcite , this match is achieved by overcoming the overcooling problem with both high resolution and a physically motivated sn feedback model . previous works consistently produced metallicities too high when compared to observations at high @xmath12 , due to overcooling and a rapid consumption of gas and subsequent metal enrichment . instead , our sn feedback scheme regulates the sfrs of our galaxies , preventing them from too quickly consuming their gas . the early sf seen in our more massive galaxies is due instead to the accretion of cold gas that can cool rapidly to the disk to form stars . despite running our simulations in a fully cosmological context that includes galaxy mergers , the dominant gas supply is from smoothly accreted gas that has never belonged to another galaxy halo . for a milky way mass galaxy with a quiescent merging history like the galaxies shown here , only @xmath025% of the accreted gas is acquired from other galaxy halos , resulting in a similar fraction of the disk stars forming from this gas . it is our goal to assess the impact that cold flow gas accretion may have in altering current models of disk galaxy formation . a direct comparison with semi - analytic models is difficult to do and beyond the scope of this paper . current sams include the possibility of rapid gas cooling for low mass galaxies . it has been suggested that this regime is similar to the cold flow gas accretion seen in simulations @xcite . @xcite performed a detailed comparison of the galics sam @xcite to the sph results of k05 , and found that the cold gas accretion rates in the two schemes were in agreement . this implies that the amount of cold gas currently available in sams may agree well with simulations , though this needs to be confirmed with further comparisons . however , the results in this paper may emphasize a separate problem within the cold flow regime . even in the case of rapid cooling in sams , the temperature of the gas is initially set to the virial temperature of the galaxy halo , and cooling times are based on this assumption . prior to the development of a hot halo , the free fall time determines when the gas reaches the disk in either the sams or the simulations . however , sams do not yet model the impact of filaments on galaxy growth . our two most massive galaxies , which are able to develop a hot halo at high redshifts ( 4 to 6 ; see also * ? ? ? * ) , are still capable of accreting cold gas that never approaches the virial temperature of the halo in filaments at high @xmath12 . these filaments allow a significant amount of cold gas mass to accumulate inside of the galaxy , even for massive galaxies capable of supporting a shock at early times . this cold gas cools quickly to the disk to form stars . this rapid settling to the disk allows for the building of stellar disks at higher redshifts than predicted if all gas is heated to the virial temperature , as in the standard model . we have shown that the discrepancy in temperatures between the shocked and unshocked gas components leads to significantly longer cooling times for the shocked gas ( note that their densities will be discrepant as well , since most of the unshocked component are found in dense filaments , exacerbating the difference in cooling times ) . if the unshocked gas had reached the virial temperature , the star formation rates in the disk would have been delayed ( in the cases shown here , until after @xmath12 = 1 ) . a study of the angular momentum content of the accreted gas will be examined in future work , but the presence of cold flow gas accretion , particularly in filaments in massive halos , allows for the building of large , massive disks at high @xmath12 , in agreement with observations . we would like to thank a. dekel , d. keres , j. dalcanton , s. white , and q. guo , for helpful conversations during this project . ab , fg , and tq were supported by nsf itr grant phy-0205413 . ab acknowledges support from a wa space grant fellowship . fg acknowledges support from a theodore dunham grant , hst go-1125 , nsf grant ast-0607819 and nasa atp nnx08ag84 g . simulations were run at tacc , sdsc , cineca , and nas .

globular clusters in extragalactic systems have established themselves as powerful diagnostics of the star formation history , dynamics , and structure of their host galaxies ( see ashman & zepf 1998 for a recent review ) . most of the observational studies of extragalactic globular cluster systems ( gcss ) have concentrated on the systems of early type galaxies , where globular clusters are easily identified against the smooth , dust free background . these studies have uncovered the characteristic properties of such globular cluster systems , as well as notable galaxy - to - galaxy variations in these properties . in comparison , information on the globular cluster systems of late - type galaxies is sparse . the use of globular cluster systems as probes of the formation and evolution of galaxies is severely limited by this scarcity of data for spiral galaxy gcss . there are two primary problems . the first is that much of the interpretation of early - type galaxy gcss has used the well - studied globular cluster system of the milky way as a benchmark . for example , the result that the globular clusters of elliptical galaxies are more metal - rich in the mean than those of spiral galaxies is largely based on the mean metallicity of milky way globulars . consequently , it is extremely important to establish whether the properties of the milky way globular cluster system are typical of spiral galaxy gcss . the second issue is that a better understanding of the gcss of late type galaxies is needed in order to fully utilize gcss as probes of the formation and early evolution of galaxies . this is self - evident for the case of late type galaxies , but is also a critical issue for constraining models of the formation and early evolution of elliptical galaxies . this is particularly important for testing the predictions of the merger model , in which elliptical galaxies are formed in the merger of spiral galaxies . in this picture , the gcss of elliptical galaxies are composite systems . one population of clusters is associated with the progenitor spirals , while a second population of clusters forms in the merger event . ashman and zepf ( 1992 ) described several testable predictions arising from this scenario . however , the limited information on the the gcss of spirals leads to uncertainties in these predictions since it is currently unclear what constitutes a typical spiral galaxy gcss . one particular problem is that the characteristic specific frequency of globular clusters around spirals is poorly known . the absence of firm constraints on the number of globular clusters contributed by the progenitor spirals to an elliptical galaxy gcs leads to uncertainties in many of the detailed predictions for these systems that follow from the merger model . one of the key results of the present study is to add to the database of globular cluster specific frequencies for spiral galaxies . in this paper , we present an imaging study of the globular cluster systems of two edge on spiral galaxies : ngc 4565 and ngc 5907 . globular clusters were already statistically detected in ngc 4565 as stellar - like objects above the mean background by van den bergh & harris ( 1982 ) and fleming et al . the former detected @xmath4 100 clusters to an equivalent limiting magnitude of @xmath5 , and found the clusters to be compatible with a surface density profile of the form @xmath6 , similar to the profile of the milky way system . the latter study determined the peak of the globular cluster luminosity function to be at v@xmath7 . fleming et al . ( 1995 ) extrapolated the observed number of clusters ( @xmath8 ) to a total number of @xmath9 globular clusters and a specific frequency of @xmath10 under their assumptions for luminosity and distance . ngc 5907 has been searched for globular clusters by harris , bothun & hesser ( 1988 ) , but no globular cluster could be detected . the authors speculated that the lack of globular clusters was related to the non existence of a central bulge . in section 2 , we present the properties of the galaxies and the observations , and describe the photometry against a highly variable background . in section 3 , the results are analyzed . in section 4 we compare the globular cluster systems to each other and to the milky way . concluding remarks are made in section 5 . we obtained wfpc2 observations of the nearby , edge on spiral galaxies , ngc 4565 and ngc 5907 ( program go 6092 ) . basic properties of the two galaxies are given in table 1 and a composite of our images is shown in fig . 1 . ngc 4565 is classified as an sb galaxy ( rc3 , hubble atlas ) with an inclination of @xmath11 ( garca - burillo et al . 1997 ) , and is at a distance of 10 mpc , based on surface brightness fluctuations and planetary nebulae ( jacoby et al . 1996 and references therein ) . ngc 5907 is classified as an sc galaxy ( rc3 , hubble atlas ) , with an inclination of @xmath12 ( morrison et al . 1994 ) , and is at a distance of 12 mpc , based on the tully fisher relation ( bottinelli et al . 1988 , schniger & sofue 1994 ) . the two galaxies are very similar to each other in luminosity , but differ considerably in their disk characteristics . ngc 4565 has a prominent thick disk , accounting for roughly @xmath13 of the total luminosity ( van der kruit & searle 1981 ) . ngc 5907 has a less prominent bulge and an upper limit to the thick disk luminosity of 0.6% of the thin disk value a relative contribution of less than @xmath14 of the thick disk of the milky way ( morrison et al . 1994 ) . for both of our target galaxies , ngc 4565 and ngc 5907 , we obtained two pointings with the wfpc2 on hst . as shown in fig . 1 , we aligned each pointing so that the major axis was parallel to the edges of two wf chips , the third wf chip and the pc sampling part of the halo . for every pointing , images in the f450w and f814w filters were obtained with total integration times ranging from 600 sec to 780 sec in the f450w filter , and of 480 sec in the f814w filter . every pointing was split in three exposures shifted by 0.5 . the basic image reduction was carried out under iraf . for every pointing , the three calibrated science images provided by the space telescope science institute were shifted by 0.5 ( using the task imshift ) and trimmed in order to be aligned . they were then combined with the task crrej that rejected the vast majority of cosmic rays.the images were transformed into fits format for the photometry . the photometry was carried out using the sextractor software ( bertin & arnouts 1996 ) . the finding parameters were set to 2 connected pixels 2.5@xmath15 above the local background ( computed in a 10 by 10 pixel mesh ) . the computation of a local background in a relatively fine grid was chosen to allow the finding criteria to be equivalent in the various parts of the image , in particular towards the disks where the background becomes very irregular . typically , down to an equivalent b magnitude of 25.5 , 0 to 2 objects were detected in both the f450w and f814w image on the pc , 8 to 18 on the wf `` halo '' field , and 130 to 340 on the wf fields including disc and bulge . the latter included a large number of stellar associations and hii regions . positions on the sky ( ra and dec ) were computed for all objects using the task metric in the stsdas package . distance from the center , along the disc , and orthogonal distance from the disc were computed using position angles of @xmath16 for ngc 4565 ( rupen 1991 ) and @xmath17 for ngc 5907 ( garca - burillo et al . 1997 ) . the photometry was carried out with 2 pixel radius apertures in order to avoid being affected by background irregularities . corrections to get the equivalent magnitude to a 0.5 aperture were determined from isolated bright objects on the images and found to be 0.35 and 0.18 mag for the pc and wf f450w measurements respectively , and 0.55 and 0.21 mag in the f814w filter , with errors of @xmath18% . additional 0.1 magnitudes were added to obtain the total magnitude of the objects ( derived for the wfpc2 point spread functions , following holtzman et al . note that at the distance of our galaxies the largest globular clusters will be resolved and these corrections underestimated . this will hardly affect the colors since the luminosity profiles in f450w and f814w are similar , but could significantly affect the individual magnitudes , as a function of the object size ( cf . kundu & whitmore 1998 , puzia et al . 1999 ) . finally , the calibration relation given in holtzman et al . ( 1995 ) were used to obtain johnson cousins b and i magnitudes . extensive artificial star experiments with artificial point sources and artificial globular clusters were carried out . for the latter wfpc2 point spread functions obtained with tiny tim ( krist & hook 1997 ) and convolved with a modified hubble law of core radius 0.1 and 0.2were used . the latter correspond to core radii of about 4 to 10 pc at the distance of 10 to 12 mpc , and lie in the upper range of the values measured in the milky way ( only 27% of the milky way cluster have a core radius larger than 4 pc in the compilation of harris 1996 ) . the resulting fwhm of our artificial objects range from 0.8 to 1.5 pixels and 1.5 to 2.5 pixels for artificial stars and globular clusters respectively , at b@xmath19 . these allowed us to define robust selection criteria for globular clusters based on the fwhm of the detected objects ( @xmath20 and @xmath21 pixels , as returned by sextractor ) . these selection criteria were applied down to b@xmath22 ; at fainter magnitudes the signal to noise was not sufficient to discriminate with high confidence between point sources and extended objects . note that down to this magnitude limit our finding algorithm is 100% complete and objects will only be missed if they are physically obscured , e.g. by dust . we further introduced a cut in color and magnitude . for the color cut , we chose b i @xmath23 , roughly 0.2 magnitudes bluer than the bluest globular clusters in the milky way ( see also fig . 3 ) . no upper limit was chosen to avoid excluding reddened objects . as a magnitude range , we chose b@xmath24 , corresponding to b@xmath25 and @xmath26 at the respective distances of ngc 4565 and ngc 5907 . as a reference , the brightest clusters in the milky way and m31 have b@xmath27 ( @xmath28 cen ) and b@xmath29 ( mayall ii ) . our upper magnitude cut was dictated by the fwhm selection , only reliable down to b@xmath22 which was used as the limiting magnitude . the peak of the globular cluster luminosity function is expected at b@xmath30 , as derived by della valle et al . ( 1998 ) , using the new hipparcos distance calibration and the list of milky way clusters of harris ( 1996 ) . this corresponds to b@xmath31 and b@xmath32 at the distance of the two galaxies , so that we pass the peak of the luminosity function in both cases . the expected number of foreground stars contaminating our images and passing our selection criteria were estimated from galactic models ( ibata , priv . com . ) and the hubble deep field to be @xmath33 per pointing . the number of contaminating galaxies was estimated from several fields of the medium deep survey ( griffiths et al . 1993 ) and our selection criteria turned out to be robust against background galaxy contamination at these bright magnitudes ( b@xmath19 ) estimated to be @xmath34 per pointing . the final samples include 40 and 25 globular clusters in ngc 4565 and ngc 5907 respectively ( see table 2 and 3 ) . all candidates ( except one in ngc 5907 identified as a star ) were confirmed by a visual inspection . the distribution of globular clusters , together with all objects detected is shown in fig . 2 . the colors of the globular cluster candidates in ngc 4565 , as well as visual inspection of the galaxy dust lane , suggests that the majority of the clusters with positive z suffer significant extinction . the identifiable regions of dust are less extended in ngc 5907 and there is less direct evidence for significant extinction , but we can not rule out reddening of the clusters in this galaxy . color magnitude diagrams for the detected globular clusters in ngc 4565 and ngc 5907 are shown in fig . 3 , together with one for milky way globular clusters ( uncorrected for reddening ) as comparison . the mean colors and dispersion for the samples are @xmath35 in ngc 4565 , @xmath36 in ngc 5907 , and @xmath37 in the milky way ( @xmath38 when corrected for reddening , corresponding to a mean metallicity of [ fe / h]@xmath39 dex ) . this tends to support the evidence that reddening is more of an issue in ngc 4565 than ngc 5907 , at least if the _ intrinsic _ colors ( @xmath4 metallicities ) of the gcss of the three galaxies are similar . one cautionary note in this regard is that the bluer mean colors of the ngc 5907 clusters is partly driven by some very blue objects amongst the globular cluster candidates . since these blue objects are not preferentially located near the disk , it is unlikely that they are young compact star clusters , but it is possible that one or two such objects maybe be included in the sample . despite these complications , the small differences ( @xmath40 ) in the mean @xmath41 values probably suggest only moderate differences in the mean metallicities ( @xmath42 dex following the conversion of couture et al . in order to estimate the total number of globular clusters around ngc 4565 and ngc 5907 , three extrapolations are required . the first is to account for globular clusters fainter than our magnitude limit . the second is for globular clusters that are not covered by our imaging field . the third is for globular clusters lying behind the disk and bulge and being obscured . we chose to estimate the counts by direct comparison with the milky way and estimate the total number of globular clusters in the two galaxies by comparing our observations with the same spatial region and equivalent magnitude limit in our galaxy . this approach has the advantage of taking into account all three extrapolations at once without propagating unnecessary errors . it is motivated by the small number of objects detected and the limited spatial coverage of our observations . these do not allow us to derive the spatial profile of either globular cluster system , necessary to estimate the spatial extrapolation . further , we would still need a model to account for the clusters behind the disk . specifically , we created a mask for each galaxy based on the areal coverage of the two wfpc2 pointings of our observations . these masks excluded regions between chips and the edges of chips where the images were trimmed . also excluded were small regions ( @xmath4 150 @xmath43 ) centered on the bulges where our artificial star experiments indicated the data to be incomplete . by imposing the relevant mask onto the milky way clusters and using an absolute magnitude cut appropriate to @xmath44 at the distances of the two galaxies , we determined the number of milky way globular clusters that would have been detected with our observations . we used positions and magnitudes of milky way globular clusters taken from the mcmaster catalog ( harris 1996 ) to create a 2-dimensional spatial distribution by projecting galactocentric @xmath45 coordinates into the @xmath46 plane . figures 4 and 5 give a visual representation of the results of this process . for each galaxy , there are four possible orientations of the mask that preserve the alignment of the pointings along the major axis of the galaxy . we therefore repeated the above exercise for all four orientations and in each case counted the number of milky way globular clusters that were `` detected '' ( see table 4 ) . the mean number of clusters from these four possibilities was used to estimate the total number of globular clusters , @xmath47 , in each galaxy , using the expression : @xmath48 where @xmath49 is the total number of globular clusters in the milky way ( @xmath2 , ashman & zepf 1998 ) , @xmath50 is the number of globular clusters observed in our target galaxies ( 40 and 25 for ngc 4565 and ngc 5907 , respectively , corrected to 38 and 23 when accounting for fore-/background contamination ) , and @xmath51 is the mean number of globular clusters from the four mask orientations ( see table 4 ) . the method relies on several assumptions that we discuss in turn . the mcmaster catalog lists 141 globular clusters , 134 of which have galactocentric coordinates , and 112 of which have de - redenned absolute b magnitudes ( which served for the magnitude cut ) . however , the clusters with no reported @xmath52 magnitudes are almost exclusively more than one magnitude fainter than the peak of the luminosity function ( based on their available absolute @xmath53 magnitudes ) and would also be undetected at the distances of our target galaxies . we assumed the total number of globular clusters in the milky way to be @xmath2 . the majority of undetected milky way globulars are assumed to be behind the galactic bulge . the counterparts of such globular clusters in ngc 4565 and ngc 5907 would also be undetected in our observations due to obscuration by disk and bulge . the scale length and scale heights of the thin disk of all three galaxies is not too different ( e.g. van der kruit & searle 1981 , morrison et al . 1994 , kent et al . we therefore assume than the number of obscured objects is the same in all galaxies ( within 20% ) . thus while our milky way sample is not complete , the missing clusters tend to be ones that we would not detect in ngc 4565 and ngc 5907 anyway . we assumed that the globular cluster luminosity functions of ngc 4565 and ngc 5907 are similar to the one of the milky way , i.e. roughly gaussian with a peak at b@xmath54 and a standard deviation of @xmath55 ( see sect . this assumption of a ` universality ' of the globular cluster luminosity functions is well supported by all recent observations ( e.g. whitmore 1997 for a recent review ) . for our assumed distances ( see table 1 ) , our limiting magnitude of b@xmath22 corresponds to @xmath56 and @xmath57 magnitudes past the peak for ngc 4565 and ngc 5907 respectively . this corresponds to @xmath58 and @xmath59 past the peak . in other words , we sampled @xmath60% and @xmath61% of the luminosity functions respectively , and had to implicitly extrapolate over the remaining fraction . this extrapolating was taken care of by a corresponding cut in the absolute magnitude in the milky way sample . we stress that down to the magnitude limit , we do not suffer from finding incompleteness due to the software finding algorithm . another caveat is the implicit assumption that the spatial profile of the milky way globular cluster system is similar to that of the globular cluster systems of ngc 4565 and ngc 5907 . for ngc 4565 , the available photographic data are consistent with this assumption ( see sect . 1 ) . for ngc 5907 , we assume it also to be the case ( but see sect . if the spatial profile of the globular cluster system of ngc 5907 was markedly shallower than that of the milky way , we would detect a higher fraction of globular clusters in the `` halo '' wf and pc fields than we actually do . on the other hand , if the spatial profile was much steeper , it would not greatly affect our estimate of @xmath47 since our areal coverage includes roughly half of all the globular clusters for density profiles comparable to that of the milky way . errors in the distances to the galaxies lead to uncertainty in both the limiting magnitude and size of the mask used to derive the total number of clusters around each galaxy . the distance to ngc 4565 is fairly well established , with three independent secondary indicators all agreeing closely with a value of 10 mpc . the distance to ngc 5907 was derived from the tully fisher relation and found to be 12 mpc by independent groups . to estimate the effects of distance variations on our derived values , we carried out the same procedure as above with the galaxies at assumed distances of @xmath62 % their preferred values , affecting both mask size and limiting magnitude . the results of this exercise are summarized in table 4 . in summary , the above considerations lead us to total numbers of globular clusters for the two galaxies of @xmath63 and @xmath64 , following the equation of sect . 3.1.1 , where @xmath50 was corrected for fore-/background contamination . the error is the statistical error only : it includes poisson errors in the observed number of globular clusters , poisson errors in the average number of milky way globular clusters in the mask ( derived from the 4 different orientations ) , and errors in the number of contaminating objects . to this error , potential systematic errors should be added . first , we estimate up to 20% difference in the number of obscured clusters between the milky way and our target galaxies . second , we assumed for the milky way a total number of clusters of @xmath2 ; for a different assumption , our counts should be adjusted accordingly . note that this systematic error does not influence any direct comparison between our two galaxies . the third systematic error comes from our assumed distances . for a 20% error in distance , our results would vary by @xmath65 for ngc 4565 and @xmath66 for ngc 5907 , mainly due to the different limiting magnitude . note that this systematic error significantly affects the total numbers of clusters , but has a much smaller effect on the derived specific frequencies ( see below ) due to the varying total magnitude of the galaxy which compensates somewhat the varying total number of clusters . finally , for ngc 5907 an additional uncertainty is present due to the lack of knowledge on the exact shape of the globular cluster density profile . thus , formally , the total number of globular clusters are @xmath67 and @xmath68 , where the first error is the statistical one , the second error the systematic one . for ngc 4565 , our result is in good agreement with that of fleming et al . ( 1995 ) , who followed a completely different approach . this gives us confidence that our approach is also valid for ngc 5907 and the previous non detection of globular cluster should probably be attributed to the poor observing conditions . one can also express these numbers in terms of a specific frequency @xmath69 , where @xmath70 is the absolute visual magnitude of the parent galaxy . using the absolute magnitudes of the two galaxies given in table 1 and the total numbers from the previous section , we obtain @xmath71 and @xmath72 for ngc 4565 and ngc 5907 , respectively . the errors include the random errors from above , as well as an error of 0.2 magnitudes in the de - reddened apparent magnitudes of the galaxies . the effect of the systematic errors would be to lower the @xmath73 values by 0.05 , should the total number of globular clusters in the milky way be @xmath74 ; and to vary @xmath73 by less than 0.1 should the distance of the galaxy vary by 20% ( the increase / decrease in total number being partly compensated by an increase / decrease in the total magnitude of the galaxy ) . the quantity @xmath73 was introduced by harris & van den bergh ( 1981 ) primarily for use in elliptical galaxies where there is little variation in the stellar population from one galaxy to another . zepf & ashman ( 1993 ) attempted to account for stellar mass - to - light ratio variations in a _ statistical _ sense by defining a parameter @xmath75 to be the number of globular clusters per unit ( @xmath76 ) stellar mass of a galaxy . conversion from galactic luminosity to stellar mass was achieved by assuming a characteristic @xmath77 for each morphological type of galaxy . this is not an ideal procedure for individual galaxies , but it does tend to minimize differences in @xmath73 generated by stellar population differences . thus the approach allows us to obtain a comparison of the number of globular clusters in our target galaxies with other late type galaxies studied to date . for the conversion from luminosity to mass , we follow zepf & ashman ( 1993 ) and use @xmath77 values of 5.4 , 6.1 and 5.0 for sa , sab - sb and sbc - sc galaxies , respectively ( cf . faber & gallagher 1979 ) . the @xmath75 value for ngc 4565 is @xmath78 if it is assumed to be of type sb and 1.2 if its type is sbc ( errors are of the order 20% , see above ) . for ngc 5097 ( type sc ) @xmath79 . we comment on these values below . a visual comparison of the detected globular clusters in ngc 4565 and ngc 5907 with the masked milky way distributions ( fig . 4 and 5 ) suggests that the globular cluster system of ngc 5907 may be more flattened than that of the milky way . it is difficult to derive shape information directly for the globular cluster system of this galaxy from our small sample of 25 objects , as well as the asymmetric geometry of the field of view . a comparison of the @xmath80-distributions of ngc 5907 with that of the masked milky way dataset suggests a difference at a 2-@xmath15 level according to a k - s test ( ngc 4565 and the milky way show no detectable difference ) . the inner regions of the milky way globular cluster system is itself known to be somewhat flattened , primarily because of the presence of a thick disk population of globular clusters ( e.g. zinn 1985 ) so our data are consistent with a flattened globular cluster distribution in ngc 5907 , but do not demand it . the previous non detection of globular clusters in ngc 5907 ( harris et al . 1988 ) could also point towards a deficiency of objects in the halo , although we can only speculate on this point . their ground based study was based on object over - densities , and was probably unable to detect clusters close to the disk . with a limiting magnitude similar to ours , the density of globular clusters in the halo would be only marginally detectable in a statistical comparison with a background field . the shape of the cluster system in ngc 5907 is of some interest since it has been reported by sackett et al . ( 1994 ) that this galaxy has a highly flattened halo ( @xmath81 ) with a flat density profile falling off roughly as @xmath82 , compared to @xmath83 for the milky way halo and halo clusters . although ngc 4565 and ngc 5907 have similar total luminosities and hubble types , there are significant differences in the disk and bulge properties of the two galaxies . if the properties of the disk and bulge of a spiral galaxy influence the formation and evolution of its globular cluster system , we would therefore expect to find differences in the gcss of these two galaxies . the metal rich globular cluster population in the milky way shares the kinematics , abundances and spatial distribution of the bulge and thick disk ( e.g. armandroff 1993 ; burkert & smith 1997 ) . in the absence of a prominent bulge and thick disk ( as in ngc 5907 ) , we could have expected the absence of metal rich globular clusters and the presence of a halo population only . the globular cluster system of ngc 5907 would have been less flattened than the one of ngc 4565 ( whereas the opposite is observed as discussed above ) , and the total number of clusters normalized to the galaxy mass would be lower by 20% to 30% in ngc 5907 compared to ngc 4565 ( estimated from the fraction of metal rich to metal poor clusters in the milky way ) . the latter could be the case ( see sect . 3.3 ) but for the contradictory reason that ngc 5907 lacks halo clusters and not disk clusters . given the uncertainties in reddening ( i.e. metallicity determination from the colors ) , we can not allocate individual clusters in these two galaxies to the disk / bulge or halo . this awaits spectroscopic measurements of individual clusters . however , the fact that the globular cluster systems of ngc 4565 and ngc 5907 are still very similar despite difference in the thick disk and bulge characteristics suggests that the processes which form the bulge and thick disk are largely unrelated to the processes which form the total globular cluster systems around late - type galaxies . as mentioned above , both ngc 4565 and ngc 5907 are similar in many respects to the milky way . not only are their morphological types similar ( sb and sbc compared to the sbc milky way ) , but also their absolute magnitudes ( @xmath84 and @xmath85 compared to @xmath86 for the milky way , derived from the tully - fisher relation by adopting @xmath87@xmath88 ) and by implication their total stellar masses . it is therefore notable , although perhaps not surprising , that all three galaxies have very similar _ numbers _ of globular clusters . independently of the total number of milky way clusters , ngc 4565 has about 20% more clusters than the milky way ( but is also slightly more massive ) , while ngc 5907 has the same number of globular clusters as the milky way within the errors . even large errors in our adopted distance would leave all numbers within 50% of each other . in terms of specific frequency , the milky way has @xmath89 for an assumed total magnitude of @xmath90 and @xmath2 globular clusters , identical to the value derived for the two other spirals . five other spirals of type sb to sc have studied globular cluster systems . their @xmath73 values are @xmath91 ( m31 , sb ) , @xmath57 ( ngc 253 , sc ) , @xmath92 ( ngc 2683 , sb ) , @xmath93 ( ngc 4216 , sb ) , @xmath94 ( ngc 5170 , sb ) , all taken from the compilation of ashman & zepf ( 1998 ) . the quoted errors do not take into account the possibility of systematic errors . the corresponding @xmath75 values ( using the mean @xmath95 values as in sect . 3.2 ) are : @xmath96 , and @xmath97 . a maximum likelihood estimation of the mean and dispersion ( including the milky way @xmath98 , ngc 4565 and ngc 5907 ) returns @xmath99 . within the large uncertainties , galaxies of type sb to sc seem to produce a similar number of globular clusters per unit mass . should ngc 5907 have a lower @xmath73 than assumed ( see sect . 3.3 ) , there would be a hint for sc galaxies to have fewer globular clusters per unit mass . an increase of the number of globular clusters per unit mass along the hubble sequence of late - type galaxies is strengthened by observations of 4 sa and sab galaxies ( ngc 3031 ( = m81 ) , ngc 4569 , ngc 4594 , and ngc 7814 ) . the respective @xmath73 values are @xmath100 , @xmath101 , @xmath102 , @xmath103 ( taken from the compilation in ashman & zepf 1998 ) , and the respective @xmath75 values are @xmath104 , @xmath105 , @xmath106 , and @xmath107 . the mean @xmath75 value for these earlier types is @xmath108 , but note that , except for m81 , these values are poorly defined ( either the number of globular clusters and/or the assumed distance is uncertain ) . we have studied the globular cluster systems around two edge - on spiral galaxies , ngc 4565 and ngc 5907 . for both these galaxies we derive a specific frequency of @xmath109 , indistinguishable from the value of the milky way gcs . the similarity of the specific frequencies is notable since the bulge - to - disk ratio and prominence of the stellar thick disk both differ considerably between the three galaxies . this result suggests that the properties of the thick disk and bulge of spiral galaxies do not have a significant influence on the building of globular cluster systems around such galaxies . there is some evidence for flattening of the globular cluster system of ngc 5907 , but we are unable to determine whether this is driven by a flattened halo population of clusters or a thick disk system . the latter possibility seems somewhat unlikely given the upper limits on any stellar thick disk in this galaxy . an interesting alternative is that the globular clusters in ngc 5907 are following the profile of a significantly flattened halo . we are very grateful to our program coordinator , max mutchler , whose work and help greatly assisted this project . p . gratefully acknowledges support from the alexander von humboldt foundation . k.m.a . and s.e.z . acknowledge support for this project from nasa through the stsci grant go-06092.01 - 94a . ra ( 2000 ) & @xmath110 & @xmath111 & rc3 dec(2000 ) & @xmath112 & @xmath113 & rc3 type & sb & sc & rc3 v@xmath114 & @xmath115 & @xmath116 & rc3 ( value corrected for internal and external reddening)(b v)@xmath114 & @xmath117 & @xmath118 & rc3 e(b v ) & 0.015 ( + internal ) & 0.011 ( + internal ) & schlegel et al . 1998 inclination & @xmath11 & @xmath12 & garca - burillo et al . 1997 , morrison et al . 1994 thick disk & 5% of b@xmath119 & @xmath120% of thin disk & van der kruit & searle 1981 , morrison et al . 1994 position angle & @xmath121 & @xmath17 & rupen et al . 1991 , garca - burillo et al . 1997 v@xmath122 & @xmath123@xmath88 & @xmath124@xmath88 & rc3 adopted distance & @xmath125 mpc & @xmath126 mpc & jacoby et al . 1996 , bottinelli et al . 1988 , schniger & sofue 1994 @xmath127 & 30.0 & 30.4 & derived from the above distance @xmath128 & @xmath129 & @xmath85 & derived from the above values kazf4565 - 1 & 12 36 13.8 & @xmath13026 00 35.3 & -11.6 & -122.4 & @xmath131 & @xmath132kazf4565 - 2 & 12 36 14.0 & @xmath13025 59 58.2 & -35.8 & -94.6 & @xmath133 & @xmath134kazf4565 - 3 & 12 36 15.7 & @xmath13026 00 41.0 & 11.1 & -108.5 & @xmath135 & @xmath136kazf4565 - 4 & 12 36 16.5 & @xmath13025 58 25.0 & -76.8 & -3.7 & @xmath137 & @xmath138kazf4565 - 5 & 12 36 17.4 & @xmath13025 59 16.8 & -31.6 & -32.2 & @xmath139 & @xmath140kazf4565 - 6 & 12 36 17.5 & @xmath13025 59 24.0 & -25.7 & -36.4 & @xmath141 & @xmath142kazf4565 - 7 & 12 36 18.0 & @xmath13025 58 51.2 & -44.7 & -9.0 & @xmath143 & @xmath144kazf4565 - 8 & 12 36 18.0 & @xmath13025 59 26.5 & -20.0 & -34.3 & @xmath145 & @xmath146kazf4565 - 9 & 12 36 18.1 & @xmath13025 59 29.8 & -15.9 & -35.0 & @xmath147 & @xmath148kazf4565 - 10 & 12 36 18.4 & @xmath13025 59 30.5 & -12.7 & -32.8 & @xmath149 & @xmath150kazf4565 - 11 & 12 36 18.5 & @xmath13025 59 26.9 & -14.7 & -29.5 & @xmath151 & @xmath152kazf4565 - 12 & 12 36 18.8 & @xmath13025 59 04.9 & -27.4 & -11.3 & @xmath153 & @xmath154kazf4565 - 13 & 12 36 18.9 & @xmath13025 59 01.7 & -28.1 & -7.7 & @xmath155 & @xmath156kazf4565 - 14 & 12 36 19.2 & @xmath13025 59 25.4 & -9.2 & -22.4 & @xmath157 & @xmath158kazf4565 - 15 & 12 36 19.7 & @xmath13026 00 04.7 & 23.2 & -45.3 & @xmath159 & @xmath160kazf4565 - 16 & 12 36 19.8 & @xmath13025 58 55.9 & -23.4 & 5.1 & @xmath161 & @xmath162kazf4565 - 17 & 12 36 19.9 & @xmath13025 59 20.4 & -5.4 & -11.3 & @xmath163 & @xmath164kazf4565 - 18 & 12 36 20.0 & @xmath13025 59 14.3 & -8.7 & -6.2 & @xmath165 & @xmath166kazf4565 - 19 & 12 36 20.2 & @xmath13025 59 56.0 & 22.0 & -34.8 & @xmath167 & @xmath168kazf4565 - 20 & 12 36 20.3 & @xmath13025 59 06.4 & -11.5 & 2.0 & @xmath169 & @xmath170kazf4565 - 21 & 12 36 20.8 & @xmath13025 58 31.8 & -30.8 & 31.8 & @xmath171 & @xmath172kazf4565 - 22 & 12 36 21.1 & @xmath13025 59 36.6 & 17.8 & -11.3 & @xmath173 & @xmath174kazf4565 - 23 & 12 36 21.2 & @xmath13025 58 34.7 & -25.3 & 33.2 & @xmath175 & @xmath176kazf4565 - 24 & 12 36 21.6 & @xmath13025 59 44.2 & 27.3 & -12.8 & @xmath177 & @xmath178kazf4565 - 25 & 12 36 21.9 & @xmath13025 58 18.1 & -29.6 & 52.2 & @xmath179 & @xmath180kazf4565 - 26 & 12 36 22.3 & @xmath13025 58 30.4 & -17.6 & 46.6 & @xmath181 & @xmath182kazf4565 - 27 & 12 36 22.4 & @xmath13025 58 40.8 & -8.8 & 40.6 & @xmath183 & @xmath184kazf4565 - 28 & 12 36 22.8 & @xmath13025 58 01.6 & -32.9 & 71.8 & @xmath185 & @xmath186kazf4565 - 29 & 12 36 23.4 & @xmath13025 59 10.0 & 20.5 & 28.7 & @xmath187 & @xmath188kazf4565 - 30 & 12 36 23.7 & @xmath13025 59 10.0 & 24.0 & 32.0 & @xmath189 & @xmath190kazf4565 - 31 & 12 36 24.3 & @xmath13025 59 08.2 & 28.1 & 38.6 & @xmath191 & @xmath192kazf4565 - 32 & 12 36 25.4 & @xmath13025 58 21.7 & 6.2 & 82.0 & @xmath193 & @xmath194kazf4565 - 33 & 12 36 25.4 & @xmath13025 58 33.2 & 14.1 & 74.0 & @xmath195 & @xmath196kazf4565 - 34 & 12 36 25.5 & @xmath13025 58 36.1 & 17.3 & 72.9 & @xmath167 & @xmath197kazf4565 - 35 & 12 36 25.9 & @xmath13025 58 47.6 & 28.7 & 68.1 & @xmath198 & @xmath199kazf4565 - 36 & 12 36 26.1 & @xmath13025 57 26.6 & -25.9 & 127.8 & @xmath200 & @xmath201kazf4565 - 37 & 12 36 26.1 & @xmath13025 58 42.2 & 27.3 & 74.1 & @xmath202 & @xmath203kazf4565 - 38 & 12 36 26.6 & @xmath13025 58 08.4 & 8.7 & 103.2 & @xmath175 & @xmath204kazf4565 - 39 & 12 36 26.9 & @xmath13025 58 00.8 & 6.1 & 111.2 & @xmath205 & @xmath206kazf4565 - 40 & 12 36 27.4 & @xmath13025 58 16.0 & 21.5 & 105.5 & @xmath177 & @xmath207 kazf5907 - 1 & 15 15 40.1 & @xmath13056 20 03.8 & -96.7 & -64.6 & @xmath208 & @xmath209kazf5907 - 2 & 15 15 46.7 & @xmath13056 19 50.9 & -51.8 & -29.7 & @xmath210 & @xmath211kazf5907 - 3 & 15 15 46.9 & @xmath13056 21 14.8 & -15.2 & -105.1 & @xmath212 & @xmath213kazf5907 - 4 & 15 15 47.3 & @xmath13056 21 31.3 & -5.6 & -118.8 & @xmath214 & @xmath215kazf5907 - 5 & 15 15 47.7 & @xmath13056 20 52.8 & -18.5 & -82.4 & @xmath216 & @xmath217kazf5907 - 6 & 15 15 49.1 & @xmath13056 20 28.0 & -18.3 & -55.0 & @xmath179 & @xmath218kazf5907 - 7 & 15 15 50.0 & @xmath13056 20 03.1 & -22.3 & -29.4 & @xmath198 & @xmath219kazf5907 - 8 & 15 15 51.5 & @xmath13056 19 28.6 & -25.5 & 7.4 & @xmath220 & @xmath221kazf5907 - 9 & 15 15 51.8 & @xmath13056 19 49.1 & -14.2 & -10.1 & @xmath222 & @xmath223kazf5907 - 10 & 15 15 52.2 & @xmath13056 20 38.8 & 9.8 & -53.6 & @xmath224 & @xmath225kazf5907 - 11 & 15 15 52.5 & @xmath13056 20 11.4 & -0.1 & -28.0 & @xmath183 & @xmath226kazf5907 - 12 & 15 15 53.1 & @xmath13056 19 01.2 & -25.0 & 37.8 & @xmath227 & @xmath228kazf5907 - 13 & 15 15 53.4 & @xmath13056 19 22.1 & -13.7 & 20.0 & @xmath229 & @xmath230kazf5907 - 14 & 15 15 54.2 & @xmath13056 20 00.2 & 8.6 & -11.6 & @xmath220 & @xmath231kazf5907 - 15 & 15 15 54.5 & @xmath13056 20 02.4 & 11.0 & -12.8 & @xmath232 & @xmath233kazf5907 - 16 & 15 15 54.8 & @xmath13056 19 39.4 & 4.0 & 9.1 & @xmath234 & @xmath176kazf5907 - 17 & 15 15 55.4 & @xmath13056 18 45.3 & -14.7 & 60.2 & @xmath235 & @xmath236kazf5907 - 18 & 15 15 55.7 & @xmath13056 19 44.4 & 12.6 & 7.8 & @xmath237 & @xmath238kazf5907 - 19 & 15 15 55.8 & @xmath13056 19 49.4 & 15.7 & 3.5 & @xmath239 & @xmath240kazf5907 - 20 & 15 15 55.9 & @xmath13056 19 28.9 & 8.0 & 22.6 & @xmath241 & @xmath242kazf5907 - 21 & 15 15 56.1 & @xmath13056 19 57.3 & 21.2 & -2.8 & @xmath243 & @xmath244kazf5907 - 22 & 15 15 57.3 & @xmath13056 19 25.3 & 17.2 & 30.9 & @xmath245 & @xmath246kazf5907 - 23 & 15 15 57.4 & @xmath13056 17 47.4 & -24.1 & 119.5 & @xmath247 & @xmath248kazf5907 - 24 & 15 15 57.6 & @xmath13056 19 10.9 & 13.4 & 45.1 & @xmath249 & @xmath236kazf5907 - 25 & 15 15 57.8 & @xmath13056 18 12.6 & -9.8 & 98.6 & @xmath141 & @xmath250 @xmath251 orientation & 29 & 24 + @xmath252 orientation & 34 & 24 + @xmath253 orientation & 36 & 26 + @xmath254 orientation & 35 & 23 + mean , dispersion & @xmath255 & @xmath256 + & & + 20% closer & @xmath257 & @xmath258 + 20% further & @xmath256 & @xmath259 +

since the late twentieth century , research on the mechanical stability of macromolecules turned a prolific field due to the advances in manipulation techniques of individual biomolecules , usually termed single - molecule experiments . one of the most important techniques is atomic force microscopy ( afm ) , in which a biomolecule is stretched between a rigid platform and the tip of the cantilever @xcite . in these experiments , the controlled parameter is either the length of the macromolecule ( length - control protocols ) or the force exerted over it ( force - control protocols ) , and its conjugated magnitude is measured . as a result , a force - extension curve ( fec ) is obtained , which characterizes the elasto - mechanical behaviour of the macromolecule and provides fundamental information about its unfolding pathway @xcite . in a typical pulling experiment , the end - to - end distance of the molecule @xmath5 is increased with a pulling rate @xmath1 , that is , @xmath6 . remarkably , the fec exhibits a sawtooth pattern @xcite showing how the macromolecule comprises several structural units or blocks , in general each one with different stability properties . each block unfolds individually causing a drop in the measured force . the unfolding pathway is , basically , the order and the way in which the structural blocks of the macromolecule unravel . recent studies show that the pulling velocity plays a relevant role in determining the unfolding pathway @xcite . different unfolding pathways are observed depending on ( a ) which of the ends ( c - terminus or n - terminus ) the domain is actually pulled from and ( b ) the pulling speed . it has been claimed that it is the inhomogeneity in the distribution of the force across the protein , for high pulling speeds , that causes the unfolding pathway to change @xcite . nevertheless , to the best of our knowledge , a theory that explains this crossover is still lacking . we focus on the unfolding pathway of protein domains composed of several stable structural units @xcite . in this respect , a good candidate is represented by the maltose binding protein ( mbp ) , a stable and well characterized protein comprising two domains . mbp has been recently employed in studies on mechanical unfolding and translocation @xcite . in particular , bertz and rief @xcite identified four intermediate states in the fec of mechanical denaturation experiments , each one associated to the unravelling of a specific unit . the authors termed such units `` unfoldons '' and determined their typical unfolding sequence . however , simulations @xcite showed that this unfolding scenario holds only for low pulling speed as the pathway depends on the rate at which the molecule is pulled : at very low pulling rates , it is the weakest unfoldon that unfolds first , while at higher rates the first unfoldon to unravel is the pulled one . the main purpose of this paper is to develop a physical theory that predicts the observed dependence of the unfolding pathway on the pulling velocity . specifically , we do so in the adiabatic limit , that is , the regime of slow pulling speeds that allow the system to sweep the whole stationary branches of the fec @xcite . also , we would like to stress that , although our approach is applied to the unfolding of a single protein , it extends to the unfolding of polyproteins comprising several domains ( modular proteins ) , which unfold sequentially @xcite . the plan of the paper is as follows . in section [ sec : model ] , we put forward our model to investigate the unfolding pathway of proteins . in general , the free energy characterizing each unit ( unfoldon or module ) of the protein is different , that is , there is a certain degree of asymmetry ( or disorder ) in the free energies . moreover , we discuss the role of thermal noise and the ( ir)relevance of the details of the device controlling the length of the protein . section [ sec : real_chain ] is devoted to the analysis of the pulling of the model , by means of a perturbative expansion in both the asymmetry of the free energies and the pulling speed . in sections [ sec : asymmetry ] and [ sec : pulling_speed ] we obtain the corrections introduced by the asymmetry and the finite value of the pulling speed , respectively . in section [ sec : critical_speed ] we show that , in the simplest situation , there naturally appears a critical pulling speed @xmath2 , below ( above ) which it is the weakest ( pulled ) unit that unfolds first . in general , when more than one unit has a different free energy , we show that for low ( high ) enough pulling velocity , it is still the weakest ( pulled ) unit that unfolds first , but other pathways are present for intermediate velocities . numerical results for some particular situations are shown in section [ sec : numerics ] . they are compared to the analytical results previously derived , and a quite good agreement is found . finally , section [ sec : conclusions ] deals with the main conclusions of the paper . the appendices cover some technical details that we have omitted in the main text . let us consider a certain protein domain comprising @xmath0 unfoldons ( or a polyprotein composed of @xmath0 , possibly different , modules ) . from now on , we will refer to these unfoldons or modules as units . when the molecule is submitted to an external force @xmath7 , the simplest description is to portray it as a one - dimensional chain , where the end - to - end extension of the @xmath8-th unit in the direction of the force is denoted by @xmath9 . in a real afm experiment , the molecule is attached as a whole to the afm device and stretched . following guardiani et al . @xcite , we model this system with a sequence of nonlinear bonds , as in figure [ fig:1 ] : the endpoints of the @xmath8-th unit are denoted by @xmath10 and @xmath11 , so that its extension @xmath9 is @xmath12 the evolution of the system follows the coupled overdamped langevin equations @xmath13 in which the @xmath14 are the gaussian white noise terms , such that @xmath15 and @xmath16 , with @xmath17 and @xmath18 being the friction coefficient of each unit ( the same for all ) and the temperature of the fluid in which the protein is immersed , respectively ( @xmath19 is the boltzmann constant ) . the global free energy function of the system is @xmath20 in , @xmath21 is the contribution to the free energy introduced by the force - control or length - control device ( see below ) , while @xmath22 is the single unit contribution to @xmath23 . the total length of the system is given by @xmath24 in force - control experiments , the applied force @xmath7 is a given function of time , whereas in length - control experiments the device tries to keep the total length @xmath25 equal to the desired value @xmath5 , also a certain function of time . the corresponding contributions to the free energy are [ eq:6b ] @xmath26 @xmath27 in which @xmath28 stands for the stiffness of the length - control device . the length is perfectly controlled in the limit @xmath29 , when @xmath30 for all times . a sketch of the model is presented in figure [ fig:1 ] . an apparently similar system , in which each module of the chain follows the langevin equation @xmath31 has been recently analyzed @xcite . in this approach , the modules are completely independent in force - controlled experiments because these langevin equations completely neglect the spatial structure of the chain . while this simplifying assumption poses no problem for the characterization of the force - extension curves in @xcite , it is not suited for the investigation of the unfolding pathway , in which the spatial structure plays an essential role . the spatial structure of biomolecules can be described in quite a realistic way by using a model proposed by hummer and szabo several years ago to investigate their stretching @xcite , but the simplified picture which follows from figure [ fig:1 ] makes an analytical approach feasible . now , we turn to look into the unfolding pathway of this system . as the evolution equations are stochastic , this pathway may vary from one trajectory of the dynamics to another . nevertheless , in many experiments @xcite it is observed that there is a quite well - defined pathway , which suggests that thermal fluctuations do not play an important role in determining it . physically , this means that the free energy barrier separating the unfolded and folded conformations at coexistence ( that is , at the critical force , see below ) is expected to be much larger than the typical energy @xmath32 for thermal fluctuations . therefore , we expect the thermal noise terms in our langevin equations to be negligible and , consequently , they will be dropped in the remainder of the theoretical approach developed in the paper . of course , if the unfolding barrier for a given biomolecule were only a few @xmath32s , the thermal noise terms in the langevin equations could not be neglected and our theoretical approach would have to be changed . in order to undertake a theoretical analysis of the stretching dynamics , one further simplification of the problem will be introduced . we consider that the device controlling the length is perfectly stiff , and the total length @xmath33 does not fluctuate for all times . we expect this assumption to have little impact on the unfolding pathway : otherwise , the latter would be more a property of the length - control device than of the chain . in fact , we show in section [ sec : numerics ] that the unfolding order is not affected by this simplification . for perfect length - control , the mathematical problem is identical to that of the force - control situation , but now the force @xmath7 is an unknown ( lagrange multiplier ) that must be calculated at the end by imposing the constraint @xmath34 . therefore , the extensions @xmath35 s obey the deterministic equations [ eq:7 ] @xmath36 we have introduced the pulling speed @xmath37 which is usually time independent . we assume that @xmath22 allows for bistability in a certain range of the external force @xmath7 , in the sense that @xmath38 @xmath39 is a double - well potential with two minima , see figure [ fig:2 ] . therefore , in that force range , each unit may be either folded , if @xmath9 is in the well corresponding to the minimum with the smallest extension , or unfolded , when @xmath9 belongs to the well with the largest extension . if the length is kept constant ( @xmath40 ) , there is an equilibrium solution of , @xmath41 and @xmath42 is calculated with the constraint @xmath43 . this solution is stable as long as @xmath44 for all @xmath8 . if all the units are identical , @xmath45 , the metastability regions of each module ( the range of forces for which the equation @xmath46 has several solutions ) coincide . therefore , we obtain stationary branches corresponding to @xmath47 unfolded units and @xmath48 folded units that have been analyzed in detail in @xcite . if all the modules are not identical , the metastability regions do not perfectly overlap since the units are not equally strong : the weakest one is that for which the equation @xmath46 ceases to have multiple solutions for a smaller force value . it is important to note that we can change all the forces @xmath49 to @xmath50 and @xmath7 to @xmath51 , we have the same system but with @xmath52 and @xmath53 instead of @xmath54 and @xmath7 , respectively . then , we may use the free energies for any common value of the force @xmath55 and interpret the lagrange multiplier as the excess force from this value to be applied to the system . a constant force only shifts the equilibrium point of a harmonic oscillator : @xmath56 must be substituted by @xmath57 . ] . let us consider the pulling of our system . we write the @xmath8-th - unit free energy as @xmath58 in which @xmath59 is the `` main '' part , common to all the units , and @xmath60 represents the separation from this main contribution . if all the units are perfectly identical , @xmath61 for all @xmath8 or , equivalently , @xmath62 . in principle , in an actual experiment , the splitting of the free energy in can be done if the free energy @xmath63 of each unit is known : we may define the common part as the `` average '' free energy over all the units , @xmath64 , and @xmath65 . from a physical point of view , the dimensionless parameter @xmath66 measures the importance of the heterogeneity in the free energies . our theory could be applied to a situation in which the free energy deviations @xmath67 were stochastic and followed a certain probability distribution , for instance to represent the slight differences among very similar units , as done in @xcite to analyze the force - extension curves . in particular , the forces @xmath68 in the evolution equations can also be split as @xmath69 as already noted above , we can use the free energies for any common value of the force @xmath55 , and interpret @xmath7 as the extra applied force from this value . in what follows , we consider @xmath59 with two , equally deep , minima corresponding to the folded ( f ) and unfolded ( u ) configurations . figure [ fig:2 ] presents a qualitative picture of the free energy and its derivative . the two minima correspond to lengths @xmath70 and @xmath71 , with @xmath72 . also the point @xmath73 at which @xmath74 is marked . it is the condition @xmath75 that essentially determines the stability threshold , as it provides the limit force @xmath76 at which the folded basin ceases to exist for the `` main '' potential . in the deterministic approximation considered here , thermal fluctuations are neglected and , for @xmath77 , the folded unit can not jump over the free energy barrier hindering its unfolding : it has to wait until , at @xmath78 , the only possible extension is that of the unfolded basin . of course , neglecting thermal noise restricts in some way the range of applicability of our results , see section [ sec : critical_speed ] for a more detailed discussion and also the numerical section [ sec : numerics ] . keeping the above discussion in mind , now we analyze the limit of stability of the different units . the asymmetry correction @xmath79 shifts the threshold force for the different units . the extension @xmath80 at which the @xmath8-th unit loses its stability is obtained by solving the equation @xmath81 , which linearized in both the displacement @xmath82 and @xmath83 reads @xmath84 noting that @xmath75 , we get that @xmath85 see [ apa ] for details . the corresponding force is @xmath86 in which we have also dropped terms of the order of @xmath87 . then , units with @xmath88 ( @xmath89 ) are weaker ( stronger ) than average . when the system is continuously pulled , the total length of the system @xmath5 has been shown to be a good reaction coordinate @xcite . therefore , on physical grounds it is reasonable to use @xmath5 to measure time and write the evolution equations as [ eq:11 ] @xmath90 , \\ \fl \gamma v_{p } \frac{d x_{i}}{dl } = -2u'(x_{i})+u'(x_{i+1 } ) + u'(x_{i-1 } ) + \xi[-2{\delta\ ! f}_{i}(x_{i})+{\delta\ ! f}_{i+1}(x_{i+1})+\delta f_{i-1}(x_{i-1 } ) ] , \nonumber \\ \qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad\qquad 1<i < n , \\ \fl \gamma v_{p } \frac{d x_{n}}{dl } = -2u'(x_{n})+u'(x_{n-1})+f + \xi[-2\delta f_{n}(x_{n } ) + { \delta\ ! f}_{n-1}(x_{n-1 } ) ] , \\ \fl f = \gamma v_{p}+u'(x_{n})+\xi\,{\delta\ ! f}_{n}(x_{n } ) . \end{aligned}\ ] ] moreover , this change of variable makes the pulling speed @xmath1 appear explicitly in the equations , allowing us to consider @xmath1 as a perturbation parameter for slow enough pulling processes . now , we consider a system in which the asymmetry in the free energies is small and which is slowly pulled . thus , is solved by means of a perturbative expansion in powers of the pulling velocity @xmath1 and the disorder parameter @xmath83 , that is , [ eq:12 ] @xmath91 @xmath92 up to the linear order in both @xmath1 and @xmath83 . the zero - th ( lowest ) order corresponds to the chain of identical units ( @xmath93 ) with a given constant length @xmath5 ( @xmath40 ) . namely , @xmath94 and @xmath95 obey the equations [ eq:13 ] @xmath96 the solution of this system is straightforward , @xmath97 the force is equally distributed among all the units of the chain in equilibrium , as expected . if we start the pulling process from a configuration in which all the units are folded and the force is outside the metastability region ( the usual situation ) , the units extensions and the applied force are @xmath98 to the lowest order . to calculate the linear corrections in @xmath83 and @xmath1 , we have to substitute and into , and equate terms proportional to @xmath83 and @xmath1 , respectively . this is done below in two separate sections : firstly , for the asymmetry contribution @xmath99 and , secondly , for the `` kinetic '' contribution @xmath100 . all the modules are not characterized by the same free energy , and here we calculate the first order correction introduced thereby . the asymmetry corrections @xmath101 obey the system of equations [ eq:18 ] @xmath102 which is linear in the @xmath101 s , and thus can be analytically solved . it is clear that our expansion breaks down when @xmath103 . this was to be expected , since we know that the stationary branch with all the modules folded is unstable when @xmath104 becomes negative for some unit @xmath8 , and to the lowest order this takes place when @xmath103 . the solution of the above system of difference equations is obtained by standard methods @xcite , with the result @xmath105 see [ ap1 ] for more details . interestingly , the force is homogeneous across the chain , since to first order in @xmath83 we have that @xmath106 = u'(\ell)+\xi \overline{{\delta\ ! f}}(\ell)=f^{(0)}+\xi \delta f. \label{eq:20}\ ] ] this is nothing but the stationary solution , up to first order in the disorder s , no correction for the lagrange multiplier ( applied force ) would appear to the first order . this is logical , up to the first order the force expression coincides with the spatial derivative of the average potential , that is , @xmath107 . ] . moreover , ( [ eq:19 ] ) implies that there are units with @xmath108 and others with @xmath109 , depending on the sign of @xmath110 . this is a consequence of our perturbation expansion , since @xmath111 for all times , as given by ( [ eq:14 ] ) , and thus @xmath112 . let us remember that we denote by @xmath73 the value of the extension at which the common main free energy reaches its limit of stability , see figure [ fig:2 ] . taking into account only the asymmetry correction , it is the weakest unit that unfolds first , since the most negative @xmath113 leads to the largest positive @xmath101 and then it is the one that first verifies the condition @xmath114 ( for a more detailed discussion , see [ apa ] ) . an alternative way of looking at this is to recall that the force corresponding to the limit of stability is smallest for the weakest unit : since the force is homogeneously distributed along the chain , it is the weakest module that first reaches its stability threshold . now we look into the `` kinetic '' correction due to the finite pulling speed @xmath1 . the zero - th order solution is given by , so that @xmath115 for all @xmath8 , and we have [ eq:21 ] @xmath116 , \\ \delta f = \gamma + u''(\ell)\delta x_{n}.\end{aligned}\ ] ] the solution to this system of linear difference equations @xcite is [ eq:22 ] @xmath117,\\ \label{eq:22b } \delta f=\frac{(n+1)(2n+1)\gamma}{6n}.\end{aligned}\ ] ] see [ ap1 ] for more details . again , @xmath118 because the zero - th order solution ( [ eq:14 ] ) gives the total length , @xmath111 for all times . is reasonable on intuitive grounds : the kinetic correction @xmath100 increases with @xmath8 because the last module is the one that is actually pulled . therefore , on the basis of only the kinetic correction , it is the last module that would unfold first because @xmath119 is the largest . thus , the condition @xmath120 is first verified for @xmath121 . it is interesting to note that the force was equally distributed for the asymmetry correction , as expressed by , but this is no longer true if we incorporate the kinetic correction . up to the the first order , @xmath122 . therefore , the force @xmath49 depends on the unit @xmath8 : for all times , it is smaller the further from the pulled unit we are . again , there is an alternative way of understanding why the last unit would unfold first if we were considering perfectly identical units ( @xmath93 ) : for any time , it would be the last unit that suffered the largest force and thus the first that reached their common limit of stability @xmath123 . if the last unit is not the weakest , there is a competition between the asymmetry and the kinetic corrections . for very low pulling speeds , in the sense that @xmath124 , the term proportional to @xmath1 can be neglected and it is the weakest unit ( the one with the largest @xmath101 ) that unfolds first , as discussed in section [ sec : asymmetry ] . on the other hand , for very small disorder , in the sense that @xmath125 , the term proportional to @xmath83 is the one to be neglected and it is the last unit ( the one with the largest @xmath100 ) that unfolds first , as also discussed in section [ sec : pulling_speed ] . therefore , different unfolding pathways are expected as the pulling speed changes . collecting all the contributions to the extensions , we have that @xmath126 we have rearranged the terms in @xmath9 in such a way that the first two terms on the rhs are independent of the unit @xmath8 , all the dependence of the length of the module on its position across the chain has been included in the last term . note that we are expanding the solution in powers of @xmath1 around the `` static '' solution , which is obtained by putting @xmath40 in . thus , the `` static '' solution corresponds to the stationary one the system would reach if we kept the total length constant and equal to its instantaneous value at the considered time . it is essential to realise that is only valid for very slow pulling , as long as the corrections to the `` static '' solution are small , and this is the reason why the limit of stability is basically unchanged as compared to the static case . in order to be more precise , we refer to this kind of very slow pulling as _ adiabatic _ pulling . a main result of our paper is that , even for the case of adiabatic pulling , there appear different unfolding pathways depending on the value of the pulling speed . in the adiabatic limit we are considering here , the pulling process has to be slow enough to make the system move very close to the stationary force - length branches , but not so slow that the system has enough time to escape from the folded basin . as discussed in @xcite , there is an interplay between the pulling velocity and thermal fluctuations . for very slow pulling velocities , the system has enough time to surpass the energy barrier separating the two minima , which leads to the typical logarithmic dependence of the `` unfolding force '' @xmath127 on the pulling speed , specifically @xmath128 @xcite . is of the order of unity , its particular value depends on the specific shape of the potential ( linear - cubic , cuspid - like , ... ) considered @xcite . ] on the other hand , as already argued at the beginning of section [ sec : real_chain ] , for _ adiabatic pulling _ , the units unfold not because they are able to surpass the free energy barrier but because the folded state ceases to exist at the force @xmath123 corresponding to the upper limit of the metastability region . the unit that unfolds first is the one for which @xmath129 for the shortest time . in light of the above , it is natural to investigate whether it is possible to determine which module is the first to unfold for a given pulling speed . to put it another way , we would like to calculate the `` critical '' velocities which separate the velocity intervals in which a specific module unfolds first . let us assume that , for a given pulling speed @xmath1 , it is the @xmath8-th module that unfolds first . all the modules @xmath130 to its left , that is , with @xmath131 , will not open first if the pulling velocity is further increased because the difference between the kinetic corrections @xmath132 increases with @xmath1 . therefore , the first module @xmath130 to unfold when the velocity is sufficiently increased it is always to its right . the velocity @xmath133 for which each couple of modules @xmath134 , @xmath135 , reach simultaneously the stability threshold is determined by the condition @xmath136 determines both the value of @xmath137 ( or time @xmath138 ) at which the stability threshold is reached and the relationship between @xmath1 and @xmath83 . implies that @xmath139 we already know that the length corresponding to the limit of stability is very close to the threshold length @xmath73 , its distance thereto being of the order of @xmath140 , as shown in [ apa ] . therefore , to the lowest order , @xmath137 can be approximated by @xmath73 , and we get @xmath141}{j(j-1)-i(i-1 ) } , \quad j > i.\ ] ] clearly , the minimum of these velocities is the one that matters : let us denote by @xmath142 the position of the module for which @xmath133 reaches its minimum value @xmath143 , @xmath144 for @xmath1 just below @xmath143 , it is the @xmath8-th module that unfolds first , but for @xmath1 just above @xmath143 , it is the @xmath130-th module that unfolds first . let us denote the weakest module by @xmath145 , that is , @xmath146 is smallest for @xmath147 . if @xmath148 is smaller than @xmath149 , the first unit to reach the stability limit is the weakest one . then , we rename the latter velocity @xmath150 , that is , @xmath151 because it is the first one of a ( possible ) series of critical velocities separating different unfolding pathways , see below . let us denote by @xmath152 the module which unfolds first in the `` second '' velocity region , @xmath1 just above @xmath153 , that is , @xmath154 . this unit ceases to be the first to unfold for the velocity @xmath155 the successive changes on the unfolding pathway take place at the critical velocities @xmath156 in which @xmath157 . this succession ends when @xmath158 : in that case , for @xmath159 , the first unit to unfold is always the pulled one . this upper critical velocity @xmath160 can be computed in a more direct way , @xmath161 consistency of the theory requires that @xmath162 ; a short proof is presented in [ apb ] . we have a trivial case for @xmath163 , when the pulled unit is precisely the weakest and it is always the first to unfold for any pulling speed . the simplest nontrivial case appears when all the modules has the same free - energy with the exception of the weakest , and @xmath164 , , and reduce to @xmath165}{n(n-1)-\alpha_{1}(\alpha_{1}-1)}.\ ] ] note that the situation is quite simple , since there exist a single critical velocity @xmath166 . for @xmath167 the weakest module unfolds first whereas for @xmath168 the last one unfolds first . in general , when the units have different free energies the situation may be more complex , as shown in the previous paragraph . there appear intermediate critical velocities , which define pulling speed windows where neither the weakest unit nor the last one is the first to unfold . in order to obtain these regions , we need to recursively evaluate . throughout this numerical section , we check the agreement between our theory and the numerical integration of the evolution equations . firstly , we discuss the validity of the simplifications introduced in the development of the theory , namely ( i ) negligible thermal noise and ( ii ) perfect length - control . secondly , we look into the critical pulling speed , showing that there appears such a critical speed in the simulations and comparing this numerical value to the theory developed before . we consider a system composed of @xmath169 unfoldons , such as the maltose binding protein @xcite , each one characterized by a quartic bistable free energy . in reduced variables , the free energies have the form @xmath170 , where @xmath171^{2},\ ] ] with @xmath172 for @xmath173 , @xmath174 , @xmath175 and @xmath176 is different from the one in @xcite ( @xmath177 ) . its only effect is a shift of the origin of the extensions , our choice implies that a positive ( negative ) sign of the extension corresponds to an unfolded ( folded ) configuration . ] . the value of the friction coefficient is , also in reduced variables , @xmath178 . we use these dimensionless reduced variables in order to make it easier to compare our results to those in @xcite . the function is one of the simplest , but reasonable , choice to describe the free energy of different unfoldons of the same protein domain . using our notation , we have @xmath179 with @xmath180 . and give us the limits of stability up to first order in the asymmetry @xmath83 , @xmath181 for this simple example , is exact . the weakest unit is the first one , because @xmath182 is the minimum value of the force at the limit of stability . for the values of the parameters we are using , @xmath183 and @xmath184 . since we are writing the free energies for a common given value of the force , we are assuming that all the units have their two minima equally deep at the same force . this assumption is made to keep things simple : the main ingredient for having an unfolding pathway that depends on the pulling speed is to have different values of the forces @xmath185 at the stability threshold for the different units . in the case we are considering , the weakest unit is the first one , while the others share the same free energy . this means that we have the simplest scenario for the critical velocity in our theoretical approach : it is always the weakest ( for @xmath3 ) or the last ( for @xmath4 ) unit that opens first , as discussed at the end of the previous section . here , ( [ eq:26a ] ) for @xmath186 and @xmath169 reduces to @xmath187 evolution of the extensions of the different units in a pulling experiment as a function of the length of the system @xmath25 . the pulling speed is @xmath188 and the length - control device has a stiffness @xmath189 . the symbols correspond to a typical realization of the langevin process with @xmath190 , whereas the lines correspond to the deterministic ( zero noise ) approximation . , scaledwidth=70.0% ] to start with , we consider the relevance of the noise terms in . in figure [ fig:3 ] , we plot the integration of the langevin equations together with the deterministic approximation @xcite for a concrete case : the free energy of the first unit corresponds to @xmath191 ( @xmath192 ) , the stiffness of the device controlling the length is @xmath189 , the temperature is @xmath190 and the pulling speed is @xmath188 . for these values of the parameters , taken from @xcite , the critical velocity in is @xmath193 , so we are considering a subcritical velocity , @xmath3 . thermal fluctuations are small , and thus the same unfolding pathway is observed in the deterministic and the majority of the stochastic trajectories . trajectories , for a subcritical pulling speed @xmath194 . ( right ) the same as in the left panel , but for a supercritical pulling speed @xmath195 . as the temperature decreases , the frequency of the deterministic unfolding pathway approaches unity in both cases.,title="fig:",scaledwidth=49.0% ] trajectories , for a subcritical pulling speed @xmath194 . ( right ) the same as in the left panel , but for a supercritical pulling speed @xmath195 . as the temperature decreases , the frequency of the deterministic unfolding pathway approaches unity in both cases.,title="fig:",scaledwidth=49.0% ] let us consider in more detail the relevance of thermal noise : from a physical point of view , it may be inferred by looking at the height of the free energy barrier at the critical force in terms of the thermal energy @xmath32 . for the values of the parameters we are using , this barrier is around @xmath196 in reduced units . this explains why thermal noise is basically negligible in figure [ fig:3 ] , in which @xmath190 . if the temperature is decreased to @xmath197 , the barrier is so high , around @xmath198 times the thermal energy , that essentially all the stochastic trajectories coincide with the deterministic one . on the other hand , if the temperature is increased to @xmath199 , the barrier in only a few , around @xmath200 , times the thermal energy , and we expect that the deterministic approximation ceases to be valid . in order to further clarify this point , we present figure [ fig : new ] . both panels display bar graphs with the frequencies with which each unit unfolds first in the stochastic trajectories obtained over @xmath201 trajectories of the langevin equations with perfect length control and different values of the temperature . in the left panel , a subcritical velocity @xmath194 is considered , so that the weakest ( first ) unit is expected to unfold first . in the right panel , the numerical data for a supercritical velocity @xmath195 are shown , for which the pulled ( fourth ) unit would unfold first . the effect of thermal noise is quite similar in both cases . for the low temperature @xmath197 , the frequency of the deterministic pathway is close to unity and , for the temperature in figure [ fig:3 ] , @xmath190 , its frequency is still very large , clearly larger than any of the others . on the other hand , for the higher value of the temperature , @xmath199 , thermal noise can no longer be neglected . in the following , we restrict the analysis to the physically relevant case in which the deterministic approximation gives a good description of the first unfolding event . in figure [ fig:4 ] ( left panel ) , we look into the same pulling experiment as before , but now we compare the deterministic evolution of the extensions for two finite values of the stiffness to the @xmath29 limit . consistently with our expectations , the unfolding pathway is not affected by this simplification . nevertheless , the control of the length of course improves as @xmath28 increases ( see right panel ) . although for the smaller values of @xmath28 the length is not perfectly controlled , the curves in the left panel , which correspond to different values of @xmath28 , are almost perfectly superimposed when plotted as a function of the real length of the system @xmath25 ( but not of the desired length @xmath5 ) . this means that the real length @xmath25 is a good reaction coordinate , as already said in section [ sec : real_chain ] . we have integrated the deterministic approximation ( zero noise ) of the langevin equations for different values of the pulling speed , and extracted from them the numerical value of the critical velocity as a function of the asymmetry @xmath202 . in order to obtain this numerical prediction , we initially set @xmath1 equal to the theoretical critical velocity given by . then , we recursively shift it by a small amount @xmath203 , such that @xmath204 , until the pathway changes . we compare the values so obtained to the theoretical expression , in figure [ fig:5 ] . we find an excellent agreement for @xmath205 , for @xmath206 there appears some quantitative discrepancies . these discrepancies stem from two points : ( i ) the perturbative expansion used for obtaining from and ( ii ) the intrinsically approximate character of , since @xmath73 gives rigorously the limit of stability only for the static case @xmath40 . therefore , we have looked for the solution of in the numerical integration of the deterministic equations . this is the dashed line in figure [ fig:5 ] , which substantially improves the agreement between theory and numerics because we have eliminated the deviations arising from point ( i ) above . in fact , for the case we have studied in the previous figures , which corresponds to a not so small asymmetry @xmath192 , the improved theory gives an almost perfect prediction for the critical velocity . now we consider a more realistic potential for the units , which has been introduced by berkovich et al . for modelling the unfolding of single - unit i27 and ubiquitin proteins in afm experiments @xcite moreover , it has also been used to investigate the stepwise unfolding of polyproteins in force - clamp conditions @xcite and their force - extension curves in @xcite . at zero force , it reads @xmath207 + \frac{k_btl_c}{4p}\left(\frac{1}{1-\frac{x}{l_c}}-1-\frac{x}{l_c } + \frac{2x^2}{l_c^2}\right ) , \label{eq : berkovich}\ ] ] that is , it is the sum of a morse and a worm - like - chain potentials , representing the enthalpic and the entropic contributions to the free energy , respectively @xcite . we take the values of the parameters from @xcite , @xmath208 nm ( persistence length ) , @xmath209 nm ( contour length ) , @xmath210k , @xmath211pn nm(@xmath212 ) , @xmath213 nm , @xmath214 . we measure force and extensions in the units @xmath215=100 $ ] pn , @xmath209 nm , respectively . accordingly , dimensionless variables are introduced with the definitions @xmath216)$ ] , @xmath217 , @xmath218 , @xmath219 . thus , a dimensionless potential is obtained , which reads @xmath220 ^ 2 - 1 + a\!\left(\frac{1}{1-x}-1-x+2x^2\right ) \right\ } , \label{eq : berk - dimless}\ ] ] note that , in order not to clutter our formulae , we have not introduced a different notation for the dimensionless potential . the values of the parameters therein are @xmath221 , @xmath222 , @xmath223 and @xmath224 . in dimensionless variables , @xmath225 ( @xmath226pn ) and @xmath227 ( @xmath228 nm ) . the relevant time scale is set by the friction coefficient @xmath17 , @xmath229= \gamma l_c/[f]$ ] . in turn , @xmath17 is given by the einstein relation @xmath230 , where @xmath231 is the diffusion coefficient for tethered proteins in solution . we consider a typical value of @xmath231 , also taken from @xcite , @xmath232 nm@xmath233/s , so that @xmath234pn nm@xmath235 s. we consider a system of @xmath236 units , again with all the units but the first being identical . then , @xmath61 , @xmath173 , and the first unit is the weakest because @xmath237 . the situation is then similar to the one we have already analyzed with the quartic potential , but there is a difference that should be noted : here , @xmath59 is the free energy at zero force , whereas for the quartic potential @xmath59 was the free energy at a force @xmath55 for which the folded and unfolded minima were equally deep . then , the force here must not be interpreted as the extra force from @xmath55 , but as the whole force that is applied to the polyprotein . on the basis of our theory , we expect the simplest situation with only one critical velocity @xmath2 , below ( above ) which the weakest unit ( the pulled unit ) unfolds first . this is also indeed the case in the numerical simulations , and we compare the theoretical and numerical critical velocities in figure [ fig:6 ] . a very good agreement is found again , up to values of the asymmetry @xmath83 of the order @xmath238 . the above discussion shows that the validity of the theory presented here is not restricted to simple potentials like the quartic one ; on the contrary , it can be confidently applied to experiments in which the units are described by realistic potentials . moreover , for the typical parameters we are using , the theoretical critical velocity @xmath2 for the berkovich potential equals @xmath239nm / s for an asymmetry @xmath240 , which can be regarded as quite a conservative estimate of the largest asymmetries for which our theory gives an almost perfect description of the unfolding pathway . interestingly , this pulling speed corresponds to the upper range of velocities usually employed in afm experiments , for instance see table i of @xcite . therefore , testing our theory in real afm experiments with modular proteins should be achievable . finally , we consider a more complex situation , in which more than one unit is different from the rest and there may exist more than one critical velocity . to be concrete , we have considered a system with @xmath236 units in which @xmath241 , @xmath237 as before but @xmath242 , with @xmath59 being the quartic potential in . in this situation , we have two different critical velocities : for very low pulling speeds , the weakest unit is the first to unfold , but there appears a velocity window inside which neither the weakest nor the pulled unit is the first to unfold . this stems from the fact that the first and the third unit reach simultaneously the limit of stability for a velocity @xmath243 that is smaller than the velocity @xmath244 for which the first and the last would do so . the physical reason behind this is the threshold force of the pulled unit being larger enough than that of the third one . we recall that @xmath133 is the velocity for which the @xmath8-th and the @xmath130-th unit reach simultaneously their limits of stability . afterwards , the third unit and the fourth attain the limit of stability in unison for a velocity @xmath245 , and the following picture emerges from our theory . using the notation introduced in section [ sec : critical_speed ] , we define two critical velocities , @xmath246 such that for @xmath247 , it is the weakest unit that unfolds first , for @xmath248 , it is the third unit that unfolds first , and , finally , for @xmath249 , the first unit to unfold is the pulled one . we check the more complex scenario described in the previous paragraph in figure [ fig:7 ] . therein , we observe that ( i ) our theory correctly predicts the existence of the three pulling regimes described above but ( ii ) even for very small asymmetries , there appear some noticeable discrepancy between theory and simulation . since the validity of the perturbative expansion for obtaining the critical velocities from the condition ( [ eq : stab ] ) is strongly supported by the accurateness of the theoretical prediction for the simplest case , see figures [ fig:5 ] and [ fig:6 ] , this discrepancy should stem from the intrinsically approximate character of the condition @xmath250 for determining the stability threshold when @xmath251 . therefore , an improvement of the present theory should involve the derivation of a more accurate condition for obtaining the stability threshold in the case of finite pulling velocity . this point , which probably makes a multiple scale analysis necessary for lengths close to the condition @xmath252 , certainly deserves further investigation . and @xmath253 as a function of the asymmetry @xmath83 . in this case , our theoretical approach is able to reproduce the existence of the three different pulling regimes , but the discrepancies between the theoretical and the numerical values for the critical velocities are larger than in the simpler cases considered in the previous figures.,scaledwidth=70.0% ] we have investigated the general properties of the unfolding pathway in pulled proteins by means of a simple model portraying the chain as a sequence of nonlinear modules . the deterministic approximation of the langevin equations controlling the time evolution of the units extensions is used therein . this is a sensible approach in our model system : note that the free energies characterising the different units are considered to be very similar , and therefore their kramers rates for thermal activated unfolding would also be very close . then , were thermal effects important , our unfolding trajectories would be essentially stochastic and we would not observe a specific unfolding pathway . nevertheless , in some recent optical tweezers experiments in which thermal fluctuations are relevant , definite pathways have also been observed . although trajectories are indeed stochastic , the pathway is well - characterised @xcite . therefore , the analysis of these experiments needs a more sophisticated theory , which takes into account thermal noise effects . moreover , it seems that there are other elements that should be incorporated , such as ( i ) the possible coupling between the different units and ( ii ) their dissimilar free energies . for example , the former may explain the existence of _ dead ends _ observed in @xcite , that is , intermediate states that do not allow the system to completely relax , whereas the latter may lead to kramers rates leading to the separation of the timescales for the different unfolding events . the equilibrium extensions of the units are governed by different free energies , which we have called asymmetry or disorder in the free energies . our theory , based on stability considerations , is able to explain the experimental observations : ( i ) for low pulling speeds , it is the weakest unit that unfolds first , ( ii ) for large enough pulling speed , it is the pulled unit that opens first . this has been done by introducing a perturbative expansion both in the asymmetry of the free energies and in the pulling speed . moreover , our approach makes it possible to identify a critical rate that separates two well - defined regimes . to the lowest order , this critical velocity has a linear dependence on the asymmetry of the potential . in spite of the crude approximations , our theory compares quite well with the numerical data even beyond the applicability regime . moreover , our results provide a guide to interpret some inversions observed in the sequence of unfolding of the stable regions of the maltose - binding protein during its mechanical denaturation @xcite . it must be stressed that to the lowest order in our theory , the system is sweeping the stationary branches of the force - extension curve . in this sense , the pulling process is very slow or _ adiabatic_. despite this adiabatic nature of the pulling process , it is not always the weakest unit to unfold first . as long as the pulling speed @xmath251 , the closer to the pulled terminal one unit is , the larger the force acting on it . this gradient in the distribution of the force across the protein , which increases with the pulling speed , makes it possible that the last unit reaches first its limit of stability . we have limited ourselves to the investigation of the first unfolding event . however , our argument can be easily generalized to the next unfolding event : the difference is that the zero - th order approximation is no longer given by all the units sweeping the all - units - folded branch but by the sweeping of the branch with one module unfolded and the remainder folded . then , a similar perturbative expansion around this zero - th order solution in powers of the asymmetry and the pulling speed would give the next unit that opens . if the biomolecule comprises several perfectly identical units , the asymmetry correction vanishes , because @xmath254 ( and thus @xmath255 ) for all the units . in that case , our theory predicts that it is always the pulled unit that unfolds first . however , even in engineered modular proteins , slight differences from module to module may be present . in fact , this has lead to the analysis of the impact of quenched disorder in the force - extension curves of biomolecules @xcite . in the present context , we may also introduce stochastic free energy deviations , following a certain probability distribution . next , our theoretical approach can be applied to this system with quenched disorder in the free energies . interestingly , evidence of dynamical disorder , that is , a fluctuating environment , has been recently brought to bear in stretching experiments @xcite . the analysis of this situation needs a more complex theory , in which the free energy landscape fluctuates in time , and is outside of the scope of the present paper . there are methods that extract the free energy landscape from experimental data of pulling experiments , even when there are intermediates @xcite . the resulting free energy is usually calculated as a function of the end - to - end distance of the molecule . nevertheless , in order to apply our theory , we do not need this global energy landscape as a function of the end - to - distance of the molecule but each unit s contribution thereto as a function of its own extension . in this regard , it is relevant to note that a similar velocity dependent unfolding pathway should also be found in modular proteins , although it has not been experimentally investigated to the best of our knowledge . in fact , we have analyzed a simple polyprotein model with a realistic potential , and observed a completely analogous behaviour . when all the modules are not identical , the weakest one will always open first for small enough pulling velocities . on the other hand , if the pulled unit is not the weakest , this will no longer be the case as the pulling speed is increased . since the free energy of each module is experimentally accessible and the critical velocity lies on the experimental range , a reliable test of our theory could be done in modular proteins . another possibility that deserves attention is to test our theory in ankyrin repeat proteins . in @xcite , the unfolding of a consensus ankyrin repeat protein , ni6c , has been investigated . this protein is composed of eight repeats : the two capping repeats are different from the six identical central ones , that at the c - terminus ( n - terminus ) is weaker ( stronger ) than the rest . pulling from the c - terminus ( weakest unit ) , lee et al . observe that the unfolding always starts from this end . this is what is expected from our theory , since when the pulled and the weakest units coincide , there is no competition between the asymmetry and kinetic terms . therefore , it seems relevant to carry out the same experiment but pulling from the n - terminus ( strongest unit ) , in which a much richer phenomenology is to be expected . very recently , sequential unfolding has been reported in a simple model @xcite , which makes it possible to understand the stepwise unfolding observed in force - clamp experiments with modular proteins @xcite . this sequential unfolding appeared as a consequence of a depinning transition near the stability threshold introduced by the coupling between nearest neighbour units . interestingly , the unfolding of the ankyrin repeat protein in @xcite does not only start from a well - defined end but it is also sequential , which may hint at the significance of this kind of short - ranged couplings in the experiment . then , it seems also relevant to analyze whether a similar sequential unfolding is present in the model developed here , when a similar short ranged interaction between neighbouring units is considered . this work has been supported by the spanish ministerio de economa y competitividad grant fis2011 - 24460 ( ap ) . cap acknowledges support by a phd fellowship from fundacin cmara ( universidad de sevilla ) . to first order in @xmath83 , the extension @xmath80 such that @xmath256 verifies @xmath257 that is , @xmath258 the corresponding force at the stability threshold is obtained from . to the lowest order in the deviations , @xmath259 because the next term , @xmath260 , is of the order of @xmath87 . therefore , the @xmath8-th module reaches its limit of stability at the time for which @xmath261 , that is , when the length per unit @xmath262 has the value @xmath263 verifying @xmath264 or , equivalently , @xmath265 we know that @xmath266 when @xmath267 . but @xmath74 and thus we can not substitute @xmath73 on the rhs of . on the other hand , this means that the dominant balance for @xmath267 involves the lhs and the first term on the rhs of . therefore , making use of @xmath268 , we get @xmath269 since @xmath270 ( see figure [ fig:2 ] ) , this means that only the units with @xmath146 smaller than the average ( that is , weaker than average ) reach the limit of stability in the limit as @xmath271 . in fact , it is the weakest unit , that is , the unit with smallest @xmath146 , that unfolds first . it is interesting to note that , in order to obtain , we have completely neglected the last term on the rhs of . since , in turn , this term stems from the last term on the rhs of , to the lowest order we are solving the equation @xmath129 . in other words , to the lowest order the stability threshold can be considered to be given by the non - disordered , zero - asymmetry case , free energy @xmath59 . for the sake of concreteness and simplicity , we have stuck to the asymmetry contribution @xmath101 in this appendix , but the same condition @xmath129 would still be valid , had we taken into account the kinetic contribution @xmath100 derived in section [ sec : pulling_speed ] . the reason is that there is also a factor @xmath272 in the denominator of @xmath100 , see , and thus both the terms coming from @xmath101 and @xmath100 are dominant against the last term on the rhs of . in this appendix , we briefly discuss a general procedure which is useful to solve linear difference equations similar to those in and . the methods for solving difference equations often resemble those used for solving analogous differential equations ; the latter may be thought of as the continuous limit of the former . both and belong to the following general class of second - order linear difference equations for @xmath273 , [ eq : b1 ] @xmath274 in which @xmath275 , @xmath276 are arbitrary functions and @xmath277 is a given constant . ( [ eq : b1b ] ) is a second - order linear difference equation , and ( [ eq : b1a ] ) and ( [ eq : b1c ] ) are its boundary conditions . it may be thought of as a discrete inhomogeneous diffusion equation : @xmath278 is the first discrete derivative , so that @xmath279 is the second discrete derivative @xcite . in complete analogy with the corresponding differential equation @xmath280 , the general solution of ( [ eq : b1b ] ) is @xmath281 in which @xmath282 and @xmath283 are two arbitrary constants . the solution of ( [ eq : b1 ] ) is , as usual , univocally determined by the boundary conditions , from which specific values for @xmath282 and @xmath283 are obtained . let us show that both and can be cast in the above form . firstly , in , it is easily seen that if we define @xmath284 , ( [ eq : b1b ] ) is obtained with @xmath285 . secondly , in , it is straightforward to identify @xmath286 and @xmath287^{-1}$ ] . a simple calculation gives the constants @xmath282 and @xmath283 in for each case , and thus the expressions for @xmath99 and @xmath100 in the main text . here , we prove that @xmath162 . it is easy to show that this inequality follows if we have that @xmath288 in which @xmath289 . due to , @xmath290 minimize @xmath291 . therefore , in particular , @xmath292 , which is readily shown to be equivalent to . hoffmann t and dougan l , 2012 chem . soc . rev . * 41 * , 4781 smith s b , cui y and bustamante c , 1996 science * 271 * , 795 lu h and schulten k , 1999 proteins struct . funct . genet . * 35 * , 453 manosas m and ritort f , 2005 biophys . j. * 88 * , 3224 - 3242 cao y , kuske r and li h , 2008 biophys . j. * 95 * , 782 carrion - vazquez m , oberhauser a f , fowler s b , marszalek p e , broedel s e , clarke j and fernandez j m , 1999 proc . usa * 96 * , 3694 fisher t e , marszalek p e and fernandez j m , 2000 nat . * 7 * , 719 liphardt j , onoa b , smith s b , tinoco i and bustamante c , 2001 science * 292 * , 733 bustamante c , bryant z and smith s b , 2003 nature * 421 * , 423 klimov d k and thirumalai d , 2000 proc . usa * 97 * , 7254 hyeon c , dima r i , and thirumalai d , 2006 structure * 14 * , 1633 liphardt j , dumont s , smith s b , tinoco i and bustamante c , 2002 science * 296 * , 1832 li m s and kouza s , 2009 j. chem . phys . * 130 * , 145102

the behavior of scattering in the limit of high energy and fixed momentum transfer is described in qcd , at least for situations in which perturbation theory applies , by the balitskii - fadin - kuraev - lipatov ( bfkl ) pomeron @xcite ( or short distance pomeron ) . attempts to test experimentally this sector of qcd have started in the last few years , mainly based on measurements of deeply inelastic events at low values of the bjorken variable @xmath1 in lepton - hadron scattering @xcite and jet production at large rapidity separations in hadron - hadron collisions @xcite . in this paper we study the possibilities for investigating qcd pomeron effects in a different context , namely in photon - photon scattering at @xmath2 colliders , where the photons are produced from the lepton beams by bremsstrahlung . the results of this paper also apply to @xmath3 or @xmath4 colliders . some aspects of this study have been presented in refs . @xcite . the quantity we consider is the total cross section for off - shell photon scattering at high energy . this can be measured in @xmath5 collisions in which both the outgoing leptons are tagged . this cross section presents some theoretical advantages as a probe of qcd pomeron dynamics compared to the structure functions for deeply inelastic scattering off a proton ( see for instance the discussion in ref . @xcite ) or a ( quasi)-real photon ( see for example ref . @xcite ) , essentially because it does not involve a non - perturbative target . unlike protons or quasi - real photons , virtual photon states can be described through perturbative wave functions . in some respects the off - shell photon cross section presents analogies with the process of scattering of two quarkonia ( or `` onia '' ) , which has been proposed as a gedanken experiment to investigate the high energy regime in qcd @xcite . in this case , non - perturbative effects are suppressed by the smallness of the onium radius . in the case of virtual photons the size of the wave function is controlled by the photon virtuality instead of the heavy quark mass . it is an interesting feature of investigations at @xmath6 colliders that this size can be tuned by measuring the momenta of the outgoing leptons . on the other hand , such experimental studies may prove to be difficult due to the smallness of the available rates . as we shall see , for large photon virtualities @xmath7 the cross section falls off like @xmath8 . an estimate of the number of events that one may expect to be available for such studies at lep200 and a future linear @xmath9 collider will be provided later in the paper . there have recently been other investigations of the high energy regime in the context of photon - photon scattering . balitskii @xcite has proposed an expansion of the scattering amplitude in the high energy limit in terms of wilson line operators . this method provides an elegant reformulation of the bfkl problem , and may prove to be useful to get beyond the leading logarithm approximation . bartels , de roeck and lotter @xcite have evaluated the photon - photon cross section at present and future @xmath6 colliders . their results are similar to those in refs . @xcite . in ref . @xcite detailed studies have been carried out for diffractive meson production and photon structure functions at lep200 . to describe the electron - positron scattering process ( fig . [ fee ] ) , we will parametrize the five - fold differential cross section as @xmath10 \hspace*{0.8 cm } . % \ ] ] here we denote by @xmath11 the fractions of the longitudinal momenta of the leptons a and b that are carried by the photons , by @xmath12 the photon virtualities , and by @xmath13 the angle between the lepton scattering planes in a frame in which the photons are aligned with the @xmath14 axis . the overall factor in the right hand side gives the distribution averaged over the angle @xmath13 , while @xmath15 , @xmath16 are the asymmetries . ( figee.eps width 8 cm ) we will start by considering the @xmath13-averaged cross section , and we will express it in the equivalent photon approximation @xcite by folding the @xmath17 cross section with the flux of photons from each lepton . this flux is proportional to the probability density for the splitting @xmath18 , and depends on the photon polarization . we will thus write @xmath19 where the transverse and longitudinal photon flux factors @xmath20 , @xmath21 are given by @xmath22 and @xmath23 , with @xmath24 , is the cross section for the scattering of two photons with polarizations @xmath25 and @xmath26 . we will proceed in the following way . in sec . [ sec : notations ] we set our notations and describe the born approximation to the photon - photon cross section at high energy . in this section we concentrate on the case of transversely polarized photons , that is , the cross section @xmath27 in eq . ( [ epaave ] ) above . in sec . [ sec : polarization ] we extend these results to include the full polarization dependence and we discuss the associated asymmetries . in sec . [ sec : summation ] we consider the summation of the leading logarithmic corrections to the photon - photon cross section due to bfkl pomeron exchange , and emphasize how the perturbative results depend on the total energy and the photon virtualities . [ sec : scales ] and [ sec : limitations ] are devoted to discussing some of the limitations of the treatment based on the bfkl equation . in sec . [ sec : scales ] we focus on the dependence of the cross section on two mass scales ( in the running coupling and in the high energy logarithms ) , which are left undetermined in a leading log analysis . in sec . [ sec : limitations ] we consider the limitations of using the bfkl approach that follow from the behavior at very large @xmath28 . [ sec : virtuality ] illustrates how to relate the summed result for the photon - photon cross section to the small-@xmath1 behavior of the photon deeply inelastic structure function . in sec . [ sec : regge ] we compare the qcd result with expectations based on traditional regge theory . in sec . [ sec : softandhard ] we consider the limit of low photon virtualities and discuss the region of transition between hard and soft scattering . the rates at the level of the @xmath29 cross section are examined in sec . [ sec : numerical ] . some concluding remarks are given in sec . [ sec : conclusions ] . we collect in appendix [ app : born ] the details of the born order calculation , and in appendix [ app : quark ] some formulas which are useful for comparing the gluon - exchange and quark - exchange contributions to the photon - photon cross section . we consider the total cross section for the scattering of two transversely polarized virtual ( space - like ) photons @xmath30 and @xmath31 , with virtualities @xmath32 and @xmath33 , in the high energy region where the center - of - mass energy @xmath34 is much larger than @xmath35 , @xmath36 . we also suppose that the photon virtualities are in turn large with respect to the qcd scale @xmath37 , so that the process occurs at short distances ( much smaller than @xmath38 ) and qcd perturbation theory applies . we work in a reference frame in which the incoming photons have zero transverse momenta , and are boosted along the positive and negative light - cone directions . for a four - momentum @xmath39 , we define the `` @xmath40 '' and `` @xmath41 '' momentum components @xmath42 , @xmath43 as @xmath44 the incoming photon momenta are parametrized as follows in a notation where @xmath45 : @xmath46 here the `` @xmath40 '' and `` @xmath41 '' components @xmath47 approximately build up the total energy @xmath28 , and are much larger than the initial virtualities @xmath48 the born contribution to this cross section is given by the high energy approximation to the reaction @xmath49 in lowest order perturbation theory . the corresponding diagrams involve the exchange of two gluons between the two quark - antiquark pairs @xcite , and are exemplified in fig . [ ftwoglu ] . we parametrize the outgoing quark momenta as @xmath50 and denote by @xmath51 the exchanged gluon momentum . in the high energy limit defined by eq . ( [ qaplusqbminus ] ) , the kinematic region which dominates the integrations is the one in which the transverse momenta flowing in the loops are of the order of the initial virtualities , and the light - cone components of the exchanged gluon are suppressed with respect to the transverse momenta by a quantity of order @xmath52 or @xmath53 , so that @xmath54 . ( figtwoglu.eps width 7 cm ) the squared amplitude for the graph in fig . [ ftwoglu ] , integrated over the final quark and antiquark phase space , for fixed transverse photon polarizations @xmath55 , @xmath56 , is given by @xmath57 } \over { \left ( ( p_a + k)^2 + i \ , \varepsilon \right ) \left ( ( p_a - q_a)^2 - i \ , \varepsilon \right ) } } \ , \ , \nonumber \\ & & \times \ , { { { \rm tr}\bigl [ { \rlap / p}_b ( i g_s t^r \gamma^\alpha)\ , i\ , ( { \rlap / p}_b - \rlap / k ) \ , ( i \ , e_b \ , e { \rlap/\varepsilon}_b ) \ , ( { \rlap / q}_b - { \rlap / p}_b + \rlap / k ) ( - i g_s t^s \gamma^\beta ) ( - i ) ( { \rlap / p}_b - { \rlap / q}_b ) ( - i e_b e { \rlap/\varepsilon}_b ) \bigr ] } \over { \left ( ( p_b - k)^2 + i \ , \varepsilon \right ) \left ( ( p_b - q_b)^2 - i \ , \varepsilon \right ) } } \;\;. \nonumber % \end{aligned}\ ] ] here @xmath58 , @xmath59 are the electric charges of the quarks in units of @xmath60 , and the indices @xmath61 run over the light quark flavors , @xmath62 . ( the flavors @xmath63 need a separate treatment . ) in the high energy approximation this amplitude takes the form @xmath64 } \over { \left [ z_a \ , ( 1-z_a ) \ , q^2_a + ( { { \bf p}}_a + { { \bf k}})^2 \right ] \ , \left [ z_a \ , ( 1-z_a ) \ , q^2_a + { { \bf p}}_a^2 \right ] \ , } } \ , \nonumber \\ & & \times \ , { { \left [ 4 z_b ( 1 - z_b ) { \bf{\varepsilon}}_b \cdot { { \bf p}}_b \ ; { \bf{\varepsilon}}_b \cdot ( { { \bf p}}_b - { \bf k } ) - { { \bf p}}_b \cdot ( { { \bf p}}_b - { \bf k } ) \right ] } \over { \left [ z_b \ , ( 1-z_b ) \ , q^2_b + ( { { \bf p}}_b - { { \bf k}})^2 \right ] \ , \left [ z_b \ , ( 1-z_b ) \ , q^2_b + { { \bf p}}_b^2 \right ] \ , } } \;\;\;\;\;\;\;\;. % \end{aligned}\ ] ] see appendix a for details of this calculation . the amplitudes for the other graphs , in which the gluons are connected to different fermion lines , can be derived from eq . ( [ m3 ] ) by using the replacements @xmath65 \to { { 1-z_a } \over { z_a } } \ , \left [ z_a \ , ( 1-z_a ) \ , q^2_a + { { \bf p}}_a^2 \right ] \ , % \ ] ] in the denominator , and @xmath66 in the numerator . ( analogous replacements hold for the momentum components of the quark @xmath67 ) . in addition , the contribution from the graphs in which the quark and antiquark lines are interchanged can be obtained by symmetrizing the above expressions with respect to @xmath68 ( and , analogously , @xmath69 ) . we add the graphs , and divide by @xmath70 to form the @xmath71 cross section @xmath72 we find @xmath73 the factors @xmath74 in this formula come from the gluon propagators . these factors multiply functions @xmath75 , which describe the coupling of the exchanged gluon to the quark - antiquark system created by the virtual photon with virtuality @xmath7 and transverse polarization @xmath76 . the explicit expression for @xmath77 is @xmath78 } \over { \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right]^2 } } - { { \left [ { { \bf p } } \cdot ( { { \bf p } } + { \bf k } ) - 4 z ( 1 - z ) { \bf{\varepsilon } } \cdot { { \bf p } } \ ; { \bf{\varepsilon } } \cdot ( { { \bf p } } + { \bf k } ) \right ] } \over { \left [ ( { { \bf p } } + { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 \right ] \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right ] } } \right\ } \;\;\;\;. \nonumber % \end{aligned}\ ] ] the functions @xmath77 can be thought of as `` color functions '' of the virtual photon since they describe the color flow in the @xmath79 states . from the point of view of light - cone perturbation theory @xcite , they correspond to the coupling of the null - plane photon wave function to gluons . in the remainder of this section we discuss the case of the average over the two transverse photon polarizations . the detailed polarization dependence of the color functions @xmath77 and the associated polarization asymmetry in the cross section are treated in sec . [ sec : polarization ] . by taking the polarization average @xmath80 we define the function @xmath81 @xmath82 we find @xmath83 \ , \nonumber \\ & ~ & \times \ , \left [ { { { { \bf p}}^2 } \over { \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right]^2 } } - { { { { \bf p } } \cdot ( { { \bf p } } + { \bf k } ) } \over { \left [ ( { { \bf p } } + { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 \right ] \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right ] } } \right ] \;\;. % \end{aligned}\ ] ] each one of the two @xmath84-dependent terms in the integrand of eq . ( [ gi1funct ] ) would lead to an ultraviolet divergent integral , but the divergence cancels in the sum , illustrating that the gluon does not couple to the color singlet @xmath85 system in the limit @xmath86 . the photon virtuality @xmath7 regularizes the denominators in the infrared region , @xmath87 . in the limit of small @xmath7 , the probability density @xmath88 / 2 $ ] for the splitting of a transversely polarized photon into a quark - antiquark pair @xmath89 correctly factors out in front of the logarithmic singularity @xmath90 associated with the region of strong ordering @xmath91 . by introducing the feynman parametrization @xmath92 \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right]^2 } } \nonumber \\ & & \hspace*{2 cm } \ , = \int^1_0 \ , d \lambda \ , { { 2 \,(1 - \lambda ) } \over { \left [ ( { { \bf p } } + \lambda \ , { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 + \lambda \ , ( 1 - \lambda ) { { \bf k}}^2 \right]^3 } } % \end{aligned}\ ] ] and carrying out the integration over the shifted transverse momentum variable @xmath93 in eq . ( [ gi1funct ] ) , one can obtain the following useful representation of the function @xmath94 as an integral over two dimensionless variables : @xmath95 \ , \left [ z^2 + ( 1 - z)^2 \right ] } \over { \lambda \ , ( 1 - \lambda ) \,{{\bf k}}^2 + z \ , ( 1-z ) \ , q^2 } } \hspace*{0.5 cm } . % \ ] ] this representation explicitly shows that the distribution @xmath96 is symmetric under interchange of the space - like boson virtualities @xmath97 and @xmath7 . in the configurations in which one of the virtualities is much smaller than the other one , this distribution is logarithmically enhanced . more precisely , we find @xmath98 \hspace*{0.5 cm } , \hspace*{0.8 cm } { { \bf k}}^2 \ll q^2 \hspace*{0.5 cm } , % \ ] ] @xmath99 \hspace*{0.5 cm } , \hspace*{0.8 cm } { { \bf k}}^2 \gg q^2 \hspace*{0.5 cm } , % \ ] ] where the coefficient in front of the logarithm is given by the first moment of the splitting density @xmath100 = { 2 \over 3 } \hspace*{1 cm } . % \ ] ] the logarithmic behavior at small @xmath97 comes from the region @xmath101 . here the quark transverse momentum is much smaller than the photon virtuality , and the quark longitudinal momentum fraction is very small , @xmath102 ( or @xmath103 ) . this region is sometimes referred to as the aligned - jet region , and corresponds to configurations in which the @xmath85 system fluctuates to large sizes @xcite . note that while for the case at hand of the @xmath104 total cross section this region contributes only a logarithmic enhancement , for the case of non - inclusive processes , such as processes involving rapidity gaps , this is expected to become the dominant contribution @xcite . the integration over the parameter @xmath105 in eq . ( [ twodimintg ] ) can be explicitly performed . this yields @xmath106 \ , , % \ ] ] where we have introduced the variables @xmath107 in fig . [ glin ] , we report the result of the numerical evaluation of the integral ( [ etaxig ] ) by plotting the function @xmath81 versus @xmath108 . ( figglin.eps width 13 cm ) the unpolarized @xmath109 cross section to born order , @xmath110 , can be obtained by inserting the result for the function @xmath81 in the general formula ( [ convsigma ] ) . it is convenient to use the representation ( [ twodimintg ] ) of @xmath81 . by substituting this representation in eq . ( [ convsigma ] ) , and carrying out the integration over the gluon transverse momentum @xmath111 , we obtain @xmath112 \int_0 ^ 1 \ , { d \ , \lambda_a } \left [ \lambda_a^2 + ( 1 - \lambda_a)^2 \right ] \ , \nonumber \\ & & \times \int_0 ^ 1 \ , { d \ , z_b } \left [ z_b^2 + ( 1 - z_b)^2 \right ] \int_0 ^ 1 \ , { d \ , \lambda_b } \left [ \lambda_b^2 + ( 1 - \lambda_b)^2 \right ] \ , \nonumber \\ & & \times { { \ln \left [ { q_a^2 \ , z_a \ , ( 1 - z_a ) \ , \lambda_b \ , ( 1 - \lambda_b ) } / \left ( { q_b^2 \ , z_b \ , ( 1 - z_b ) \ , \lambda_a \ , ( 1 - \lambda_a ) } \right ) \right ] } \over { q_a^2 \ , z_a \ , ( 1 - z_a ) \ , \lambda_b \ , ( 1 - \lambda_b ) - q_b^2 \ , z_b \ , ( 1 - z_b ) \ , \lambda_a \ , ( 1 - \lambda_a ) } } \hspace*{0.5 cm } . % \end{aligned}\ ] ] this formula provides an expression for the cross section in the large @xmath28 limit in terms of dimensionless integrals , which allows us to study the dependence on the energy and mass scales . to this order in perturbation theory the cross section has a constant behavior with the energy @xmath28 . the cross section depends on the mass scales @xmath113 and @xmath114 only . factoring out an overall scale factor @xmath115 in eq . ( [ sigmarepr ] ) , we are left with a function @xmath116 of the ratio @xmath117 : @xmath118 the function @xmath116 can be computed by performing numerically the integrations over the dimensionless variables @xmath14 s and @xmath105 s . the result is plotted in fig . [ s0 ] . ( figs0.eps width 13 cm ) notice that the dependence of @xmath116 on @xmath119 is rather mild when @xmath119 is near 1 . thus , in this region , @xmath116 in eq . ( [ tildesigmadef ] ) could be treated as a constant , so that @xmath120 when , in contrast , the two virtualities are widely disparate from each other , either @xmath121 or @xmath122 , the function @xmath116 has a strong dependence on @xmath119 , vanishing linearly ( modulo logarithmic enhancements ) with either @xmath123 or @xmath119 , respectively . this provides the overall change of scale in the cross section @xmath124 @xmath125 we will come back to this later on and examine the details of these behaviors numerically ( see sec . [ sec : virtuality ] ) . we now study the dependence of the color functions @xmath77 and of the photon - photon cross section on the polarization of the virtual photon . working in the frame defined by eq . ( [ qaqb ] ) , and denoting by @xmath126 , @xmath127 the lightlike unit vectors @xmath128 we introduce the following decomposition of the polarization tensor for the photon @xmath129 : @xmath130 the index @xmath131 in the first term in the right hand side of eq . ( [ gmunu ] ) runs over the transverse polarizations . we describe these polarizations by using the linear basis @xmath132 from the second term in the right hand side of eq . ( [ gmunu ] ) , we define the longitudinal polarization vector as @xmath133 the last term in eq . ( [ gmunu ] ) does not contribute because of current conservation . we thus have @xmath134 formulas analogous to eqs . ( [ gmunu])-([gmunubis ] ) , but with @xmath135 replaced by @xmath136 , hold for the photon @xmath137 . let us first consider the case of transverse polarizations . it is convenient to introduce the tensor amplitude @xmath138 by rewriting eq . ( [ capitalg ] ) in the form @xmath139 the explicit expression for @xmath138 can be found in appendix a. for polarization indices @xmath140 we also introduce the notation @xmath141 we can parametrize the transverse components of the tensor @xmath142 in terms of two scalar functions @xmath143 and @xmath144 : @xmath145 the function @xmath81 represents the unpolarized color function , which we have discussed in the previous section ( eqs . ( [ avg ] ) , ( [ gi1funct ] ) ) . the function @xmath144 carries the information on the polarization dependence . its explicit expression is @xmath146 ^ 2 } } + { { { { \bf p } } \cdot ( { { \bf p } } + { \bf k } ) - 2 \ , ( { { \bf p } } + { \bf k } ) \cdot { \bf k } \ ; { { \bf p } } \cdot { \bf k } / { { \bf k}}^2 } \over { \left [ ( { { \bf p } } + { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 \right ] \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right ] } } \right ] \;\;. % \end{aligned}\ ] ] the integrals in eq . ( [ g21 ] ) can be handled using the same procedure as in the unpolarized case : @xmath147 \ , , \nonumber % \end{aligned}\ ] ] where we have used the variables defined in eq . ( [ etaxi ] ) . correspondingly , the cross section for scattering transversely polarized photons , eq.([convsigma ] ) , can be decomposed as @xmath148 \hspace*{0.6 cm } . % \ ] ] the polarization average @xmath149 has been given to order @xmath150 in the previous section ( eq . ( [ sigmarepr ] ) ) , and the asymmetry @xmath151 to the same order @xmath150 is given in terms of the color function @xmath152 as @xmath153 numerical results are reported in figs . [ glin ] and [ s0 ] , where we plot @xmath152 versus @xmath154 , and show the dependence of @xmath151 on the incoming photon virtualities . unlike @xmath81 , @xmath152 is not logarithmically enhanced in the regions @xmath155 , @xmath156 . this is related to the fact that the splitting function associated with @xmath152 has zeros at the endpoints of the spectrum in the longitudinal momentum fraction , @xmath157 and @xmath158 ( see eq . ( [ g22 ] ) ) , whereas the unpolarized splitting function goes to a finite constant . the asymmetry @xmath151 associated with the photon - photon scattering process , calculated here in the born approximation , contributes to the asymmetry @xmath16 in eq . ( [ dsdphi ] ) at the level of the @xmath159 scattering . numerical results for the @xmath159 process will be given in sec . [ sec : numerical ] . we now move on to the case of the longitudinal polarization . let us consider the longitudinal - longitudinal color function : @xmath160 following the lines of the calculation described in detail for the case of transverse photons , we find @xmath161 ^ 2 } } - { { q^2 } \over { \left [ ( { { \bf p } } + { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 \right ] \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right ] } } \right ] \;\;. % \end{aligned}\ ] ] this contribution equals the color function @xmath152 given above . this can be seen by introducing an integral over a feynman parameter @xmath105 , then integrating over the transverse momentum @xmath162 in eq . ( [ gll1 ] ) . we get @xmath163 as noted above , @xmath152 has no logarithms at small @xmath164 . this corresponds to the absence of aligned - jet terms for longitudinally polarized photons @xcite . contributions from longitudinally polarized photons enter the @xmath159 cross section ( [ epaave ] ) . they will be included in the numerical estimates that we give in sec . [ sec : numerical ] . finally , we consider the interference contribution between longitudinal and transverse polarizations : @xmath165 in the high energy approximation in which we are working , this contribution vanishes . this can be seen explicitly by writing the color function in the general form @xmath166 and computing the invariant function @xmath167 . we get @xmath168 interference terms of the kind in eq . ( [ gtl ] ) would give rise to the asymmetry @xmath15 ( see eq . ( [ dsdphi ] ) ) at the level of the @xmath169 scattering process . we thus see that this asymmetry vanishes in the high energy approximation . we have seen that the two - gluon exchange mechanism gives rise to a constant @xmath170 total cross section at large @xmath28 , @xmath171 . to higher orders in perturbation theory , the iteration of gluon ladders in the @xmath172-channel promotes this constant to logarithms , and the perturbative expansion of the cross section at high energy has the form @xmath173 \hspace*{0.6 cm } , \hspace*{0.8 cm } l = \ln ( s / q^2 ) \hspace*{0.6 cm } , % \ ] ] where @xmath7 is a scale of the order of the initial photon virtualities , the sum represents the series of the leading logarithms to all orders in the strong coupling @xmath174 , and the dots stand for non - leading terms . to study the high energy behavior , it is convenient to analyze the cross section in its mellin - fourier moments , defined by @xmath175 where the @xmath176-integral goes along a contour parallel to the imaginary axis and to the right of any singularities in @xmath177 . we see from this definition that a constant behavior of the cross section with the energy @xmath28 is generated by a simple pole in the moments @xmath177 at @xmath178 , while powers of logarithms are generated by multiple poles at @xmath178 . the inverse of ( [ mellin ] ) is @xmath179 to sum the leading logarithmic terms , the basic observation of bfkl @xcite is that the logarithms arise when multiple soft gluons are emitted into the final state . these gluons have transverse momenta @xmath180 of the same order as @xmath35 and @xmath36 and have strongly ordered rapidities , lying between the rapidities of the quarks in photon @xmath181 and the quarks in photon @xmath182 . along with emission of real gluons , one also includes the exchange of corresponding virtual gluons . [ fac ] illustrates how the two gluon exchange graph of fig . [ ftwoglu ] is generalized to allow for multiple gluon emission . the quarks comprising photon @xmath181 couple to a gluon with momentum @xmath183 , while the quarks comprising photon @xmath182 couple to a gluon with momentum @xmath184 . gluons can be exchanged or emitted into the final state inside the subgraph labeled @xmath185 . in fact , the gluons that carry momenta @xmath183 and @xmath184 are , in general , combinations of soft gluons that carry a net color octet charge , that is , reggeized gluons @xcite . from a kinematic viewpoint , we can treat these as being equivalent to ordinary perturbative gluons . we consider the unpolarized @xmath186 cross section and adopt the following notation for the diagram in fig . [ fac ] : @xmath187 here @xmath188 represents the four - point green function for gluons @xmath189 and @xmath190 , while the functions @xmath191 represent the quark loops . we have made the following approximation . we note that the quark loop for photon @xmath181 depends sensitively on the minus component , @xmath192 , of the momentum that enters the quark loop via gluon @xmath181 . on the other hand , the quarks in photon @xmath182 have very large minus components of momenta . furthermore , because of strong rapidity ordering , all the other gluons inside of @xmath185 have minus components of momenta that are much larger than @xmath192 . thus we can neglect @xmath192 everywhere except in the quark loop for photon @xmath181 . then we include the integration over @xmath192 in the definition of @xmath193 . similarly we neglect @xmath194 everywhere except in the quark loop for photon @xmath182 and we include the integration over @xmath194 in the definition of @xmath195 . ( figfac.eps width 8 cm ) consider eq . ( [ genfig ] ) specialized to the case of two gluon exchange . the approximate gluon four - point function @xmath196 is trivial in this case : @xmath197 we identify @xmath198 then we recover the earlier result ( [ convsigma ] ) . for a generic term in @xmath199 , the transverse momenta @xmath200 and @xmath201 are independent variables , while the factor @xmath202 turns into @xmath203 . taking moments with respect to @xmath28 and changing integration variables to @xmath204 , @xmath205 and @xmath206 gives the structure @xmath207 having evaluated @xmath208 at some fixed perturbative order , we are interested in the poles of @xmath177 at @xmath178 . there are poles from the small @xmath209 and @xmath210 ends of the integrations over @xmath209 and @xmath210 . to evaluate @xmath177 without losing any powers of @xmath211 , we approximate @xmath212 the pole associated with the integration over @xmath209 can be thought of as arising from the integration over the rapidity of the final state gluon with the largest rapidity . almost all of the momentum fraction @xmath209 is taken by this gluon . since , in the leading logarithmic approximation , this gluon has a rapidity that is much less than that of the quarks in photon @xmath181 , the momentum fraction @xmath209 is negligible compared to 1 . similarly , the pole associated with the integration over @xmath210 can be thought of as arising from the integration over the rapidity of the final state gluon with the most negative rapidity . the poles from the large @xmath105 end of the integration over the function @xmath208 can be thought of as arising from integrations over final state gluons with intermediate rapidities . the leading logarithmic result can thus be written in the form @xmath213 where @xmath214 is the bfkl function that describes the interaction of gluons with fast moving colored systems moving in opposite directions and obeys the bfkl equation @xcite . it is normalized so that at order @xmath215 we have @xmath216 the solution to the bfkl equation can be written to all orders in @xmath174 in the form @xmath217 where @xmath218 the function @xmath219 is determined by solving the eigenvalue problem for the bfkl kernel and is given by @xmath220 with @xmath221 being the euler @xmath221-function . the bfkl function ( [ gareprf ] ) has poles at @xmath178 order by order in perturbation theory . ( [ ktfac ] ) shows that the poles in the @xmath104 cross section are generated from the ones in @xmath222 by integrating the color functions over @xmath180 . while these functions are specific to the off - shell photon probe , the function @xmath223 is universal . the same function contributes to the small @xmath1 behavior of the cross sections in hadron - initiated processes @xcite via the high energy factorization formulas . by inserting the representation ( [ gareprf ] ) in eq . ( [ ktfac ] ) , and scaling out the dependence on the photon virtualities , we get @xmath224 where @xmath225 is defined as the following @xmath226-transform of the photon color function @xmath227 @xmath228 the explicit expression of the function @xmath229 can be determined by using the representation ( [ twodimintg ] ) and performing the integral transform . the result reads @xmath230 eq . ( [ gareprsi ] ) , together with the explicit formulas ( [ chi ] ) , ( [ viexp ] ) , gives the leading logarithmic result for the moments of the @xmath231 total cross section . it sums the @xmath211 poles to the accuracy @xmath232 , for any @xmath233 . the lowest order perturbative contribution , @xmath234 , can be recovered from the summed formula ( [ gareprsi ] ) by expanding the denominator @xmath235 to the zeroth order in @xmath174 . the simple pole @xmath236 is the mellin transform of unity , and one can check numerically that the @xmath237-integral @xmath238 agrees with eq . ( [ sigmarepr ] ) and fig . [ s0 ] . in general , multiple pole contributions @xmath239 to the cross section are obtained by retaining higher orders in the @xmath174-expansion of eq . ( [ gareprsi ] ) . we see that the general structure of the coefficients of the leading logarithmic series comes from both @xmath219 and @xmath240 : the former is a universal function associated with the bfkl pomeron , while the latter describes the coupling of the pomeron to a specific physical source . the total cross section is obtained from eq . ( [ gareprsi ] ) by taking the inverse mellin - fourier transform ( [ mellin ] ) . by evaluating the @xmath176-integral from the residue at the pole @xmath241 , one gets @xmath242 note that this result depends on two mass scales , the scale @xmath243 in @xmath174 and the scale @xmath7 in the high energy corrections , whose reliable determination would require a next - to - leading analysis . we discuss the uncertainties in the leading log result associated with these scales in secs . [ sec : scales ] and [ sec : numerical ] . in the limit @xmath244 the integral ( [ energy ] ) is dominated by the region near @xmath245 , where the function @xmath246 has a saddle point . in the saddle point approximation one obtains @xmath247 with @xmath248 eq . ( [ saddle ] ) shows the asymptotic power behavior characteristic of the qcd pomeron , @xmath249 with @xmath250 . the pre - factor that determines the normalization of the asymptotic cross section , on the other hand , depends on the off - shell photon probe , and is controlled by the value of the function @xmath251 at @xmath252 . if the two photon virtualities are significantly far apart , corrections of order @xmath253 need to be taken into account when one calculates the large @xmath28 limit . in this case , one finds that the position of the saddle point is shifted to @xmath254 . apart from corrections in the pre - factor , the net effect of evaluating the integral ( [ energy ] ) around the shifted saddle point is to multiply the expression in the right hand side of eq . ( [ saddle ] ) by the factor @xmath255 \;\;\ ; . % \ ] ] the cross section acquires a gaussian modulation in @xmath256 with a width that grows like @xmath257 . ( figxsect.eps width 13 cm ) in the general case , the integral ( [ energy ] ) can be performed numerically . in fig . [ xsect ] we show the result as a function of @xmath258 for a given choice of the values of the photon virtualities and the strong coupling . for comparison we also plot the saddle point formula . as the energy increases the two curves get closer . however , in the range considered , the sub - asymptotic contributions are still significant ( about @xmath259 at @xmath260 , @xmath261 at @xmath262 ) . the large size of the corrections to the saddle point approximation can be mainly traced back to the fact that the function @xmath263 itself is rather sharply peaked around @xmath264 . this is illustrated in fig . [ chi05 ] , where we see that , for instance , for @xmath265 and @xmath266 , the width of the pomeron factor is still not small compared to the width of the factor associated with the off - shell photon color function . this effect accounts for most of the shift in the normalization of the cross section between the asymptotic and exact evaluation of the leading logarithmic sum . ( figchi05.eps width 13 cm ) using the same method described above for the cross section averaged over the two transverse photon polarizations , one can derive summed results for each photon polarization @xmath267 and for the polarization asymmetry @xmath151 . denoting by @xmath25 , @xmath268 and @xmath26 , @xmath269 the polarization indices for photons @xmath129 and @xmath137 , we write the factorization formula for the moments of the polarized cross section @xmath270 as @xmath271 the color functions @xmath77 have been discussed in sec . [ sec : polarization ] . the full solution for the green s function @xmath272 of the bfkl pomeron reads @xcite @xmath273 with @xmath274 the function @xmath223 in eq . ( [ gareprf ] ) , relevant to the case of the unpolarized cross section , is given by the term @xmath275 in eq . ( [ gareprfm ] ) . consider the product @xmath276 of the asymmetry times the averaged cross section , with the asymmetry @xmath151 defined in eq . ( [ unplusass ] ) . due to the azimuthal integration only the terms with @xmath277 in eq . ( [ gareprfm ] ) contribute . substituting the color function @xmath152 ( see eq . ( [ gij ] ) ) in eq . ( [ ktfacpol ] ) , one finds for the @xmath176 moments @xmath278 where @xmath279 is defined from the @xmath226-transform of the function @xmath280 analogous to ( [ videf ] ) , and has the expression @xmath281 by inverse mellin - fourier transformation , one obtains the summed formula for the asymmetry in the energy space : @xmath282 where @xmath149 is given by eq . ( [ energy ] ) . as in the case of the unpolarized cross section , the asymptotic behavior in the limit @xmath244 is determined by the saddle point approximation to the integral in eq . ( [ energyasy ] ) . in this case we observe a negative power law with the energy @xmath28 , controlled by the value of @xmath283 at the saddle point , @xmath284 , indicating that the angular correlations between the two photons tend to be washed out when bfkl pomeron exchange dominates . the contributions to the photon - photon cross section from longitudinally polarized photons are given by formulas analogous to eq . ( [ energy ] ) in terms of different combinations of the functions @xmath285 and @xmath286 , and the same function @xmath246 : @xmath287 unlike @xmath288 in eq . ( [ viexp ] ) , @xmath289 has simple poles at @xmath290 ( instead of double poles ) , corresponding to the non - logarithmic behavior of the color function @xmath152 at @xmath291 , @xmath156 noted in sec . [ sec : polarization ] . replacing functions @xmath285 by functions @xmath286 accounts for the different size of the longitudinal cross sections with respect to the purely transverse one . roughly , one finds @xmath292 the result ( [ energy ] ) for the @xmath104 cross section depends on two mass scales which can not be determined in a leading logarithmic analysis : the mass @xmath243 at which the running coupling @xmath174 is evaluated , and the mass @xmath7 that provides the scale for the high energy logarithms . the former can be thought of as being associated with the integrations over the transverse momenta in the loops contributing to higher order diagrams , while the latter stems from the longitudinal integrations . a reliable determination of these scales would require a next - to - leading order calculation . lacking this , we provide here qualitative arguments to relate these scales with the physical hard scales of the problem . in sec . [ sec : numerical ] we will use these relations to examine numerically the dependence of the cross section on the scale choices . a possible choice of the scale @xmath243 in the strong coupling is based on the prescription of ref . @xcite . to apply this prescription , we consider the lowest order gluon exchange contribution , which we have discussed in sec . [ sec : notations ] . this is given ( see eq . ( [ convsigma ] ) ) by the integral in @xmath293 of two @xmath77 factors , each of which is proportional to @xmath174 , @xmath294 . we first compute the quark loop contribution to the gluon propagator , renormalized in the @xmath295 scheme : @xmath296 \;\;\;\ ; , % \ ] ] where @xmath297 is the contribution to @xmath298 from one quark loop , and @xmath299 . inserting a quark loop into the gluon propagator in the lowest order diagram amounts to replacing @xmath300 by @xmath301 % \ ] ] ( as in @xcite this can be regarded as a contribution to an effective coupling @xmath302 ) . now , following @xcite we choose the scale @xmath243 so that the quark loop contribution vanishes after integrating over @xmath97 : @xmath303 this is the same procedure that applies in the case of the abelian theory . using the representation ( [ twodimintg ] ) for @xmath77 , we are led to calculate an integral of the form @xmath304 exploiting the symmetry of the integrand under the transformation @xmath305 and interchange of the variables @xmath306 , @xmath307 , one can show that the condition ( [ vanish ] ) is satisfied by @xmath308 note that , when the two photon virtualities @xmath113 and @xmath114 are far apart from each other , the prescription @xcite picks out a scale for @xmath174 which is neither of the order of the big virtuality nor of the order of the small one , but rather is proportional to the geometric mean @xmath309 . the value of the proportionality coefficient @xmath310 in eq . ( [ cmu ] ) is specific to the subtraction scheme ( @xmath295 ) chosen to define the quark loop insertion . the argument given above applies to the factors of @xmath174 that appear in the born approximation to the high energy cross section . the result ( [ energy ] ) , however , also contains a dependence on the running coupling through the higher order factor @xmath311 associated with the solution of the bfkl equation . for the scale in @xmath174 in this case one does not have such a simple argument as the one described above . for the numerical estimates in sec . [ sec : numerical ] we will make the assumption that the same value of @xmath243 also controls the running coupling in the bfkl factor . in sec . [ sec : numerical ] we will check the numerical effect of varying this scale . we now consider the mass @xmath7 that provides the scale for the large energy logarithms @xmath312 ( see eq . ( [ expansion ] ) ) . to estimate this scale , we observe that the rapidity of gluons exchanged in the rungs of the bfkl ladders should lie between the rapidity @xmath313 of the quark @xmath314 ( produced by the photon @xmath129 ) and the rapidity @xmath315 of the quark @xmath67 ( produced by the photon @xmath137 ) . this gives rise to integrations over the rapidity intervals @xmath316 the logarithms of the energy @xmath28 are generated precisely by these integrals . estimating the size of the rapidity intervals thus allows us to estimate the scale @xmath7 . expressing the `` @xmath40 '' and `` @xmath41 '' momentum components of the quark @xmath314 as @xmath317 we write its rapidity as @xmath318 similarly , for the quark @xmath67 we have @xmath319 taking the difference between these rapidities gives @xmath320 the average transverse momenta @xmath321 , @xmath322 carried by the quarks are of the order of the photon virtualities @xmath35 , @xmath36 . for the longitudinal momentum fractions @xmath323 , @xmath324 , we assume a typical maximum value of the order @xmath325 in the high energy region . using these estimates , we obtain @xmath326 we thus identify the scale @xmath327 the transverse - momentum integrations in the factorization formula ( [ ktfac ] ) are dominated by values of @xmath328 of the order of the photon virtualities . ( for the purpose of this section we will take the photon virtualities to be equal , @xmath329 . ) as a result , for sufficiently off shell photons the dominant contribution to the cross section comes from short distances , and the evaluation of eq . ( [ ktfac ] ) gives rise to a finite result in perturbation theory . however , there are limitations on the perturbative treatment that are intrinsic to the bfkl equation . these limitations come from the region of very high @xmath28 . although an accurate understanding of these effects is an open problem @xcite that goes beyond the scope of this work , one can nevertheless make some rough estimates . we discuss them in this section . in order to keep the notation simple , we do not distinguish here between the scales @xmath330 , @xmath331 , and @xmath332 , calling all of these simply @xmath7 . it is known from the structure of the bfkl equation that , even if the incoming @xmath333 , @xmath334 are large , say , @xmath335 , the typical transverse momenta in the gluon ladders contributing to the function @xmath336 ( see eq . ( [ gareprf ] ) ) may diffuse away from @xmath7 as the energy becomes very large . the diffusion coefficient in @xmath337 can be read directly from the asymptotic solution to the bfkl equation ( or , equivalently , from the exponential term ( [ incldiff ] ) in the @xmath104 cross section ) and is given by @xmath338 as a result , the distribution of the transverse momenta in the bfkl ladders is a gaussian in @xmath337 centered around @xmath7 with a width proportional to @xmath339 @xcite . with increasing @xmath28 the distribution broadens , and one becomes sensitive to the region of small transverse momenta . a self - consistency check of the perturbative treatment requires that the contribution from transverse momenta of the order of @xmath340 be suppressed . in order to keep away from momenta of this order one has to have @xmath341 identifying @xmath174 with the strong coupling evaluated at the scale @xmath7 , @xmath342 , one gets @xmath343 for any given @xmath7 , this can be read as an upper bound on the domain of energies @xmath28 in which we expect the perturbative approach based on the bfkl pomeron to be reliable . observe that , for small values of @xmath174 , this is not so stringent a constraint . however , because the bfkl function @xmath246 has a large second derivative at the saddle point , the numerical value of the coefficient @xmath344 is small . therefore , the limit ( [ difflimit1 ] ) may become relevant unless @xmath7 is very large . the bfkl equation is also known to give rise to violation of the unitarity bound at asymptotically large energies . the growth of the cross section predicted by the bfkl equation can not continue indefinitely , and unitarity corrections must arise to slow it down . roughly , these effects are expected to become important when the calculated cross section is bigger than the naive geometrical cross section @xmath345 . a careful discussion may be found in ref . the simplest estimate , @xmath346 , yields , using eq . ( [ saddle ] ) , @xmath347 where we have neglected factors from the term in @xmath348 in eq . ( [ saddle ] ) . assuming , as we did before , @xmath174 to be evaluated at the scale @xmath7 , we can write @xmath349 in the term @xmath181 we collect contributions arising from the factor in the square root in eq . ( [ unitlimit ] ) as well as from the factor @xmath285 in eq . ( [ saddle ] ) . the different functional dependence on @xmath174 in the right hand sides of the inequalities ( [ difflimit1 ] ) and ( [ unitlimit1 ] ) implies that , for small enough values of @xmath174 , the unitarity limit ( [ unitlimit1 ] ) is more stringent than the diffusion limit ( [ difflimit1 ] ) . this suggests that for sufficiently high @xmath7 it should be possible to study unitarization in a purely perturbative context @xcite . on the other hand , the coefficients @xmath344 and @xmath350 are significantly different in size this may make the two bounds ( [ difflimit1 ] ) and ( [ unitlimit1 ] ) rather comparable for moderate values of @xmath7 . in addition , inspection of eqs . ( [ saddle ] ) , ( [ unitlimit ] ) suggests that the term @xmath181 in eq . ( [ unitlimit1 ] ) is not necessarily negligible , and may contribute to push the onset of unitarity corrections further away . as to the impact on experimental studies at future @xmath29 colliders , we observe that unitarity corrections should set in when the cross section has grown to be much larger than the born value @xmath351 . at a future @xmath29 collider one may thus expect to see the rise of the cross section with @xmath258 . possibly , one may see this rise slow as unitarity corrections become important @xcite . if we let @xmath113 be much larger than @xmath114 in eq . ( [ energy ] ) , we obtain a result that describes small @xmath1 deeply inelastic scattering from a transversely polarized photon @xcite whose virtuality @xmath114 is sufficiently large to allow the use of perturbation theory to analyze its decomposition into quarks . the structure function @xmath352 ( which is the same as @xmath353 ) is related to the virtual photon - photon cross section by @xmath354 with @xmath355 thus eq . ( [ energy ] ) gives a leading logarithmic summation for @xmath356 at small @xmath1 : @xmath357 the scale @xmath7 is not fixed by a leading logarithmic calculation . a sensible choice for @xmath7 based on a qualitative argument has been discussed in sec . [ sec : scales ] . for the purpose of the present section we simply leave it undetermined . consider the perturbative expansion of eq . ( [ f1 ] ) , @xmath358 \ , v_1 ( \gamma ) \ , v_1 ( 1 - \gamma ) \hspace*{0.1 cm } . % \ ] ] the functions @xmath263 , @xmath359 and @xmath219 have poles at integer values of @xmath237 . thus this expression contains contributions proportional to @xmath360 times logarithms , @xmath361 times logarithms , and so forth . we extract the leading twist contribution , the part proportional to @xmath360 times logarithms , by rewriting the integral as an integral over a contour @xmath362 that encircles the singularity at @xmath363 plus an integral over a contour from @xmath364 to @xmath365 . the integral over the contour @xmath362 is the leading twist contribution . we discard the integration over the contour from @xmath366 to @xmath365 , which contains the higher twist contributions . thus the leading twist contribution to @xmath356 is @xmath367 \ , v_1 ( \gamma ) \ , v_1 ( 1 - \gamma ) \nonumber\\ & = & { 1 \over ( 2 \pi)^3 \alpha } \ { \lower 6 pt \hbox{\vbox{\hbox{$\rm res$ } \kern -6 pt \hbox{$\scriptstyle \,\gamma = 0 $ } } } } \ , \left ( { { { q}^2_a } \over { { q}^2_b } } \right)^{\gamma } \left [ 1 + { \bar \alpha_s } \ , \chi ( \gamma ) \ln\left(q_a^2 \over x\ , q^2 \right ) + \cdots \right ] \ , v_1 ( \gamma ) \ , v_1 ( 1 - \gamma ) \hspace*{0.1 cm } . % \end{aligned}\ ] ] it is straightforward to determine the perturbative coefficients . the function @xmath368 has a double pole at @xmath363 , as does the function @xmath369 . thus there are three powers of @xmath370 at leading order in @xmath174 , which is @xmath150 since @xmath371 . we find @xmath372 \hspace*{0.2 cm } , \hspace*{0.4 cm } \zeta ( 2 ) \simeq 1.64 \hspace*{0.2 cm}. % \end{aligned}\ ] ] if we restore the summation of the leading logs of @xmath373 , we obtain @xmath374 this is not a simple function . however , one can easily perform the integration numerically . one interesting result is that we observe what used to be called `` precocious bjorken scaling '' . that is , the leading twist approximation begins to be accurate at values of @xmath113 that are not really very large compared to the only other scale in the problem , @xmath114 . we illustrate this in fig . [ twist ] , working at order @xmath150 . we compare the leading twist approximation that applies for @xmath375 , the leading twist approximation that applies for @xmath376 , and the full result . first we define @xmath377 and plot @xmath378 versus @xmath379 . then we define @xmath380 and plot @xmath381 finally , we plot @xmath382 at the leading order in @xmath174 without approximation . we see that the leading twist approximation for @xmath375 is quite good down to @xmath383 , while the leading twist approximation that applies for @xmath384 is quite good down to @xmath385 . in fact , at @xmath386 , both approximations are quite good . a similar behavior is observed if we include higher orders in @xmath174 . ( figtwist.eps width 13 cm ) the qcd result for @xmath387 scattering can be compared with expectations for the structure of the high energy cross section based on traditional regge theory @xcite . in regge theory , to analyze the elastic scattering of particles @xmath181 and @xmath182 one considers the singularity structure of the amplitude @xmath388 in the complex angular momentum plane . the simplest case is a pole in the angular momentum @xmath389 plane located at a position dependent on @xmath172 , @xmath390 . one then obtains the asymptotic behavior @xmath391 for @xmath244 and @xmath172 fixed , and the amplitude takes the factorized form @xmath392 here @xmath393 , @xmath394 are functions of the transferred momentum , associated respectively with the couplings of particles @xmath181 and @xmath182 to the reggeon whose trajectory is @xmath395 . for the case of the @xmath104 total cross section , this would correspond to the structure @xmath396 where we have used the optical theorem to relate the total cross section to the imaginary part of the forward elastic amplitude , and we have let @xmath397 and @xmath398 depend on the photon virtualities in the case of off - shell photons . however , it has been long known , from various phenomenological and theoretical considerations @xcite , that this structure can not be exactly true , and strong - interaction scattering at high energy has to have a more complicated singularity structure than a pole , such as moving or fixed cuts . in this case , one does not expect the factorized form for the cross section to hold . if we now turn to the qcd result ( see eq . ( [ gareprsi ] ) ) , we may ask what kind of singularities the amplitude has in the angular momentum plane . the bfkl pomeron is known to give rise to a ( fixed ) branch point singularity @xcite . to see this , let us consider the moments of the @xmath387 cross section . at leading level , we are allowed to identify the moment @xmath176 with the complex angular momentum @xmath389 ( [ gareprsi ] ) is written in terms of an integral in the complex @xmath237-plane . the integrand has the pole @xmath399 and one factor of @xmath400 for the coupling of the gluon system to the quarks in each photon . thus the integrand has a factorized structure . to understand the angular - momentum singularity structure , one should see what becomes of the pole @xmath401 after @xmath237-integration . for @xmath402 , the @xmath237-integral is well approximated by the residue at the rightmost pole @xmath403 to the left of the integration contour . as we have seen in the previous section , numerically this approximation turns out to be fairly good down to values of @xmath404 just above @xmath405 . in this approximation one gets @xmath406 \;\ ; , \;\;\ ; c_n = - { { v_1 ( { \bar \gamma}_n ) \ , v_1 ( 1- { \bar \gamma}_n ) } \over { { \bar \alpha}_s \ , \chi^{\prime } ( { \bar \gamma}_n ) } } \;\;\;. % \ ] ] the leading pole @xmath407 is determined by the equation @xmath408 . at @xmath409 it has a square - root branch point singularity : @xmath410 correspondingly , the residue @xmath411 is also singular : @xmath412 therefore , the qcd result implies a branch point rather than a simple pole in the angular momentum plane . as a consequence , we do not obtain a regge - factorized form for @xmath413 . it is of interest , however , to see by how much this factorization is violated . we first consider the asymptotic formula ( [ saddle ] ) , obtained by using the saddle point approximation around @xmath414 . in this case , provided the scale @xmath7 is fixed by @xmath415 , as discussed in sec . [ sec : scales ] , and the qcd coupling @xmath174 is fixed , an approximate form of regge factorization is recovered . the piece which violates this factorization in eq . ( [ saddle ] ) is proportional to the square root of a logarithm , and is therefore a slowly varying function . a more substantial source of factorization breaking comes in when one takes into account the correction due to the position of the saddle point drifting away from @xmath416 , see eq.([incldiff ] ) . for the exact leading log result ( [ energy ] ) , a possible way to quantify the deviation from the regge - factorized behavior is to look at the ratio @xmath417 ^ 2 } \over { \sigma_{\gamma^ * \ , \gamma^ * } ( s , q^2_a , q^2_a ) \ , \sigma_{\gamma^ * \ , \gamma^ * } ( s , q^2_b , q^2_b ) } } \hspace*{0.5 cm } . % \ ] ] if regge factorization holds , this quantity should be equal to @xmath405 . on the other hand , from eq . ( [ energy ] ) we see that @xmath418 goes like @xmath419 for @xmath375 , and @xmath420 for @xmath376 . that is , for @xmath35 and @xmath36 sufficiently far apart the regge - factorized form breaks down . ( figratio.eps width 13 cm ) in fig . [ fratio ] , we plot @xmath418 as a function of the ratio of the photon virtualities @xmath421 for different values of the energy . it is interesting to observe that , for typical parameter values , @xmath418 varies by not more than 40% when @xmath421 varies from 0.1 to 10 . we take this as an indication that an approximate regge factorization holds numerically for the exact integral ( [ energy ] ) if @xmath421 is not too large or too small . the calculation of the high energy photon - photon cross section discussed so far is a perturbative calculation , based on the dominance of short distances for large photon virtualities . when the photon virtualities decrease , one goes out of the region of validity of the perturbative approach . as the photons go near the mass shell , the high energy scattering process is expected to become dominated by soft interactions . in this regime one is not able to calculate in qcd , and in order to have a ( phenomenological ) description of the cross section , one rather has to rely on models for strong - interaction scattering based on regge theory . for on - shell photons , the regge factorization hypothesis allows us to relate the photon - photon total cross section to the photon - proton and proton - proton cross sections , as follows @xmath422 assuming the values @xmath423 , @xmath424 , one gets @xmath425 . for virtual photons with small @xmath35 and @xmath36 , the fall - off of the cross section can be estimated from vector meson dominance : @xmath426 as the photon virtualities increase , the cross section , instead of continuing to fall like @xmath427 , should begin to fall more slowly . at large photon virtualities ( of the order of a few @xmath428 , or bigger ) , it should go over to the perturbative scaling behavior in eq . ( [ energy ] ) , @xmath429 at fixed @xmath258 . note that the @xmath430 behavior could not be obtained in the framework of the regge factorization ( [ ppgap ] ) even if one assumed perturbative scaling in each one of the photon cross section factors , that is , even if one assumed @xmath431 . this is the counterpart ( at the level of hadronic cross sections ) of the effect of the deviation from unity observed for the ratio @xmath418 in the previous section ( see eq . ( [ reggeratio ] ) ) . in fact , experimental data on the @xmath432 cross section for large photon virtuality are now available in the region of high energies from the measurements of small-@xmath1 deeply inelastic scattering at hera . the above observation amounts to saying that , even if one used the data for @xmath433 , the relation ( [ ppgap ] ) would not lead to the correct perturbative qcd result for the @xmath104 cross section . in fig . [ loglog ] we show a log - log plot of the curves corresponding to the soft and perturbative formulas for the @xmath7-behavior of the cross section @xcite . for the former , we use eqs . ( [ ppgap])-([vmd ] ) , and for the latter we take the born approximation to eq.([energy ] ) . in this plot we are interested in emphasizing the dependence of the cross section on @xmath7 at fixed @xmath258 . for this reason we limit ourselves to the lowest order formulas , and do not include the higher order summation of the @xmath434 terms that give an enhancement at large energies . it is understood that such high - energy corrections affect both the soft and perturbative curves , giving rise to `` soft - pomeron '' and `` hard - pomeron '' effects in the two cases . ( figloglog.eps width 13 cm ) the region of intermediate values of @xmath7 in fig . [ loglog ] ( @xmath435 of the order of @xmath436 ) is where the transition from the soft - scattering regime to the hard - scattering regime is expected to occur . the mechanism through which this happens is not theoretically under control at present , and one may consider trying to estimate the cross section in this region by interpolating between the two curves . in the next section we will study the prospects for investigating high energy @xmath170 scattering at @xmath2 colliders , and we will discuss which regions in @xmath7 in fig . [ loglog ] can likely be accessed experimentally at lep200 and a future @xmath29 collider . the cross section for high energy virtual photon scattering can be measured in @xmath437 collisions in which the outgoing leptons are tagged . the cross section for the electron - positron scattering process is obtained by folding the @xmath109 cross section with the flux of photons from each lepton . consider the four - fold differential @xmath437 cross section averaged over the angle between the lepton scattering planes , eq . ( [ epaave ] ) . to get an estimate of the rates available to study bfkl effects in virtual photon scattering at @xmath437 colliders of the present and next generation , we integrate this cross section over a region @xmath438 determined by cuts that we discuss below : @xmath439 we choose \i ) @xmath440 , @xmath441 , where @xmath442 is a few gev , in order that the coupling @xmath174 be small , and that the process be dominated by the perturbative contribution ; \ii ) @xmath443 , in order that the high energy approximation be valid . we discuss the parameter @xmath444 below . note that , with these criteria , the photon virtualities @xmath7 lie in a range @xmath445 in which the equivalent photon approximation ( eq . ( [ epaave ] ) ) is expected to work fairly well . on one hand , kinematical corrections of order @xmath446 are suppressed in this range . on the other hand , contributions of order @xmath447 to the @xmath448 splitting process can be neglected . to choose a value for the parameter @xmath444 , we compare the gluon exchange contribution with contributions that are suppressed by a power of @xmath28 @xcite . we consider first the two gluon exchange graph , for which @xmath449 for large @xmath28 . taking the case @xmath450 , we have from fig . [ s0 ] @xmath451 next , we consider the leading order ( electromagnetic ) contribution to @xmath452 , occurring via quark exchange , for which @xmath453 . in appendix [ app : quark ] we report results for this subprocess . for @xmath454 and large enough @xmath28 , the corresponding cross section is well approximated by the formula : @xmath455 demanding that the gluon exchange graph give a larger contribution than the quark exchange graph leads to the requirement @xmath456 for typical perturbative values , @xmath457 , we get @xmath458 . we will therefore use @xmath459 as a standard value for @xmath444 . we note that for @xmath460 , one is surely entitled to drop terms in the gluon exchange graphs that are suppressed by powers of @xmath461 , as we have done . we thus compute the integrated rate @xmath462 in eq . ( [ intrate ] ) using the results given in sec . [ sec : summation ] for the photon - photon cross section , and setting the scales in the running coupling and in the high energy logarithms according to the prescriptions discussed in sec . [ sec : scales ] . the dependence of @xmath462 on the lower bound @xmath463 on the photon virtualities is shown by the `` summed '' curve in fig . [ f500 ] for the energy of a future @xmath29 collider . [ f200 ] shows the cross section for the lep collider at cern operating at @xmath464 . ( fig500.eps width 13 cm ) ( fig200.eps width 13 cm ) the dashed and solid lines in figs . [ f500 ] and [ f200 ] correspond to the result of using , respectively , the born and the summed expressions for the photon - photon cross section . at the values of @xmath465 considered in the figures , summation effects enhance the rates significantly in the range of @xmath442 of a few @xmath466 . as @xmath442 increases , lowest order perturbation theory gets closer and closer to the fully summed prediction , as a result of both @xmath174 becoming small and the phase space closing up for the high energy logarithms . in figs . [ f500 ] and [ f200 ] , we also plot separately the contribution to the cross section from purely transverse photons , that is , the contribution from the term in @xmath467 in eq . ( [ epaave ] ) . we see that this contribution accounts for about three quarters of the full cross section . for values of the cuts @xmath468 , @xmath469 , we find @xmath470 at lep200 energies , and @xmath471 at the energy of a future collider . these cross sections would give rise to about @xmath472 events at lep200 for a value of the luminosity @xmath473 , and about @xmath474 events at @xmath475 for @xmath476 . the choice of the cuts that can realistically be implemented is affected by experimental constraints . in particular , the lowest photon virtualities that can be reached are limited by the angular acceptance of the detector , according to the relation @xmath477 , where @xmath478 is the beam energy , @xmath479 the angle of the tagged lepton , and @xmath1 is the momentum fraction of the emitted photon . in this situation , the value @xmath480 , for which the rates ( [ vallep ] ) , ( [ valnlc ] ) are given , implies detecting leptons scattered through angles down to about @xmath481 at lep200 , which is close to the range of the current luminosity monitors at the lep experiments @xcite . for a future 500 gev collider , @xmath482 corresponds to a minimum angle of about @xmath483 . it appears that working down to such an angle will be difficult but not impossible @xcite . if instead we take @xmath484 , the minimum angle is @xmath485 . then the cross section is about @xmath486 , corresponding to about @xmath487 events . as stated earlier , the numerical results given above depend on the choice of the scales in @xmath174 and in the high energy logarithms that enter the photon - photon cross section . for the calculations described above , we have used the prescriptions given in sec . [ sec : scales ] . different scale choices are possible , and they would affect the predictions at the next - to - leading logarithmic order , which is beyond the present theoretical accuracy . we can use the variation of the results with the scale choices to get an estimate of the uncertainties associated with unknown sub - leading corrections . we can vary the two scales @xmath243 and @xmath7 ( see eqs . ( [ cmu ] ) , ( [ cqu ] ) ) independently . an illustration of this is reported in fig . here we compare the result of fig . [ f500 ] with the curve obtained by multiplying the scale in @xmath174 by a factor of @xmath488 , @xmath489 , and the curve obtained by reducing the scale in the high energy logarithms by a factor of @xmath488 , @xmath490 . the band between these two curves indicates that the uncertainty on the leading logarithmic result is fairly large , and emphasizes the need for improving the accuracy of the calculations at high energy . ( figband.eps width 13 cm ) the plots of the cross section versus @xmath463 shown in figs . [ f500 ] and [ f200 ] illustrate the expected dependence of the photon - photon cross section on the photon virtualities . if we fix @xmath442 we can look at the dependence on the photon - photon c.m . energy @xmath491 . it is useful to use @xmath492 and the photon - photon rapidity @xmath493 as variables instead of @xmath494 and @xmath495 . then we define @xmath496 using eq . ( [ epaave ] ) , we can write @xmath497 as @xmath498 where we omit three similar terms . we see that @xmath499 is very directly related to the @xmath500 cross section . we plot @xmath501 at @xmath502 in fig . [ hats200 ] for @xmath503 and in fig . [ hats500 ] for @xmath504 . here we choose @xmath505 and @xmath506 . in each case , we show a curve for the born level cross section and another for the full bfkl cross section . we also show the cross section arising from the scattering of ( transversely polarized ) photons via quark exchange instead of gluon exchange . we see that , with our choice of cuts , quark exchange scattering is suppressed . ( eeshat2.eps width 13 cm ) ( eeshat5.eps width 13 cm ) for @xmath507 the cross section shows a strong dependence on the cut @xmath508 . with our choice of @xmath509 and with @xmath510 , the cross section is forced to vanish for @xmath511 . as @xmath512 increases , the effect of this cut on the @xmath35 and @xmath36 integrations becomes less and less important . since @xmath513 is independent of @xmath514 at born level , the cross section begins to flatten out as @xmath512 increases to about @xmath515 . for larger values of @xmath512 , the born cross section decreases because of the influence of the photon flux factor @xmath516^{1/2})^2 $ ] . for the summed bfkl curve , the growth of @xmath517 overcomes the effect of the photon flux factor , so that the cross section rises with @xmath512 . the curves for @xmath503 and @xmath518 are similar . the main difference is that at @xmath519 there is more available range for @xmath512 . our plots are for @xmath502 . the available range of @xmath520 is @xmath521 . thus the cross section integrated over @xmath520 goes to zero as @xmath522 . we see from the results presented above that at a future @xmath29 collider it should be possible to probe the effects of pomeron exchange in a range of @xmath7 where summed perturbation theory applies . one should be able to investigate this region in detail by varying @xmath35 , @xmath36 and @xmath523 independently . at lep200 such studies appear to be more problematic mainly because of limitations in luminosity . even with a modest luminosity , however , one can access the region of relatively low @xmath7 in the graph of fig . [ loglog ] if one can get down to small enough angles . this would allow one to examine experimentally the transition between soft and hard scattering . we now move on to the angular distribution for the @xmath159 scattering cross section , and consider the asymmetries @xmath15 , @xmath16 introduced in eq . ( [ dsdphi ] ) . as pointed out in sec . [ sec : polarization ] , @xmath15 is zero at leading order . on the other hand , @xmath16 is given by an equivalent - photon formula in terms of the asymmetry @xmath151 for the @xmath104 scattering process discussed in sec . [ subsec : asym ] . this reads @xmath524 with @xmath151 and @xmath149 being given in eqs . ( [ energyasy ] ) , ( [ energy ] ) . we can use the same cuts discussed earlier in this section to integrate eq . ( [ epaasy ] ) , and thus define @xmath525 where @xmath462 is the integrated rate in eq . ( [ intrate ] ) . performing the integral numerically , we find that the asymmetry @xmath526 is very small . as noted in sec . [ sec : summation ] , the role of the summed bfkl terms is that of reducing the magnitude of the asymmetry with respect to the born order result . at a 500 gev collider , in the range of the angular and energy cuts previously described , we find @xmath527 . we observe that spin effects in photon - photon scattering at high energies are interesting , but the predicted asymmetries are either zero or small . understanding the behavior of high energy hadron reactions from a fundamental perspective within qcd is an important goal of particle physics . as we have shown in this paper , virtual photon scattering @xmath528 at high energies , @xmath529 , provides a remarkable window into pomeron physics . the total cross section can be studied as a function of the space - like mass of each incident projectile . most importantly , the process can be investigated in the regime where the photons both have large virtuality , so that one can use the framework of perturbative qcd . compared to tests of the qcd pomeron behavior based on deeply inelastic structure functions , the measurement of the total cross section for sufficiently off - shell photons is free from the long - distance ambiguities related to the structure of the hadronic target . on the other hand , unlike tests based on associated jet production in lepton - hadron or hadron - hadron collisions , the @xmath530 measurement is fully inclusive and therefore it does not depend on specifying the details of the final state . the scattering of highly virtual photons can be described as the interaction of two incident color singlet @xmath79 pairs of small transverse size interacting through multiple gluon exchange . we have studied this reaction both in the born approximation ( corresponding to two - gluon exchange ) and also with the inclusion of the higher - order summation encompassed by the bfkl equation . the cross section at high energies and large virtuality takes a factorized form in transverse coordinates . however , it does not factorize simply into separate functions of @xmath113 and @xmath114 , which reflects the cut structure of the bfkl pomeron in the complex angular momentum plane . we have also examined the background contribution from quark exchange , a process which is power - law suppressed at high energy . according to this analysis , the @xmath531 cross section falls off at high virtuality only as @xmath532 , where @xmath533 . the rate for sensitive tagged - lepton experiments at high energy @xmath534 or @xmath535 colliders is thus not negligible . in particular it appears that the main features of the perturbative qcd predictions , such as the energy dependence , the factorization properties of the cross section , the scaling laws in @xmath113 , @xmath114 , as well as the polarization and azimuthal correlations can be tested in detail at a high - energy and high - luminosity next linear collider . we have also found that an interesting first look at virtual photon scattering can be obtained from the tagged lepton events measured in the luminosity monitors of present experiments at lep200 . more precisely , we estimate that , in the region of photon virtualities where summed perturbation theory is expected to apply , there should be several hundred events at lep200 , and about @xmath536 events at a future 500 gev collider with an integrated luminosity of @xmath537 . we also find that the enhancement due to bfkl pomeron terms over the born cross section is sizable , and should be visible particularly in the @xmath538-distribution of the cross section , with @xmath523 . the dependence of the cross section on the photon virtualities @xmath35 and @xmath36 is perturbative , and can be predicted in the framework of the bfkl equation . these predictions can be tested by measuring the angles of the recoil leptons . both the case in which the two photon virtualities are varied together ( @xmath539 ) and the case in which they are kept far apart ( @xmath540 ) are of interest . in the second case one gets to observe the structure function of a virtual photon at small bjorken-@xmath1 . the spin structure is rich , but hard to observe . most of the observable cross section comes from the scattering of two transversely polarized photons . for this part of the cross section , there is an asymmetry in the angular distribution of the outgoing leptons , but this asymmetry is less than @xmath541 . in the region of low photon virtualities ( @xmath542 smaller than a few @xmath466 ) , the photon - photon cross section becomes dominated by soft interactions . here one can not use a perturbative analysis . on the other hand , one may explore experimentally at what scales the breakdown of the perturbative result occurs , and how this is connected to the onset of the phenomenological `` soft - pomeron '' behavior . the theory that is available at present is leading logarithmic and therefore is affected by rather large uncertainties . these uncertainties can be parametrized in terms of two mass scales , the transverse scale that controls the running coupling and the longitudinal scale associated with the high energy logarithms . a next - to - leading logarithmic calculation would help determine these scales . such a calculation could make the theoretical predictions much more precise . at the largest values of @xmath514 , new effects related to unitarity and diffusion may become important . if so , an improved theory that deals with these effects would be testable at a future @xmath29 collider . we are grateful to j. bjorken and a. mueller for discussions and for their interest in this work . we thank d. strom for useful advice . this work was supported in part by the united states department of energy grants de - ac03 - 76sf00515 and de - fg03 - 96er40969 . we start with the expression ( [ m1 ] ) for the amplitude corresponding to the graph in fig . [ ftwoglu ] . the overall charge factor in eq . ( [ m1 ] ) is @xmath543 where we have used the color trace @xmath544 . we use the mass shell constraints on the final quarks , @xmath314 and @xmath67 , to eliminate the integrals over their `` @xmath41 '' and `` @xmath40 '' components , respectively , thus obtaining @xmath545 @xmath546 we use the mass shell constraints on the antiquarks to eliminate the integrals over the `` @xmath41 '' and `` @xmath40 '' components of the exchanged momentum @xmath233 , as follows : @xmath547 @xmath548 note that in eqs . ( [ kminus ] ) , ( [ kplus ] ) we have neglected terms of order @xmath549 and @xmath550 with respect to unity , consistently with the high energy approximation . ( [ m1 ] ) can then be rewritten as @xmath551 } \over { \left ( ( p_a + k)^2 + i \ , \varepsilon \right ) \left ( ( p_a - q_a)^2 - i \ , \varepsilon \right ) } } \ , \nonumber \\ & ~ & \times \ , { { { \mbox { \rm tr } } \left [ \ , { \rlap / p}_b \ , \gamma^\alpha \ , ( { \rlap / p}_b - \rlap / k ) \ , { \rlap/\varepsilon}_b \ , ( { \rlap / q}_b - { \rlap / p}_b + \rlap / k ) \ , \gamma^\beta \ , ( { \rlap / p}_b - { \rlap / q}_b ) \ , { \rlap/\varepsilon}_b \right ] } \over { \left ( ( p_b - k)^2 + i \ , \varepsilon \right ) \left ( ( p_b - q_b)^2 - i \ , \varepsilon \right ) } } \;\;. % \end{aligned}\ ] ] we now re - express the denominators and numerators of the amplitude ( [ m1 ] ) in the high energy limit . with the neglect of terms of order @xmath549 and @xmath550 , the denominators take the form @xmath552 @xmath553 and analogously @xmath554 @xmath555 in the numerator , light - cone gluon polarizations are dominant at high energy , and therefore we are led to calculate the product of traces @xmath556 \ ; \nonumber \\ & ~ & \times \ , { \mbox { \rm tr } } \left [ \ , { \rlap / p}_b \ , \gamma^{- } \ , ( { \rlap / p}_b - \rlap / k ) \ , { \rlap/\varepsilon}_b \ , ( { \rlap / q}_b - { \rlap / p}_b + \rlap / k ) \ , \gamma^{- } \ , ( { \rlap / p}_b - { \rlap / q}_b ) \ , { \rlap/\varepsilon}_b \right ] \;. \nonumber % \end{aligned}\ ] ] the result reads @xmath557 substituting eqs . ( [ denoa1])-([denob2 ] ) and ( [ trres ] ) in eq . ( [ m2 ] ) , we obtain the expression ( [ m3 ] ) for the amplitude corresponding to the graph in fig . [ ftwoglu ] . the total cross section is arrived at by adding the contributions from the other graphs according to the replacements described in the text below eq . ( [ m3 ] ) , and dividing by @xmath70 . the result reads @xmath558 } \over { \left [ z_a \ , ( 1-z_a ) \ , q^2_a + ( { { \bf p}}_a + { { \bf k}})^2 \right ] \ , \left [ z_a \ , ( 1-z_a ) \ , q^2_a + { { \bf p}}_a^2 \right ] } } \right . \nonumber \\&&\hspace*{1 cm } \left . - { { \left [ 4 z_a ( 1 - z_a ) ( { \bf{\varepsilon}}_a \cdot { { \bf p}}_a ) ^2\ ; - { { \bf p}}_a^2 \right ] } \over { \left [ z_a \ , ( 1-z_a ) \ , q^2_a + { { \bf p}}_a^2 \right]^2 } } \right\ } \nonumber \\ & ~ & \times \ , \left\ { { { \left [ 4 z_b ( 1 - z_b ) { \bf{\varepsilon}}_b \cdot { { \bf p}}_b \ ; { \bf{\varepsilon}}_b \cdot ( { { \bf p}}_b - { \bf k } ) - { { \bf p}}_b \cdot ( { { \bf p}}_b - { \bf k } ) \right ] } \over { \left [ z_b \ , ( 1-z_b ) \ , q^2_b + ( { { \bf p}}_b - { { \bf k}})^2 \right ] \ , \left [ z_b \ , ( 1-z_b ) \ , q^2_b + { { \bf p}}_b^2 \right ] } } \right . \nonumber \\&&\hspace*{1 cm } \left . - { { \left [ 4 z_b ( 1 - z_b ) ( { \bf{\varepsilon}}_b \cdot { { \bf p}}_b ) ^2\ ; - { { \bf p}}_b^2 \right ] } \over { \left [ z_b \ , ( 1-z_b ) \ , q^2_b + { { \bf p}}_b^2 \right]^2 } } \right\ } . \nonumber % \end{aligned}\ ] ] this coincides with eq . ( [ convsigma ] ) in the text once the explicit expression ( [ capitalg ] ) for @xmath77 is used . using eqs . ( [ m2])-([trplusminus ] ) , the tensor @xmath559 introduced in eq . ( [ curlyg ] ) takes the form @xmath560 } \over { \left [ ( { { \bf p } } + { { \bf k}})^2 + z \ , ( 1-z ) \ , q^2 \right ] \ , \left [ { { \bf p}}^2 + z \ , ( 1-z ) \ , q^2 \right ] } } + \ , symm . \ , \right\ } \;\;\ ; , \nonumber % \end{aligned}\ ] ] where the additive symmetric terms are obtained from the replacements given in sec . [ sec : notations ] . the gluon exchange diagrams discussed in the text provide the dominant contribution to the photon - photon cross section in the high energy limit . they give rise to constant ( in born order ) or logarithmic ( in higher orders ) terms at large @xmath28 in the cross section . this appendix is concerned with quark exchange contributions , which vanish in the large energy limit . we examine quark exchange in order to estimate the energy at which gluon exchange becomes dominant . the leading - order term of quark - exchange type comes from the purely electromagnetic process @xmath561 this contribution is suppressed by a power of @xmath28 at high energies , @xmath562 . to express the cross section for this process , we parametrize the incoming photon momenta as in eq . ( [ qaqb ] ) , and introduce the variables @xmath563 in terms of which the total energy @xmath28 has the expression @xmath564 in the high energy region one has @xmath565 , and @xmath566 as in eq . ( [ qaplusqbminus ] ) . the cross section for the process ( [ qexreac ] ) to order @xmath567 , averaged over the transverse photon polarizations , has the form @xmath568 \hspace*{0.8 cm } , % \ ] ] where @xmath569 and @xmath570 are rational functions of @xmath571 , and are plotted in fig . [ qfun ] versus the variable @xmath572 for different values of the ratio @xmath573 . the @xmath574-dependence of the cross section ( [ sqexunpol ] ) is reported in fig . [ qsig ] for the case of equal virtualities . the cross section vanishes at the kinematic threshold @xmath575 , it has a maximum around @xmath576 , then it falls off and vanishes for @xmath577 ( corresponding to high energy ) like @xmath578 , according to the asymptotic formula @xmath579 the power suppression with @xmath574 at small @xmath574 is the one expected from the exchange of a spin-@xmath416 line in the @xmath172-channel . the logarithmic enhancement is associated with the integration over the region of small angles at the splitting vertex @xmath580 in the limit of small photon virtuality . the behavior of the cross section is qualitatively the same in the case of unequal virtualities . lipatov , sov . j. nucl . * 23 * , 338 ( 1976 ) ; e.a . kuraev , l.n . lipatov and v.s . fadin , sov . jetp * 45 * , 199 ( 1977 ) ; i. balitskii and l.n . lipatov , sov . j. nucl . * 28 * , 822 ( 1978 ) . h. abramowicz , plenary talk at ichep96 ( warsaw , july 1996 ) , in proceedings of the xxviii international conference on high energy physics , eds . z. ajduk and a.k . wroblewski , world scientific , p.53 . d0 collaboration , phys . * 77 * , 595 ( 1996 ) . brodsky , talk at workshop on high energy @xmath29 colliders , brookhaven national laboratory , may 1996 . f. hautmann , talk at ichep96 ( warsaw , july 1996 ) , preprint oits 613/96 , in proceedings of the xxviii international conference on high energy physics , eds . z. ajduk and a.k . wroblewski , world scientific , p.705 . brodsky , f. hautmann and d.e . soper , phys . lett . * 78 * , 803 ( 1997 ) . p. aurenche , g.a . schuler et al . , report on @xmath581 physics " in proceedings of the workshop physics at lep2 " , eds . g. altarelli , t. sj " ostrand and f. zwirner , cern 1996 - 01 , vol.1 , p.291 . mueller , nucl . * b415 * , 373 ( 1994 ) ; a.h . mueller and b. patel , _ ibid . _ * b425 * , 471 ( 1994 ) . i. balitskii , nucl . phys . * b463 * , 99 ( 1996 ) . j. bartels , a. de roeck and h. lotter , phys . lett . b * 389 * , 742 ( 1996 ) . v.m . budnev , i.f . ginzburg , g.v . meledin and v.g . serbo , phys . rep . * 15 c * , 181 ( 1975 ) . low , phys . d * 12 * , 163 ( 1975 ) ; s. nussinov , phys . lett . * 34 * , 1286 ( 1975 ) , phys . d * 14 * , 246 ( 1976 ) ; j.f . gunion and d.e . soper , _ ibid . _ * 15 * , 2617(1977 ) . bjorken , j. kogut and d.e . soper , phys . d * 3 * , 1382 ( 1971 ) . bjorken and j. kogut , phys . d * 8 * , 1341 ( 1973 ) . frankfurt and m. strikman , phys . * 160 * , 235 ( 1988 ) . s. catani , m. ciafaloni and f. hautmann , phys . b * 242 * , 97 ( 1990 ) , nucl . * b366 * , 135 ( 1991 ) ; j.c . collins and r.k . ellis , _ ibid . _ * b360 * , 3 ( 1991 ) . lepage and p.b . mackenzie , phys . d * 48 * , 2250 ( 1993 ) ; s.j . brodsky , g.p . lepage and p.b . mackenzie , _ ibid . _ * 28 * , 228 ( 1983 ) . bjorken , preprint slac - pub-7341 , presented at snowmass 1996 summer study on new directions for high energy physics , e - print archive hep - ph/9610516 . j. bartels and h. lotter , phys . b * 309 * , 400 ( 1993 ) . mueller , nucl . * b437 * , 107 ( 1995 ) . schuler and t. sj " ostrand , zeit . c * 68 * , 607 ( 1995 ) , phys . b * 376 * , 193 ( 1996 ) ; m. gl " uck , e. reya and m. stratmann , phys . d * 51 * , 3220 ( 1995 ) . collins , _ an introduction to regge theory and high energy physics _ , cambridge university press , cambridge , 1977 . opal collaboration , contributed paper pa03 - 007 at ichep96 ( warsaw , july 1996 ) ; j.a . lauber ( opal ) , in proceedings of the xxviii international conference on high energy physics , eds . z. ajduk and a.k . wroblewski , world scientific , p.725 . s. kuhlman et al . , physics and technology of the next linear collider , snowmass 1996 report , e - print archive hep - ex/9605011 .

a basic problem in database theory is _ query evaluation in probabilistic databases _ : given a ( boolean ) query @xmath0 and a probabilistic database @xmath1 , where each tuple has a given probability , compute the probability of the lineage of @xmath0 over @xmath1 . the problem is computationally hard , even for fixed queries in simple syntactic forms @xcite . the _ lineage of a ( boolean ) query @xmath0 over a database @xmath1 _ is a monotone boolean function @xmath2 over the tuples in @xmath1 that accepts a subset @xmath3 of tuples of @xmath1 if and only if @xmath0 is true in @xmath3 . a standard approach to probabilistic query evaluation is _ query compilation _ * chapter 5 ) . here , the lineage @xmath2 , given as a boolean circuit , which is computable in polynomial time if @xmath0 is fixed , is implemented within a succinct circuit class where its probability is efficiently computable . in other words , to avoid computing the probability of a circuit , which is hard , the circuit is first _ compiled _ to a tamer form . circuit classes supporting tractable probability computation , and model counting in particular , are central in _ knowledge compilation _ @xcite ; particular emphasis is posed on the hierarchy of deterministic decomposable circuits . a circuit is _ decomposable _ if its and gates represent independent probabilistic events @xcite , and _ deterministic _ if its or gates represent exclusive probabilistic events @xcite : probability computation is then feasible in linear time on deterministic decomposable circuits . aimed at a detailed syntactic classification of tractable cases of probabilistic query evaluation , jha and suciu amply explored the compilability of various classes of queries into various classes of deterministic decomposable circuits , fruitfully bridging database theory and knowledge compilation @xcite . in this context , they studied the compilability of queries whose lineages have small circuit treewidth into decision diagrams , obdds in particular @xcite ; a study we continue in this article . an _ ordered binary decision diagram _ ( obdd ) is a deterministic read - once branching program where every path from the root to a leaf visits the boolean variables in the same order @xcite . the obdd size of a boolean function is the size ( number of nodes ) of its smallest obdd implementation . the width of an obdd is the largest number of nodes labeled by the same variable , and the obdd width of a boolean function is the smallest width attained by its obdd implementations . unifying several known tractable cases of the probability computation problem , jha and suciu introduce a structural parameter for boolean functions , called _ expression width _ @xcite or _ circuit treewidth _ @xcite , that measures , for any boolean function , the smallest treewidth of a circuit computing the function . they show that a boolean function of @xmath4 variables and circuit treewidth @xmath5 has obdd size @xmath6 where @xmath7 is a fast growing ( double exponential ) function @xcite . the bound is tight in the sense that there are boolean functions of @xmath4 variables and circuit treewidth @xmath5 whose obdd size is @xmath8 @xcite . the bound ( [ eq : intr-1 ] ) gives polynomial size obdd implementations for circuits of bounded treewidth , but the degree of the polynomial depends ( badly ) on the treewidth . however , as jha and suciu show @xcite , restricting to functions of small _ circuit pathwidth _ resolves this issue . indeed , a boolean function of @xmath4 variables and circuit pathwidth @xmath5 has obdd width @xmath9 , hence obdd size @xmath10 and conversely , every boolean function of obdd width @xmath5 has circuit pathwidth @xmath11 . therefore , _ a class of boolean functions has bounded circuit pathwidth if and only if it has bounded obdd width . _ as jha and suciu conclude , the quest naturally arises for a similar characterization of bounded circuit treewidth . the quest involves , for starters , identifying a circuit class ideally as tractable as obdds but more succinct , and therefore capable of matching the bound ( [ eq : intr-2 ] ) . natural candidates , like fbdds or even nondeterministic read - once branching programs fail @xcite . the question is natural and nontrivial . compared to the substantial understanding of the compilability of cnf circuits parameterized by treewidth , or even cliquewidth @xcite , the parameterized compilability of general circuits is relatively unexplored and poorly understood ; which is unsatisfactory because , in theory , the circuit treewidth of a class of boolean functions can be bounded on general circuits and unbounded on cnfs ( * ? ? ? * example 2.9 ) ; and in practice , query lineages are often presented by circuits rather than by cnfs @xcite . the only bound on the size of a compilation for a circuit that avoids a dependence on its treewidth in the exponent , like in ( [ eq : intr-2 ] ) as opposed to ( [ eq : intr-1 ] ) , is a compilation of _ size _ @xmath12 circuits into _ decomposable _ forms of size @xmath13 by petke and razgon @xcite , where @xmath14 is an exponential function . this compilation , however , lacks two features that are either needed by or desirable in its intended application to query compilation . the crucial missing feature is that decomposable circuits , in the absence of determinism , do not support model counting ( nor , then , probability computation ) . besides , in the upper bound ( [ eq : intr-3 ] ) , the size of the compilation depends on the size of the circuit , @xmath12 , not just on the number of its variables , @xmath4 , and the former can be much larger than the latter . indeed , petke and razgon ask whether decomposable forms of size linear in @xmath4 are attainable for circuits of bounded treewidth ( * ? ? ? * section 5 ) . in the first part of the article ( section [ sect : part1 ] ) , we show that _ a class of boolean functions has bounded circuit treewidth if and only if it has bounded sdd width _ , which perfectly complements the aforementioned characterization of circuit pathwidth via obdd width by jha and suciu . more precisely , we prove the following ( theorem [ th : linear - sdd - size ] and surrounding discussion ) . result 1 . : : _ a boolean circuit of @xmath4 variables and treewidth @xmath5 has sdd width @xmath9 , thus sdd size @xmath15 where @xmath7 is a triple exponential function . conversely , every boolean function of sdd width @xmath5 has circuit treewidth @xmath11 . _ introduced by darwiche @xcite , _ sentential decision diagrams _ ( sdds ) are a relaxation of obdds based on a generalized form of shannon decomposition . an obdd respecting the variable ordering @xmath16 takes a _ binary decision _ of the form @xmath17 , where @xmath18 and @xmath19 are obdds respecting the variable ordering @xmath20 . intuitively , based on a binary case distinction on @xmath21 , the obdd executes subobdds respecting the subordering @xmath20 . an sdd respecting the variable tree @xmath22 , whose left and right subtrees split the variables in two disjoint blocks @xmath23 and @xmath24 , takes a _ sentential decision _ of the form @xmath25 . here , based on an @xmath12-ary ( exhaustive and disjoint ) case distinction on @xmath23 , implemented by the @xmath12 sdds @xmath26 respecting the left subtree of @xmath22 , the @xmath12 sdds @xmath27 respecting the right subtree of @xmath22 are executed . sdds are theoretically very robust , being essentially as tractable as obdds @xcite but exponentially more succinct @xcite . they are also appealing in practice since an sdd compiler is reasonable to design and implement ( as opposed to an fbdd compiler for instance , whose design is already fairly elusive ) . indeed , available sdds compilers already yield more succinct sdds than obdds , leveraging the additional flexibility offered by variable trees compared to variable orders @xcite . moreover , sdds have canonical forms , and hence carry a natural notion of width which in particular implies , if bounded , linear size implementations , exactly as obdd width does for obdds @xcite . thus , quite remarkably , our study unveils that circuit treewidth is characterized by an independently introduced , theoretically solid , and practically useful notion of circuit width , namely , sdd width . ( 0,0 ) # 1#2#3#4#5 @font ( 3284,2549)(6189,-10343 ) ( 7164,-9063)(0,0)[lb ] ( 7793,-8476)(0,0)[lb ] ( 7111,-10231)(0,0)[lb ] ( 7059,-9650)(0,0)[lb ] figure [ fig : bf ] illustrates the compilability panorama for boolean functions relative to bounded circuit pathwidth / obdd width , bounded circuit treewidth / sdd width , and polynomial obdd and sdd size . the class @xmath28 contains all boolean functions of obdd width @xmath29 , and similarly for sdds ; the class @xmath30 contains all boolean functions of circuit treewidth @xmath29 , and similarly for cpw and circuit pathwidth . we have @xmath31 jha and suciu leave open the question whether the circuit treewidth of a boolean function is computable ( * ? ? ? * section 6 ) . using the fact that satisfiability of mso - sentences is decidable on graphs of bounded treewidth @xcite , we answer the question positively ( proposition [ prop : seese ] ) . result 2 . : : _ the circuit treewidth of a boolean function is computable . _ ( 0,0 ) # 1#2#3#4#5 @font ( 3284,2549)(6189,-10343 ) ( 7762,-8479)(0,0)[lb ] ( 7059,-9650)(0,0)[lb ] ( 7111,-10231)(0,0)[lb ] ( 7164,-9063)(0,0)[lb ] in the second part of the article ( section [ sect : part2 ] ) , we study the implications of our compilability results in query compilation , refining the picture drawn by jha and suciu for unions of conjunctive queries ( ucqs ) with and without inequalities @xcite . we prove the following statement ( theorem [ th : main - ineq ] ) . result 3 . : : _ a union of conjunctive queries with or without inequalities containing inversions has lineages of exponential deterministic structured size . _ introduced by dalvi and suciu @xcite , _ inversion freeness _ is a syntactic property of ucqs and ucqs with inequalities that implies compilability of their lineages in constant width ( linear size ) obdds and polynomial size obdds @xcite . on the other hand , if a query contains inversions , then it has lineages with large obdd @xcite , and even sdd @xcite , implementations . _ structuredness _ is a strong form of decomposability where not only for every and gate the circuits leading into the gate are defined on disjoint sets of variables @xcite , but their variables are partitioned accordingly to an underlying variable tree @xcite . as alluded in their informal description above , sdds as well as obdds are special deterministic structured forms . therefore result 3 formally generalizes analogous previous incompilability results for sdds and obdds @xcite . the proof has the main and sole merit to combine proof ideas of jha and suciu together with lower bound techniques for deterministic structured circuits based on single partition communication complexity @xcite . a careful inspection of the proof shows that result 3 also exponentially separates disjunctive normal forms ( dnfs ) , and even prime implicant forms ( ips ) , from structured deterministic negation normal forms ( nnfs ) . in this interpretation result 3 settles a special case of the much harder problem of separating dnfs ( and ips ) and deterministic decomposable nnfs ( d - dnnfs ) ; which , thanks to the recently established separation of decomposable nnfs ( dnnfs ) and d - dnnfs @xcite , is the last open question about the relative succinctness of the compilation languages considered in the classic article by darwiche and marquis @xcite . figure [ fig : ucq ] and figure [ fig : ucqneq ] give an overview of query compilability for lineages of ucqs with and without inequalities . for lineages of ucqs we have that @xmath32 because by result 3 inversions imply large structured deterministic forms , hence large sdds ; on the other hand , inversion freeness implies constant width obdds @xcite , so that @xmath33 . the picture for lineages of ucqs with inequalities is @xmath34 because again result 3 implies large sdds in the presence of inversions , and inversion freeness implies polynomial size obdds @xcite ; thus @xmath35 . in a symmetric fashion , jha and suciu conjecture that , for lineages of ucqs with inequalities , it also holds that @xmath36 . ( 0,0 ) # 1#2#3#4#5 @font ( 3284,2549)(6189,-10343 ) ( 7762,-8479)(0,0)[lb ] ( 7164,-9063)(0,0)[lb ] ( 7111,-10231)(0,0)[lb ] ( 7059,-9650)(0,0)[lb ] our bound ( [ eq : int-4 ] ) amounts to a vast improvement of the available bounds . compared to ( [ eq : intr-1 ] ) , it attains linear size versus ( large degree ) polynomial size compilation on bounded circuit treewidth classes . compared to ( [ eq : intr-3 ] ) , it answers abundantly the quest for linear size decomposable forms for boolean functions of bounded treewidth : our forms are not just linear size and decomposable , but even deterministic and structured . pushing the dependency of the compilation size down from @xmath12 ( the size of the given circuit , as in ( [ eq : intr-3 ] ) ) to @xmath4 ( the number of its inputs , as in ( [ eq : int-4 ] ) ) required an entirely new compilation idea . of boolean variables of the query lineage @xmath2 is the ( large ) number of tuples in the database @xmath1 , and @xmath12 is the size of the circuit implementation of @xmath2 . roughly @xmath37 , where @xmath38 is the size of the query @xmath0 . hence , avoiding a dependence on @xmath12 means obtaining a bound where the degree of the polynomial is a universal constant , not just independent of the circuit treewidth of @xmath2 , but also independent of @xmath0 . ] the idea used by petke and razgon @xcite to obtain ( [ eq : intr-3 ] ) was the following . given a circuit @xmath39 of @xmath40 variables and @xmath41 gates , first compute its tseitin cnf @xmath42 ; the circuit treewidth of the latter is ( linearly ) related to the former . to obtain a decomposable form for @xmath39 , existentially quantify the ( gate ) variables @xmath43 in a decomposable form @xmath44 for @xmath42 : @xmath45 this indirect approach via tseitin forms , however , introduces two critical issues . on the one hand , the size of @xmath44 depends on @xmath46 , so that the size of the compiled form will depend on @xmath12 ( as opposed to depending on @xmath4 only ) . on the other hand , for @xmath47 to have size polynomial in that of @xmath44 , the latter can not be deterministic @xcite ; hence the resulting compiled form will not be deterministic either . our compilation approach avoids tseitin forms and compiles the circuit directly ; it relies on a new insight on boolean functions , and more specifically on the combinatorics of their _ subfunctions _ or _ cofactors _ @xcite . dually to the notion of cofactor , we introduce ( definition [ def : factor ] ) the notion of _ factor _ of a boolean function @xmath48 , that is , a function @xmath49 whose models correspond exactly to the assignments of @xmath24 generating some cofactor of @xmath50 . we then show ( lemma [ lemma : factors ] ) that first , the rectangle @xmath51 formed by multiplying any two factors @xmath49 and @xmath52 of @xmath48 is either disjoint from @xmath50 , that is @xmath53 , or contained in @xmath50 , that is @xmath54 . and @xmath52 is a boolean function over @xmath55 whose models are exactly those assignments of @xmath55 whose restriction to @xmath24 ( @xmath56 ) models @xmath57 ( @xmath58 ) . ] and second ( lemma [ lemma : drc ] ) , the pairs of factors @xmath49 and @xmath52 of @xmath48 satisfying the latter condition , call them implicants , form a disjoint rectangle cover of @xmath50 , that is , @xmath59 where the disjunction is deterministic and the conjunctions are decomposable ( structured , indeed ) . we then elaborate on the main technical lemma of jha and suciu ( * ? ? ? * lemma 2.12 ) to turn the above structural insight into a compilation of small size . we show that a circuit of small treewidth computing a function @xmath50 naturally delivers a variable tree where the number of cofactors of @xmath50 generated by assigning the variables below every node in the tree is small ( lemma [ lemma : jhasuciu ] ) . it follows that the disjoint rectangle covers described above are small for every factor of @xmath50 ; we then obtain the desired compilation by an inductive construction up the variable tree ( lemma [ lemma : dsdnnf - sub ] and theorem [ th : factorized - forms - main ] ) . indeed result 1 is proved for a more basic canonical deterministic structured class of circuits ( theorem [ th : factorized - forms - main ] and proposition [ prop : fiw - ctw ] ) , which is equivalent to sdds as far as the boundedness of their widths , and reduces to obdds in the special case of linear variable trees ; the class is of independent interest and gives a fresh structural insight into sdds ( see also the conclusion ) . our construction , significantly shorter to describe and easier to analyze than its precursors , effectively encompasses the construction by jha and suciu in that , if carried out in the special case of circuit pathwidth , it compiles a circuit of @xmath4 variables and pathwidth @xmath5 into an obdd ( not just an sdd ) of width @xmath9 and size @xmath60 . the article is organized as follows . the required notions from knowledge compilation and communication complexity are given in section [ sect : prel ] . the part of the article devoted to circuit treewidth and sentential decision ( section [ sect : part1 ] ) deals first with the introduction and the development of the notion of factor and factor width for a boolean function , and the relation of the latter with circuit treewidth ( section [ sect : fwbf ] ) ; next , it presents the actual compilations in canonical deterministic structured nnfs and sdds ( section [ sect : impl - sdnnf ] ) . section [ sect : part2 ] is devoted to query compilation . we present questions and directions for future research in section [ sect : concl ] . for every integer @xmath61 , we let @xmath62=\{1,\ldots , n\}$ ] . we refer the reader to a standard source for the notions of treewidth , tree decomposition , and nice tree decomposition @xcite . we consider ( boolean ) circuits over the standard basis , namely dags whose non - source nodes , called internal gates , are unbounded fanin conjunction ( @xmath63 ) and disjunction ( @xmath64 ) gates and fanin @xmath65 negation ( @xmath66 ) gates , and whose source nodes , called input gates , are pairwise distinct variables or constants ( @xmath67 and @xmath68 ) . a designated sink node is called the output gate . a circuit is in _ negation normal form _ , in short an _ nnf _ , if its negation gates are only wired by input gates . the size @xmath69 of a circuit @xmath70 is the number of its gates . let @xmath23 be a finite set of variables . a circuit @xmath70 over @xmath23 is a circuit whose input gates are labelled by variables in @xmath23 or by constants . a circuit @xmath70 on @xmath23 computes a boolean function @xmath71 over the boolean variables @xmath23 , @xmath72 in the usual way . we let @xmath73 denote the models of @xmath70 and @xmath74 . two circuits @xmath70 and @xmath75 over @xmath23 are equivalent , in symbols @xmath76 , if @xmath77 ; we also write @xmath78 or say that @xmath70 computes @xmath50 if @xmath79 . for a gate @xmath14 in a circuit @xmath70 over @xmath23 , we let @xmath80 denote the subcircuit of @xmath70 rooted at @xmath14 . in particular , @xmath81 if @xmath14 is the output gate of @xmath70 . for a circuit @xmath70 over variables @xmath23 and a gate @xmath82 , we let @xmath83 denote the variables appearing at input gates of @xmath80 . let @xmath14 be an @xmath64-gate in a circuit @xmath70 , and let @xmath84 and @xmath85 be two distinct gates wiring @xmath14 in @xmath70 . then @xmath14 is called _ deterministic _ if @xmath86 , viewing each circuit involved in the equation as a circuit over @xmath87 . the determinism of @xmath14 in @xmath70 implies that the two subcircuits @xmath88 and @xmath89 are independent in the sense that @xmath90 where each circuit involved in the equation is viewed as a circuit over @xmath87 . a circuit where all @xmath64-gates are deterministic is called _ let @xmath24 be a finite nonempty set of variables . a _ variable tree _ ( in short , a _ vtree _ ) for the variable set @xmath24 is a rooted , ordered , binary tree @xmath22 whose leaves correspond bijectively to @xmath24 ; for simplicity , we identify each leaf in @xmath22 with the variable in @xmath24 it corresponds to . for technical convenience , we slightly relax the standard definition not requiring for a vtree to be a full binary tree . for every internal node @xmath91 of the vtree @xmath22 with two children , we let @xmath92 and @xmath93 denote resp . the left and right child of @xmath91 . moreover , we denote by @xmath94 the subtree of @xmath22 rooted at node @xmath91 , and by @xmath95 ( the variables corresponding to ) the leaves of @xmath96 . let @xmath70 be a circuit over the variable set @xmath23 , and let @xmath22 be a vtree for the variable set @xmath24 . let @xmath14 be a fanin @xmath97 @xmath63-gate in @xmath70 , having wires from gates @xmath84 and @xmath85 , and let @xmath98 have two children @xmath92 and @xmath93 . we say that @xmath14 is _ structured by @xmath91 _ if @xmath99 and @xmath100 . we say that @xmath70 _ structured by @xmath22 _ if each @xmath63-gate in @xmath70 ( has fanin @xmath97 and ) is structured by some node in @xmath22 . a circuit is called _ structured _ if it is structured by some vtree . a class of structured nnfs is _ canonical _ if , for every boolean function @xmath101 and vtree @xmath102 with @xmath103 , if two circuits @xmath70 and @xmath75 in the class both compute @xmath50 and are structured by @xmath22 , then they are syntactically equal ( not just semantically equivalent ) . note that if a gate @xmath82 is structured , then it is also _ decomposable _ , i.e. , every two distinct gates @xmath84 and @xmath85 wiring @xmath14 satisfy @xmath104 . the decomposability of @xmath14 in @xmath70 implies that the two subcircuits @xmath88 and @xmath89 are independent in the sense that @xmath105 when viewing each circuit involved in the equation as a circuit over its own variables . a _ sentential decision diagram _ , in short _ sdd _ , is a deterministic structured nnf @xmath70 over @xmath23 of the form @xmath106 } ( p_i \wedge s_i ) \text{,}\ ] ] structured by a vtree @xmath102 , @xmath103 , such that the following holds . there exists a node @xmath98 with two children @xmath107 and @xmath108 structuring each @xmath63-gate appearing in ( [ eq : sdd - def ] ) , the @xmath109 s are sdds over @xmath110 structured by @xmath111 , the @xmath112 s are sdds over @xmath113 structured by @xmath114 , and moreover : * @xmath115}p_i \equiv \top$ ] ; * @xmath116 for all @xmath117 in @xmath118 $ ] . constants ( @xmath67 and @xmath68 ) are sdds ( over any variable set ) structured by any vtree , and a literal ( @xmath119 or @xmath120 ) is an sdd ( over any variable set containing @xmath119 ) structured by any vtree containing @xmath119 . sdds become canonical forms if , in addition to the above , the following holds @xcite : * @xmath121 for all @xmath117 in @xmath118 $ ] . let @xmath23 be a finite set of variables . a partition of @xmath23 is a sequence of pairwise disjoint subsets ( blocks ) of @xmath23 whose union is @xmath23 . let @xmath122 be a partition of @xmath23 . for @xmath123 and @xmath124 , we let @xmath125 denote the assignment of @xmath23 whose restriction to @xmath126 equals @xmath127 for @xmath128 . also , for @xmath129 and @xmath130 , we let @xmath131 . a _ ( combinatorial ) rectangle _ over @xmath23 is a boolean function @xmath132 over the boolean variables @xmath23 such that there exist a partition @xmath122 of @xmath23 and boolean functions @xmath133 for @xmath128 such that @xmath134 . we also call a subset @xmath135 of @xmath136 a rectangle over @xmath23 , with underlying partition @xmath122 , if there exists a rectangle @xmath137 , with underlying partition @xmath122 , such that @xmath138 . let @xmath139 be a boolean function over the boolean variables @xmath23 . a finite set @xmath140\}$ ] of rectangles over @xmath23 is called a _ rectangle cover _ of @xmath50 if @xmath141 } \mathsf{sat}(r_i ) \text{;}\ ] ] the rectangle cover is called _ disjoint _ if the union in ( [ eq : rect - cover ] ) is disjoint . disjoint rectangle covers and deterministic structured nnfs are tightly related . @xcite[th : struct ] let @xmath70 be a ( deterministic ) structured nnf computing a function @xmath139 and respecting a vtree @xmath22 for @xmath23 . for every node @xmath98 , @xmath50 has a ( disjoint ) rectangle cover of size at most @xmath69 where each rectangle has underlying partition @xmath142 . let @xmath101 be a boolean function , and let @xmath122 be a partition of @xmath23 where @xmath143 . the _ communication matrix _ of @xmath50 relative to @xmath122 , denoted by @xmath144 is a boolean matrix whose rows and columns are indexed by boolean assignments of @xmath145 and @xmath146 , resp . , and whose @xmath147th entry equals @xmath148 . we regard communication matrices as matrices over the reals . a basic fact in communication complexity is that the rank of the communication matrix is a lower bound on the size of disjoint rectangle covers of a function . * section 4.1)[th : rklb ] let @xmath149 be a partition of the variables of a function @xmath50 , where @xmath143 . every disjoint rectangle cover of @xmath50 into rectangles with underlying partition @xmath149 contains at least @xmath150 rectangles . a typical application of the above statement is the _ disjointness function _ , @xmath151 where @xmath152 and @xmath153 . it is folklore that @xmath154 i.e. , the communication matrix of @xmath155 relative to @xmath156 has full rank ( * ? ? ? * exercise 7.1 ) . thus every disjoint rectangle cover of @xmath155 into rectangles with underlying partition @xmath157 has at least @xmath158 rectangles . in this section , we introduce the notion of factor width of a boolean function , relate it with its circuit treewidth , and show that , when parameterized by its factor width , a boolean function admits a linear size compilation into some natural classes of canonical deterministic structured nnfs ( including sdds ) . we introduce the notion of _ factor width _ , and recall from the literature the notion of circuit treewidth @xcite . let @xmath159 be a boolean function over a finite set of variables @xmath23 . for a set of variables @xmath24 we use the notation @xmath160 to display a partition of the variables of @xmath50 into the two blocks @xmath161 and @xmath162 . it is intended that if @xmath163 and @xmath164 , then @xmath165 . the _ cofactor _ ( or _ subfunction _ ) of @xmath50 induced by @xmath163 is the boolean function @xmath166 such that @xmath167 for all @xmath164 . a function @xmath168 is called a _ cofactor of @xmath101 relative to @xmath162 _ if it is equal to the cofactor of @xmath50 induced by some @xmath163 . [ ex : cofactors ] let @xmath169 be boolean implication , i.e. @xmath170 iff @xmath171 , for all @xmath172 . the cofactors of @xmath50 relative to @xmath173 , induced by boolean assignments of @xmath119 , are @xmath174 and @xmath175 . the cofactors of @xmath50 relative to @xmath119 , induced by boolean assignments of @xmath173 , are @xmath176 and @xmath177 . the cofactors of @xmath50 induced by boolean assignments of both @xmath119 and @xmath173 are @xmath178 and @xmath179 . the only cofactor of @xmath50 induced by boolean assignments of no variables is @xmath180 itself . we introduce notation to denote the cofactor of a boolean function @xmath101 generated by replacing , in a partition of @xmath23 , the variables in certain blocks by constants . let @xmath181 be a partition of @xmath23 , let @xmath182 $ ] , and let @xmath183 for all @xmath184 . we write @xmath185 where @xmath186 if @xmath184 and @xmath187 otherwise , for all @xmath188 $ ] , to denote the cofactor of @xmath50 induced by @xmath189 , i.e. the boolean function over the variables @xmath190\setminus l } y_i$ ] defined , for every @xmath190\setminus l } ( b_i \colon y_i \to \{0,1\})$ ] , by @xmath191\setminus l } b_i \mapsto f(b_1,\ldots , b_l)\text{.}\ ] ] [ ex : notation ] let @xmath50 be as in example [ ex : cofactors ] , so that @xmath139 for @xmath192 . let @xmath193 . we write @xmath194 . let @xmath195 . we write @xmath196 to denote the cofactor of @xmath50 relative to @xmath162 induced by @xmath197 . if @xmath198 , then @xmath199 . let @xmath200 . intuitively , a factor of @xmath201 is a function @xmath49 whose models correspond exactly to the assignments of @xmath24 that induce some fixed cofactor @xmath202 of @xmath50 . [ def : factor]let @xmath24 and @xmath23 be finite sets of variables and let @xmath203 be a boolean function . a boolean function @xmath204 is called a _ factor of @xmath101 relative to @xmath24 _ if there exists a cofactor @xmath205 of @xmath50 such that @xmath206 the factors of @xmath50 relative to @xmath24 are denoted by @xmath207 . [ ex : factors ] let @xmath169 be boolean implication as in example [ ex : cofactors ] . the function @xmath208 is a factor of @xmath50 relative to @xmath119 , because there exists a cofactor of @xmath50 relative to @xmath173 , namely @xmath209 , such that @xmath210 iff @xmath211 . the function @xmath212 is a factor of @xmath50 relative to @xmath119 , because there exists a cofactor of @xmath50 relative to @xmath173 , namely @xmath213 , such that @xmath210 iff @xmath211 . note that cofactors and factors of a boolean function relative to a variable set are , in general , distinct . [ ex : fneqcof ] let @xmath169 be boolean implication as in example [ ex : cofactors ] . then @xmath208 is a factor of @xmath50 relative to @xmath119 ( example [ ex : factors ] ) , but it is not a cofactor of @xmath50 relative to @xmath119 , since the only cofactors of @xmath50 relative to @xmath119 are equivalent to @xmath120 and @xmath214 ( example [ ex : cofactors ] ) . note that , by definition [ def : factor ] , @xmath215 but we insist on @xmath24 being an arbitrary set of variables for technical convenience . moreover , again by definition [ def : factor ] , @xmath216 and the union is disjoint . in words , @xmath207 naturally determines a partition of @xmath217 whose blocks , of the form @xmath218 for @xmath219 , correspond to the cofactors of @xmath50 relative to @xmath220 . finally , we introduce the notion of factor width of a boolean function . let @xmath139 be a boolean function and let @xmath22 be a vtree for @xmath221 . the _ factor width of @xmath50 relative to @xmath22 _ , in symbols @xmath222 , is defined by @xmath223 the _ factor width _ of @xmath50 is defined by @xmath224 the _ treewidth of a circuit _ @xmath70 , in symbols @xmath225 , is the treewidth of the undirected graph underlying ( the directed acyclic graph underlying ) @xmath70 . the _ circuit treewidth _ @xmath226 of a boolean function @xmath50 is the minimum treewidth of a circuit computing @xmath50 . a crucial fact in our development is that the factor width of a boolean function is bounded above by a function of its circuit treewidth . the proof is a revisitation of ( * ? ? ? * lemma 2.12 ) . [ lemma : jhasuciu ] for all boolean functions @xmath50 , @xmath227 let @xmath70 be a treewidth @xmath228 circuit computing the boolean function @xmath101 . let @xmath135 be a _ tree decomposition of the gates of @xmath70 , witnessing treewidth @xmath228 ; without loss of generality , the root of @xmath135 is the empty bag , therefore each input gate of @xmath70 ( i.e. , each variable in @xmath23 ) is forgotten exactly once in @xmath135 . we associate to @xmath135 a vtree @xmath22 for @xmath23 as follows . let @xmath229 be a set of fresh variables in a bijective correspondence with the leaves of @xmath135 . label the leaves of @xmath135 by pairwise distinct ( dummy ) variables in @xmath229 . for every variable @xmath230 , append a fresh leaf labelled @xmath119 to the node forgetting @xmath119 in @xmath135 . the resulting tree @xmath22 is a vtree for @xmath231 . for every @xmath98 , let @xmath232 denote the variables in @xmath23 appearing in @xmath96 ( or equivalently , the variables in @xmath96 that are not dummy variables ) . in light of ( [ eq : dummyvars ] ) , to show that @xmath233 it is sufficient to prove that @xmath234 matches the bound for all @xmath98 . if @xmath91 is a leaf in @xmath22 , then @xmath235 for some @xmath230 and @xmath236 , or @xmath91 is labelled by some dummy variable in @xmath229 , so that @xmath237 and @xmath238 . otherwise @xmath91 is a bag @xmath239 in @xmath135 and in this case @xmath232 contains the variables in @xmath23 forgotten by nodes in the subtree of @xmath135 rooted at @xmath91 . for a gate @xmath14 in @xmath70 , let @xmath240 be the gates in @xmath239 with a directed path to @xmath14 in ( the dag underlying ) @xmath70 whose intermediate gates are not in @xmath239 . namely , @xmath241 iff @xmath242 and there exists a directed path @xmath243 in the dag underlying @xmath70 such that @xmath244 . let @xmath14 be a gate in @xmath70 . we freely identify @xmath80 , the subcircuit of @xmath70 rooted at @xmath14 , with the boolean function on @xmath23 it computes . [ cl : main - cl ] for every @xmath82 , it holds that @xmath245 be a set of fresh variables . let @xmath246 denote the circuit obtained from @xmath80 by transforming each gate @xmath241 into an input gate labelled by the variable @xmath247 . we distinguish two cases depending on whether or not @xmath14 is in @xmath239 . if @xmath248 , then observe that @xmath249 , or @xmath250 ; otherwise , if @xmath251 uses variables in both @xmath232 and @xmath252 then , in the graph underlying @xmath70 , there exists a path from @xmath232 to @xmath252 not intersecting @xmath239 , contradicting the properties of @xmath135 . if @xmath249 then , for every @xmath253 , @xmath254 is a function of @xmath43 among at most @xmath255 possibilities . otherwise , if @xmath250 , then every @xmath253 yields the same function @xmath256 of @xmath257 . since each variable @xmath247 in @xmath251 represents the subcircuit @xmath258 of @xmath70 for @xmath241 , it follows that @xmath259 if @xmath260 , then each proper subcircuit of @xmath251 uses only variables in @xmath261 or only variables in @xmath257 ; otherwise we obtain a contradiction as above . let @xmath262 and @xmath263 be a bipartition of the immediate subcircuits of @xmath251 such that the @xmath264 s only use variables in @xmath261 and the @xmath265 s only use variables in @xmath257 . if @xmath14 is an @xmath63-gate , then @xmath266 for every @xmath253 , @xmath267 is a function @xmath268 over @xmath43 ; on the other hand , every @xmath253 yields the same function @xmath269 over @xmath257 , namely @xmath270 . therefore , for all @xmath253 , @xmath254 is a function over @xmath257 of the form @xmath271 where @xmath268 is a function over @xmath43 , so that @xmath272 recalling that each variable @xmath247 in @xmath251 represents the subcircuit @xmath258 of @xmath70 for @xmath241 , we have @xmath259 the cases where @xmath14 is a @xmath64-gate or a @xmath66-gate are similar , and the claim is proved . for all @xmath82 , let @xmath273 . by induction on @xmath274 we prove that for all @xmath260 it holds that latexmath:[\[\label{eq : ind } for the base case , let @xmath260 be such that @xmath276 . then @xmath277 where the first inequality holds by ( [ eq : rec ] ) as @xmath278 in this case . for the inductive case , let @xmath260 be such that @xmath279 . then , resp . by ( [ eq : rec ] ) and the induction hypothesis as @xmath280 for all @xmath241 , @xmath281 but @xmath282 where @xmath283 justifies the last inequality . we now conclude the proof . let @xmath14 be the output gate of @xmath70 , i.e. , @xmath284 . then , justifying the first and second inequalities resp . by ( [ eq : rec ] ) and by ( [ eq : ind ] ) , @xmath240 and @xmath285 , @xmath286 and we are done . we conclude the section observing that circuit treewidth is computable , thus answering a question posed by jha and suciu ( * ? ? ? * section 6 ) . [ prop : seese ] the circuit treewidth of a boolean function is computable . let @xmath287 be a variable set . say that a graph ( with loops ) @xmath57 _ implements _ a circuit over @xmath23 iff there exists a circuit @xmath70 over @xmath23 such that @xmath57 is ( isomorphic to ) the graph whose edge set @xmath288 is defined relative to @xmath70 as follows : * for every wire @xmath289 in @xmath70 , add to @xmath288 a path @xmath290 , @xmath291 , @xmath292 from @xmath14 to @xmath293 with a loop on @xmath85 , where @xmath84 and @xmath85 are fresh vertices . * add to @xmath288 a loop on the output gate @xmath14 of @xmath70 . * let @xmath294 for every @xmath295-gate @xmath14 in @xmath70 , add to @xmath288 a @xmath296-star @xmath297 , @xmath298 , @xmath299 centered at @xmath14 , where @xmath300 are fresh vertices . intuitively a graph ( with loops ) @xmath57 implementing a circuit @xmath70 is a faithful representation of @xmath70 in the vocabulary of graphs , where the arcs and labels used to represent @xmath70 in the vocabulary of circuits are suitably expressed by edges ( and loops ) . let @xmath50 be a boolean function over @xmath23 . it is a tedious but straightforward exercise to write an mso sentence @xmath301 that is true on a graph ( with loops ) @xmath57 iff @xmath57 implements a circuit @xmath70 on @xmath23 computing @xmath101 ( * ? ? ? * examples 4.10 , 4.13 , and 4.18 ) . also note that , if the graph ( with loops ) @xmath57 implements a circuit @xmath70 , then the treewidth of @xmath57 is equal to the treewidth of @xmath70 . now , let @xmath5 be an upper bound on the circuit treewidth of @xmath50 ( for instance , the treewidth of the dnf whose terms are exactly the models of @xmath50 ) . seese proves that , given an mso sentence on the vocabulary of graphs , it is decidable whether it is satisfied by a graph ( with loops ) of treewidth @xmath5 @xcite . we therefore cycle for @xmath302 until we find @xmath303 such that @xmath301 is modeled by a graph of treewidth @xmath304 . we show that boolean functions of small factor width have implementations of small width within natural canonical subclasses of deterministic structured nnfs ( including canonical sdds ) ; conversely , small width implementations within such circuit classes imply small circuit treewidth . thus a class of boolean functions has bounded circuit treewidth iff it has bounded width implementations in some natural , canonical classes of deterministic structured nnfs , including sdds . we introduce the notion of _ factorized implicant _ of a boolean function @xmath48 , roughly @xmath305 where @xmath24 and @xmath56 are disjoint sets of variables , @xmath306 ( resp . , @xmath307 ) is a factor of @xmath50 relative to @xmath24 ( resp . , @xmath56 ) . the first key insight is that the rectangle formed by multiplying any two factors @xmath308 and @xmath309 of @xmath48 is either disjoint from @xmath50 or contained in @xmath50 . [ lemma : factors ] let @xmath139 be a boolean function . let @xmath24 and @xmath56 be disjoint subsets of @xmath23 . let @xmath269 , @xmath57 , and @xmath58 be factors of @xmath50 relative to @xmath55 , @xmath24 , and @xmath56 , respectively . then either @xmath310 or @xmath311 if @xmath312 , then let @xmath313 and @xmath314 be such that @xmath315 . let @xmath316 and @xmath317 . it suffices to show that @xmath318 . below , @xmath319 . assume for a contradiction that @xmath320 . therefore , by definition , the cofactors of @xmath50 induced by @xmath321 and @xmath322 are distinct , i.e. , @xmath323 on the other hand , by definition , @xmath324 implies that @xmath325 , and similarly @xmath326 implies that @xmath327 . in particular , @xmath328 and @xmath329 . then @xmath330 a contradiction . intuitively , a factorized implicant of a function @xmath48 is a pair of factors @xmath308 and @xmath309 entirely contained in @xmath50 , as in ( [ eq : rect - cont ] ) . formally , let @xmath139 be a boolean function . let @xmath269 , @xmath57 , and @xmath58 be factors of @xmath50 relative to @xmath55 , @xmath24 , and @xmath56 , resp . , where @xmath24 and @xmath56 are disjoint subsets of @xmath23 . then @xmath331 is a _ factorized implicant of @xmath269 relative to @xmath332 _ if @xmath57 and @xmath58 satisfy ( [ eq : rect - cont ] ) relative to @xmath269 . we denote by @xmath333 the set of factorized implicants of @xmath269 relative to @xmath332 . the second key insight is that the factorized implicants of @xmath48 naturally induce a disjoint rectangle cover for @xmath48 . [ lemma : drc ] let @xmath139 be a boolean function and let @xmath269 be a factor of @xmath50 relative to @xmath55 , where @xmath24 and @xmath56 are disjoint subsets of @xmath23 . then @xmath334 forms a disjoint rectangle cover of @xmath269 , i.e. , @xmath335 and the union is disjoint . a circuit interpretation of lemma [ lemma : drc ] is that @xmath336 where @xmath337 , @xmath338 , and @xmath339 are circuits using variables in @xmath55 , @xmath24 , and @xmath56 computing @xmath269 , @xmath57 , and @xmath58 , respectively . the @xmath63-gates are decomposable as @xmath24 and @xmath56 are disjoint , and by lemma [ lemma : drc ] the @xmath64-gate is deterministic . we claim that the union on the right of ( [ eq : main - node-1 ] ) is disjoint . indeed if @xmath340 and @xmath341 are distinct implicants in @xmath333 , then @xmath342 or @xmath343 . if @xmath342 , then @xmath344 because distinct factors of @xmath50 relative to @xmath24 have disjoint models by ( [ eq : part ] ) . similarly , if @xmath343 , then @xmath345 therefore , @xmath346 , and we are done . we prove the equality in ( [ eq : main - node-1 ] ) . for the nontrivial inclusion ( @xmath347 ) , let @xmath348 be in @xmath349 . by ( [ eq : part ] ) , there exist @xmath219 such that @xmath350 and @xmath351 such that @xmath352 . then @xmath353 . it follows by lemma [ lemma : factors ] that @xmath354 . then @xmath355 and @xmath197 is contained in the union on the right . the above insight can be exploited recursively to implement boolean functions within a natural , canonical class of deterministic structured forms . let @xmath50 be a function and let @xmath22 be a vtree , both over the variables @xmath23 . for every node @xmath98 and every factor @xmath269 of @xmath50 relative to @xmath232 , we construct a circuit @xmath356 as follows . if @xmath91 is a leaf of @xmath22 , then @xmath235 for some variable @xmath230 . there are two cases . either @xmath357 , or @xmath358 . in the former case , @xmath359 and @xmath360 in the latter case , @xmath361 and @xmath362 if @xmath91 is a node of @xmath22 with children @xmath107 and @xmath108 , we put @xmath363 finally we put @xmath364 where @xmath365 is the root of @xmath22 ; note that @xmath50 itself is a factor of @xmath50 relative to @xmath23 ; its models induce the cofactor @xmath366 of @xmath50 , the identically @xmath65 function ( over @xmath367 ) . [ lemma : dsdnnf - sub ] let @xmath50 be a boolean function and let @xmath22 be a vtree , both over the variables @xmath23 . let @xmath98 and let @xmath368 . the following holds . * @xmath356 is a deterministic structured nnf respecting the vtree @xmath96 . * @xmath356 computes @xmath269 . the proof is a routine induction on the depth of @xmath91 in @xmath22 . the base case holds by inspection of ( [ eq : base-1])-([eq : base-3 ] ) . the inductive case holds by inspection of ( [ eq : main - node ] ) , using ( [ eq : main - node-1 ] ) in lemma [ lemma : drc ] and the induction hypothesis ; indeed , note that the disjunction arising in ( [ eq : main - node ] ) is deterministic because the union in ( [ eq : main - node-1 ] ) is disjoint , and the conjunctions arising in ( [ eq : main - node ] ) are structured by @xmath96 ( namely the left and right conjuncts are over variables @xmath369 and @xmath370 resp . , where @xmath107 and @xmath108 are the children of @xmath91 in @xmath22 ) by the induction hypothesis . note that , by lemma [ lemma : dsdnnf - sub ] , it follows that @xmath371 is a deterministic structured nnf computing @xmath50 , canonical in that it is uniquely determined by the vtree @xmath22 and @xmath50 ; the notion of _ factorized implicant width _ of @xmath50 relative to @xmath22 arises naturally . let @xmath50 be a boolean function and let @xmath22 be a vtree , both over the variables @xmath23 . factorized implicant width of @xmath50 relative to @xmath22 _ , in symbols @xmath372 , is defined by is an @xmath63-gate arising from ( [ eq : main - node ] ) , we say that @xmath14 is _ structured _ by @xmath98 . ] @xmath373 the _ factorized implicant width _ of @xmath50 , in symbols @xmath374 , is defined by @xmath375 relative to its factorized implicant width , a boolean function ( of @xmath4 variables ) has linear ( in @xmath4 ) size compilations into canonical deterministic structured forms . [ th : factorized - forms - main ] a boolean function @xmath50 of @xmath4 variables and factorized implicant width @xmath5 has canonical deterministic structured nnfs of size @xmath376 . let @xmath23 be the variables of @xmath50 , so that @xmath377 , and let @xmath22 be a vtree for @xmath23 witnessing factorized implicant width @xmath5 for @xmath50 . the circuit @xmath371 in ( [ eq : cft ] ) is a canonical deterministic structured nnf computing @xmath50 by lemma [ lemma : dsdnnf - sub ] . moreover , we claim that @xmath371 has size @xmath376 . the @xmath4 leaves of @xmath22 contribute at most @xmath378 input gates and @xmath4 @xmath66-gates in @xmath371 . the @xmath379 internal nodes of @xmath22 contribute each at most @xmath5 @xmath63-gates ( by the definition of factorized implicant width ) , and each such gate is linked with at most @xmath380 @xmath64-gates . hence @xmath371 contains at most @xmath381 gates . we conclude the section showing that a class of boolean functions has bounded circuit treewidth iff it has bounded factorized implicant width . it is sufficient to prove that the factorized implicant width of a boolean function is bounded below and above by computable functions of its circuit treewidth . for the upper bound , we have @xmath382 where the first inequality is justified by the observation that every @xmath63-gate @xmath14 in @xmath371 , structured by a node @xmath98 , corresponds to a pair of factors of @xmath50 , and the second inequality follows by lemma [ lemma : jhasuciu ] . for the lower bound , we verify that small factorized implicant width implies small circuit treewidth . [ prop : fiw - ctw ] for all boolean functions @xmath50 , @xmath383 let @xmath22 be a vtree for @xmath23 such that @xmath384 we claim that @xmath385 , so that @xmath386 by lemma [ lemma : dsdnnf - sub ] . for every gate @xmath387 , let @xmath388 denote the closed neighborhood of @xmath14 in the undirected graph underlying @xmath371 . we define a tree decomposition for the undirected graph underlying @xmath371 , as follows . the bags of the tree decomposition have the form @xmath389 for all @xmath98 . the root of the tree decomposition is @xmath390 . the bag @xmath391 has an arc to the bag @xmath392 iff @xmath107 has an arc to @xmath108 in @xmath22 . by definition , @xmath393 contains all the @xmath63-gates structured by the node @xmath98 ; there are at most @xmath5 such gates in @xmath371 by definition , and each such gate has indegree @xmath97 and outdegree @xmath65 by construction ( therefore degree @xmath380 in the undirected graph underlying @xmath371 ) . hence @xmath394 . we check the desired properties . since every wire of @xmath371 enters or leaves an @xmath63-gate and every @xmath63-gate is structured by some @xmath98 , the edges of the undirected graph underlying @xmath371 are covered by the tree decomposition . moreover let @xmath14 be a gate of @xmath371 occurring in two distinct bags @xmath391 and @xmath392 . then @xmath395 where @xmath84 is a @xmath63-gate structured by @xmath107 and @xmath85 is a @xmath63-gate structured by @xmath108 . by construction of @xmath371 , either @xmath107 has an arc to @xmath108 in @xmath22 or @xmath108 has an arc to @xmath107 in @xmath22 . hence @xmath391 and @xmath392 are adjacent . we show that the notion of factorized implicant lies at the core of ( and provides fresh insight on ) the canonical construction of sdds for boolean functions . we prepare the actual description of the construction in two steps . the first step yields , by a straightforward generalization of lemma [ lemma : drc ] , a factorized implicant decomposition reminiscent of ( [ eq : syntactic - fif ] ) for unions of factors . [ lemma : union - drc ] let @xmath139 be a boolean function , let @xmath24 and @xmath56 be disjoint subsets of @xmath23 , and let @xmath396 . then @xmath397 forms a disjoint rectangle cover of @xmath398 . in terms of circuits , the statement means that @xmath399 where @xmath337 , @xmath338 , and @xmath339 are as in ( [ eq : syntactic - fif ] ) , the @xmath64-gate is deterministic , and the @xmath63-gates are decomposable . we have @xmath400 where the second equality follows by applying lemma [ lemma : drc ] to @xmath269 . we claim that the union is disjoint . indeed , by ( [ eq : part ] ) , distinct factors of @xmath50 relative to @xmath55 , in particular those in @xmath401 , have disjoint models ; moreover , by lemma [ lemma : drc ] , distinct implicants of a factor have disjoint models . the second step enforces the properties of a proper sentential decision over the factorized implicant form ( [ eq : syntactic - sdd ] ) given by lemma [ lemma : union - drc ] , as follows . for each @xmath219 , let @xmath402 and observe that @xmath403 where empty disjunctions are implemented by @xmath67 , is equivalent to @xmath404 and is a sentential decision form as the factors of @xmath50 relative to @xmath24 partition @xmath405 . however , ( [ eq : noncanonical - sdd - form ] ) is not a canonical form because distinct @xmath406 and @xmath407 in @xmath207 can give @xmath408 . let @xmath409 for all @xmath410 $ ] , and @xmath411 it is readily observed that @xmath412 is equivalent to @xmath413 and moreover : * @xmath414}\bigvee_{p \in \mathcal{p}_i } c_p$ ] ; * @xmath415 for @xmath117 in @xmath118 $ ] ; * @xmath416 for @xmath117 in @xmath118 $ ] . we now use the above development to describe a recursive construction of a canonical sdd for a given boolean function . let @xmath50 be a boolean function and let @xmath22 be a vtree , both over @xmath23 . for every node @xmath98 and every subset @xmath401 of factors of @xmath50 relative to @xmath232 , we construct a circuit @xmath417 , as follows . if @xmath91 is a leaf of @xmath22 , then @xmath235 for some variable @xmath230 . there are two cases . either @xmath357 , or @xmath358 . in the former case , @xmath359 and : @xmath418 ; @xmath419 . in the latter case , @xmath361 and : @xmath418 ; @xmath420 ; @xmath421 ; @xmath422 . if @xmath91 is a node of @xmath22 with children @xmath107 and @xmath108 , we put @xmath423 finally we put , where @xmath365 is the root of @xmath22 , @xmath424 [ lemma : sdd - constr - correct ] let @xmath50 be a boolean function and let @xmath22 be a vtree , both over the variables @xmath23 . let @xmath98 and let @xmath425 . * @xmath417 is a canonical sdd respecting the vtree @xmath96 . * @xmath417 computes @xmath426 . by induction on the depth of @xmath91 in @xmath22 . the base case holds by construction . the inductive case holds inspection of ( [ eq : cvh ] ) , using ( [ eq : canonical - sdd - form ] ) together with ( sd1)-(sd3 ) , ( [ eq : noncanonical - sdd - form ] ) , lemma [ lemma : union - drc ] , and the induction hypothesis . therefore @xmath427 is the canonical sdd computing @xmath50 , uniquely determined by the vtree @xmath22 . we recall the notion of _ sentential decision width _ of @xmath50 relative to @xmath22 . [ def : sdd - width ] let @xmath50 be a boolean function and let @xmath22 be a vtree , both over @xmath23 . the _ sentential decision width of @xmath50 relative to @xmath22 _ , in symbols @xmath428 , is defined by @xmath429 the _ sdd width _ of @xmath50 , in symbols @xmath430 , is defined by @xmath431 it is well known that obdds are canonical sdds respecting linear vtrees , i.e. vtrees where every left child is a leaf @xcite ; in this case , the notion of sdd width in definition [ def : sdd - width ] reduces to the usual notion of obdd width @xcite . moreover , as a boolean function ( of @xmath4 variables ) has linear ( in @xmath4 ) obdd size parameterized by its obdd width , likewise it has linear sdd size parameterized by its sdd width . [ th : linear - sdd - size ] a boolean function @xmath50 of @xmath4 variables and sdd width @xmath5 has canonical sdd size @xmath376 . let @xmath23 be the variables of @xmath50 , so that @xmath377 , and let @xmath22 be a vtree for @xmath23 witnessing sdd width @xmath5 for @xmath50 . the circuit @xmath427 in ( [ eq : sdd - root ] ) is a canonical sdd by lemma [ lemma : sdd - constr - correct ] . moreover , we claim that @xmath427 has size @xmath376 . the @xmath4 leaves of @xmath22 contribute at most @xmath432 input or negation gates in @xmath371 . the @xmath379 internal nodes of @xmath22 contribute each at most @xmath5 @xmath63-gates ( by the definition of sdd width ) , and each such gate is linked with at most @xmath380 @xmath64-gates . hence @xmath371 contains at most @xmath433 gates . we conclude observing that , for classes of boolean functions , bounded circuit treewidth and bounded sdd width collapse . indeed , on the one hand , the sdd width of a boolean function @xmath50 is bounded above by a computable function of its circuit treewidth , namely , @xmath434 since in the canonical sdd @xmath427 for @xmath22 every @xmath63-gate @xmath14 structured by a node @xmath98 corresponds to a pair of sets of factors of @xmath50 ( plus @xmath67 ) , and @xmath435 is bounded above by @xmath226 as in lemma [ lemma : jhasuciu ] . on the other hand , for all boolean functions @xmath50 , along the lines of proposition [ prop : fiw - ctw ] , @xmath436 by combining ( [ eq : fiw - ub])-([eq : fiw - lb ] ) and ( [ eq : sdd - ub])-([eq : sdd - lb ] ) , the factorized implicant width ( resp . , sdd width ) of a boolean function is squeezed between computable functions of its sdd width ( resp . , factorized implicant width ) . in this section , we show that inversions in unions of conjunctive queries , with or without inequalities , imply large deterministic structured circuits for their lineages . let @xmath437 be a relational vocabulary . union of conjunctive queries ( ucqs ) with inequalities _ @xmath0 is a disjunction of existentially closed conjunctions of atoms @xmath438 and inequalities @xmath439 , where @xmath440 and @xmath119 , @xmath173 , @xmath441 are variables , @xmath410 $ ] . we call @xmath0 a ucqs if it does not contain inequalities . the _ lineage of a boolean query @xmath0 over a database @xmath1 _ is a boolean function @xmath2 whose boolean variables are the tuples in @xmath1 such that , for every subdatabase @xmath442 , it holds that @xmath443 iff @xmath444 , where @xmath445 is defined by @xmath446 iff @xmath447 . a _ lineage of a boolean query _ is a lineage of the query over some database . we prove the main result . for all @xmath448 , let @xmath449 \}$ ] , @xmath450 \}$ ] , @xmath451 \}$ ] for @xmath452 $ ] , and @xmath453}z^i$ ] . for @xmath454 $ ] , let : @xmath455 } ( x_{l } \wedge z^1_{l , m})\text{,}\\ h^i_{k , n}(z^i , z^{i+1 } ) & = \bigvee_{l , m \in [ n ] } ( z^i_{l , m } \wedge z^{i+1}_{l , m})\text{,}\\ h^k_{k , n}(z^k , y ) & = \bigvee_{l , m \in [ n ] } ( z^k_{l , m } \wedge y_{m})\text{.}\end{aligned}\ ] ] in ( * ? ? ? * proposition 7 ) and ( * ? ? ? * theorem 3.9 ) jha and suciu show the following , resp . for ucqs and ucqs with inequalities . [ lemma : inv - impl - h ] let @xmath0 be a ucqs with or without inequalities . if @xmath0 contains an inversion of length @xmath456 , then for every @xmath61 there exist a lineage @xmath101 of @xmath0 on @xmath457 variables and assignments @xmath458 for @xmath459 and @xmath460 such that @xmath461 as we only need the implication stated in lemma [ lemma : inv - impl - h ] , we omit the technical definition of the notion of inversion @xcite . the following statement unifies and generalizes analogous results by jha and suciu for ucqs with inequalities vs. obdds ( * ? ? ? * theorem 3.9 ) and by beame and liew for ucqs vs. sdds ( * ? ? ? * theorem 4.6 ) . [ th : main - ineq ] let @xmath0 be a ucqs with or without inequalities . if @xmath0 contains an inversion of length @xmath456 , then for every @xmath61 there exists a lineage @xmath50 of @xmath0 on @xmath457 variables whose deterministic structured nnf size is @xmath462 . let @xmath0 be a query with inequalities . if @xmath0 contains an inversion of length @xmath456 , then by lemma [ lemma : inv - impl - h ] for every @xmath61 there exist a lineage @xmath101 of @xmath0 on @xmath457 variables and assignments @xmath458 for @xmath459 and @xmath460 such that @xmath463 . let @xmath70 be a deterministic nnf of size @xmath464 computing @xmath50 , structured by the vtree @xmath22 . then by the properties of deterministic structured nnfs dblp : conf / aaai / pipatsrisawatd08 , it holds that @xmath465 is a deterministic nnf of size @xmath466 that computes @xmath467 and is structured by @xmath22 , for all @xmath460 . by lemma [ lemma : main - inversion ] , there exists @xmath468 such that @xmath469 has size @xmath470 . therefore @xmath70 has size @xmath462 . the proof idea is that , if a query @xmath0 contains inversions , then it has a lineage @xmath2 of which each @xmath471 is a cofactor ( @xmath460 ) . if @xmath70 is a small deterministic form for @xmath2 respecting a vtree @xmath22 , then small deterministic structured forms for each @xmath471 , _ all _ respecting the vtree @xmath22 , can be mined from @xmath70 by suitably assigning its inputs . but this is impossible for communication complexity reasons ( lemma [ lemma : main - inversion ] ) . [ lemma : main - inversion ] for every vtree @xmath22 for @xmath472 and every family @xmath473 of deterministic structured nnfs , where @xmath469 is structured by @xmath22 and computes @xmath471 ( @xmath460 ) , there exists @xmath468 such that @xmath469 has size @xmath462 . we let @xmath232 denote the variables in @xmath474 , @xmath475 denote the variables in @xmath476 , and @xmath477 denote the variables in @xmath478 . [ cl : find - node ] there exists @xmath98 such that @xmath479 . let @xmath480 , for all @xmath98 . let @xmath481 be a root - leaf path in @xmath22 such that , letting @xmath482 , @xmath483 for all @xmath484 . let @xmath188 $ ] be minimum such that @xmath485 so that @xmath486 by construction , @xmath487 hence , letting @xmath488 , we have @xmath489 and we are done . by claim [ cl : find - node ] , let @xmath98 be such that @xmath479 . let @xmath490 and @xmath491 . assume without loss of generality that @xmath492 ; otherwise the argument is similar . it follows by the choice of @xmath91 that @xmath493 and that @xmath494 we enter a case distinction . the first case is covered by the following claim . [ cl : h0-hard ] if there exists @xmath495 $ ] such that for all @xmath496 $ ] it holds that @xmath497 implies @xmath498 , then @xmath499 has size @xmath500 . for @xmath495 $ ] , let @xmath501 . by hypothesis , there exists @xmath495 $ ] such that @xmath502 . write @xmath503 then @xmath504 is a deterministic nnf structured by @xmath22 of size @xmath505 @xcite . it follows from theorem [ th : struct ] that @xmath506 has a disjoint rectangle cover of size at most @xmath507 where each rectangle has underlying partition @xmath508 . by the choice of @xmath91 , it holds that @xmath232 contains @xmath509 variables in @xmath23 . for the sake of notation , say that @xmath510 , so that @xmath511 . then , since @xmath512 , we have that @xmath513 note that @xmath514 is the complement of the disjointness function @xmath515 in ( [ eq : df ] ) . therefore the complement of the communication matrix of @xmath506 relative to @xmath508 is equal ( up to a permutation of rows and columns ) to the communication matrix of @xmath516 , i.e. , @xmath517 where @xmath65 denotes the @xmath518 all-@xmath65 matrix . therefore , by ( [ eq : rkdf ] ) and basic linear algebra , @xmath519 hence @xmath520 therefore , by theorem [ th : rklb ] and ( [ eq : lb - compl - df ] ) , every disjoint rectangle cover of @xmath506 into rectangles with underlying partition @xmath508 contains at least @xmath521 rectangles . summarizing , @xmath506 has a disjoint rectangle cover of size at most @xmath507 where each rectangle has underlying partition @xmath508 , but every such rectangle cover contains at least @xmath521 rectangles . hence @xmath522 and we are done ( recall ( [ eq : nx - lb ] ) ) . the second ( and complementary ) case is covered by the following claim . [ cl : hkp - hard ] if for all @xmath495 $ ] there exists @xmath496 $ ] such that @xmath497 and @xmath523 , then there exists @xmath524 $ ] such that @xmath525 has size @xmath462 . define a set @xmath135 of pairs @xmath526 as follows . for each @xmath495 $ ] such that @xmath527 , choose @xmath496 $ ] such that @xmath523 , and add @xmath526 to @xmath135 . note that @xmath528 , let @xmath529 be such that @xmath530 iff @xmath531 and @xmath532 . also , let @xmath533 note that @xmath534 form a partition of @xmath135 , so that @xmath535 which implies that there exists @xmath524 $ ] such that @xmath536 , and we are done . let @xmath537 $ ] . for all @xmath530 , it holds that @xmath538 and @xmath532 . let @xmath539 and @xmath540 . write @xmath541 and let @xmath542 so that @xmath543 is a deterministic nnf structured by @xmath22 of size @xmath544 @xcite . it follows from theorem [ th : struct ] that @xmath543 has a disjoint rectangle cover of size at most @xmath545 where each rectangle has underlying partition @xmath546 . since @xmath547 , we have that @xmath548 and along the lines of claim [ cl : h0-hard ] we obtain @xmath549 and we are done . the third inequality holds by plugging the convex function @xmath550 in jensen s inequality @xmath551 . claim [ cl : h0-hard ] and claim [ cl : hkp - hard ] imply the statement . we have related the circuit treewidth of a boolean function with the width of its sdd implementation ( and more generally its width in natural canonical classes of deterministic structured forms ) , and we have incorporated constant width sdds and polynomial size sdds in the panorama of query compilation for union of conjunctive queries with and without negations . the comparison of theorem [ th : main - ineq ] and ( * ? ? ? * theorem 4.6 ) reiterates the question about the relative succinctness of deterministic structured forms and sdds @xcite . as beame and liew observe , a natural candidate function for an exponential separation is the _ indirect access storage _ ( _ isa _ ) function , which is known to have large obdds but whose small deterministic structured forms deviate substantially from the sdd syntax ( * ? ? ? * section 6 ) . however , as we prove in appendix [ app : isaproof ] , isa has small sdd size , which unfortunately leaves us with no candidates for a separation . the canonical structured deterministic forms induced by factorized implicants , introduced in section [ sect : fimpldec ] , deserve in our opinion both a direct investigation in the framework of the knowledge compilation map @xcite , and a thorough comparison with the data structures used in _ factorized databases _ , which are more than just reminiscent of structured deterministic forms @xcite . the question remains whether bounded circuit treewidth lineages imply bounded obdd width for ucqs with inequalities , as conjectured by jha and suciu @xcite . another intriguing conjecture is that sdds with or gates of bounded fanin are quasipolynomially simulated by obdds . the containment of bounded width sdds in polynomial size obdds , discussed in the introduction and obtained in section [ sect : part1 ] , imply a polynomial simulation of bounded width sdds , which have indeed bounded fanin ors , by obdds . this research was supported by the fwf austrian science fund ( parameterized compilation , p26200 ) . 10 a. amarilli , p. bourhis , and p. senellart . provenance circuits for trees and treelike instances . in _ proc . of icalp 2015 _ , pages 5668 . springer , 2015 . a. amarilli , p. bourhis , and p. senellart . tractable lineages on treelike instances : limits and extensions . in _ proc . of pods 2016 _ , pages 355370 . acm , 2016 . p. beame and v. liew . new limits for knowledge compilation and applications to exact model counting . in _ proc . of uai 2015 _ , pages 131140 . auai press , 2015 . s. bova . sdds are exponentially more succinct than obdds . in _ proc . of aaai 2016 _ , pages 929935 . aaai , 2016 . s. bova , f. capelli , s. mengel , and f. slivovsky . on compiling cnfs into structured deterministic dnnfs . in _ proc . of sat 2015 _ , pages 199214 . springer , 2015 . s. bova , f. capelli , s. mengel , and f. slivovsky . knowledge compilation meets communication complexity . in _ proc . of ijcai 2016 _ , pages 10081014 . ijcai / aaai , 2016 . r. e. bryant . graph - based algorithms for boolean function manipulation . , 35(8):677691 , 1986 . a. choi and a. darwiche . dynamic minimization of sentential decision diagrams . in _ proc . of aaai 2013_. aaai , 2013 . n. n. dalvi and d. suciu . the dichotomy of conjunctive queries on probabilistic structures . in _ proc . of pods 2007 _ , pages 293302 . acm , 2007 . n. n. dalvi and d. suciu . efficient query evaluation on probabilistic databases . , 16(4):523544 , 2007 . a. darwiche . decomposable negation normal form . , 48(4):608647 , 2001 . a. darwiche . on the tractable counting of theory models and its application to truth maintenance and belief revision . , 11(1 - 2):1134 , 2001 . a. darwiche . new canonical representation of propositional knowledge bases . in _ proc . of ijcai 2011 _ , pages 819826 . ijcai / aaai , 2011 . a. darwiche and p. marquis . a knowledge compilation map . , 17:229264 , 2002 . g. v. den broeck and a. darwiche . on the role of canonicity in knowledge compilation . in _ proc . of aaai 2015 _ , pages 16411648 . aaai , 2015 . j. flum and m. grohe . . springer , 2010 . e. grdel , y. gurevich , and c. hirsch . the complexity of query reliability . in _ proc . of pods 1998 _ , pages 227234 . acm , 1998 . t. j. green , g. karvounarakis , and v. tannen . provenance semirings . in _ proc . of pods 2007 _ , pages 3140 . acm , 2007 . a. k. jha and d. suciu . knowledge compilation meets database theory : compiling queries to decision diagrams . in _ proc . of icdt 2011 _ , pages 162173 acm , 2011 . a. k. jha and d. suciu . on the tractability of query compilation and bounded treewidth . in _ proc . of icdt 2012 _ , pages 249261 . acm , 2012 . a. k. jha and d. suciu . knowledge compilation meets database theory : compiling queries to decision diagrams . , 52(3):403440 , 2013 . s. jukna . . springer , 2012 . t. kloks . . springer , 1994 . s. mengel . parameterized compilation lower bounds for restricted cnf - formulas . in _ proc . of sat 2016 _ , pages 312 . springer , 2016 . d. olteanu . factorized databases : a knowledge compilation perspective . , 2016 . u. oztok and a. darwiche . a top - down compiler for sentential decision diagrams . in _ proc . of ijcai 2015 _ , pages 31413148 . aaai , 2015 . k. pipatsrisawat and a. darwiche . new compilation languages based on structured decomposability . in _ proc . of aaai 2008 _ , pages 517522 . aaai , 2008 . t. pipatsrisawat and a. darwiche . a lower bound on the size of decomposable negation normal form . in _ proc . of aaai 2010_. aaai , 2010 . i. razgon . on obdds for cnfs of bounded treewidth . in _ proc . of kr 2014_. aaai , 2014 i. razgon . on the read - once property of branching programs and cnfs of bounded treewidth . , 75(2):277294 , 2016 . i. razgon and j. petke . cliquewidth and knowledge compilation . in _ proc . of sat 2013 _ , pages 335350 . springer , 2013 . d. seese . the structure of models of decidable monadic theories of graphs . , 53(2):169195 , 1991 . d. suciu , d. olteanu , r. christopher , and c. koch . . morgan & claypool publishers , 2011 . i. wegener . . siam , 2000 . let @xmath5 and @xmath12 be positive integers such that @xmath552 . the _ indirect access storage _ ( in short , _ isa _ ) function on @xmath553 variables @xmath554 also displayed as @xmath555 accepts input @xmath556 also displayed as @xmath557 iff , letting @xmath558 be the number whose binary representation is @xmath559 and @xmath560 be the number whose binary representation is @xmath561 , it holds that @xmath562 . let @xmath565 be the vtree for variables @xmath566 formed by a right - linear subtree @xmath567 whose left leaves correspond to the variables in @xmath568 and whose ( unique ) right leaf @xmath91 is the root of a left - linear subtree @xmath569 whose ( unique ) left leaf corresponds to @xmath570 and whose right leaves correspond ( in a postorder traversal ) to @xmath571 . for instance , the vtree @xmath572 is depicted in figure [ fig : isaproof ] . we liberally identify the leaves of @xmath573 with their labels ( so that we simply call @xmath119 the leaf of @xmath573 labelled by the variable @xmath119 ) . for @xmath452 $ ] , we let @xmath574 denote the node in @xmath573 whose left child is @xmath575 ; for @xmath576 $ ] , we let @xmath577 denote the node in @xmath573 whose right child is @xmath578 . a term on @xmath579 is a conjunction of literals on @xmath579 . and a term containing both literals of a variable is denoted by @xmath67 . ] we call a term on @xmath579 _ small _ if it contains at most @xmath580 distinct variables . note that the number of small terms on @xmath579 is @xmath581 since @xmath582 . we now construct an sdd @xmath70 computing @xmath563 and respecting @xmath573 , where each @xmath63-gate structured by a node of the form @xmath577 ( @xmath576 $ ] ) conjoins one small term on @xmath579 and an input gate ( namely , a literal on @xmath577 or a constant ) . as @xmath70 has at most @xmath583 input gates , it follows that the number of @xmath63-gates in @xmath70 structured by nodes of the form @xmath584 ( @xmath576 $ ] ) is @xmath585 moreover , @xmath70 is such that the number of @xmath63-gates structured by nodes of the form @xmath574 ( @xmath452 $ ] ) is @xmath586 since @xmath587 . it follows immediately that @xmath70 has size polynomial in @xmath4 . indeed , as each @xmath63-gate in @xmath70 contributes a constant number of @xmath64-gates , @xmath70 contains at most @xmath564 internal gates . also , @xmath70 has at most @xmath583 input gates . hence @xmath588 the upper part of @xmath70 is isomorphic to an obdd respecting the order @xmath589 and having @xmath590 source gates . each such source , say @xmath591 , corresponds to the boolean assignment @xmath592 , @xmath452 $ ] , of the variables in @xmath568 and implements the cofactor @xmath593 as an sdd respecting the vtree @xmath569 , as follows . we start observing that each cofactor in ( [ eq : isacof ] ) is expressible as a sentential decision , respecting the root node of @xmath569 and involving only small terms on @xmath579 ( and literals on @xmath594 ) . case @xmath598 : in this case , we have to implement @xmath599 , which is equivalent to @xmath600 here @xmath601 corresponds to the term @xmath602 where @xmath603 represents @xmath560 in binary , and @xmath604 , @xmath605 , @xmath606 $ ] . for all @xmath607 and @xmath608 $ ] we say that _ orbits on @xmath304 _ if @xmath610 represents @xmath611 in binary . if @xmath609 orbits on @xmath304 , we let @xmath609 orbits on @xmath304 denote the term @xmath612 by direct inspection , @xmath599 is equivalent to a sentential decision @xmath613 as in ( [ eq : sdd - def ] ) , namely , @xmath614 where @xmath615 ranges over all pairs such that @xmath609 orbits on @xmath304 . here , the interesting @xmath63-gates ( marked with @xmath616 ) are structured by the node @xmath617 in @xmath573 . moreover , the @xmath109 s are small terms as they contain @xmath618 variables by construction . [ ex : hec ] assume @xmath619 , so that we compute @xmath620 . in this case @xmath621 and @xmath622 . the following lists the disjuncts corresponding to @xmath623 for @xmath624 ( we use @xmath625 and @xmath626 as shortenings ) : @xmath627 if @xmath615 gives @xmath628 , then the corresponding subdisjunction reduces to one disjunct only ; and , if @xmath615 gives @xmath629 , then the corresponding subdisjunction reduces to two disjuncts only , as the following example illustrates . case @xmath634 : say that @xmath635 represents @xmath636 . we implement @xmath637 , which is equivalent to @xmath638 where the notation is as in the previous case ; note that @xmath639 . an equivalent sentential decision of the form ( [ eq : sdd - def ] ) is obtained by disjoining @xmath640 and the following : @xmath641 where the interesting @xmath63-gates ( marked with @xmath616 ) are structured by the node @xmath617 in @xmath573 . moreover , the @xmath109 s are small terms as they contain @xmath580 variables by construction . the construction implements each gate @xmath591 by the sentential decision given by claim [ cl : base ] . we now claim that the construction can continue recursively by implementing the resulting @xmath109 s as sdds respecting subtrees of the subtree of @xmath573 rooted at @xmath648 . [ cl : isarec ] let @xmath649 be a small term on @xmath579 . then @xmath649 is equivalent to a sentential decision @xmath596 of the form ( [ eq : sdd - def ] ) , where the @xmath63-gates are structured by some @xmath650 ( @xmath576 $ ] ) and the @xmath109 s are small terms . let @xmath609 be the assignment of @xmath657 to @xmath658 such that @xmath659 by direct inspection , @xmath649 is equivalent to the sentential decision @xmath613 of the form ( [ eq : sdd - def ] ) @xmath660 where @xmath197 ranges over the assignments of @xmath657 in @xmath658 distinct from @xmath609 .

the analysis of cosmic large - scale structure in the distribution of galaxies contributed significantly to establish the current cosmological concordance model . with the increasing degree of perfection of the standard two - point correlation analysis , the related measurements reached unprecedented precision . however , the available and upcoming galaxy redshift survey data @xcite can not merely be characterized by its two - point correlation properties . already the filamentary structure that we see in those surveys indicates that the higher - order correlations play an important role . there have been different approaches to use this information for improving our knowledge about the constituents and evolution of the universe . the most direct approach of course is to simply measure the higher - order correlations directly . this has been done for both simulated and observed data starting for early catalogs with @xcite and @xcite . especially measurements of the three - point function have been standard for recent surveys @xcite and provide useful complimentary information @xcite . for higher orders , however , there are computational and conceptual challenges . on the computational side , increasing the order of the correlation function measured unleashes the curse of combinatorics . measuring all pairs in current surveys is feasible , for all triangles there are methods to do it @xcite , but beyond that it becomes increasingly hard . and then , from the conceptual side , even if measuring the full @xmath7-point function was possible , it would require modeling and covariances for a large number of data bins . already for a modest number of bins for each parameter of the @xmath7-point function this is a challenge . therefore , it is worthwhile to consider other approaches that contain information on certain aspects of the higher - order clustering without referring to the full correlation functions . since the beginning of large - scale structure analysis , a large number of such methods have been used , including moments of counts in cells , void probability @xcite , structure functions on minimal spanning trees , wavelet methods , the genus , etc . recently also approaches that attempt to bring the higher - order information back into the two - point correlation function through non - linear transforms were used like a logarithmic transformation @xcite . the method ( re-)considered in this paper is using minkowski functionals . they also also fall into the category of methods that provide condensed access to higher - order information . this information is encoded in the morphology of the density field . minkowski functionals access it by quantifying extended regions of the field through their geometrical properties like volume and surface ( see sec . [ sub : minkowski - functionals ] ) . for the construction of these extended regions , two different methods have been widely used in the past : in the first application of minkowski functionals for analyzing large - scale structure in the galaxy distribution in ( see @xcite for a short motivation and @xcite for a brief overview ) , the regions considered were defined by the union of balls around every galaxy location . we shall use the same method , referred to as germ - grain model , here and explain the details in sec . [ sub : germ grain ] . after a couple of applications in the analysis of galaxy and cluster catalogs , the germ - grain model has been less used in recent years . the second and most popular form of the application of minkowski functionals works with extended regions that are defined by iso - contours of the field under investigation . it has been used for the higher - order contributions to the dark matter overdensity field @xcite , for the weak lensing shear field @xcite , for local morphology of cosmic structure and for the morphology of reionization bubbles @xcite . another very active field of recent application is the study of iso - temperature contour maps of the cosmic microwave background ( cmb ) , where minkowski functionals helped to establish stringent bounds on the non - gaussianity of the cmb . so the iso - contour minkowski functionals are a well established tool in various domains of cosmology . however , as we shall discuss in sec . [ sub : germ grain ] , the germ - grain model accesses different aspects of the higher - order distribution and can be better suited for the analysis of galaxy redshift surveys . we have demonstrated some of these advantages already in the analysis of the sloan digital sky survey ( sdss ) data release seven ( dr7 ) luminous red galaxies ( lrgs ) in @xcite . in the present paper , we concentrate more explicitly on the higher - order aspects of the minkowski functionals . to this end , we exploit the analytically known connection to the integrals over higher - order correlation functions . using direct measurements of the two - point correlation function , we can isolate the contribution of higher - order correlation functions to the minkowski functionals . this allows us to quantify the evidence for non - gaussianity of the galaxy distribution ( see @xcite and @xcite for related recent studies of non - gaussianity with isodensity contour functionals ) . the analysis is demonstrated with the final data from the baryon oscillation spectroscopic survey ( boss ) @xcite that is now publicly available within the dr12 of sdss - iii . being the largest spectroscopic galaxy redshift data set yet observed , it provides an ideal testbed for the analysis and reveals the strong presence of higher - order correlations already at redshifts of @xmath8 . we proceed as follows : sec . [ sec : minkowski - functionals - in - the ] recalls the properties of minkowski functionals in general and the germ - grain model in particular . we write out the connection of the functionals to higher - order correlations explicitly in [ sub : densdef ] and lay out in [ sub : density - dependence ] , how knowing the concrete form allows us to isolate the part of the functionals that is only dependent on higher - orders . we explain the subtleties of a gaussian reference model in the discrete case in [ sub : gauss - poisson ] and recall the basic properties of our minkowski code in [ sub : samplingcode ] . in sec . [ sec : data ] we turn to the description of the sdss dr12 data used . [ sub : cmass ] summarizes the data and mocks we are using . in [ sub : ceometry ] , we address issues related to the imperfections in the data , constructing a custom angular mask that defines a high quality region . in [ sub : radial ] we turn to the radial distribution of the data and choose the analysis volume and the radial grid . in [ sub : weights ] we finally discuss the effect of the regions that can not be accounted for by our custom mask . sec . [ sec : higherorders ] then turns to the core of the analysis . in [ sub : mfdensities ] we present the minkowski functionals of the mocks and the data and explain how we transform the data to be able to isolate the higher - order only part . this part is then examined in [ sub : highsubtract ] . in [ sub : seriescoefficients ] we demonstrate that the contributions to this higher - order part range at least up to the @xmath9-point function . to wrap up we discuss in [ sub : voidfraction ] an interesting aspects about the galaxy distribution using the base minkowski functional to study the volume in extreme voids in the observed field . we conclude in sec . [ sec : conclusion ] . in this section , we shall provide the basic notions necessary for understanding the quantities we used in our analysis . we summarize some key points already covered in the dr7 analysis paper @xcite , but with an emphasis on the crucial points that are the source for common misconceptions . minkowski functionals ( often called mfs later on ) quantify the morphology of extended bodies . they are defined as scalar functionals that map the shape of a body to a set of real numbers . a quite surprising result of integral geometry @xcite is that any such scalar functional , defined for bodies in euclidean @xmath10-dimensional space , that fulfills some basic requirements ( motion invariance , additivity and conditional continuity ) can be expressed as a linear combination of @xmath11 base functionals . there are different choices for the normalization of the base functionals . we will use one where the four mfs that exist for @xmath12 are denoted @xmath13 and related to geometrical properties of the body as follows : @xmath14 here , @xmath15 is the volume of the body , @xmath16 is its surface area , @xmath17 the integral mean curvature of the surface , and @xmath18 the euler characteristic ( the integral gaussian curvature of the surface ) . instead of using the functionals @xmath19 directly , we will mainly work with the mf densities @xmath20 : @xmath21 here @xmath22 is the volume within the survey boundary . in the practical calculation of the @xmath20 discussed in sec . [ sub : samplingcode ] , @xmath22 is the volume of the region that is more than two ball radii away from our custom angular mask of fig . [ fig : custommask ] . as mentioned , mfs are only defined for extended bodies . for the analysis of large - scale structure , however , the galaxies are usually treated as points . for points , all quantities in eq . ( [ eq : minkrelgeo ] ) are , a priori , trivial . for a meaningful analysis of large - scale structure , one therefore has to smooth the point distribution in some manner to define extended regions . the two most popular choices for defining extended regions are : 1 . isodensity contours 2 . the germ - grain model the first approach attempts to reconstruct the underlying density field from the observed galaxies and to define the extended bodies as the interior of the surfaces of constant density . the value of this density threshold can then be varied and the functionals are studied as a function of this parameter . in the second approach , the extended bodies are constructed by surrounding every galaxy ( the germ ) by an extended body ( the grain ) , which in our case is a ball of a certain radius , and calculating the functionals for the union of all these balls . the diagnostic parameter in this case is the common radius of the balls . [ fig : boolean ] shows how a structured body emerges with increasing radius . are surrounded by balls of a common radius @xmath23 . then the mfs of the body formed by the union of all the balls are studied as a function of @xmath23 . the parts of the body outside of the survey region are not taken into account . with increasing @xmath23 , more and more balls intersect and a complex structure develops . through the single and multiple intersections of neighboring balls , it encodes information about the @xmath7-point correlation hierarchy . plot from @xcite.,scaledwidth=45.0% ] only those parts of the structure are taken into account that lie within the survey boundaries , in fig . [ fig : boolean ] illustrated as a cube . even though the basic analysis tool is the same in both cases , the properties of the two approaches are quite different . the isodensity contours , at least in theory , are more directly linked the underlying physical density field , whereas the germ - grain radii are purely diagnostic parameters that merely specify the scale at which the correlations are probed . regarding the temporal evolution , the mfs for isodensity contours turn out to be relatively independent of the ( linear ) growth of the density field , if the value of the threshold density is adjusted accordingly . on the germ - grain side , there is no corresponding relabeling and , therefore , as we also shall see below , they depend significantly on the amplitude of the power spectrum . besides these conceptual differences , there are some technical advantages to the calculation of germ - grain mfs : 1 . the evaluation directly uses the point distribution and does not use a smoothing procedure that is an art by itself . the use of balls does not introduce additional ( smoothing ) parameters . the construction of smooth surfaces can be noisy if the density is low . the rougher smoothing procedure in the germ - grain model on the other hand always constructs some well defined surface . for both smoothing methods there have been detailed analysis how the values of the measured functionals are related to the parameters of the underlying density distribution . for the germ - grain case we recall this relation in the following section . due to the simplicity of the geometrical construction as a union of intersecting balls , it is relatively straightforward to relate the germ - grain mfs to correlation properties of the analyzed point distribution . for the case of a poisson distribution this has been achieved in @xcite , with the only relevant parameter being the density of the random sample . in the presence of structure , the dependence on higher order correlations has first been demonstrated in . in the following , we shall use the form derived in @xcite and detailed in the appendix of @xcite . in these works it has been shown that the ensemble average of the germ - grain mf densities defined in eq . ( [ eq : minkdens ] ) , for the case of balls as grains , can be expressed in an analytic manner as a function of the quantities @xmath24 and @xmath25 . @xmath24 is the ensemble average density of the point distribution studied , and the @xmath25 are the unnormalized modified mfs from the dr7 paper @xcite . the analytic relation reads @xmath26 and provides the implicit definition of the @xmath25 . in the rest of the paper we will drop the explicit ensemble averaging brackets and use @xmath27 as a shorthand , as we never refer to the mf density of one individual realization . we discuss this aspect in more detail in sec . [ sub : samplingcode ] . for a poisson distribution , the @xmath25 are simply the mfs @xmath28 of the balls of common radius @xmath23 that we use to obtain extended bodies : @xmath29 in the poisson case eq . ( [ eq : minkdensdef ] ) reduces to the result of @xcite . for a point distribution with structure , the @xmath25 are modified by weighted integrals of the dimensionless higher - order connected correlation functions @xmath30 . this modification is what we are interested in in this paper . it is convenient to define it as a multiplicative dimensionless correction to the poisson value , i.e. to study the quantities @xmath31 we will refer to @xmath32 as the modified mfs . they can be easily obtained from the quantities we measure , i.e. the average densities @xmath20 , by inverting eqs . ( [ eq : minkdensdef ] ) : @xmath33 in the analysis below , we will mainly use these modified mfs instead of the densities @xmath20 , as they are more directly related to the correlation properties of the point distribution via ( [ eq : minkcorrcon ] ) . from @xcite the relation of @xmath32 to the correlations of order @xmath34 , @xmath35 , is of the form @xmath36 with @xmath37 being simply the mfs of the _ intersection _ of @xmath34 balls . @xmath38 denotes a ball around the origin and @xmath39 a ball around the point @xmath40 . in our model all intersecting balls have the same radius @xmath23 . for general germ - grain models , this is not a necessary requirement . one could imagine to increase the weight of some points by assigning them larger radii than other points . using elliptic or even more irregular grains would also be possible . however , this would complicate the interpretation as we expect that it would lead to a more complicated relation of the functionals to the correlation hierarchy than the one of eq . ( [ eq : minkcorrcon ] ) . the advantage of the @xmath32 of being more directly related to the correlation properties is counterbalanced by an increase in the error . in @xcite we found weaker constraints when using the @xmath25 as compared to the @xmath20 . the main reason for this is the additional error that is introduced by removing the exponential suppression @xmath41 from @xmath42 in ( [ eq : minkdensdef ] ) . for @xmath43 , for example , @xmath44 so the relative error for the fraction of the volume not covered by any ball enters also for the higher functionals . this spoils to some extent the good precision that can be obtained for @xmath42 , because @xmath45 has to be measured in the full volume and @xmath42 only on the surface . especially for small @xmath46 it becomes costly to determine @xmath45 well enough . we will therefore restrict our analysis of the @xmath32 to parameter regions where @xmath46 is larger than @xmath47 . the modified mfs @xmath32 of eq . ( [ eq : minkcorrcon ] ) not only depend on our diagnostic parameter , i.e. the radius @xmath23 of the balls that we are using to smooth the galaxy distribution , but also on the sample density . the rather complicated @xmath23 dependence is entirely encoded in the weight functions @xmath48 in the integrals over correlation functions . the dependence on the mean density of the galaxy sample @xmath24 , however , is of an easier form . it is a simple power series in @xmath24 @xmath49 with coefficients @xmath50 and @xmath51 that contain all of the radius and correlation function dependence . @xmath52 is the volume of a ball @xmath53 of radius @xmath23 . we choose to define the second index of @xmath54 such that it directly corresponds to the order of the integrated correlation function . the formal split into @xmath24 dependence and correlation function dependence in eq . ( [ eq : powerseriesdecomp ] ) allows us two things : first , we can remove the poisson part and the part that depends on the two - point correlation function , if we measure the sample density and the two - point correlations independently . this leaves us with a quantity @xmath55 that only includes higher - order contributions . this will prove the significance of their contribution to the mfs in sec . [ sub : highsubtract ] . second , we can extract the integrals over correlation functions by fitting a polynomial to the density evolution of @xmath56 . this allows to define a fitting formula for the form of the mfs in the sdss by giving the values and covariances of the coefficients @xmath57 . we shall use this in sec . [ sub : seriescoefficients ] . we close with the remark that , in general , a finite number of coefficients is not sufficient to describe the full shape of a power series of the sort ( [ eq : powerseriesdecomp ] ) . for cooked up point distributions with small @xmath30 , but a single large @xmath58 at some higher @xmath59 , giving only the lower coefficients will not determine @xmath32 well . for the assumed structure of the universe as the product of standard structure formation , however , we find that the coefficients @xmath60 are well behaved and the series @xmath32 converges reasonably well ( see sec . [ sub : seriescoefficients ] ) . in order to illustrate the relations ( [ eq : minkcorrcon ] ) , let us consider the simplest non - trivial case . in analogy with the continuous case , one can define a gaussian point distribution by the absence of all higher - order connected correlation functions . in this case the hierarchy of correlations in ( [ eq : minkcorrcon ] ) is truncated after the linear term and we have @xmath61 the weight functions @xmath62 are non - zero only for the range of @xmath63-values for which the balls intersect . for a coordinate system centered around the origin , i.e. the center of @xmath53 , the @xmath40-values therefore lie in a sphere of radius @xmath64 . a configuration of balls of radius @xmath23 therefore probes the correlation function up to a scale @xmath64 . we shall therefore plot the radial dependence as a function of the diameter @xmath65 instead of the radius . for isotropic correlations ( [ eq : minkgaussexp ] ) becomes @xmath66 the shape of the weights @xmath67 is shown in fig . [ fig : integrationwindow ] . and their fourier transforms can be found in @xcite ( in the fourier form of @xmath68 in eq . ( 16 ) there is a typo : the prefactor of the integral should be @xmath69 instead of @xmath70 ) . ] of eq . ( [ eq : minkgaussexp-1 ] ) , for a ball radius of @xmath71 . all functions have been normalized to their maximal value . the functionals of higher index @xmath72 probe the correlation function at larger distances . ( same plot already in @xcite ) [ fig : integrationwindow],scaledwidth=45.0% ] let us just recall that the weights are such that @xmath73 and @xmath74 have a particularly interesting form . @xmath73 directly contains the amplitude of fluctuations on the scale of the balls @xmath75 , @xmath76 and @xmath74 is a simple integral over the two - point function @xmath77 which could therefore be recovered from the derivative of @xmath74 . let us pause here and elaborate a bit on the gauss - poisson process as such ( see @xcite for a more complete discussion of what follows ) . it is tempting to think of the truncated quantities ( [ eq : minkgaussexp-1 ] ) as being the mfs of a point distribution that represents the gaussian limit of an actual point distribution that has higher - order correlations . however , in general there will not be a gauss - poisson point process that actually realizes these mfs , as in order for a gauss - poisson process to exist , @xmath78 and @xmath24 have to fulfill certain conditions . already in the continuum case , the clustering amplitude is limited by the fact that the matter density can only be positive . the density distribution , therefore , is in any case only approximated by a gaussian . this approximation of course is , excellent for early times and large scales , but when matter fluctuations @xmath75 grow too large it breaks down . this is also the point where higher - order correlations become important . in the same manner , the clustering amplitude of a gauss - poisson point process is limited . in this case , however , also the mean density of the point sample @xmath24 plays a role . for low enough density a thinning of the point distribution goes to an approximated poisson distribution . this effect can compensate larger fluctuations and it can be shown @xcite that the corresponding constraint is @xmath79 where @xmath80 are the fluctuations in the number density of cells @xmath81 stemming from structure ( @xmath82 are the full fluctuations including the poisson term ) , @xmath83 is the volume of one of these cells , and @xmath84 the mean number of particles in such a cell . this constraint has two interesting implications : first , we can already tell from the amplitude of the two - point function if the field will have a significant amount of higher order clustering . when we evaluate ( [ eq : fluctconstraint ] ) for the structure that we measure in the boss galaxies , we find that the constraint actually is violated . second , to study the gaussian limit on large scales it is not sufficient to simply increase the domain size ( the ball size in our case ) . in some cases , where the fluctuations on a region decay faster than its volume the constraint indicates that the structure for larger cells becomes more gaussian . in general , however , one will have to reduce the number density . in the continuous case , by smoothing over larger and larger regions , the amplitude of the fluctuations between these regions decreases and the gaussian approximation becomes better and better . in the discrete case , the analogous coarse graining procedure involves a subsampling of the density field . in the same way that the number of regions is reduced in the continuous case when averaging the density over larger cells , individual points of the downsampled point distribution represent larger regions . we will demonstrate this effect in sec . [ sub : mfdensities ] where we show the mfs for a couple of densities . the lower this density becomes , the better the gaussian model fits . in addition to the requirement ( [ eq : fluctconstraint ] ) , @xcite also showed that the correlation function of a gauss - poisson point process has to satisfy @xmath85 and can not be negative @xmath86 any violation of ( [ eq : fluctconstraint ] ) ( [ eq : xiconst ] ) already implies the existence of higher - order correlations @xmath30 with @xmath87 . we note finally that the condition ( [ eq : fluctconstraint ] ) prevents ( [ eq : v0-sigma ] ) and therefore the volume fraction @xmath45 in ( [ eq : minkdensdef ] ) from becoming negative . so we could have derived a similar constraint on a gauss - poisson process from the requirement that @xmath88 . the procedure we use to measure the functionals is the same as in the dr7 analysis @xcite . as the quantities that are related to the hierarchy of correlation functions via ( [ eq : minkdensdef ] ) are , strictly speaking , the ensemble averages of the mf densities @xmath20 , we approximate these averages by creating randomly subsampled realizations of the point distribution and averaging over their mf densities . mathematically speaking there is no guarantee that these averages should converge to those of an ensemble of independent realizations ( see also @xcite ) , but practically the procedure is working quite well . for studies on small scales one could in principle use subfields of the full survey as independent realizations and perform an average that is then more likely to converge ( even though it also requires the assumption of ergodicity and is only valid approximately as neighboring regions are not completely independent ) . for the scales that we are interested in , however , there is only one realization of the survey and therefore we use the subsampling . for details on how we generate the samples see the end of sec . [ sub : radial ] . the calculation of the @xmath20 for a given downsampled realization is performed by chipmink ( code for high speed investigation of partial minkowski functionals ) , introduced in the dr7 paper @xcite . in a first step , it generates neighbor lists for every galaxy in the sample . galaxies are considered as neighbors if they are closer than twice the maximal radius of the balls . to determine @xmath89-@xmath90 , the code considers galaxies that are more than @xmath64 away from the boundary . using the list of neighbors the code calculates the uncovered surface , integrated mean curvature and euler characteristic of the ball put around the galaxy . these partial functionals are then summed to give the global @xmath20 . the volume used for obtaining the densities from the summed @xmath91 is the effective part that is more than @xmath64 away from the boundary . for the volume functional @xmath45 another set of neighbor lists is generated that contains all galaxies within @xmath92 around points from a random sample filling the same volume as the data , but also eventual holes in the data that are not treated as internal boundaries . around each primary random point we generate @xmath4 random points in spheres of increasing radius @xmath93 . the volume fraction covered by the balls is then estimated by counting the fraction of the secondary randoms that have a distance of less than @xmath23 to any of the galaxies in the neighbor list of the primary random point . as compared to the dr7 code , we made a couple of enhancements . besides improved memory management , the code can now handle boundaries of arbitrary shape in ra - dec . it can use data with non - constant density , uses a larger fraction of points for radii smaller than @xmath94 , and samples the @xmath95-plane more efficiently than the earlier version . we plan to make also the current version publicly available . having completed the discussion of the methods , we now turn to a description of the data . the galaxy sample we are analyzing was observed by the sloan digital sky survey ( sdss ) @xcite using the 2.5-meter sloan telescope @xcite . photometric data used for target selection was obtained using a drift - scanning mosaic ccd camera @xcite and information on the camera , photometry and photometric pipeline can be found in @xcite . all the photometry was re - processed and released in the eighth data release @xcite . based on the so obtained target catalog , the spectroscopic redshifts are taken by the spectrographs described in @xcite . the final step of spectroscopic data reduction and redshift determination is presented in @xcite . the data analyzed here is contained in the twelfth data release ( dr12 ) @xcite of sdss data and was obtained during the third stage of sdss @xcite within the baryon oscillation spectroscopic survey ( boss ) @xcite . it is described in detail in @xcite . the full data set consists of the spectroscopic redshifts of 836,347 galaxies . the area probed is divided in a northern and a southern part and two different target redshift ranges . for our analysis we use the largest contiguous part of the survey , which is the northern cmass sample . given the technical progress in the boundary treatment of our code , mentioned in sec . [ sub : samplingcode ] , it would now also be possible to use it for the combined sample analyzed in @xcite . the mock catalogs we are using for estimating errors and covariances were constructed using two different methods . one is based on the quick particle mesh technique ( qpm ) described in @xcite and the other one is a hybrid method @xcite which employs what the authors call `` augmented '' second - order lagrangian perturbation theory . it was used for generating the multidark(md)-patchy mocks @xcite . we shall work with both mock catalogs in our analysis , but concentrate on the md - patchy mocks . , @xmath96 and @xmath97 away from the nearest boundary.[fig : custommask],scaledwidth=45.0% ] for taking the data of the northern cmass sample , @xmath98 objects were targeted over an area of @xmath99 . out of these , @xmath100 turned out to be galaxies and obtained redshifts from good boss spectra @xcite . they are distributed over @xmath101 . their redshifts range approximately from @xmath102 to @xmath103 . the central region of redshift @xmath104-@xmath105 that is used in most of the cmass analysis ( see e.g. @xcite ) contains @xmath106 galaxies . the md - patchy and qpm mock catalogs are constructed such that they trace the number density of the data . they are available for the central range of @xmath104-@xmath105 . their angular distribution fills the full survey geometry . for matching the angular distribution of the data as well , regions of bad data quality or missing observations can be masked out . the qpm mocks contain on average @xmath107 galaxies . after application of the full veto mask the average number is @xmath108 . for the md - patchy mocks , the unmasked number of galaxies is @xmath109 and the masked one @xmath110 . our custom radial selection of a redshift range of @xmath111 , explained in section [ sub : radial ] , reduces the number of galaxies that we use to @xmath0 . in order to compare the data to the galaxy samples created in the mock catalogs , it would be sufficient to calculate the mfs for the masked mocks and the data . however , ideally we would like to derive quantities that are independent of effects of the survey boundary . this is not necessary for the direct comparison , but desirable for later comparison of the data with other simulations without having to apply the survey mask to them . when estimating the correlation function , the effect of boundaries ( inner boundaries such as holes as well as the outer boundary ) can be taken into account by using random samples with the exact same boundaries as the data . the excess of data over random pairs in a given bin is then solely due to structure . in the case of the mfs , unfortunately there is no simple analogue to this approach . we tried to correct the mfs of the masked mocks with the mfs of a random distribution and the same mask . a comparison with the unmasked mocks showed that this correction is not effective in the mf case . in order to minimize the influence of boundaries , we therefore choose to restrict our analysis to regions with good data quality . this is achieved by using only galaxies that are more than @xmath64 away from the closest boundary , where @xmath23 is the common radius of the balls around every galaxy ( see sec . [ sub : germ grain ] ) . of course , especially for large @xmath23 , this reduces the volume probed and so the number of galaxies that can be used . we therefore have to find a compromise as to which boundaries we include and which we can ignore without affecting the result too much . as discussed in detail in @xcite , there are at least seven different effects that lead to bad or missing data . for each effect there is a veto mask that can be used in the correlation function analysis . the effects are : * the central bolt obscures the center of the plate * ly@xmath112 targets are given priority in fiber assignment * no observations around bright stars * no observations around bright objects * non - photometric conditions * spectra with bad seeing * spectra degraded by extinction the regions obscured by the first three effects are quite small individually , but relatively evenly distributed over the sky . including them , and especially the @xmath64 region around them , would reduce the usable area by a large amount , so we can not correct for their effect . the fourth effect is also present all over the sky , however , the obscured regions have a wide range of sizes . therefore we can exclude the largest ones . . the points indicate the minimal value of the density of the 1000 qpm mocks in the respective bin . the reference density with respect to which we define our downsampling is @xmath113 . this combination of reference density and redshift range maximizes the number of downsampled galaxies in the boxed blue shaded region . we also use the points above the reference density by downsampling always from the full density.[fig : radialselection],scaledwidth=45.0% ] the combination of the lower four masks is shown as the black regions in fig . [ fig : custommask ] . of those regions we select those with an area on the sky of more than about @xmath114 . we densely sample these regions and their boundaries with points that are shown in blue . these are used to determine the distance of a galaxy from the nearest boundary . the projected holes created by ball diameters @xmath64 of @xmath115 , @xmath96 and @xmath116 are shown as the regions not filled by orange , blue and yellow color . at these diameters the effective volume left is @xmath117 , @xmath118 and @xmath119 , respectively . for the largest scales probed of @xmath120 , only @xmath121 remains . as the mask we constructed does not take into account all imperfections , we have to test the extent to which the result still depends on the survey boundaries . in order to study this , we removed the galaxies from the mocks that fall into one of the seven masks described above . then we determined the distance of the remaining galaxies from our boundaries that are shown in fig . [ fig : custommask ] . we calculated their germ - grain mfs and those of the galaxy in the corresponding cubic mocks periodically wrapped to remove any boundary . the resulting comparison is shown in fig . [ fig : maskcorrection ] . it also contains the case where we use the full veto mask to remove the unobserved galaxies , but only take the outer boundaries into account in the calculation of the mfs . we find that our custom mask greatly improves the match with the boundaryless case as compared to not taking inner boundaries into account . after the correction the periodic mfs are now within the error bars of the sky mfs . the sky values of @xmath73 in fig . [ fig : maskcorrection ] have been corrected for a systematic @xmath122 offset from the periodic case . we shall apply the same correction throughout the rest of the paper . as the offset has been derived from a comparison of the sky mocks to the periodic mocks , one could think of using the deviations of fig . [ fig : maskcorrection ] to apply density and scale dependent corrections to both the sky mocks and the data , to empirically remove the effect of the boundary . we decided not to apply this correction as it would make the results for the mfs of the data completely dependent on this particular set of mocks . from fig . [ fig : maskcorrection ] , we also see that our correction is largest for the fourth functional @xmath74 and least important for @xmath73 . this reflects the property of the integration kernel of fig . [ fig : integrationwindow ] . with increasing index @xmath72 , the functionals probe larger and larger scales and are hence more sensitive to the large scale boundary effects we removed . in addition to the angular selection , we also apply some radial cuts . as described in sec . [ sub : density - dependence ] we want to study the mfs of the galaxy distribution as a function of ball radius , but also sample density . a controlled variation of sample density requires a roughly constant density throughout the sample . otherwise the summation of partial mfs ( see sec . [ sub : samplingcode ] ) from different regions of the survey would combine @xmath20 values effectively measured at different densities , removing the clear distinction between the @xmath24 and the @xmath30 dependence in eq . ( [ eq : minkdensdef ] ) . to generate a homogeneous sample we downsample the density to a common value for all radial shells . the concept is shown in fig . [ fig : radialselection ] . -@xmath24-space . for the smaller diameters there are @xmath123 density points , for the largest diameter used in the analysis @xmath124 , there are @xmath125 . the lines of constant @xmath45 are calculated for a poisson point process and correspond to \{0.99 , 0.95 , 0.75 , 0.55 , 0.35 , 0.15 , 0.05 , 0.01 } from top to bottom.[fig : densitygrid],scaledwidth=45.0% ] there , we plot the binned density distribution of the qpm mocks . for each bin we show the minimum of the 1000 qpm mocks . using this density distribution , we choose a reference density such that the number of points in the constant density region of the considered redshift range is maximal . this density is @xmath126 and corresponds to @xmath127 galaxies . the total number of galaxies in this redshift range is @xmath0 . as we used the minimum of all mocks , we can be sure that we can downsample from this reference density and obtain the same constant density for every single mock realization . the sampling points in radius and density that we used are shown in fig . [ fig : densitygrid ] . for ball diameters up to @xmath128 , there are @xmath123 density points ranging from @xmath129 to @xmath130 of the reference density . for diameters above @xmath128 we decrease the maximal density and therefore the number of density points . the reason for this is two - fold : first , as discussed in sec . [ sub : densdef ] , for very small fractions of uncovered volume ( i.e. large radius and/or high density ) , the precision of the determination of @xmath45 is no longer high enough to give reasonable error bars for the derived quantities @xmath32 . second , the runtime is much worse if the galaxies have a large numbers of neighbors . the density cutoff in fig . [ fig : densitygrid ] is therefore roughly chosen along the lines of constant @xmath45 . [ fig : densitygrid ] includes these lines for the poisson case , which we used for a first estimate of reasonable density ranges . from the range of diameters shown in fig . [ fig : densitygrid ] , we choose @xmath131 to @xmath132 for our analysis . for smaller scales than @xmath131 we expect to pick up too much effects from the ignored holes in the survey and for larger scales than @xmath132 , the volume far enough from the boundary becomes too small . for our independent measurement of the correlation functions of the data and the mocks , we chose to use those regions that are more than @xmath133 away from the closest boundary that we treat . the structure that we measured for these regions is shown in fig . [ fig : meancorrfun ] . the concrete monte carlo procedure we use is then to randomly select galaxies from the full number of galaxies in our redshift range ( and _ not _ a pre - downsampled reference density as in the dr7 paper ) for each @xmath134-point of the grid in fig . [ fig : densitygrid ] , to obtain a sample with the constant target density in each radial bin . we then calculate the mfs for this configuration . we repeat the random selection and average the resulting mfs over these different realizations of the subsampled point distribution . the number of realizations is largest for the smallest densities and ranges from @xmath135 for a fraction of @xmath129 to @xmath136 for a fraction of @xmath130 . scatter of the mocks.[fig : meancorrfun],scaledwidth=45.0% ] in addition to the regions that could not be observed or had bad data quality and that therefore had to be masked out , there are also imperfections in the rest of the observed survey volume . even though the completeness of the survey is high , see @xcite , there are still galaxies that have not obtained robust redshifts ( no - z ) . in addition due to the physical size of the fibers of the spectrograph , galaxies separated by less than @xmath137 can not be observed simultaneously ( close pairs ) . in the correlation function analysis , these holes are accounted for by upweighting nearby galaxies . in our case this can not be done as easily . in principle one could use imaging data for the angular position and the redshift of nearby galaxies to mimic the procedure in the correlation function case and fill in the holes . however , as we are already ignoring a large number of holes and looking at relatively large scales , we decided not to take these holes into account and to mask them in the mocks consistently . however , we include another class of weights correcting for spurious density fluctuations due to stellar density and seeing . for the correlation function analysis , these have been combined into a systematic weight ( see @xcite ) . we use this systematic weight to adjust the probability of selecting a galaxy when we downsample to a given target density . as the modifications are below @xmath138 and not factors of @xmath139 to @xmath125 as in the case of the no - z and close pairs weights , the fluctuations can be accounted for relatively easily by this procedure . in this section we first present the measured values for the observables @xmath20 introduced in sec . [ sub : densdef ] . we study their radial and density dependence based on the md - patchy mocks . then we turn to the analysis of the higher - order contributions . for this analysis , we need the @xmath32 that are obtained from the measured @xmath20 through eqs . ( [ eq : inverse - mink ] ) . from the @xmath32 we can then simply subtract the two - point part of the expansion ( [ eq : minkcorrcon ] ) to extract the @xmath140 part ( [ eq : defmodminkhigher ] ) that only depends on higher orders . we will see that it contributes significantly to the full @xmath32 . finally we analyze the coefficients @xmath60 of ( [ eq : minkcorrcon ] ) order by order and compare them to correlation function integrals . as mentioned in sec . [ sub : samplingcode ] , the quantities calculated by chipmink are the the mf densities @xmath20 . their dependence on the ball diameter is shown in fig . [ fig : minfundens ] and based on the mean of @xmath141 md - patchy realizations . we only used @xmath141 mocks in this case because fig . [ fig : minfundens ] is mainly for illustration and the large range of diameters is not optimal for the evaluation of the @xmath32 as discussed at the end of sec . [ sub : densdef ] . in order to illustrate also the density dependence , we plot @xmath142 different densities ranging from @xmath129 to @xmath130 of our reference density of @xmath126 ( the same density points as in fig . [ fig : densitygrid ] , however with an increased maximal radius ) . for the most extreme cases of @xmath129 and @xmath130 , we also plot the mfs of a poisson point process as dashed lines . for the high density case , the functionals for the mocks are clearly non - poissonian , whereas in the low density case the downsampling erased already a lot of the structure and the difference to a poisson sample is less important . let us qualitatively describe the difference for the euler characteristic @xmath90 ( @xmath143 ) . the faster drop of the functional with structure indicates that the individual balls aggregate earlier , which reduces the number of independent components @xmath144 faster than for the poisson case . then the holes @xmath17 in the union of multiple balls prevail up to larger radii and in the end there are less cavities @xmath145 but at larger radii . for our quantitative analysis in sec . [ sec : higherorders ] we will mainly use the modified mfs @xmath32 , eq . ( [ eq : minkcorrcon ] ) . therefore , we use eqs . ( [ eq : inverse - mink ] ) to transform from the initial observables @xmath146 to @xmath147 . fig . [ fig : modminkfun ] shows the dependence of the @xmath32 on the ball diameter for the same densities as in fig . [ fig : minfundens ] , but now for 900 md - patchy mocks . in these plots the convergence of the functionals to those of the poisson point process with decreasing sample density is more visible than in fig . [ fig : minfundens ] . however , given the small error bars , even for the lowest density of @xmath148 the structure is well detected . the differing number of radial points for different densities illustrates our grid of points sampling the @xmath134 plane of fig . [ fig : densitygrid ] . for the highest density , @xmath125 radial bins of @xmath149 are possible until the survey volume is significantly filled , for the lowest density there are @xmath150 bins ( of which only the first @xmath151 are shown here ) . ccccccccc + & & & & & + + & @xmath152 & @xmath153 & & @xmath152 & @xmath154 & & @xmath152 & @xmath154 + @xmath155 & 1810 & 41.1 & & 171 & 10.1 & & 29.8 & 1.31 + @xmath156 & 1120 & 31.7 & & 88.1 & 5.93 & & 32.1 & 1.54 + @xmath157 & 454 & 19.1 & & 46.9 & 2.93 & & 29.6 & 1.28 + @xmath158 & 182 & 10.6 & & 29.7 & 1.30 & & 23.5 & 0.694 + we will now turn to the comparison of the density dependence of the modified mfs @xmath32 in the mocks and in the sdss data . instead of probing the grid of fig . [ fig : densitygrid ] for a fixed density as in fig . [ fig : modminkfun ] , we show in fig . [ fig : modminkdensdep ] the @xmath32 as a function of the sampling density for a ball diameter of @xmath133 . the form of @xmath32 in this direction of parameter space is easier to describe than in the @xmath159 direction , due to the power series form of the @xmath24-dependence , see eq . ( [ eq : powerseriesdecomp ] ) . in the plots of fig . [ fig : modminkdensdep ] , the poisson - case corresponds to the horizontal line at @xmath160 , thanks to the normalization of the @xmath32 in eq . ( [ eq : minkcorrcon ] ) . the data are clearly different from @xmath160 , showing that there is structure in the galaxy distribution . the straight lines that depart from the poisson case at zero density show the @xmath32 truncated after the term linear in @xmath24 , see eq . ( [ eq : minkgaussexp ] ) . they have been calculated by measuring the two - point correlation functions for each of the @xmath161 qpm and @xmath162 md - patchy mocks and integrating them with the corresponding weight functions @xmath163 of eq . ( [ eq : minkgaussexp-1 ] ) . the plot shows the average and the @xmath164 fluctuations between the different mock realizations . the average of the correlation functions that formed the basis of these lines is shown in fig . [ fig : meancorrfun ] . the clear deviation of the full @xmath32 of the data and the mocks from these lines shows that the integrals over higher - order correlation functions in the series expansion of @xmath32 actually do play a significant role for the structure in the sdss galaxies . this becomes even more apparent in the plots of fig . [ fig : modminksubtract ] . cccccc + & & & + + & @xmath152 & @xmath153 & & @xmath152 & @xmath154 + @xmath155 & 10300 & 97.3 & & 290 & 5.50 + @xmath156 & 11600 & 104 & & 282 & 5.22 + @xmath157 & 12300 & 107 & & 221 & 2.76 + @xmath158 & 12500 & 108 & & 252 & 4.04 + for these we use the integrated two - point correlations and subtract them off the corresponding @xmath32 for each mock individually . from these subtracted quantities @xmath140 ( eq . ( [ eq : defmodminkhigher ] ) ) we calculate the mean of the mocks and the covariances of the density points . the plots of the residuals show , that the higher - order contributions to the modified mfs are quite well described by the structure in the md - patchy mocks . the data happens to lie close to the @xmath164 line of the fluctuations between the @xmath162 mocks for all four mfs . it is quite encouraging that the mocks that were constructed to match the two- and three - point correlations only also capture what is going on in the higher orders quite well . as the data points in both fig . [ fig : modminkdensdep ] and fig . [ fig : modminksubtract ] are strongly correlated , we need to take into account the full covariance matrix in order to estimate the significance of the remaining difference . the results of this calculation are shown in tab . [ tab : chisqmockdat ] . for the curves of fig . [ fig : modminkdensdep ] and fig . [ fig : modminksubtract ] with @xmath123 density points , we first evaluate the @xmath3 values and then convert their @xmath165-values into those of a two sided deviation for a gaussian , which we express as multiples of one standard deviation . we inflate the error bars by multiplying the inverse covariance matrix by a factor of @xmath166 , in order to account for the finite number of mocks we use for estimating the covariance matrix @xcite . we find that the md - patchy mocks are a much better match of the data than the qpm mocks which are about @xmath167 away from the data in this case . for the higher components only , i.e. the central column of tab . [ tab : chisqmockdat ] , the agreement is even better and does not exceed @xmath168 for any of the functionals . so also quantitatively with covariances included the impression from the residual plots of fig . [ fig : modminksubtract ] that the typical deviation is around @xmath169 is correct . when we combine all the information of the relevant @xmath170 points in our sampling grid of fig . [ fig : densitygrid ] , i.e. ball diameters from @xmath131 to @xmath132 and their complete density dependence , we find in tab . [ tab : chisqradcomb ] that the deviation of the higher - order md - patchy modified mfs is at most @xmath171 . for some diameters as in tab . [ tab : chisqmockdat ] the agreement is nearly perfect . this is very encouraging , as the mocks have only been generated to match the two- and three - pt . given the combined values of tab . [ tab : chisqradcomb ] , it is clear that there is still some room for improvement in the construction of realistic mocks to match the higher orders even better . regarding the deviation of the @xmath140 from the case of vanishing higher - order correlations ( the zero line in fig . [ fig : modminksubtract ] ) , the conclusion is clear . there is strong evidence for higher - order correlations in the mocks and the data as seen by the mfs . even though this is not surprising , as at least the three - point function of the sdss data is well known ( see e.g. @xcite and @xcite ) , it is still interesting to see the statistical power that we have at hand for testing models with different higher - order correlations . for all of the quantities of tabs . [ tab : chisqmockdat ] and [ tab : chisqradcomb ] , one should keep in mind that the different mfs are not independent . so in order to derive one single value that would quantify the mismatch between mocks and data , one would also here have to take the covariances into account . for a single radius this is possible . for the full data set , we are limited by the number of mocks that we analyzed as an estimate of the @xmath172 covariance matrix would be noisy . finally , let us illustrate the scale dependence of the higher order contributions . [ fig : modminksubtractrad ] shows the deviation of the full @xmath32 ( [ eq : minkcorrcon ] ) from the two - point part ( [ eq : minkgaussexp ] ) , i.e. @xmath140 of ( [ eq : defmodminkhigher ] ) . so it is the same quantities as in fig . [ fig : modminksubtract ] , but now as a function of ball diameter for four fixed densities @xmath173 . the higher the density , the larger is the deviation from the two - point part , because the higher - order contributions play a more important role in the @xmath24-power series . in the radial direction , the functionals probe the higher - order correlations at different scales . for the two - point part we have seen in fig . [ fig : integrationwindow ] that the integration kernel picks up correlations at larger scales for increasing functional index @xmath72 . this is most probably also the case for the higher - order correlations ( we explicitly checked it for the three - point integrals ) and so @xmath174 probes the largest scales that we accessed in this analysis . the decreasing slope of the radial dependence reflects a similar behavior in the three point integral of @xmath175 . for even larger scales , the latter integral becomes negative and the difference to the two - point case decreases again . having found a significant contribution of higher - order correlations to the mfs , we now want to study which orders actually contribute . using the relation in eq . ( [ eq : minkcorrcon ] ) we can decompose the functionals in different contributions from higher orders by fitting a polynomial of order @xmath7 to the density dependence of the @xmath32 as described in sec . [ sub : density - dependence ] . this decomposition is not necessary for using the higher - order mf information in the comparison of the model to the data , as we have shown in the previous section , but it illustrates the convergence properties of the @xmath32 series . in fig . [ fig : higherfit ] we evaluate the convergence of the fit for increasing polynomial order @xmath7 . of eq . ( [ eq : minkintegr ] ) . to the @xmath24-dependence of @xmath43 at a diameter of @xmath176 we fit polynomials in @xmath24 of increasing order taking the full correlations of the points into account and extract the @xmath177 from the best fit coefficients . the linear to third order polynomials in the large panel give a terrible fit already visually . the fourth and fifth order in the `` residual '' part of the plot indicate convergence at 5.-order . the corresponding @xmath3 values can be found in tab . [ tab : higherfitchi ] and confirm this fact.[fig : higherfit],scaledwidth=45.0% ] the deviations of the fitted model from the data are quantified in tab . [ tab : higherfitchi ] . from both presentations it is clear that the linear term that only includes the two - point coefficient @xmath178 is a very bad fit to the density evolution . the table indicates that only after the inclusion of the fifth term , which corresponds to integrals over the six - point function , the polynomial fit captures the shape of the modified mfs well . however , this only means that we have some sensitivity of the mfs of the mocks to contributions from up to the six - point function . it does not mean that these contributions are also significantly different from the ones in the data . in fact , as we have found in sec . [ sub : highsubtract ] the higher - order contributions of the mocks and the data actually agree relatively well . in order to evaluate how strongly the coefficients @xmath60 of eq . ( [ eq : powerseriesdecomp ] ) are biased by truncating the power series at a finite @xmath7 , we plot in fig . [ fig : evwithfitorder ] the dependence of the mean of the best fit values of @xmath179 , @xmath180 and @xmath181 of the @xmath162 md - patchy mocks on the polynomial order @xmath7 for a fit at a diameter of @xmath133 . as high values of @xmath7 also increase the chance of picking up spurious fluctuations , we decided to put a prior on the higher order coefficients . to this end we fit a three - parameter pad model to the density evolution of the mean mfs extracted from @xmath6 periodic md - patchy cubes . taylor expanding this model provides some model for the higher - order coefficients . we choose a gaussian prior around the so obtained values with a standard deviation corresponding to the value of the mean @xmath182 . we only apply this prior to coefficients beyond the four - point function coefficient @xmath183 . the first two coefficients @xmath178 and @xmath184 are already well measured without any prior , and for @xmath183 it is sufficient to regularize the higher orders . fig . [ fig : evwithfitorder ] shows that the value of the first coefficient @xmath178 is independent of @xmath7 as soon as @xmath7 is high enough that we have a good fit , i.e. @xmath185 from fig . [ fig : higherfit ] . the second coefficient @xmath184 is also largely independent of fit order , but needs a higher fit order @xmath7 than @xmath178 to be unbiased . for @xmath183 there is still some remaining uncertainty from the fit order , which is , however , still much smaller than the error from the scatter of the mocks . [ fig : errorellipse ] shows evolution of the mean value together with the @xmath164 contours of the parameters @xmath180 and @xmath181 for different @xmath7 starting at @xmath186 , to include also the covariances between the parameters . and @xmath181 of the fit to @xmath187 of fig . [ fig : higherfit ] for a diameter of @xmath176 . the leftmost point is for a third order polynomial and the order is increasing from left to right with a maximum order of @xmath188 . the error bars indicate @xmath164 intervals . starting from the ninth order , error bars include all higher order points.[fig : errorellipse],scaledwidth=45.0% ] below the order @xmath189 that provides the first good fit in tab . [ tab : higherfitchi ] , the best fit values are strongly dependent on fit order . for orders starting at @xmath190 the shift becomes much smaller than the scatter of the mocks . for the following plots , we will use polynomials of order @xmath191 for the @xmath43 fit at @xmath192 . to determine the fit order for the radii with a smaller number of density points , i.e. beyond @xmath128 , we constructed and analyzed test data . these were created from a pad model fit to the cut sky mocks and provide a model with known coefficients . then we applied the fitting procedure to these test data and found that for high enough @xmath7 , the coefficients we put in were recovered within the errors estimated by the fluctuations from the mocks . from the results of these test data fits , we determined for every diameter the minimum @xmath7 needed . for the larger diameters this value is lower due to the decreasing contribution of higher orders . the reason for this decrease is the steep scale dependence for the higher - order coefficients , an example of which we will see in fig . [ fig : threeptcoeffs ] for the three - point term ( see also fig . 14 of @xcite for the scale dependence of the coefficients of @xmath73 for a log - normal model ) . the decomposition into @xmath7-point components also allows us to test to which extent the deviation of the data from the mocks in fig . [ fig : modminkdensdep ] could be simply the result of a misnormalization of the power spectrum . the green line in fig . [ fig : modminkdensdep ] therefore shows the mean of the md - patchy mocks for a reduction in the power of @xmath193 . it has been obtained by taking the coefficients of the @xmath9-th order fit of fig . [ fig : higherfit ] and multiplying the integrated correlation functions with the power of the reduction factor that corresponds to the number of power spectra involved , i.e. @xmath194 for @xmath35 . of course this is only a very simplified model , given that the fit is not unbiased , especially for higher orders , but it should still provide a useful approximation . from the result in fig . [ fig : higherfit ] we see that at least approximately it removes the offset of the data from the mocks and makes them agree even better . this seems to be consistent with the analysis of the power spectrum in @xcite . our redshift range roughly corresponds to their second redshift bin , for which their fig . 1 indicates that the data also is slightly lower than the mean of the md - patchy mocks . cccc + fit order & coeff . & @xmath152 & p - value + @xmath160 & @xmath179 & 7859 & @xmath195 + @xmath139 & @xmath180 & 1509 & @xmath196 + @xmath197 & @xmath181 & 306 & @xmath198 + @xmath199 & @xmath200 & 47 & @xmath201 + @xmath189 & @xmath202 & 11.6 & @xmath203 + for the two cases @xmath178 and @xmath184 where the convergence in fig . [ fig : evwithfitorder ] is convincing , we compare in figs . [ fig : twoptcoeffs ] and [ fig : threeptcoeffs ] the extracted coefficients to reference models . [ fig : twoptcoeffs ] shows the two - point coefficients @xmath178 . for those , @xmath204 coincides with @xmath205 ( see eq . ( [ eq : v0-sigma ] ) ) and @xmath206 corresponds to the volume averaged correlation function ( see eq . ( [ eq : v3-xi ] ) ) . in addition to the average of the coefficients extracted from the md - patchy mocks , we show the coefficients of the data and the expected coefficients derived from integrals over the average two - point correlation function that we measured independently . the overall agreement of the two independent determinations is reassuring . it provides a confirmation that the fit to @xmath56 can extract individual terms from the power series ( [ eq : minkcorrcon ] ) . there is , however , for some of the functionals a relatively large offset . it is not clear if this offset is due to residual bias in the fitting method , or if our method to estimate the correlation function of a mock does not include the boundaries in the same way than the mfs . [ fig : threeptcoeffs ] then shows the @xmath184 . the connection to other known quantities is not as apparent as in the two - point case and we did not obtain an independent estimate of the three - point correlation integrals . what is shown in fig . [ fig : threeptcoeffs ] instead is the three - point function model of @xcite . it is based on leading - order perturbation theory and includes redshift space distortions . for the plots in fig . [ fig : threeptcoeffs ] we fixed the three bias parameters to @xmath207 , @xmath208 and @xmath209 . the @xmath210-value is the one we used in fig . [ fig : twoptcoeffs ] and the @xmath211-value is obtained from eq . ( 23 ) in @xcite and based on local lagrangian biasing @xcite . we did not attempt to fit for @xmath212 and so the lines in fig . [ fig : threeptcoeffs ] do not contain any free parameter . given that , the match of the model with the extracted coefficients is encouraging . in addition to being related to the hierarchy of @xmath7-point correlation functions , the mfs also contain information about the cosmic web . for example , the partial functionals probe the local morphology around the galaxy sites , and for an isodensity contour smoothing `` shapefinders '' have been defined to quantify the amount of filamentary vs. planarity of the local density field @xcite . for the global germ - grain mf densities that we study here , there are different quantities that can be extracted . the euler characteristic @xmath213 , for example , indicates for large balls how many extreme voids ( i.e. regions that are further away from any galaxy than @xmath92 ) there are in the galaxy distribution . the area functional @xmath89 gives an idea of their area and @xmath45 their volume . more generally , the comparison of the void fraction @xmath46 to the one of a poisson distribution of the same density indicates how inhomogeneously these void regions are distributed . in fig . [ fig : freevolumefraction ] , we therefore show @xmath46 as a function of the ball diameter and compare it to the corresponding poisson case . for a ball of diameter @xmath159 , where the first values correspond to the lower ( blue ) points and the values restarting at @xmath197 to the upper ( red ) points . the densities are @xmath214 and @xmath215 . especially for the higher density the solid poisson lines lie way below the values for the data , so the structure leads to a significantly higher void fraction than we would expect in an unclustered field.[fig : freevolumefraction],scaledwidth=45.0% ] especially for the higher density of @xmath216 the difference is quite remarkable . for a diameter of @xmath128 , the volume is a factor of @xmath217 larger than in the poisson case . for the most extreme case of @xmath132 , it is nearly @xmath218 . in this latter case the expected number of galaxies in a ball of diameter @xmath132 would be @xmath219 . with the standard poisson fluctuation one would expect a fraction of @xmath220 to be uncovered . including the observed structure however , there is @xmath221 of the @xmath222 in our analyzed region that remain uncovered . of course one has to make sure that this conclusion is not significantly affected by the holes in the survey . we tested this using the periodic cubic md - patchy mocks and plot the resulting void fractions as purple and green points in fig . [ fig : freevolumefraction ] . for the region that we evaluated , the cubic mocks for the difference is small , below @xmath223 for @xmath224 and below @xmath225 for @xmath226 . however , for the more extreme diameters the deviations will grow , but we do not expect that this affects the general picture . in this paper , we perform a new analysis that combines minkowski functionals with the standard two - point correlation function to measure the amount of non - gaussianity of the density field . we find in sec . [ sub : highsubtract ] that the significance of the higher - order contribution to the mfs is of the order of @xmath227 ( see tab . [ tab : chisqradcomb ] and fig . [ fig : modminksubtract ] ) . this demonstrates that this part of the mfs can be well enough measured to give meaningful constraints on the higher - order properties of the galaxy distribution . given this constraining power it is reassuring that we find from fig . [ fig : modminksubtract ] and tab . [ tab : chisqmockdat ] that the md - patchy mocks , that were constructed to match the two- and three - point properties of the measured galaxy distribution , match fairly well to the higher - order mfs . this indicates that as far as the higher - orders accessible by the mfs are concerned , the mock galaxy catalogs are quite good . there is still some room for improvement as the combined deviation is as large as @xmath228 for some functionals ( see tab . [ tab : chisqradcomb ] ) , but this might be addressed by a @xmath193 shift in the normalization of the mock power spectrum ( see fig . [ fig : modminkdensdep ] ) . on the technical side , our careful boundary treatment minimizes the effects of the survey geometry on our measurement of the modified mfs ( see fig . [ fig : maskcorrection ] ) . the corrected values lie within the error bars of a boundaryless periodic box . for the qpm mocks , we find a much larger deviation ( beyond @xmath229 ) from the data than for md - patchy ( see tab . [ tab : chisqmockdat ] ) . as in this case already the two - point correlations are overnormalized in the range of scales that we are interested in ( see fig . [ fig : meancorrfun ] ) , a reduction of the amplitude of the initial gaussian field could also help in this case . we show that in the the higher - order part of the mfs calculated for the sdss dr12 cmass galaxies contains at least information up to the integrated @xmath9-point function ( see tab . [ tab : higherfitchi ] ) . this underlines the power of the mfs to bring condensed higher - order information into a simple functional form . aw thanks antonio cuesta for providing support with the qpm correlation functions . we thank graziano rossi and thomas buchert for valuable comments on the manuscript . the work of aw was supported by the german research organization dfg , grant no . wi 4501/1 - 1 . dje is supported by u.s . department of energy grant de - sc0013718 and as a simons foundation investigator . 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 . t. , 1995 , in mcket j. p. , gottlber s. , mller v. , eds , large scale structure in the universe robust morphological measures for large scale structure . p. 156 , http://arxiv.org/abs/astro-ph/9412061[astro-ph/9412061 ] , http://adsabs.harvard.edu/abs/1995lssu.conf..156b[ads ] hadwiger h. , 1957 , vorlesungen ber inhalt , oberflche und isoperimetrie . die grundlehren der mathematischen wissenschaften in einzeldarstellungen mit besonderer bercksichtigung der anwendungsgebiete ; 93 , springer , http://cosmologist.info/isbn/3-540-02151-5 , [ isbn : 3 - 540 - 02151 - 5 , ] m. , schmalzing j. , buchert t. , wagner h. 1996 , in weiss a. , raffelt g. , hillebrandt w. , von feilitzsch f. , buchert t. , eds , astro - particle physics the significance of the fluctuations in the iras 1.2 jy galaxy catalogue .. pp 8398 , http://adsabs.harvard.edu/abs/1996app..conf...83k[ads ] r. et al . , 2001 , in harnden jr . f. r. , primini f. a. , payne h. e. , eds , astronomical data analysis software and systems x vol . 238 of astronomical society of the pacific conference series , the sdss imaging pipelines . p. 269 , http://arxiv.org/abs/astro-ph/0101420[astro-ph/0101420 ] , http://adsabs.harvard.edu/abs/2001aspc..238..269l[ads ] m. c. et al . , 2011 , in american astronomical society meeting abstracts # 218 the log - density as a better cosmological density variable . p. # 233.04 , http://adsabs.harvard.edu/abs/2011aas...21823304n[ads ] m. , buchert t. 1996 , in weiss a. , raffelt g. , hillebrandt w. , von feilitzsch f. , buchert t. , eds , astro - particle physics applications of minkowski - functionals to the statistical analysis of dark matter models . , http://arxiv.org/abs/astro-ph/9509014[astro-ph/9509014 ] , http://adsabs.harvard.edu/abs/1996app..conf..251p[ads ] j. , kerscher m. , buchert t. 1996 , in bonometto s. , primack j. r. , provenzale a. , eds , dark matter in the universe minkowski functionals in cosmology . p. 281 , http://arxiv.org/abs/astro-ph/9508154[astro-ph/9508154 ] , http://adsabs.harvard.edu/abs/1996dmu..conf..281s[ads ]

recently considerable effort has been devoted to understand the isospin physics because of its paramount importance in determining the structure and dynamics of finite nuclei @xcite and also in the astrophysical contexts such as supernova dynamics , protoneutron star evolutions and the structure and properties of neutron star @xcite . in stable nuclei , the neutron - proton asymmetry , @xmath0 , is about 0.2 , where @xmath1 and @xmath2 are the neutron and proton number densities . in forthcoming rare - isotope accelerator experiments the magnitude of @xmath3 could be well above 0.24 . in contrast , a large @xmath4 is reached in the dense interior of neutron stars @xcite . whereas , very small asymmetry of @xmath5 is attained near the crust of ( proto)neutron stars at densities @xmath6-@xmath7 @xmath8 @xcite . in order to understand the composition and structure of stars over this extreme values of density and asymmetry requires precise knowledge of the density dependence of nuclear symmetry energy @xmath9 . several theoretical studies have demonstrated @xcite that @xmath9 represents the leading coefficient of an expansion of the total energy , @xmath10 ) , for a neutron - rich system with respect to asymmetry : @xmath11 . the symmetric nuclear matter equation of state , @xmath12 , has been already well constrained over a wide density range from analysis of a large body of data within non - relativistic models @xcite and relativistic models @xcite . in contrast , though the nuclear symmetry energy at the normal nuclear matter density @xmath13 @xmath8 is well constrained to @xmath14 mev , its value and trend at lower and especially at higher densities are poorly known @xcite . various theoretical calculations based on microscopic @xcite and/or phenomenological many - body approaches @xcite , though reproduce the well - constrained value of @xmath15 , their high density predictions of symmetry energy are extremely diverse . while certain models predict a monotonically increasing symmetry energy with density , other model results show that @xmath9 increases initially up to about @xmath16 and decreases thereafter . these diverse predictions in the density dependence in the symmetry energy lead to large uncertainty in the incompressibility of symmetric nuclear matter which mainly stems due to insufficient knowledge about the isospin dependence of in - medium nuclear effective interactions . recently some progress has been achieved by consistently constraining the symmetry energy to @xmath17 at subsaturation densities from analysis of isospin diffusion @xcite and isoscaling @xcite data in intermediate energy heavy ion collisions and from the study of neutron skin of several nuclei @xcite . within a relativistic mean - field ( rmf ) model @xcite , the symmetry energy has been constrained to a small range @xcite from analysis of neutron skin thickness of several nuclei across the periodic table . further , the slope parameter and the incompressibility for the constrained @xmath9 was found @xcite consistent with that extracted from isospin diffusion and isoscaling data . in an effort to pindown the precise density dependence of symmetry energy , it has been noted @xcite that the skin thickness of neutron - rich heavy nucleus , for example @xmath18pb , are linearly correlated with the symmetry pressure at a density slightly below the normal nuclear matter value . since the same pressure supports a neutron star at supranormal densities against gravitational collapse , suggests an underlying correlation between neutron skin of heavy nuclei and neutron star radii @xcite . within the rmf models , the constrained @xmath9 demonstrated @xcite that the neutron skin in @xmath18pb tend to yield smaller neutron star radii . furthermore , within the constrained range of symmetry energy , significant sensitivity on several features of liquid - gas phase transition @xcite in hot asymmetric nuclear matter was found @xcite . in particular , the boundary and area of the liquid - gas coexistence region , the maximal isospin asymmetry and the critical values of pressure and isospin asymmetry all of which systematically increase with increasing overall softness of @xmath9 . it is thus instructive to explore the effects of symmetry energy in the rmf model , that has been limited to a narrow range from analysis of skin data , on the composition and structure of hot protoneutron and cold neutron star matter that encompasses a large density range . a newly born neutron star or protoneutron star ( pns ) @xcite is formed in the aftermath of a successful supernova explosion and subsequent gravitational collapse of the core of a massive star . due to short mean free path of neutrinos of @xmath19 cm ( compared to pns radii of about 10 km ) , the neutrinos are trapped temporarily within the neutrino - sphere of the star . the pns are thus rich in leptons , mostly @xmath20 and @xmath21 , and quite hot with central temperatures of @xmath22 mev . thus the star matter has the interesting possibility to undergo liquid - gas phase transition close to the surface of the star @xcite , the characteristics of which is mainly controlled by the symmetry energy at subsaturation densities . in the much longer kelvin - helmholtz evolution stage of about 10 sec , the neutrinos escape from the interior , and the hot and lepton - rich pns changes into a cold and deleptonized neutron star . the loss of neutrinos enforce the electrons and protons to combine resulting in a neutron - rich neutron star matter . consequently , the symmetry energy at supranormal densities plays an important role in determining the composition , in particular the possible presence of strange hyperons , the structure ( mass and radius ) and the long - term cooling of the neutron star . in this paper we study the role of isospin on the liquid - gas phase transition in hot lepton - rich protoneutron star and the properties of ( proto)neutron star within the accurately calibrated parameter set fsugold @xcite in the relativistic mean - field ( rmf ) model . for this purpose , we restrict the model density dependence of symmetry energy @xmath9 within a narrow range that reproduced the skin thickness of several nuclei @xcite . at the supranormal densities the contribution of the hyperon degrees of freedom to the equation of state of the star is considered . at very low densities well below the saturation value , that is relevant to the crust , few - body correlations become important . the system can then minimize its energy via the formation of medium - modified light clusters @xcite which eventually dissolve into homogeneous nuclear matter at higher densities due to pauli blocking . the paper is organized as follows : in sec . ii we introduce the rmf model for baryons along with the medium - modified clusters . the construction of the liquid - gas phase transition in the hot protoneutron star matter is presented . in sec . iii the numerical results for the structure and composition of the star matter are discussed . iv summarizes the results . we consider the relativistic mean - field ( rmf ) model with all the charge states of the baryon octet ( @xmath23 ) and their antiparticles . light clusters ( @xmath24 ) , that are treated as quasi - particles , are also included as explicit degrees of freedom . the properties of these clusters are modified by medium effects via the temperature - dependent shifts in the binding energies as treated in ref . @xcite . the interaction lagrangian density in the nonlinear rmf model @xcite for a system of baryons , clusters and leptons for protoneutron star matter is given by @xmath25 where @xmath26 is spin-1/2 fields for the particles ( @xmath27 ) , @xmath28 is the spin-0 field for the alpha particle , and @xmath29 is the spin-1 field for deuteron all of which are coupled to the @xmath30 mesons . the field strength tensors for the vector mesons are @xmath31 and @xmath32 . the nonlinear @xmath33-meson couplings ( @xmath34 ) soften the symmetric nuclear matter equation of state ( eos ) at around @xmath16 , while its high density part is softened by the self - interactions ( @xmath35 ) for the @xmath36-meson field . the sum on @xmath37 is over the non - interacting electrons and muons ( @xmath20 and @xmath38 ) and their antiparticles in the star . the covariant derivative @xmath39 of the particle @xmath40 with mass , proton , neutron numbers ( @xmath41 ) involves interaction with the vector mesons . of course for hyperons ( @xmath42 ) the ( @xmath41 ) drop out in eq . ( [ covd ] ) and hereafter . @xmath43 is the isospin projection of the baryon and cluster charge state @xmath40 . the effective mass of the @xmath40th species is @xmath44 where the vacuum rest mass of the clusters @xmath45 is given by @xmath46 . for the vacuum binding energy @xmath47 of the clusters we adopt the experimentally measured values . note the clusters which are predominant at densities @xmath48 are treated as structureless point particles . the scalar isovector @xmath49 ( the @xmath3 meson ) that splits the effective masses of the baryons has not been included in the present study as its effects on the star properties were found to be minimal @xcite . the medium dependent shift of the binding energy @xmath50 that appears only for the clusters are adopted from ref . @xcite . for nuclei embedded in nuclear matter , the shift has been calculated by solving the in - medium schrdinger equation which leads to the ground state energy @xcite @xmath51 the shift due to self - energy , @xmath52 , is already contained in the effective particle mass in the rmf model . the coulomb shift , @xmath53 , that can be evaluated within the wigner - seitz approximation gives a small value for the light clusters and thus neglected . the dominant contribution emerge from the pauli shift , @xmath54 , that has been evaluated in the perturbation theory with jastrow and gaussian approaches for @xmath55 and ( @xmath56 ) , respectively . we adopt the empirical form for ref . @xcite @xmath57 \delta b_i(t ) , \ ] ] where @xmath58 and @xmath59 . the quantity @xmath60 can be found from eqs . ( 26)-(27 ) in ref . @xcite . with increasing temperature and/or density a cluster will dissolve when its total binding energy @xmath61 vanishes . in the above expression the total densities for neutrons , @xmath62 , and protons , @xmath63 are contributions from baryons ( @xmath64 ) and clusters ( @xmath65 ) . the composition of the star is determined by the requirements of charge neutrality and @xmath66-equilibrium conditions @xmath67 , @xmath68 , where @xmath69 and @xmath70 are the baryons and @xmath37 is a lepton ( @xmath20 or @xmath38 ) or their respective antiparticles at finite temperature . with clusters ( @xmath71 ) , these weak processes correspond to @xmath72 and @xmath73 . the chemical potential of a species @xmath40 with baryon number @xmath74 , charge @xmath75 , and lepton number @xmath76 in chemical equilibrium is in general @xmath77 the three independent chemical potentials @xmath78 , @xmath79 and @xmath80 can be determined from total baryon number , charge , and lepton number conservations leading to @xmath81 for a deleptonized neutron star , the ( anti)neutrinos escape freely so that @xmath82 . the remaining chemical potentials , @xmath78 and @xmath79 , can be obtained from conservations of total baryon and charge densities @xmath83 when neutrinos are trapped , the matter ( inside the neutrino - sphere ) on a dynamical time scale is characterized by a fixed lepton fraction ( number of leptons per baryon ) , @xmath84 , for each flavor @xmath85 . in fact , muons are absent in a pns that leads to @xmath86 and @xmath87 in eq . ( [ chem ] ) . consequently , the composition of a hot pns is characterized by a constant value of @xmath88 , which in the late stage reaches a value of @xmath89 . the number density for a fermionic species @xmath90 at temperature @xmath91 @xmath92 , \ ] ] where @xmath93 is the spin of the @xmath40th species and the distribution function for particles and antiparticles ( referred to as @xmath94 sign ) is as usual @xcite @xmath95^{-1}.\ ] ] the density of the bosonic species @xmath96 is obtained from @xcite @xmath97 where @xmath98 stems from the condensed bosons in the ground state . the bose - einstein distribution follows as @xmath99^{-1 } .\ ] ] for interacting baryons and clusters , the effective energy and chemical potentials are respectively @xmath100 and @xmath101 , where @xmath102 and @xmath103 are the timelike and isospin three component of the vector @xmath36-meson and vector - isovector @xmath104-meson . for field - free lepton , these quantities reduce to @xmath105 and @xmath106 . the field equations for the mesons ( @xmath107 ) are derived in the usual way @xcite for the lagrangian of eq . ( [ rmf ] ) . herein the contribution from the cluster densities to the source terms in the meson fields are consistently incorporated by noting that in the present lagrangian the couplings @xmath108 of the meson fields to the baryons and clusters ( @xmath109 ) are constant ( density independent ) unlike that in ref . @xcite . at finite temperature and baryon density , the energy density and pressure corresponding to the lagrangian of eq . ( [ rmf ] ) can be obtained from the thermodynamical potential @xmath110 @xcite as @xmath111 \nonumber\\ & & + \sum_{i = d,\alpha } \left [ \frac{2j_i+1}{(2\pi)^3 } \int d^3k \ : e^*_i(k ) g_i(k ) + m^*_i\tilde{\rho}_i \right ] , \end{aligned}\ ] ] @xmath112 \nonumber\\ & & + \sum_{i = d,\alpha } \frac{2j_i+1}{3(2\pi)^3 } \int d^3k \ : \frac{k^2}{e^*_i(k ) } g_i(k ) .\end{aligned}\ ] ] the rmf model is applied to study the liquid - phase ( lgp ) transition in hot protoneutron star matter . such a coexistence phase may occur at subsaturation densities where the star is composed of nucleons ( @xmath113 ) , clusters ( @xmath65 ) and leptons only . a single - phase system at temperature @xmath114 and density @xmath104 is stable against lgp separation if its free energy density @xmath115 is lower than the coexistence liquid ( @xmath116 ) and gas ( @xmath117 ) phase : @xmath118 the total baryon density , @xmath119 , and the isospin density , @xmath120 , for a @xmath113-@xmath65-@xmath121-@xmath21 system satisfy @xmath122 \rho_3 & = & ( 1-\chi ) \rho_3^l + \chi \rho_3^g , \end{aligned}\ ] ] where @xmath123 being the fraction of the total volume occupied by the gas phase . the stability condition , eq . ( [ free ] ) , implies the inequalities @xcite @xmath124 and @xmath125 . these inequalities are equivalent to @xmath126 \left(\frac{\partial \mu_p}{\partial \delta}\right)_{t , p } < 0 ~ & { \rm or } & ~ \left(\frac{\partial \mu_n}{\partial \delta}\right)_{t , p } > 0 . \label{chm}\end{aligned}\ ] ] the first inequality indicates mechanical stability which means a system at positive isothermal compressibility remains stable at all densities . the second inequality corresponds to chemical stability which suggests that energy is required to alter the concentration in a stable system while maintaining temperature and pressure fixed . when one of these conditions get violated , a two - phase system is energetically favorable . the coexistence liquid - gas phase is governed by the gibbs criteria for equal pressures and chemical potentials in the two phases with different densities but at the same temperature : @xmath127 \mu_i(t,\rho^l ) & = & \mu_i(t,\rho^g ) , \label{gbs2 } \end{aligned}\ ] ] where the chemical equilibrium condition refers to all the species ( @xmath128 ) in the system . since we are dealing with free leptons , it is evident from eq . ( [ gbs2 ] ) that the lepton density in the liquid and gas phases are identical and thus forms a common background in the baryonic matter . else for interacting leptons , @xmath129 , and thus the lepton number conservation would result in an additional condition similar to those of eq . ( [ den ] ) . the original fsugold @xcite , with an isovector coupling @xmath130 , produces a soft symmetric nuclear eos with an incompressibility @xmath131 mev . in addition to the ground state properties of several nuclei , this model reproduces the strengths of the giant monopole resonance in @xmath132zr and @xmath18pb . the nucleonic matter symmetry energy for this lagrangian at temperature @xmath133 is given by @xmath134 where @xmath135 , and the fermi momentum and effective mass for nucleons are @xmath136 and @xmath137 , respectively . the effective @xmath104-meson mass is @xmath138 . in asymmetric nuclear matter , the density dependence of symmetry energy has been studied @xcite by simultaneously varying @xmath139 and @xmath140 so that for all combinations of ( @xmath139 , @xmath140 ) the symmetry energy is fixed at @xmath141 mev at an average density @xmath142 corresponding to @xmath143 @xmath144 . note the symmetry energy even at the saturation density has not been accurately constrained experimentally and only an average @xmath145 at @xmath16 is constrained by the binding energy of nuclei . thus by following the above prescription @xcite one ensures accurate binding energies and proton densities of heavy nuclei such as @xmath18pb . in fact , only those @xmath139 values were selected that reproduced the two measured observables mentioned above . for the present study we adopt @xmath146 since the corresponding @xmath147 and their slopes and curvatures are in reasonable agreement with that extracted from neutron skin thickness of several nuclei as well as the isoscaling and isospin diffusion data @xcite . it may be noted that with increasing @xmath139 the density dependence of symmetry energy becomes _ overall _ softer , i.e. , softer at supranormal densities and stiffer at @xmath148 . within this limited @xmath139 range , considerable effect on the features of liquid - gas phase transition was also observed in hot asymmetric nuclear matter @xcite . the hyperon - meson coupling constants , @xmath149 and @xmath150 , are strongly correlated by the potential depth of the hyperon @xmath151 in saturated nuclear matter @xcite as @xmath152 we use su(6 ) symmetry for the vector coupling constants @xmath153 the @xmath149 couplings are adjusted to reproduce the estimated hypernuclear potential depth @xmath154 at @xmath16 . from detailed studies of @xmath155 interaction , an attractive potential depth of @xmath156 mev has been estimated @xcite so that the observed bound @xmath157 hypernuclear states are reproduced . several @xmath158 hypernuclear states @xcite and free @xmath158s produced in reactions @xcite suggest an attractive potential of @xmath159 mev . though the @xmath157 and @xmath158 potential depths have been well studied , however , data on @xmath160 hypernuclei are scarce and ambiguous due to strong @xmath161 decay . while previous studies of hyperon stars have used an attractive @xmath160 potential ( with @xmath162 mev ) that are based on the observed bound @xmath163he hypernucleus , recent analysis @xcite of @xmath164 atomic data point to a sizable repulsive nuclear @xmath160 potential of depth @xmath165 mev . for the present study we adopt this repulsive depth of the @xmath160 potential . this ambiguity will modify considerably the hypernuclear composition in the star , which however , has smaller effects on the bulk properties of @xmath66-equilibrated star matter . we begin this section by presenting in fig . [ yp ] the fractions , @xmath166 , of free protons and clusters @xmath167 at various temperatures for symmetric nuclear matter ( snm : left panels ) and neutrino - trapped protoneutron star ( pns : right panels ) matter in the fsugold relativistic mean - field model for the isovector coupling @xmath168 . the overall trend exhibited by symmetric matter is quite similar to that for neutron - rich protoneutron star matter . at very low densities @xmath169 where the meson fields and the baryon chemical potentials drop rapidly the pns behaves as snm with @xmath170 ( see also fig . [ ypns ] ) . with the formation of the clusters , the threshold density of which is quite sensitive to the system temperature , the free proton fraction decreases rapidly . at a given temperature , the two - nucleon correlation results in the formation of deuteron first , followed by three - particle @xmath171 and finally the @xmath172 particle . at @xmath173 mev , the strongly bound alpha is seen to be the most abundant particle over a wide density range as many - body correlations gain prominence with density . moreover , with increasing temperature , the @xmath172 fraction decreases while other cluster abundances increase ; the peak positions gradually shifting to higher densities . consequently with increasing @xmath114 , the minimum of @xmath174 increases and shifts to higher densities and then decreases again at @xmath175 mev when deuterons dominate . this was also observed in the rmf model with density - dependent couplings @xcite . note that at a given temperature and density the ( free ) neutron - rich supernova matter has somewhat smaller cluster fractions compared to the symmetric nuclear matter . the clusters eventually dissolve due to pauli blocking at higher densities @xmath176 @xmath8 resulting in a homogeneous matter with @xmath170 for snm ( left panel ) and a neutron - rich @xmath177-@xmath178-@xmath121-@xmath179 pns matter ( right panel ) . we now consider the possibility of liquid - gas phase ( lgp ) transition in hot and neutrino - trapped protoneutron star consisting of neutrons , protons , electrons and electron neutrino . figure [ pbound ] shows the boundary for the existence of the coexistence phase at various temperatures in the fsugold model for the isovector coupling @xmath180 . the results are for electron lepton fraction @xmath181 ( and muon lepton number @xmath86 ) which approximately corresponds to the value at the onset of trapping . at a nucleon density of @xmath182 @xmath8 , the critical temperature for @xmath130 is @xmath183 mev which is close to that for symmetric nuclear matter @xcite . whereas a softer nuclear symmetry energy @xmath145 at subsaturation density ( @xmath184 ) , gives a somewhat smaller @xmath185 . in fact , the critical temperature divide the coexistence boundary into a low - density gas phase ( on the left ) and a high - density liquid phase ( on the right ) and the liquid - gas phase coexist within the boundary . due to trapping , the net electron concentration is quite high , and the matter at the low baryon density phase is controlled by non - degenerate electrons . as a consequence , the nuclear symmetry energy corresponding to @xmath184 and 0.03 has practically no effect on the gas phase boundary . on the other hand , due to dominant contribution of the degenerate nucleons in the relatively high - density liquid phase , larger effects from symmetric pressure in the stiffer @xmath145 for @xmath130 , shifts the liquid - phase boundary to a slightly higher density ( see inset of fig . [ pbound ] ) . the protoneutron star matter is found to exhibit a lgp transition even at very low densities at a high temperature of @xmath186 mev . with the inclusion of light clusters ( solid line in fig . [ pbound ] ) , the softening of the eos shifts the low - density coexistence phase boundary to a higher value at temperatures @xmath187 mev where clusters become abundant . on the other hand , the high - density boundary that is dominated by free nucleons is relatively unaffected . thus the clusters effectively reduce the width of the phase coexistence region and increases the critical temperature to @xmath188 mev for @xmath130 case . in fig . [ ypns ] , the particle fractions are shown for neutrino trapped matter at an electron lepton fraction @xmath189 at temperatures of @xmath190 and 2 mev for uniform matter ( solid line ) and in the coexistence liquid - gas phase ( dashed line ) . one of the major effects of trapping is to enhance the net electron fraction @xmath191 over the entire baryon density range ; the magnitude of @xmath191 gradually increases with @xmath192 . considering first uniform matter ( solid line ) , at @xmath169 @xmath8 , the energy and pressure of the hot protoneutron star matter is entirely dominated by the non - degenerate electrons because of their small mass . due to charge neutrality , the rapid rise of net electrons at this low density enforces an equally large enhancement of net number of free protons and a consequent drop in free neutron density due to baryon number conservation ( see figs . [ ypns](a ) and [ ypns](c ) ) . this tends to make the lepton - rich matter in this density regime highly symmetric with identical free neutron and proton fractions . for uniform matter at higher densities , clusters dominate the particle yield ( compared to free protons and neutrons especially at high @xmath114 ; ( figs . [ ypns](b ) and [ ypns](d ) ) so that the charge neutrality is maintained primarily by the bound protons in these clusters . compared to @xmath193 ( fig . [ yp ] ) , larger electron fraction in the present @xmath194 case enhances the cluster production via charge neutrality . at all temperatures of interest in the supernova dynamics , we find @xmath172 and deuteron to be the dominant clusters . thus , within this density range the star matter is also highly symmetric but due to abundant symmetric cluster formation . with the onset of liquid - gas phase transition , it is important to note from fig . [ ypns ] that cluster formation drops drastically ( dashed lines ) . this stems from vanishing cluster contribution from the liquid part of the mixed phase that has a density well above the transition value where the clusters dissolve , as well as due to small cluster fraction in the gas phase that has a density smaller compared to the total @xmath104 . thus at small @xmath195 mev in the coexistence phase , @xmath174 is effectively enhanced and becomes identical to @xmath191 . this leads to the formation of @xmath177-@xmath178-@xmath121-@xmath179 matter that tends to be rather symmetric even in the coexistence phase . however , with increasing @xmath114 , as the cluster fractions are large and the onset of phase transition shifts to higher @xmath104 ( see fig . [ pbound ] ) , considerable amount of clusters can occur in the low - density uniform phase of a hot protoneutron star . the formation of clusters and near symmetrization of the pns matter has two major consequences . first , during the evolution of the supernova matter , the abundant light nuclei produced within the neutrino - sphere @xcite at densities @xmath196 @xmath8 may be ejected outside and contribute to the nucleosynthesis of heavy elements . second , the decrease in the pressure due to the nuclear symmetry energy overwhelms the total increase from the thermal and leptonic pressures . this makes the equation of state softer with or without phase transition compared to the neutrino - free matter . it may be noted from fig . [ ypns ] , that at a fixed @xmath192 , anti - neutrino @xmath197 dominates over its particle counterpart at lower densities and/or in the coexistence region . moreover , the trapping density is also shifted to a higher value in the coexistence phase . thus , apart from neutrino transport , the exact location of the neutrino - sphere is quite sensitive to the influence of the coexistence phase in the supernova matter . figure . [ asym ] , shows the variation of neutron - proton asymmetry @xmath198 with baryon density at several fixed lepton fraction @xmath199 that correspond to the deleptonization epoch of the supernova at @xmath190 mev . considering the @xmath130 set , the dip seen in the asymmetry for uniform matter ( solid lines ) at @xmath200 @xmath8 is due to cluster formation that increases with @xmath199 . a further rapid drop in the asymmetry at @xmath169 @xmath8 results from lepton dominance in the star . thus , even in the late stage of neutrino diffusion ( at small @xmath192 ) , abundant nuclei can be produced in the near - symmetric matter @xcite . with the onset of liquid - gas phase transition at @xmath201 @xmath8 ( dash - dotted line ) , the asymmetry is found to decrease faster compared to the uniform matter . this stems from dominant contribution from the liquid part of the mixed phase which exhibits a dropping asymmetry at @xmath202 ( solid lines ) plus a fraction from the very low - density region of the gas phase that is nearly - symmetric . both these effects systematically favor a free energy with small @xmath3 . thus , compared to uniform matter , the supernova matter at @xmath186 mev is found to be nearly symmetric in the coexistence phase at larger densities near the neutrino - sphere and the pns behaves as @xmath177-@xmath178-@xmath121-@xmath179 matter . of course at higher temperatures , the cluster contribution would be enhanced and the onset of coexistence phase is also delayed to a higher density ; see figs . [ ypns](c)-(d ) . at a given @xmath104 , this causes the pns to be somewhat less symmetric compared to that at lower temperatures . the sensitivity of @xmath145 on asymmetry @xmath3 has been also studied by considering the @xmath203 set which gives an overall stiffer density dependence of symmetry energy . in this case , for uniform matter ( dashed line ) , a larger ( smaller ) @xmath3 is obtained at densities smaller ( larger ) than the normal nuclear matter value . however , the effects of symmetry energy is negligible both at small densities that are dominated by leptons , and also in the mixed phase ( dotted line ) . figure [ sneos ] depicts the equation of state ( eos ) of supernova matter at @xmath204 for uniform matter and in the coexistence phase for @xmath205 ( solid lines ) and 0.03 ( dotted lines ) with clusters . at supranormal densities , the matter is characterized by repulsive @xmath36-meson interaction that causes a rapid increase in the eos . at about the normal density of @xmath206 @xmath8 , the uniform matter shows a minimum for relatively small temperatures . this is a manifestation of the saturation properties of nuclear matter when thermal effects contribute modestly . a larger @xmath145 for the @xmath168 at @xmath207 ( dotted line ) gives a smaller binding energy . in fact , the sensitivity of symmetry energy on the eos persists till the contribution from free nucleons and/or clusters dominate up to a density of @xmath208 @xmath8 . in the coexistence phase , the ( free ) energy drops appreciably with decreasing density at small temperatures . as mentioned above , this is a consequence of the liquid part that contributes to enhance the binding of the system . the overall qualitative features found in the thermodynamic quantities with clusters are similar to those without clusters for the supernova matter ( not shown ) . compared to nucleons only supernova matter , the inclusion of cluster degrees of freedom reduces somewhat the total free energy . further , the total pressure is found to be small at @xmath6 @xmath8 when most of the cluster species contribute collectively . the pressure eventually becomes stiffer when only the massive alpha contributes dominantly at @xmath200 @xmath8 ( see inset of fig . [ eos ] ) . consequently , clusters delay the onset of the coexistence phase ( situated in the soft low @xmath104 regime of the eos ) to a higher density compared to that without clusters . we now consider the constitution and structure of beta - equilibrated and charge neutral matter in the superdense regime . it may be noted that thermal effects contribute negligibly in this density regime where the nucleons become degenerate . in fig . [ nph ] , the abundances of baryons and leptons as a function of density are shown for the fsugold parameter sets @xmath184 and 0.03 with and without the trapped neutrinos . the composition of the matter should be quite sensitive to the symmetry energy @xmath145 as it relates to the chemical potential via @xmath209 @xcite which in turn , determines the hyperon production threshold density by the condition @xmath210 , where @xmath211 is the energy density of the baryon species @xmath64 . for neutron star matter with @xmath212 , the stiffer density dependence of symmetry energy at @xmath184 causes larger enhancement of electron ( and hence proton ) fraction compared to @xmath130 case . however , with the appearance of negatively charged @xmath213 hyperon which competes with leptons in maintaining charge neutrality , the lepton ( @xmath20 and @xmath38 ) concentrations begin to fall . consequently , the sensitivity of @xmath145 on the composition is diminished . the charge neutral @xmath214 hyperon then appears , which occurs at a somewhat earlier density for the stiff symmetry energy compared to the soft @xmath130 set . trapped neutrinos have a large influence on the composition of the protoneutron star as evident from fig . [ nph ] for @xmath215 . the appearance of hyperons is now delayed to higher densities since in the above mentioned threshold condition , the term @xmath216 turns out to be positive . with the emergence of @xmath157 hyperon , the dropping of proton and thereby the electron fraction is now compensated by the rise of @xmath21 abundances so that @xmath199 is maintained constant . a softer symmetry energy ( @xmath130 ) is found to favor a significantly larger @xmath217 . incidentally , at very high densities @xmath216 becomes negative . the impact of this alteration in the lepton chemical potentials is the emergence of positively charged @xmath218 before the @xmath219 hyperon . from the central densities of the maximum mass stars ( see table i ) , it is evident that for the repulsive potential potential of @xmath220 mev used here , @xmath160 hyperons will not be present in the star matter at all . on the other hand , if we adopt an attractive interaction potential of @xmath221 mev , as used in previous studies , the @xmath164 hyperon will appear at a density of @xmath222 followed by @xmath223 and @xmath218 in the star matter . in such a situation @xmath158s will not appear in neutrino - trapped matter . the equation of state ( eos ) , i.e. pressure @xmath224 versus energy density @xmath225 is displayed in fig . [ eos ] for the neutrino - free matter ( @xmath226 ) and matter with an electron lepton fraction @xmath215 corresponding to that of fig . [ nph ] . for nucleons - only star , trapped neutrinos makes the eos softer as the decrease in the symmetry pressure due to enhanced proton fraction exceeds the increase in leptonic pressure . compared to the stiff @xmath184 set , a smaller symmetry energy / pressure at high density for the @xmath130 set obviously produces a softening in the eos . however , as a soft @xmath9 favors a large @xmath21 fraction @xcite , the softening is reduced in the lepton - rich matter . with the inclusion of hyperons , the eos becomes softer in both lepton - rich and lepton - poor matter . this is essentially caused by the release of the fermi pressure of the nucleons into the hyperonic degrees of freedom , and also due to decrease of pressure exerted by the leptons as these are replaced by negatively charged hyperons @xcite . however in a protoneutron star , the trapped - neutrinos inhibit the onset of hyperons ( see fig . [ nph ] ) which makes the eos with hyperons stiffer compared to the neutrino - free matter @xmath227 a reversal of behavior relative to the eos for nucleons - only star matter . as evident , the comparative stiffness increases for the larger @xmath145 ( @xmath184 ) that allows relatively slightly larger hyperon population in the cold catalyzed neutron star . [ cols= " < , < , < , < , < , < , < , < , < " , ] the differences in the eos should be reflected in the structure of the stars , namely their masses and radii . while most of the mass originate from the dense interior of the star , the crust has a negligible contribution of @xmath228 to the total mass . in contrast , about @xmath229 of the radius originates from the eos at @xmath230 and thus should be strongly influenced by the possible liquid - gas phase transition at this density regime . at first , the star sequence is calculated with the low density equation of state taken from baym - pethick - sutherland ( bps ) of ref . @xcite ; the effects of the coexistence and pure phase with clusters are discussed later . here , the overall eos is constructed by smoothly matching ( without any discontinuity in pressure , baryon and energy densities ) the bps equation of state at @xmath231 @xmath8 , followed by the mixed phase eos comprising of leptons and free nucleons only for the inner crust of the star in the rmf model , and finally with the rmf model eos at @xmath232 , with or without hyperons . the static star sequence obtained by solving tolman - oppenheimer - volkoff equations @xcite are shown in fig . [ mr ] for the different equations of state . listed in table i are the maximum masses @xmath233 , corresponding radii @xmath234 , and the central baryonic densities @xmath235 . with free nucleons and leptons - only star , the trapped - neutrinos give smaller maximum mass @xmath233 ( compared to @xmath212 case ) as the pressure from the stiff eos can support larger masses against gravitational collapse . with the introduction of hyperons , the reversal of behavior observed in the eos , i.e. trapping leads to a stiffer eos , is manifested in larger maximum mass protoneutron stars . note in general , the inclusion of hyperons leads to a relatively smaller mass stars due to softening of the eos . the larger radii in stars with hyperons are a consequence of weaker gravitational attraction from smaller masses that causes the stars to be large and diffuse . for the neutrino - free stars , though @xmath130 gives smaller masses than @xmath184 , the corresponding star radii are however found to be smaller in the former case . the baryonic interactions in this model raises the maximum mass from @xmath236 for the eos of a free gas @xcite by a factor of about two for nucleons - only stars . however , the inclusion of hyperons in cold neutrino - free stars yield maximum masses that are smaller than the current observational lower limit of @xmath237 imposed by the larger mass of the binary pulsar psr @xmath238 @xcite . the inset of fig . [ eos ] shows the eos for hot matter with entropy per baryon @xmath239 and @xmath199=0.4 for uniform matter ( dashed line ) and in the coexistence liquid - gas mixed phase ( solid line ) . for clarity , the eoss corresponding only to the high - density ( liquid ) regime of the mixed phase at volume fraction @xmath240 ( see eq . ( [ free ] ) ) are shown . in fact , @xmath241 can be as low as @xmath242 near the low - density ( gas ) phase boundary . figure [ mr ] illustrates the effect of liquid - gas phase transition at low density on the mass - radius sequence of free nucleons and clusters - only protoneutron stars for @xmath130 . though the maximum masses are identical with @xmath243 the radii in the sequence of stars are all different . in particular , a stiff eos obtained in the mixed phase from gibbs construction gives consistently large radii protoneutron stars . compared to the bps @xcite equation of state used above at @xmath231 @xmath8 , the rmf models in general provide much stiffer eoss at the crust that result in stars with much larger radii . in this article , within an accurately calibrated relativistic mean field model where the nuclear symmetry energy has been constrained from measurements of neutron skin thickness of finite nuclei , the structure and properties of neutron and protoneutron stars are studied . for hot and lepton - rich protoneutron stars we examine the possible liquid - gas phase transition near the crust comprising of free nucleons and light clusters . compared to free nucleon abundances , light clusters are found to dominate the particle yield at moderate and high temperatures in an uniform supernova matter . further , the fraction of trapped neutrino is shown to play an important role in making the supernova matter more symmetric via enhanced production of deuteron and alpha clusters . with liquid - gas phase transition , the yields of clusters drop drastically at temperatures @xmath195 mev in the coexistence phase ; here near symmetrization over a wide density @xmath244 @xmath8 is achieved by the free nucleons that are not bound in the clusters . at these low densities , symmetry energy has modest impact on the coexistence phase boundaries and on the equations of state of matter compared to thermal effects and on the number of trapped neutrinos . the influence of symmetry energy on hyperon production in the dense interior of cold neutron stars and hot lepton - rich protoneutron stars are studied . within the present knowledge of the potential depth of hyperons in bulk nuclear matter , we find a repulsive potential of @xmath160 hyperons inhibits its appearance in the star matter . neutrino trapping delays the appearance of hyperons due to relative disposition of chemical potentials of various species . a softer symmetry energy in general result in smaller chemical potentials that delays further the onset of hyperons and thereby reduces the hyperonization of the star matter . the emergence of hyperons soften the nuclear equation of state considerably in the fsugold model . this results in stars with very small masses especially for the cold deleptonized matter . we also find , that compared to uniform matter , a stiffer equation of state in the coexistence liquid - gas phase near the crust of hot lepton - rich matter yield stars with larger radii . the ongoing and future hypernuclear programs can provide accurate information of the potential depth of hyperons in bulk nuclear matter , which in conjunction with better mass and radius measurements of neutron stars could constrain the nuclear symmetry energy more accurately . p. danielewicz , r. lacey and w.g . lynch , science * 298 * , 1592 ( 2002 ) . v. baran , m. colonna , v. greco and m. di toro , phys . rep . * 410 * , 335 ( 2005 ) . li , l .- w . chen and c.m . ko , phys . rep . * 464 * , 113 ( 2008 ) . n.k . glendenning , astrophys . j. * 293 * , 470 ( 1985 ) ; phys . d * 46 * , 1274 ( 1992 ) . m. prakash , i. bombaci , m. prakash , p.j . ellis , j.m . lattimer , and r. knorren , phys . rep . * 280 * , 1 ( 1997 ) . a. steiner , m. prakash , j.m . lattimer and p.j . . rep . * 411 * , 325 ( 2005 ) . lattimer and m. prakash , science * 304 * , 536 ( 2004 ) ; phys . rep . * 442 * , 109 ( 2007 ) . j. schaffner and i.n . mishustin , phys . c * 53 * , 1416 ( 1996 ) . a. burrows and j.m . lattimer , astrophys . j. * 307 * , 178 ( 1986 ) . w. keil and h.t . janka , astron . & astrophys . * 296 * , 145 ( 1995 ) . blaizot , phys . rep . * 64 * , 171 ( 1980 ) . m. prakash and k.s . bedell , phys . * c * 31 , 1118 ( 1985 ) . brown , phys . lett . * 85 * , 5296 ( 2000 ) . brueckner , s.a . coon , j. dabrowski , phys . rev . * 168 * , 1184 ( 1967 ) . p.j . siemens , nucl . a141 * , 225 ( 1970 ) . a. akmal , v.r . pandharipande and d.g . ravenhall , phys . c * 58 * , 1804 ( 1998 ) . b.d . serot and j.d . walecka , adv . * 16 * , 1 ( 1986 ) . et al_. , phys . lett . * 78 * , 1022 ( 1997 ) . ko , and z.z . ren , phys . lett . * 78 * , 1644 ( 1997 ) . chen , c.m . ko , and b .- a . li , phys . lett . * 94 * , 032701 ( 2005 ) ; phys . c * 72 * , 064309 ( 2005 ) . shetty , s.j . yennello and g.a . souliotis , phys . rev . c * 76 * , 024606 ( 2007 ) . m. centelles , x. roca - maza , x. vias , and m. warda , phys . * 102 * , 122502 ( 2009 ) . b.k . sharma and s. pal , phys . lett . * b682 * , 23 ( 2009 ) . lalazissis , j. knig , and p. ring , phys . c * 55 * , 540 ( 1997 ) . todd and j. piekarewicz , phys . lett . * 95 * , 122501 ( 2005 ) . horowitz and j. piekarewicz , phys . lett . * 86 * , 5647 ( 2001 ) ; phys . c * 66 * , 055803 ( 2002 ) . h. mller and b.d . serot , phys . c * 52 * , 2072 ( 1995 ) . p. wang , phys . c * 61 * , 054904 ( 2000 ) ; p. wang , d.b . leinweber , a.w . thomas and a.g . williams , nucl . a748 * , 226 ( 2005 ) . b.k . sharma and s. pal , phys . c * 81 * , 064304 ( 2010 ) . pons , s. reddy , m. prakash , j.m . lattimer and j.a . miralles , astrophys . j. * 513 * , 780 ( 1999 ) . c. ishizuka , a. ohnishi , and k. sumiyoshi , nucl * a723 * , 517 ( 2003 ) . a.s . botvina and i.n . mishustin , phys . lett . * b584 * , 233 ( 2004 ) . g. rpke , phys . c * 79 * , 014002 ( 2009 ) . s. typel , g. rpke , t. klhn , d. blaschke , and h.h . wolter , phys . c * 81 * , 015803 ( 2010 ) . f.j . fattoyev and j. piekarewicz , phys . c * 82 * , 025810 ( 2010 ) . h. pais , a. santos , and c. providncia , phys . c * 80 * , 045808 ( 2009 ) . millener , c.b . dover , and a. gal , phys . c * 38 * , 2700 ( 1988 ) . dover and a. gal , ann . ( n.y . ) * 146 * , 309 ( 1983 ) . t. fukuda _ et . al . _ , phys . rev . c * 58 * , 1306 ( 1998 ) . p. khaustov _ et . c * 61 * , 054603 ( 2000 ) . batty , e. friedman , and a. gal , prog . suppl . * 117 * , 227 ( 1994 ) . e. friedman and a. gal , phys . rep . * 452 * , 89 ( 2007 ) . tolman , phys . rev . * 55 * , 364 ( 1939 ) ; j.r . oppenheimer and g.m . volkoff , _ ibid _ * 55 * , 374 ( 1939 ) . j.m . weisberg and j.h . taylor , phys . * 52 * , 1348 ( 1984 ) . g. baym , c.j . pethick , and p. sutherland , astrophys . j. * 170 * , 299 ( 1971 ) .

for almost a century now after its discovery , the schwarzschild solution describing a static , spherically symmetric vacuum spacetime in einstein s theory of general relativity ( gr ) has been a valuable tool for a wealth of theoretical and experimental studies , including in particular weak - field tests of the theory . for more than half a century , the validity of the solution beyond the schwarzschild radius @xmath0 and the consequent existence of bhs was thought to be a mathematical artifact of the field equations . this picture has changed dramatically in more recent decades . crucial events in this sense were the mathematical discovery of the kerr solution describing rotating bhs @xcite , and the astrophysical x - ray and radio observations in the 1960s and 1970s . cosmic x - ray sources , first detected in 1963 @xcite , were soon associated with compact stellar objects and , in particular , with mass exchange from an ordinary star onto a compact object in binary systems @xcite . this interpretation received further support when the x - ray source cyg x-1 was identified with the supergiant boib star hd 226868 @xcite , which established cyg x-1 as the first observed x - ray binary . over the next few years , the explanation of these systems in terms of accretion onto compact stellar objects in binary systems became widely accepted ( see @xcite for a review ) . measurements of the binary period and radial velocity provide lower limits on the mass of the compact companion . the values obtained for some x - ray binaries comfortably exceed the upper mass limit of about @xmath1 for neutron stars . recent investigations of cyg x-1 , for example , predict @xmath2 for the compact member @xcite . also starting in the early 1960s , optical and radio observations identified `` star like '' objects with surprisingly large redshifts @xcite . if interpreted as cosmological in origin , the redshifts implied enormous distances and , accordingly , luminosities . over the next decade the cosmological origin of these `` quasi - stellar '' sources ( or quasars ) became clear , and accretion onto supermassive bhs ( smbhs ) @xcite was accepted as the most plausible explanation of their energetics @xcite . by now , observations of stellar dynamics near the centers of galaxies and iron k@xmath3 emission line profiles provide strong evidence that many galaxies harbor bhs at their centers @xcite . bh masses are generally closely correlated to the bulge velocity dispersion and luminosity @xcite . for the milky way , astrometric and radial velocity observations of short - period stars orbiting the galactic center @xcite predict an unseen gravitational mass of @xmath4 . electromagnetic observations provide strong but _ indirect _ evidence for the existence of astrophysical bhs . a more direct way of probing the bh nature of astrophysical compact objects is offered by the newly emerging field of gw astronomy . first - generation ground - based laser - interferometric detectors ( such as ligo , virgo and geo600 ) have reached design sensitivity , and advanced versions of these detectors will be operational within a few years @xcite . one of the most prominent sources of detectable gws for these detectors is the inspiral and coalescence of stellar mass and , possibly , intermediate - mass bh ( imbh ) binaries . at lower frequencies , space - based laser interferometric observations @xcite have the potential to complement ground - based efforts with high signal - to - noise ratio observations of massive bh binaries . space - based detectors are guaranteed to observe galactic binaries containing white dwarfs and/or neutron stars , and they may also detect the inspiral of stellar compact objects into massive bhs . due to the weak interaction of gws with any type of matter , including the detectors , digging physical signals from the noisy data stream represents a daunting task and heavily relies on so - called _ matched filtering _ techniques @xcite . matched filtering is a common choice to search for signals of known form in noisy data , and works by cross - correlating the actual `` signal '' ( i.e. , the detector s output ) with a set of theoretical templates . accurate modeling of the gw sources thus plays a vital role in maximizing the scientific output from these experimental efforts . event rate estimates for imbhs and stellar - mass bh binaries are very uncertain , but advanced earth - based detectors are expected to observe several events per year ( see @xcite for a review ) . _ noted that recent metallicity measurements require revisions in the population synthesis models used for event rate estimates , suggesting that stellar - mass bh - bh binaries may be the first gw sources to be detected @xcite . the observations of two possible x - ray binary precursors of bh - bh binaries provide a much needed observational constraint on compact binaries containing at least one bh , suggesting again that bh - bh binary mergers may be more common than anticipated @xcite . population synthesis models have large uncertainties , related e.g. to the poorly known common envelope phase , and some models would necessarily be ruled out if advanced ligo does _ not _ detect bh binaries . in this sense , gw detectors are _ guaranteed _ to put constraints on compact binary formation in the near future . schutz @xcite also pointed out that coherent data analysis using three or more detectors may further increase event rates by factors of a few , making prospects of doing astrophysics with earth - based detectors even brighter . estimates for event rates of smbh binaries detectable by space - based interferometers range from @xmath5 to @xmath6s per year , with a most likely estimate of @xmath7 @xcite . over the past decade , bh modeling has also attracted a great deal of interest in the context of high - energy physics . according to the gauge / gravity duality , gravitational physics in anti - de sitter ( ads ) backgrounds describes field theories on the boundary of that spacetime @xcite . according to this duality , bhs in ads are dual to an equilibrium field theory , and therefore interacting bhs or bhs off - equilibrium may represent interesting physics from the dual point of view . in fact , there is some evidence that some of the physics behind quark - gluon plasma formation in heavy - ion collisions is captured by this duality , which on the gravitational side corresponds to shock wave collisions and bh formation in ads @xcite . understanding bh spacetimes is also of interest for high - energy physics in connection with extra - dimensional scenarios . solutions to the hierarchy problem in physics have been developed where all standard - model interactions are confined to a 3 + 1 dimensional brane embedded in a spacetime with large extra dimensions or an extra dimension with a warp factor @xcite . in these models gravity propagates in the entire higher - dimensional spacetime ( the `` bulk '' ) , and the fundamental planck scale could be as low as @xmath8 . these energy scales are accessible to experiments such as the large hadron collider ( lhc ) . if the fundamental planck scale is so low , thorne s hoop conjecture suggests the exciting possibility of bh production in high - energy parton - parton collisions @xcite , i.e. a direct observable signature of the existence of extra - dimensions . accurate modeling of energy and momentum losses in the form of gws during the collisions and determination of the bh formation cross section will be crucial for the analysis of data from such experiments ( see e.g. @xcite ) , and they should be amenable to classical calculations in @xmath9-dimensional gr @xcite . finally , the understanding of bh dynamics under extreme conditions is fascinating from a mathematical - physics point of view . questions such as bh stability , cosmic censorship , etcetera can only be addressed by solving the full nonlinear set of the field equations . for many of these scenarios , it is convenient to divide the coalescence of a bh binary in the framework of gr into three stages which mirror the different tools used for their modeling : ( i ) the inspiral phase , where the interaction between the individual holes is still sufficiently weak to justify the use of post - newtonian ( pn ) techniques @xcite ; ( ii ) the final orbits , plunge and merger , which are governed by the strong - field regime of einstein s equations and can be described only by fully numerical simulations ; ( iii ) the ringdown of the remnant bh , which is amenable to a perturbative treatment @xcite . in this article we will focus on the second , strong - field stage , on the numerical tools dedicated to its study , and results obtained following the 2005 breakthroughs @xcite . as we shall see , however , a comprehensive understanding of bh dynamics requires a close interplay of numerical and analytical studies , and we will discuss in various cases the interface between numerical relativity and approximation techniques . following a brief summary of the framework employed in numerical relativity ( sec . [ sec : numerics ] ) , in secs . [ sec : gws]-[sec : highenergy ] we will present the main results obtained from numerical studies of bhs in the context of gw detection , astrophysics and high - energy physics , respectively . we conclude in sec . [ sec : conclusions ] with a discussion of future applications of numerical relativity . at the end of each section we will provide references to the literature for further reading . * _ notation : _ * we shall be using geometrical units , such that the gravitational constant @xmath10 and the speed of light @xmath11 . we denote spacetime indices @xmath12 by greek letters , and spatial indices @xmath13 by latin letters . in einstein s gr , spacetime is modelled as a @xmath9-dimensional manifold with a metric @xmath14 of lorentzian signature @xmath15 . with the exception of sec . [ sec : highenergy ] we will restrict our discussion to @xmath16 , i.e. three spatial and one time dimension . the metric is determined by the einstein equations @xmath17 with @xmath18 in vacuum . while analytic solutions ( as , for example , the schwarzschild and kerr solutions and the friedmann - lematre - robertson - walker solution ) have been found for systems with high symmetry , solutions for dynamical configurations without special symmetries generally require the use of numerical tools . for this purpose the einstein equations need to be cast as an _ initial value formulation _ such that the cauchy - kowalewski theorem guarantees a uniquely determined evolution of appropriately chosen initial data . in the case of einstein s equations , it is not obvious that such a formulation can be obtained : a straightforward calculation shows that eq . ( [ eq : einstein ] ) contains second time derivatives of the metric components @xmath19 , but not of @xmath20 . this fact mirrors the coordinate or gauge freedom characteristic of gr , and implies the existence of so - called _ constraint _ equations which impose conditions on the metric components and their first derivatives on each spatial slice with constant coordinate time @xmath21 . in the following we will discuss two approaches which provide an initial value formulation of the einstein equations and have led to successful , long - term stable numerical evolutions : the generalized harmonic ( gh ) formulation employed in pretorius breakthrough bh binary simulations @xcite and the canonical arnowitt - deser - misner ( adm ) split @xcite , reformulated by york @xcite , which forms the starting point for the _ moving puncture _ breakthroughs @xcite . * _ adm 3 + 1 split : _ * spacetime is decomposed into a one - parameter family of three - dimensional spatial slices . we choose coordinates adapted to this formulation , such that the coordinate time @xmath21 labels each slice and @xmath22 denotes points on the slice . on each hypersurface , there exists a unique timelike , future pointing unit normal field @xmath23 which defines a projection operator @xmath24 onto the hypersurface . the geometry of the hypersurface is completely determined by the _ three - metric _ or _ first fundamental form _ the coordinate freedom is represented by the shift vector @xmath26 and the lapse function @xmath27 ( with @xmath28 ) , which relate spatial coordinates on ( and measure separation in proper time between ) different hypersurfaces . the projections @xmath29 , @xmath30 , @xmath31 of the einstein equations then lead to six evolution equations for the three - metric @xmath32 , three momentum constraints and the hamiltonian constraint : @xmath33 where @xmath34 , @xmath35 and @xmath36 denote the ricci tensor , ricci scalar and covariant derivative associated with the three - metric , and the extrinsic curvature @xmath37 has been introduced to obtain a first - order system in time . it turns out that this formulation of the einstein equations is only _ weakly hyperbolic _ @xcite , and therefore not suitable for long - term stable numerical evolutions . motivated by these stability problems , several modifications of the adm system have been investigated . the most popular is the baumgarte - shapiro - shibata - nakamura ( bssn ) formulation @xcite , which rearranges the degrees of freedom via a split of the extrinsic curvature into trace and trace - free parts , a conformal transformation and promotion of the contracted conformal christoffel symbols to the status of independent variables : @xmath38 the resulting set of evolution and constraint equations is given in eqs . ( 15)-(30 ) of @xcite . subject to minor modifications , such as replacing the conformal factor by @xmath39 or adding the auxiliary constraint arising from the definition of @xmath40 in eq . ( [ eq : bssn ] ) , this formulation is employed in the present generation of moving puncture evolution codes @xcite which fix the gauge via `` 1+log '' slicing and the @xmath41-driver condition @xcite . * _ gh formulation : _ * harmonic coordinates are defined by the condition @xmath42 , where @xmath43 represents the scalar wave operator . these coordinates have played an important role in the analysis of the cauchy problem in gr @xcite . in harmonic coordinates , the einstein equations take on a manifestly hyperbolic form , which allows for a generalization to arbitrary coordinate or gauge choices @xcite . the first step is to introduce four arbitrary source functions @xmath44 such that the coordinates obey @xmath45 and treat these functions as independent evolution variables . then one considers the gh system @xmath46 where @xmath47 . ( [ eq : gh ] ) is equivalent to the einstein equations , subject to the validity of the constraint ( [ eq : h ] ) . in expanded form , the gh system is given by @xmath48 and the constraints @xmath49 are preserved by virtue of the bianchi identities provided that @xmath50 and @xmath51 vanish on the initial hypersurface @xmath52 . a key ingredient in the numerical evolution of the gh system is the addition of a constraint damping term @xmath53 c_{\mu}$ ] to the right hand side of eq . ( [ eq : ghexpanded ] ) @xcite . here , @xmath54 is a non - vanishing timelike vector field . finally , it is interesting to note that the hamiltonian and momentum constraints are automatically satisfied on the initial hypersurface if @xmath55 @xcite . the coordinates were determined in pretorius initial simulations by @xmath56 and the evolution of the metric proceeds according to eq . ( [ eq : ghexpanded ] ) with the constraint damping term of gundlach _ et al . _ @xcite . for further discussion of gauge choices in the ghg system , see also @xcite . the spectral code originally developed by the caltech - cornell group ( see @xcite and references therein ) employs a first - order version of the gh system @xcite with dual - coordinate frames and bh excision . furthermore , the first - order gh formulation facilitates the specification of constraint - preserving boundary conditions @xcite . during the inspiral phase , they evolve the gauge functions @xmath57 such that they remain constant in a frame comoving with the bhs , but switch to a dynamic evolution during the plunge and ringdown ; see sec . iii in @xcite . while more complex in structure , this framework enables their code to generate bh evolutions with exceptional accuracy @xcite . * _ initial data : _ * the construction of initial data requires solving the hamiltonian and momentum constraints . here we only summarize the key concepts ; we refer the reader to cook s review article @xcite for details . most work on the construction of initial data is based on the york - lichnerowicz split @xcite , which involves a conformal transformation of metric and extrinsic curvature and separates the latter into trace and trace - free part . more recently , the _ thin - sandwich _ construction @xcite , which replaces the extrinsic curvature by the time derivative of the metric , has become a popular alternative . in either case , the resulting elliptic equations simplify substantially under the assumption of conformal flatness and a spatially constant @xmath58 . while this formalism provides a convenient method to solve the constraint equations , we still need to ensure that the initial data represent a physically realistic system , typically two bhs with specific spins and momenta . this can be achieved by generalizing the schwarzschild solution , which is obtained in the above framework in conformally flat form with conformal factor @xmath59 . a generalization to initial data of @xmath60 bhs starting from rest is directly obtained by the construction of misner @xcite or brill and lindquist @xcite . remarkably , an analytic solution for the extrinsic curvature can still be obtained for bhs with initial linear momenta @xmath61 and spins @xmath62 @xcite . by applying a compactification to the internal asymptotically flat region , brandt & brgmann @xcite arrived at the so - called _ puncture data _ construction , which is the method of choice for most numerical evolutions employing the bssn formulation . in order to overcome an upper limit of @xmath63 for the dimensionless spin of bhs in puncture type initial data @xcite , lovelace _ et al . _ @xcite generated initial data based on a generalization of the kerr - schild solution @xcite and were thus able to evolve a bh binary with spin magnitude 0.95 . in alternative to generalizing analytically known bh solutions , the presence of horizons in the initial data can be encoded in the form of boundary conditions for the metric and extrinsic curvature @xcite as determined by the isolated horizon framework @xcite . initial data along these lines have been constructed in @xcite , and form the starting point of most of the numerical evolutions using the gh system ( see e.g. @xcite ) . * _ diagnostics : _ * extracting physical information from numerical simulations of the einstein equations is nontrivial for two reasons . first , the evolved variables are dependent on the coordinate conditions ; second , it is often not possible to define local quantities familiar from newtonian physics . the first difficulty requires the calculation of gauge - independent variables . the second difficulty is alleviated by the _ isolated horizon _ framework @xcite , which facilitates the calculation of bh mass and spin in the limit of isolated bhs ; in practice , this framework is often applied when the bhs are farther apart and their interaction is considered negligible . local properties of the bhs are encoded in their apparent horizon @xcite . in particular , the irreducible mass can be expressed in terms of the apparent horizon area : @xmath64 . in the limit of an isolated bh , the spin can be derived from the integration of the rotational killing vectors over the horizon according to eq . ( 8) of @xcite . in practice it has been found that using _ flat - space _ rotational killing vectors yields reasonable results @xcite , but we also note the improved method to compute approximations to killing vectors in @xcite and an alternative method @xcite based on weyl scalars to compute quasi - local bh quantities without solving the killing equation . it is important to bear in mind , however , the approximate nature of any spin calculation in bh binary simulations . in comparison , it is more straightforward to extract global quantities of the spacetime , most notably the total mass - energy @xmath65 and the linear and angular momentum @xmath66 , @xmath67 @xcite . these are given by surface integrals at spatial infinity , see e.g. eqs . ( 7.15 ) , ( 7.56 ) and ( 7.63 ) in @xcite . arguably the most important diagnostic quantity in bh simulations is the gw signal . the most common method to extract gws is based on the newman - penrose formalism @xcite and derives the complex newman - penrose scalar @xmath68 from contraction of the weyl tensor with suitably chosen tetrad vectors @xcite , sec . iii a in @xcite . it is often convenient to decompose @xmath68 into multipoles @xmath69 using spherical harmonics of spin - weight @xmath70 : @xmath71 . contemporary numerical codes typically evaluate @xmath68 at finite coordinate radius . this results in systematic errors , due to ambiguities in the tetrad choice and the neglect of gw backscattering . while these errors can be reduced by extracting gws at a sequence of radii and extrapolating to infinite radius ( @xmath68 asymptotically falls off @xmath72 @xcite ) , a cleaner method is to calculate @xmath68 at infinity as provided by _ cauchy - characteristic extraction _ methods @xcite . from @xmath68 , one straightforwardly obtains the energy , linear and angular momentum radiated in the form of gws ; see e.g. eqs . ( 49)-(51 ) in @xcite . in gw data analysis it is more common to work with the wave strain @xmath73 decomposed into multipoles : @xmath74 . in practice , after integrating @xmath68 twice in time the signal can be severely affected by nonlinear drifts , which can be controlled to a significant extent by performing the integration in the frequency domain @xcite . alternatively , gws can be extracted by viewing the metric in the far field regime as a perturbation of the schwarzschild spacetime and employing the formalism of regge , wheeler @xcite and zerilli @xcite . one thus obtains the regge - wheeler - moncrief and zerilli - moncrief master functions @xmath75 , @xmath76 , which can be converted into the multipolar components of the gw strain @xmath77 according to eq . ( 49 ) of @xcite . for a comprehensive summary of the numerical framework for evolving einstein s field equations , we refer the reader to the books by alcubierre @xcite and baumgarte and shapiro @xcite . further details , including mathematical aspects of the equations and characteristic techniques , can be found in the review articles @xcite . bh binary systems represent one of the most promising sources of detectable gws . the parameters of a bh binary are commonly divided into _ intrinsic _ and _ extrinsic _ @xcite . intrinsic parameters characterize physical properties of the system , such as the total mass @xmath78 , the mass ratio @xmath79 , the individual bh spins @xmath80 , @xmath81 and the orbital eccentricity . extrinsic parameters such as sky position , distance , orbital inclination angle , arrival time and initial phase of the wave , in contrast , depend on the source location relative to the observer , and do not directly enter the gw source modeling process . the majority of numerical studies of bh binary spacetimes performed to date have focussed on comparable mass - ratios @xmath82 and moderate spin magnitudes @xmath83 , with particular emphasis on spins ( anti-)aligned with the orbital angular momentum . we note , however , the following explorations outside this `` best charted '' subset of the parameter space . circular binaries with mass ratios up to @xmath84 have been studied in @xcite ; comparisons for this value of @xmath85 with fully perturbative calculations have been obtained in the limit of head - on collisions in @xcite , who report a discrepancy of @xmath86 in gw energy and momentum , probably in large part due to the discretization error of the numerical simulations . the first direct comparison of numerical results with leading - order self - force prediction was studied in the context of periastron advance in bh binaries in @xcite . bh binaries with spins @xmath87 have been evolved in @xcite , but exceeding an apparent barrier of @xmath88 @xcite requires departure from the conformal flatness assumption @xcite . the present maximum is @xmath89 for aligned spins evolved for 12.5 orbits by lovelace _ finally , we note that emission of gws efficiently circularizes bh binary orbits @xcite . bh binaries are therefore expected to have vanishing eccentricity by the time they enter the frequency window of ground - based detectors , and for this reason most work on gw source modeling has focused on the quasi - circular limit . the derivation of bh momenta , including a small radial component , for quasi - circular initial data in numerical relativity is based either on integrating the pn equations of motion @xcite or iterative procedures using several numerical simulations @xcite . the decay of eccentricity during the inspiral phase as well as periastron advance has been measured in numerical simulations in @xcite . for numerical studies of bh binaries with significant eccentricity as well as arguments why eccentric binaries may represent relevant sources of gws after all , especially in the more extreme mass ratio regime and for space - based detectors , we refer the reader to @xcite . * _ gravitational waveforms from bh binaries : _ * for illustration , we display in fig . [ fig : hlm ] the real parts of the @xmath90 and @xmath91 multipoles of the gw strain @xmath77 obtained for the inspiral and coalescence of a binary and @xmath91 obtained from the inspiral and merger of a nonspinning binary with mass - ratio @xmath92 . the dashed red curve represents pn predictions matched to the numerical ( solid black ) signal . the vertical dotted lines mark the time when the frequency of the @xmath93 multipole has reached @xmath94.,height=238 ] of nonspinning bhs with mass ratio @xmath92 @xcite . in the course of the inspiral , both amplitude and frequency of the gw signal gradually increase . close to merger , defined as formation of a common apparent horizon , the gw amplitude reaches a maximum . eventually it drops exponentially as the merged hole rings down to a stationary kerr state : see e.g. fig . 18 in @xcite . the qualitative features of the gw signal emitted by different types of binaries can be summarized as follows . ( i ) the inspiral of two nonspinning , equal - mass bhs is the case most intensively studied by numerical relativity . the wave signal is dominated by the @xmath93 multipolar component , which carries @xmath95 of the total radiated energy . the longest numerical simulation by scheel _ @xcite covers 16 orbits and reports a ratio of final to initial bh mass @xmath96 and a final spin @xmath97 . ( ii ) as the mass - ratio decreases , higher - order modes become more prominent @xcite and the fractional energy in @xmath93 drops to about @xmath98 in the limit @xmath99 @xcite . motivated by the strong relation of phase and frequency of different @xmath100 multipoles , baker _ _ @xcite view the @xmath100 radiation modes as being generated by the corresponding momenta of an _ implicit rotating source _ ; see also @xcite . ( iii ) gw multipoles with odd @xmath101 are suppressed in the equal - mass limit due to symmetry . ( iv ) bh binaries with spins aligned with the orbital angular momentum emit stronger gw signals due to an effect sometimes referred to as _ orbital hang - up _ : equal - mass binaries with @xmath102 radiate about twice as much energy and angular momentum when compared to the nonspinning case @xcite . for anti - aligned spins the gw energy decreases by a similar factor . ( v ) orbital eccentricity introduces a nonmonotonic behavior of the gw frequency ; cf . fig . 6 in @xcite . ( vi ) spin - spin and spin - orbit couplings cause a precession of the orbital plane in the case of bh binaries with spins that are not aligned or anti - aligned with the orbital angular momentum . this precession manifests itself in a modulation of the gw amplitude emitted in a fixed angular direction : cf . fig . 1 in @xcite . * _ tests of general relativity : _ * numerical simulations of binary bh mergers provide an opportunity to study the nonlinearities of the theory and to test the kerr nature of astrophysical bhs . if gr is the correct theory of gravitation , all bhs in the universe should be described by the kerr solution @xcite , which depends on two parameters : the mass @xmath78 and the dimensionless spin @xmath103 . the remnant of a binary bh merger settles down to the kerr solution by emitting gravitational radiation at characteristic ( complex ) quasinormal frequencies @xmath104 that depend only on @xmath103 . here @xmath100 are the usual angular indices , and @xmath105 is the `` overtone number '' : modes with small @xmath105 have a longer damping time and should dominate the radiation @xcite . the measurement of the real and imaginary part of a single quasinormal mode contains , in principle , enough information to determine the mass and spin of the bh . together with the accurate determination of the individual bh masses from the inspiral phase this already allows for tests of gr , by testing the gr prediction for mass loss during inspiral and merger and at the same time the strong field dynamics of gr . the measurement of _ any _ other frequency or damping time then provides a further interesting test of the kerr nature of the final bh @xcite . the feasibility of these tests depends on the characteristics of a given gw detector and on the relative excitation of the modes ( see @xcite for detailed studies ) . quantifying the excitation of quasinormal modes for generic initial data is a formidable technical problem @xcite , but numerical merger simulations allow sensible estimates of the relative quasinormal mode amplitudes @xcite . the relative mode amplitude depends on the binary parameters , and therefore , in principle , the measurement of a multi - mode signal with single or ( better ) multiple gw detectors can be used to measure the binary mass ratio or the inclination of the final bh spin with respect to the line of sight . we also remark that the increased signal - to - noise ratio coming from the merger has been shown to _ improve _ the bounds on alternative theories of gravity that could come from observations of the inspiral only . keppel and ajith , for example , discussed this possibility in the context of massive graviton theories , where the graviton mass would modify the dispersion relation of gws @xcite . formulations of the evolution equations in alternative theories of gravity are in their infancy @xcite and the numerical exploration of bh binaries in alternative theories of gravity has just started @xcite . this is an interesting line of research , and in the near future we may have more concrete ways to quantify deviations from gr in strong - field mergers . * _ gw template banks : _ * the key challenge in gw data analysis is to accurately predict and isolate the features of gravitational waveforms in the data stream from gw detectors . this is necessary in order to ( i ) detect the presence of a gw signal of astrophysical origin , and ( ii ) determine the parameters of the emitting source . these goals can be achieved by matched filtering , i.e. by cross - correlating the data stream @xmath106 ( composed of detector noise @xmath105 plus a potential gw signal @xmath77 ) with a bank of theoretical waveform templates @xmath107 , where the indices @xmath108 @xmath109 denote the @xmath110 intrinsic and extrinsic source parameters . for this purpose it is imperative to construct a large set of accurate numerical waveforms which cover the relevant parameter space . this task far exceeds the capacity of purely numerical methods . if we assume a seven - dimensional parameter space one mass ratio and three spin parameters per hole a mere ten waveforms per parameter would correspond to @xmath111 waveforms . furthermore , using binary waveform models including the long inspiral phase plus merger optimizes the signal - to - noise ratio in gw observations , and the inclusion of merger and ringdown has been shown to provide significant improvements in source localization and distance calculation @xcite . because current numerical simulations cover only about 30 cycles of the inspiral , the generation of complete waveforms requires the combined use of pn and numerical methods . the construction of such gw template banks currently proceeds along either of the following two paths . \(i ) the _ effective - one - body _ ( eob ) approach @xcite maps the dynamics of the two - body problem in gr into the motion of a particle in an effective metric . the components of the effective metric are currently determined to 3pn order . the eob method improves upon this model by using additional _ pseudo - pn _ terms of higher order between @xmath112 and @xmath113 @xcite . ] which are not derived from pn expressions , but calibrated via comparison with numerical results @xcite . further improvements come from using a resummed version of the pn expanded results and from modeling nonadiabatic effects in the inspiral : see e.g. sec . iv in @xcite . the inspiral - plunge waveform resulting from this construction is then matched to a merger - ringdown signal composed of a superposition of quasinormal oscillation modes of a kerr bh . the total number of free parameters varies between the individual eob models currently investigated , but in all cases , the general strategy is to calibrate these parameters by comparison with a finite number of numerical simulations . ( ii ) _ phenomenological waveform templates _ are based on _ hybrid waveforms _ : the early inspiral is modeled by pn techniques , and the resulting signal is matched ( within a specified window ) to a numerical waveform describing the last orbits , merger and ringdown . the resulting set of hybrid waveforms are then approximated by a model containing a number of phenomenological parameters which are mapped to the physical parameters of the binaries , such as the mass ratio @xmath85 . the phenomenological models initially developed for nonspinning binaries in @xcite , and then extended to the case of spins aligned with the orbital angular momentum @xcite , are based on hybrid waveforms matched in the _ time domain _ , whereas the more recent study by santamaria _ et al . _ @xcite performs the matching in the frequency domain . in spite of such differences in their construction , the final result of all these phenomenological models are closed - form analytic expressions for the waveforms in the frequency domain . an exploration of phenomenological models for equal - mass binaries with spin precession is given in @xcite . both types of template banks are employed in the analysis of gw data . the recent search for gws from binary bh inspiral , merger and ringdown @xcite used phenomenological models for injection , and an eob model for injection and for the search templates . this search constrained the rate of mergers of binaries with individual masses in the range @xmath114~m_{\odot}$ ] to be no more than @xmath115 at 90% confidence . numerical waveforms have also been used inside the gw community - wide _ ninja _ project @xcite to study the sensitivity of existing gw search algorithms used in the analysis of observational data . for this purpose , numerical relativity waveforms were injected into a simulated data stream designed to mimic ground - based detector noise . the algorithms detected a significant fraction of the injections , but likely require further development , in particular for the purpose of measuring source parameters @xcite . a further community wide effort , the _ nrar _ project @xcite , is dedicated to a systematic exploration of the complete bh binary parameter space to develop optimally - calibrated template families to be used in gw data analysis . one of the most exciting prospects of gw detection is the idea of looking for electromagnetic counterparts to the gw signal . while a _ network _ of earth - based detectors is required to localize a gw source by triangulation ( see e.g. @xcite ) , a single space - based detectors like lisa may suffice to localize low - redshift sources and determine their luminosity distance . the inclusion of merger and/or of higher multipolar components of the radiation in the waveform models has been shown to provide significant improvements in source localization and distance determination @xcite . for this reason it is very important to construct phenomenological models including higher multipoles of the full inspiral / merger / ringdown waveform . the eob model has recently been extended in this direction @xcite . * _ accuracy requirements : _ * clearly , the two approaches for template construction mentioned above require accurate numerical waveforms . quantifying these accuracy requirements is a nontrivial task . numerical uncertainties due to finite resolution are directly tested by _ convergence analysis _ , and the error incurred by extracting gws at finite radius can be estimated using extrapolation of results from various radii to infinity @xcite . the uncertainties in phase and amplitude thus derived , however , depend on the alignment of the waveforms in phase and time ; see e.g. fig . 2 in @xcite . because offsets in phase and time are free parameters in gw data analysis , it is more convenient to measure accuracy and agreement of different waveforms in terms of quantities which take such alignment into account by construction . such a measure is obtained from the inner product between two waveforms @xmath116 and @xmath117 used in the theory of parameter estimation @xcite : @xmath118 where the tilde and asterisk denote fourier transform and complex conjugate , respectively , and @xmath119 is the one - sided power spectral density of the detector strain noise @xcite . in practice , this inner product is to be understood as maximized over constant offsets in time and phase ( @xmath120 and @xmath121 ) between the two waveforms . the signal - to - noise ratio that would be obtained for a physical signal @xmath122 and a model waveform @xmath123 is then given by @xmath124 . it is related to the optimal signal - to - noise ratio @xmath125 for a perfect model waveform by the _ mismatch _ @xcite : @xmath127 to leading order in @xmath128 , this definition of the mismatch implies @xcite @xmath129 where @xmath130 is that part of the waveform error parallel to @xmath122 . based on these definitions , accuracy requirements are obtained as follows . two waveforms differing by @xmath131 will be indistinguishable for a detector if @xmath132 @xcite , such that acceptable waveform models to be used in parameter estimation must lie within a sphere of unit radius of the exact waveform . for the purpose of event detection , we note that the fractional loss in detection of gw signals due to imperfect templates is proportional to the third power of the reduction in the snr , i.e. @xmath133 . a value of @xmath134 is typically deemed acceptable , corresponding to a mismatch @xmath135 . for detection efficiency , there arises , however , a complication due to the discrete nature of the template bank . as discussed in detail in sec . ii b of @xcite , the effective mismatch in a real gw search is given by a contribution due to the actual modeling procedure , plus a term which accounts for the discrete spacing in the parameter space : see their fig . 3 and eq . template banks adopted in ligo searches use 0.3 for the latter @xcite and , in consequence , lindblom _ at al . _ @xcite recommend a maximum mismatch @xmath136 . in terms of the right - hand side of eq . ( [ eq : mm2 ] ) , dubbed _ inaccuracy function _ in ref . @xcite , we summarize the accuracy requirements as ( cf . ( 10 ) in ref . @xcite ) @xmath137 accuracy measurements employing the mismatch or @xmath128 have been used to study numerical , semi - analytic and hybrid waveforms . according to the samurai project @xcite , relative mismatches of the @xmath138 mode between a variety of purely numerical waveforms for binaries with total mass @xmath139 are better than @xmath140 . _ @xcite found errors in the construction of hybrid waveforms to be dominated by pn contributions . numerical waveforms should be long enough to enable matching at lower frequencies , where pn approximations are more accurate : @xmath141 orbits before merger for the equal - mass , nonspinning case , and @xmath142 orbits for binaries with spins @xmath143 . @xcite report significantly more demanding requirements of @xmath144 orbits for nonspinning , equal - mass binaries . similarly , boyle @xcite concludes that longer nr waveforms or more accurate pn predictions will be required . the discrepancy largely arises from the use of more stringent accuracy thresholds : @xmath136 in @xcite ( instead of @xmath145 in @xcite ) for detection , and @xmath146 for parameter estimation . most recently , ohme _ et al . _ @xcite have argued that maximising the match over intrinsic source parameters as well as phase and time shifts may reduce length requirements on numerical waveforms . a comparison of phenomenological and eob template banks based on the inaccuracy function has been performed by damour _ et al . _ @xcite . for advanced ground - based detectors , they conclude that current phenomenological models deviate from the eob ( considered as a target model ) beyond acceptable thresholds over a wide range of the parameter space , for both , detection and parameter estimation . further reading on bh simulations in the context of gw physics is available in @xcite . astrophysical observations have provided the strongest evidence to date that bhs exist and play an important role in many physical processes in our universe . astrophysical observations of x - ray binaries or quasars , as mentioned in sec . [ sec : intro ] , have a good deal to tell us about bhs . in recent years , numerical relativity simulations of bhs have also deepened our insight into a variety of astrophysical systems . we will first summarize what we have learned from `` traditional '' electromagnetic observations about the populations of bhs we can expect to exist in the universe . then we will discuss some astrophysical implications of numerical relativity . * _ expected bh populations : _ * we have already mentioned the two main classes of bhs identified by astrophysical observations . ( i ) solar - mass bhs with @xmath147 are usually found in x - ray binaries @xcite and are formed as the end - product of the evolution of massive stars . ( ii ) smbhs with @xmath148 are believed to reside in most active galactic nuclei ( agn ) @xcite . the assembly of these smbhs is likely to result from a combination of bh mergers and accretion of surrounding matter over cosmological timescales . evidence for a third class of bhs with @xmath149 ( imbhs ) is tentative at present @xcite . gw frequencies are inversely proportional to the system s total mass , and therefore bh masses play a crucial role in binary bh detectability . earth - based detectors , such as ligo and virgo , have an optimal sensitivity band corresponding to stellar - mass bhs and imbhs , while the planned space - based detector lisa is most sensitive to high - mass imbhs and smbhs . as discussed in the previous section , the expected gw pattern from bh binaries depends on the masses and spins of the binary members . we shall further see below that the gravitational recoil imparted upon the remnants of coalescing binaries strongly depends on spins and mass ratio . so , what do `` traditional '' electromagnetic observations tell us about bh masses and spins ? current spin estimates for accreting , stellar mass bhs are usually obtained by modeling the shape of their x - ray spectrum or by analyzing the skew in fe k@xmath3 emission lines . the resulting estimates are all , to some extent , model - dependent , but frequently yield moderate values @xmath150 and , for some cases , values close to the kerr limit @xmath151 ; see e.g. @xcite and references therein . theoretical arguments and observations suggest that stellar - mass bhs in binaries _ retain the spin they had at birth _ : neither accretion nor angular momentum extraction are likely to change significantly their mass or spin @xcite . for smbhs , spin estimates relying on the shape of the fe k@xmath3 line yield values ranging from moderate @xmath152 ( e.g. @xcite to near - critical spins ( e.g. @xcite ) . smbhs are expected to grow by a combination of mergers and accretion . their spin depends sensitively on the details of these processes and of their growth @xcite . smbh assembly via bh mergers does not appear to be able to account for the large observed spin values ; this may indicate that accretion dominates over mergers @xcite . in any case , bh spins encode the history of their formation , and it would be extremely useful to have detailed knowledge of the bh spin distribution function . ultimately , such a detailed census of bh parameters is one of the key targets of future gw observations @xcite . for the purpose of numerical modeling of bhs , present measurements indicate that all spin magnitudes ( @xmath153 ) should be covered . we have already noted that stellar - mass bhs typically have masses in the range @xmath154~m_{\odot}$ ] . unfortunately , there are no confirmed observations of stellar - mass _ binary _ bh candidates , so estimates of mass ratios must rely on theoretical models . for stellar - mass binaries , population synthesis codes suggest that @xmath85 should always be quite close to unity @xcite . measurements of smbh masses are obtained by observations of stellar motion near galactic centers , as mentioned in sec . [ sec : intro ] , as well as motion of gas discs @xcite , applications of the virial theorem to the velocity dispersion of stars @xcite and reverberation mappings applied to more distant agns @xcite . an exhaustive list of galaxies with smbh mass measurements in the range @xmath155 is presented by graham _ the impact of different smbh assembly models on the mass and mass ratio distribution of detectable binaries has been discussed by various authors . the general consensus is that mass ratios @xmath156 ( and down to @xmath157 ) should be common @xcite . in contrast with the case of stellar - mass bhs , there is by now some observational evidence for smbh binaries @xcite . these observations are not sufficient to tightly constrain smbh merger rates , but they are at least broadly consistent with merger - tree models predicting tens to hundreds of events during the typical lifetime of a space - based gw detector @xcite . a more speculative kind of source for earth - based detectors consists of the intermediate mass ratio inspiral of a compact object ( neutron star or bh ) into an imbh . another promising source for advanced earth - based interferometers ( albeit with highly uncertain event rates ) are imbh - imbh mergers . the initial inspiral of these binaries could be detected via space - based interferometers , while the ringdown phase is in the optimal bandwidth for second- and third - generation detectors such as advanced ligo and et , that could therefore be used for `` follow - up '' ringdown searches @xcite . as in the case of bh spins , the observations imply that numerical relativity needs to cover a wide range in @xmath85 . this can be done most efficiently by bridging the gap between numerical studies and the perturbative modeling of extreme - mass - ratio binaries ( e.g. @xcite ) . from this discussion , it is clear that a detailed knowledge of the spin evolution as well as the generation of gravitational recoil in bh binary mergers is important for understanding the cosmological evolution of smbhs over cosmological times . we will discuss these effects in turn . * _ black - hole spins resulting from mergers : _ * prior to the 2005 numerical relativity breakthroughs , it was known that `` minor mergers '' ( @xmath158 ) of a large , rotating bh with an isotropic distribution of small objects would tend to spin down the hole @xcite . numerical merger simulations showed that the merger of comparable - mass , nonspinning bhs leads to a final kerr bh with spin parameter @xmath159 . the simulations were followed by the development of several models to predict the spin of merger remnants as a function of the binary parameters for generic mass ratios and spins . the first studies focussed on performing numerical evolutions of the last few orbits of bh binaries , and used the results to calibrate formulae that map the initial binary parameters to values for the spin of the merged hole @xcite . it became clear , however , that the binary inspiral up to the last orbits has the potential to significantly affect the spin distribution ( e.g. @xcite ) and therefore should be included , for example via pn modeling , in the derivation of maps from initial parameters to the merger remnant properties @xcite . predictions for the final spin based on the extrapolation of test - particle calculations to finite mass ratios have been developed in @xcite , and show remarkably good agreement with numerical results in the comparable - mass regime . a comprehensive review of all spin formulae is beyond the scope of this article , but we refer the reader to the review in @xcite and the discussion in sec . v of @xcite . if we assume an ensemble of bh binaries with initially randomly oriented spins , the final spin distribution is peaked around @xmath160 @xcite . @xcite reported spin flips by @xmath161 with respect to the initial spin direction of the larger hole ; spin flips of this magnitude could provide an explanation for x - shaped radio sources @xcite . _ @xcite demonstrated that spin precession over the course of a long inspiral of thousands of orbits tends to align ( antialign ) the binary bh spins with each other if the spin of the more massive bh is initially partially aligned ( antialigned ) with the orbital angular momentum , thus increasing ( decreasing ) the average final spin . spin precession is stronger for comparable masses , and it could produce significant spin alignment before merger for both smbhs and stellar - mass bh binaries . * _ gravitational recoil : _ * one of the most spectacular results obtained from numerical bh binary simulations is the quantitative prediction of the magnitude of _ gravitational recoils _ ( or _ kicks _ ) . in the 1960s it was realized that the emission of linear momentum via gws must impart a kick on the source due to momentum conservation @xcite . in the 1980s , the kick generated in compact binary inspirals and plunges was studied in the framework of pn theory and bh perturbation theory @xcite . however , the astrophysical relevance of the gravitational recoil following bh binary mergers remained an open question until the recent numerical relativity breakthroughs . for the case of an equal - mass , nonspinning bh binary , the net linear momentum emitted in gws vanishes due to symmetry . nonzero recoils are therefore only generated in systems where this symmetry is broken through ( i ) a mass ratio @xmath162 ( equivalently , a symmetric mass - ratio parameter @xmath163 ) or ( ii ) nonvanishing spins . for zero spins , early numerical studies in the range @xmath164 $ ] found that the kick velocity is well approximated by @xcite @xmath165 this result ( whose functional form is inspired by fitchett s analytical work @xcite ) was confirmed by simulations for smaller mass ratios ( @xmath166 , @xcite ) and by analytical work @xcite . the maximal recoil obtained from eq . ( [ eq : nonspinningkick ] ) is @xmath167 for @xmath168 . small orbital eccentricity @xmath169 should introduce corrections proportional to @xmath170 @xcite . spinning binaries are characterized by seven free parameters ( the mass ratio plus three components for each bh spin ) , and numerical studies inevitably focussed on subsets of the parameter space . it soon became clear that the spin interaction dominates over mass ratio effects . the first studies for equal - mass binaries with spins parallel to the orbital angular momentum revealed kicks of several hundreds km / s , with an extrapolated maximum of @xmath171 for extremal spin magnitudes @xcite . shortly thereafter , the discovery of the so - called _ superkicks _ marked one of the most surprising outcomes of numerical relativity : binaries for which the spins are perpendicular to the orbital angular momentum and anti - aligned with each other can generate kicks of _ thousands _ of km / s , with an extrapolated maximum @xmath172 for near - extremal spin magnitudes @xcite . most recently , bh binaries with spins partially aligned with the orbital angular momentum but whose spin projections into the orbital plane still correspond to the superkick configurations have been found to result in even larger maximum recoils of up to @xmath173 km / s @xcite , so - called `` hang - up kicks '' . many galaxies harbor smbhs at their centers , and galaxy mergers should generally lead to the merger of their central bhs @xcite . typical escape velocities range from @xmath174 for dwarf galaxies up to @xmath175 for giant elliptic galaxies @xcite . large kicks would displace or eject the merged hole from its host , with possibly observable consequences : a softening of the stellar density gradient in the galactic nucleus , off - center radio - loud active galactic nuclei , off - nuclear x - ray sources in nearby galaxies and the generation of electromagnetic signals via interaction of the bh with its gaseous environment @xcite . bh ejection represents a potential obstacle for bh growth via merger , and thus puts constraints on merger - history models , which must be able to explain the assembly of smbhs by redshifts @xmath176 @xcite . observed redshifts of broad - line relative to narrow - line regions in quasar spectra may be interpreted as a smoking gun of bh ejection due to gravitational recoil @xcite , but alternative interpretations ( such as a bh binary , or superposed emission regions from two interacting galaxies ) are possible @xcite . the interest in astrophysical consequences of large recoils led to numerical studies of the bh parameter space @xcite . phenomenological kick formulas inspired by pn studies @xcite and similar to eq . ( [ eq : nonspinningkick ] ) were proposed to map the input parameters of a given binary configuration to the final kick velocity . available numerical results span mass ratios in the range @xmath177 and spin magnitudes @xmath178 , and are well described by eq . ( 2 ) of ref . @xcite ( see also @xcite ) . the apparent incompatibility between large recoils and the existence of bhs at galactic centers can be resolved by mechanisms that would _ align _ the individual bh spins with the orbital angular momentum of the binary . one such mechanism are gas torques in the so - called `` wet '' ( gas - rich ) mergers , that would produce partial alignment `` early on '' in the inspiral phase @xcite . if partial alignment in gas - rich mergers is the norm , as suggested also by the spin measurements discussed above , pn spin effects will lead to _ further _ alignment of the spins with the orbital angular momentum , significantly reducing the typical values of recoil velocities @xcite . this mechanism has even been shown capable of efficient suppression of large recoil velocities in the case of the above mentioned hang - up kicks @xcite . * _ merger simulations with matter and the blandford - znajek effect : _ * many numerical relativity groups are presently working on binary bh simulations in the presence of matter . most of this work is trying to understand the signature of electromagnetic counterparts to binary bh mergers : for example , such a signature could be produced by merger events @xcite or recoiling bhs @xcite `` shocking '' the surrounding accretion disks . the georgia tech group presented the first simulations in full gr of equal - mass , spinning bhs merging in a gas cloud . they found that shocks , accretion and relativistic beaming can produce electromagnetic signatures correlated with gws , especially when spins are aligned with the orbital axis @xcite . later work by the group focussed on hot , radiatively inefficient accretion flows . in this case , bh binaries exhibit a flare followed by a sudden drop in luminosity associated with the merger , and quasi - periodic oscillations correlated with the gws during inspiral @xcite . the urbana group simulated equal - mass , nonspinning bh binaries embedded in gas clouds under different assumptions for the motion of the binary with respect to the gas . they found evidence that the accretion rate and luminosity due to bremsstrahlung and synchrotron emission would be enhanced with respect to a single bh of the same mass as the binary , possibly being detectable by lsst @xcite . recent work suggest that the circumbinary disk surrounding bh binaries should not produce detectable electromagnetic counterparts @xcite . however , the _ magnetic fields _ produced by the circumbinary disk could affect the binary dynamics @xcite , produce stronger electromagnetic counterparts @xcite and extract energy from the orbiting bhs , which ultimately merge within the standard blandford - znajek scenario , generating electromagnetic emission along dual or single jets that could be observable to large distance @xcite ( but see also @xcite ) . for further reading , we recommend the following reviews . an excellent summary of bh mass and spin measurements and ( possible ) evidence for event horizons based on `` traditional '' electromagnetic astronomy is given by narayan @xcite . the use of electromagnetic observations of bhs and neutron stars to probe strong - field gravity is reviewed by psaltis @xcite . a more extended summary of numerical results on gravitational recoil can be found in zlochower _ _ @xcite . for a more general review of numerical bh simulations we recommend centrella _ the breakthroughs in numerical relativity have opened the door to tackle many fundamental problems in bh physics and to address questions of wider interest . a new golden era in bh physics is just starting . we summarize below what we think have been the main developments in this relatively short period of time . * _ mathematical physics and fundamental issues . _ * a decades - old problem in gr concerns the high - energy collision of two bhs . due to the dominance of the gravitational interaction at large energies , this process is thought to describe general trans - planckian scattering of particles : at very large center of mass ( cm ) energies all interactions are `` frozen '' while gravity is boosted , thus all the details about internal structure of the colliding particles should be washed out . this is further supported by thorne s hoop conjecture ( recently tested in highly dynamical situations @xcite ) , which predicts bh formation from generic high - energy collisions of particles @xcite . because the final bh horizon cloaks all details about the structure of the colliding objects , _ matter does not matter _ and one can for simplicity study bhs as representing a wide class of colliding objects @xcite . for large cm energies , bh collisions are arguably the most violent and nonlinear process one could conceive of . evolving einstein s equations in such extreme regimes poses new problems , such as the need to deal with all the different scales involved , and raises several questions . a fundamental issue concerns cosmic censorship . in four spacetime dimensions , bhs have an upper bound on their angular momentum , given by , while that of a spinning top @xmath179 , so ( even though the event horizon generators move at the speed of light ) bhs rotate `` slowly '' , as measured by their kerr parameters . ] @xmath180 is it possible to tune the impact parameter in such a way as to produce a final object spinning above the kerr bound , or does nature somehow conspire to always radiate enough angular momentum as to produce a bh ? this problem has been addressed in recent years @xcite . > b_{\rm scat}$ ] ) , a prompt merger ( @xmath181<b^*$ ] ) and a nonprompt merger ( @xmath182<b_{\rm scat}$ ] ) . taken from ref . , width=151 ] we show the trajectory of two equal - mass , nonspinning , colliding bhs in fig . [ fig : crit_traj ] , for cm velocities of around @xmath183 . we notice three distinct regimes . for impact parameters above the _ scattering threshold _ ( @xmath184 , the gravitational interaction between the two colliding bhs is relatively weak and they scatter to infinity . for impact parameters below the _ threshold of immediate merger _ ( @xmath185 ) , the bhs promptly collide and merge into a single bh . the transition between these two states for @xmath186 is a nonprompt merger , which puts the bhs in a so - called zoom - whirl orbit , i.e. the number of orbits @xmath187 . the results show that delayed mergers are necessary in order for the system to radiate excess angular momentum , so that the end product obeys ( [ kerrbound ] ) . the radiated energy and angular momentum increase strongly as the impact parameter approaches the threshold of immediate merger , as can be seen in fig . [ fig : e_b ] . for _ head - on _ collisions , numerical results suggest that a fraction @xmath188 of the cm energy is radiated in the limit of very large cm energy @xcite . for finely tuned impact parameters , the total radiated energy ( in gws ) can be of order @xmath189 for a collision with @xmath190 in the cm frame , while the spin of the final hole can reach @xmath191 of the bound ( [ kerrbound ] ) . an open issue is how much cm energy can be radiated for larger boosts . finally , peak luminosities of around @xmath192 are attained in these simulations . this number is orders of magnitude above any electromagnetic phenomena and close to the maximum conjectured possible value @xcite . the scattering threshold as a function of boost was studied by shibata _ et al . _ and sperhake _ _ @xcite . for the range of boosts studied it is well approximated by @xmath193 here @xmath78 is the total adm mass and @xmath194 the velocity of each hole as measured in the cm frame . * _ higher - dimensional black holes . _ * although bh solutions and their properties should in principle fall under the `` mathematical physics '' category , the activity in this specific field has been so intense that it deserves a subsection of its own . asymptotically flat higher - dimensional black objects have a much richer structure than their four - dimensional counterparts . for instance , spherical topology is not the only allowed topology for objects with a horizon . one can also have , _ e.g. _ , black rings , with a donut - like topology . remarkably , these two different horizon topologies coexist for certain regions in phase - space @xcite . explorations of the stability of general higher - dimensional bhs are in their infancy . generically it has been conjectured that , for @xmath195 , ultra - spinning myers - perry bhs will be unstable @xcite . this instability has been confirmed by an analysis of linearized axisymmetric perturbations in @xmath196 @xcite . clearly , the study of the nonlinear development of these instabilities requires numerical methods , such as the ones reviewed here . a study of this type was very recently presented for nonaxisymmetric perturbations in @xmath197 @xcite , where it was found that a single spinning five dimensional myers - perry bh is unstable , for sufficiently large rotation parameter ( confirming previous conjectures @xcite ) . general equilibrium states in anti - de sitter backgrounds only recently started to be explored : nonaxisymmetric solutions have finally been built @xcite , confirming previous conjectures about their existence @xcite , and braneworld bhs are now also starting to be explored @xcite . fully dynamical situations are still uncharted territory . * _ tev - scale gravity scenarios . _ * the above studies are of direct relevance to some high - energy scenarios . an outstanding problem in high - energy physics is the extremely large ratio between the four dimensional planck scale , @xmath198 gev , and the electroweak scale , @xmath199 gev . it has been proposed that this _ hierarchy problem _ can be resolved by confining the standard model to a brane in a higher dimensional space @xcite . in such models , the fundamental planck scale the energy at which gravitational interactions become strong could be as low as 1 tev . thus , high - energy colliders , such as the large hadron collider ( lhc ) , may directly probe strongly coupled gravitational physics @xcite . in fact , such tests may even be routinely available in the collisions of ultra high - energy cosmic rays with the earth s atmosphere @xcite , or in astrophysical bh environments @xcite ( for reviews see @xcite ) . the production of bhs at trans - planckian collision energies ( compared to the fundamental planck scale ) should be well described by using classical gr extended to @xmath9 dimensions @xcite . the challenge is then to use the classical framework to determine the cross section for production and , for each initial setup , the fractions of the collision energy and angular momentum that are lost in the higher - dimensional space by gw emission . this information will be of paramount importance to improve the modelling of microscopic bh production in event generators . the first models for bh production in parton - parton collisions used a simple black disk approach to estimate the cross section for production @xcite . as we already described , only recently exact results for highly relativistic collisions where obtained in four dimensions , using numerical relativity techniques @xcite . the extension to higher - dimensional spacetimes is a topic of current investigations @xcite . the first numerical results concerning low - energy collisions have been reported last year @xcite , initial data for boosted bh binaries have been constructed using an extended puncture approach in @xcite and some results for high - energy collisions have been presented in @xcite . * _ ads / cft and holography . _ * in 199798 , a powerful new technique known as the ads / cft correspondence or , more generally , the gauge - gravity duality , was introduced and rapidly developed @xcite . this holographic correspondence provides an effective description of a nonperturbative , strongly coupled regime of certain gauge theories in terms of higher - dimensional classical gravity in ads backgrounds . in particular , equilibrium and nonequilibrium properties of strongly coupled thermal gauge theories are related to the physics of higher - dimensional bhs , black branes and their fluctuations . these studies revealed intriguing connections between the dynamics of bh horizons and hydrodynamics , and offer new perspectives on notoriously difficult problems , such as the bh information loss paradox , the nature of bh singularities and quantum gravity . numerical relativity in anti - de sitter backgrounds is bound to contribute enormously to our understanding of the gauge - gravity duality and is likely to have important applications in the interpretation of observations @xcite . for instance , in the context of the gauge - gravity duality , high - energy collisions of bhs have a dual description in terms of _ a ) _ high - energy collisions with balls of deconfined plasma surrounded by a confining phase , and _ b ) _ the rapid localized heating of a deconfined plasma . these are the type of events that may have direct observational consequences for the experiments at brookhaven s relativistic heavy ion collider ( rhic ) @xcite . numerical relativity in anti - de sitter is particularly difficult , and so far only very special situations have been handled @xcite . among these , we note the recent numerical proof that pure ads space is unstable against collapse to bhs @xcite . bhs in the context of high - energy physics have also been discussed by pretorius @xcite and in the white paper @xcite . a recent review of numerical simulations of black strings in five spacetime dimensions is given by lehner and pretorius @xcite . an overview about numerical results on bhs in @xmath200 dimensions is given in @xcite . finally we note kanti s review on bhs at the lhc @xcite . we conclude this review with a brief summary of the most urgent problems to be tackled by numerical relativity in the three main areas in the near future . * _ gravitational wave physics : _ * in order to meet tight accuracy requirements in gw template generation , especially for parameter estimation , we either need longer numerical waveforms or a denser spacing of templates in the parameter space . both approaches will require computational resources that grow non - linearly as a function of the accuracy , either because of the slow nature of the inspiral at larger bh separations ( see sec . 2 in @xcite for a quantitative discussion ) or because of the dimensionality of the parameter space . a second major task is the extension of existing template models to generic binaries with precessing spins and/or smaller mass ratios , bridging the gap to perturbative modeling of extreme mass ratio binaries . * _ astrophysics : _ * while existing formulae for the kick and final spin resulting from the coalescence of bhs have been helpful for astrophysical studies , the calibration of generic models as proposed by boyle _ @xcite requires a more comprehensive exploration of the parameter space . the relatively young field of numerical relativity simulations of bhs surrounded by accretion disks will likely improve our insight into expected optical counterparts to bh binary mergers , and thus provide a vital tool for the area of _ multi - messanger astrophysics_. * _ high - energy physics : _ * arguably the most urgent task is the determination of the scattering cross - section and the loss of energy and momentum in gws in trans - planckian scattering of bhs in _ arbitrary _ dimensions . in view of the stability problems encountered in present studies , it also appears desirable to obtain a better understanding of the well - posedness of the formulations currently employed for such studies . second , the modeling of bhs in generic spacetimes such as de sitter and anti - de sitter is still in its infancy , even if preliminary studies of bhs in non - asymptotically flat spacetimes are encouraging . a better understanding of boundary issues , well - posedness of the formulations and diagnostics such as wave extraction tools and horizon finders will be required to open up this uncharted and fertile ground of research . * acknowledgements : * u.s . acknowledges support from the ramn y cajal programme and grant no . fis2011 - 30145-c03 - 03 of the ministry of education and science of spain , the fp7-people-2011-cig grant _ cbheo 293412 _ , nsf grants phy-0601459 , phy-0652995 and the sherman fairchild foundation to caltech . e.b.s research is supported by nsf grant phy-0900735 and by nsf career grant phy-1055103 . this work was supported by the dybho-256667 erc starting grant , and by fct - portugal through projects ptdc / fis/098025/2008 , ptdc / fis/098032/2008 , cte - ast/098034/2008 , cern / fp/116341/ 2010 and cern / fp/123593/2011 . this research was supported by the dybho256667 erc starting grant , the nrhep295189 fp7people2011irses grant by allocations through the teragrid advanced support program at the san diego supercomputing center and the national institute for computational sciences under grant phy-090003 , the centro de supercomputacin de galicia ( cesga ) under project numbers icts - cesga-175 , icts - cesga-200 and icts - cesga-221 , the barcelona supercomputing center ( bsc ) under project aect-2011 - 2 - 0006 , aect-2011 - 3 - 0007 , aect-2012 - 1 - 0008 , the deisa extreme computing initiatives deci-6 and deci-7 prace-2ip project dybho .

in particle physics there is a number of well - established principles ( four - dimensional lorentz invariance , renormalizability , gauge symmetries , etc . ) , which are believed to underlie a consistent quantum field theory . all these principles are built into the standard model ( sm ) and its great experimental success is a major justification for their validity . however , in continuing search for the consistent description of an emerging physics beyond the standard model , one often tries to ease some of these criteria . it is then important to make sure that such modifications do not contradict to the observable data , as well as to find observable signatures of a new model . since 1950s @xcite , one uses the gauge symmetry as a main principle for constructing interactions between elementary particles . it is known , however , that in theories with chiral couplings ( i.e. when left and right - handed fermions have different charges with respect to a gauge group ) , one may encounter a situation when the classically conserved gauge current acquires an anomalous divergence at one loop . it is known that such anomalies destroy consistency of the theory , making it non - unitary @xcite . meson into two @xmath0-quanta ( famous abj - anomaly @xcite ) , the baryon and lepton number non - conservation in electroweak theory @xcite , and proton decay in the presence of magnetic monopole @xcite . ] in the standard model , where the couplings of fermions are chiral , charges of all particles can be chosen to ensure the absence of all gauge anomalies @xcite . it may happen , however , that anomaly - free theory looks like an anomalous one below certain energy scale . an example of such a theory would be an electroweak theory at the energy scale below the mass of top - quark . then anomaly is canceled by means of wess - zumino like terms @xcite . among many possible extensions of the standard model there are theories with extra dimensions . their main idea is that our 4-dimensional physics is embedded into a theory in a higher - dimensional space . the standard model fields are then realized as zero modes of a higher - dimensional ones . ( an example is given by an approach , often dubbed _ brane - world _ models , where zero modes of both matter and gauge fields are localized on the 4-dimensional defect , called _ brane _ @xcite ) . once our 4-dimensional world is just a low - energy sector of a bigger theory , the anomaly analysis changes drastically . for example , if a full theory is anomaly free ( say , it is formulated in an odd number of space - time dimensions , where all interactions are vector - like ) , then one has no reason to expect separate anomaly cancellation for the brane fields ( for an explicit example of a brane - world theory with an anomaly on the brane , see @xcite ) . if the theory on the brane is anomalous , then there is a specific type of _ brane - bulk interaction_. it is described by chern - simons - like terms in the low - energy effective action in the bulk . these terms are not gauge invariant in the presence of a brane . therefore , they generate currents , flowing to the brane and thus ensuring the gauge invariance of the full theory . such mechanism is known for a long time and is often called _ anomaly inflow _ it is quite general and appears in many ( physically very distinct ) problems : in quantum hall effect @xcite , dimensional `` brane '' , embedded into @xmath1 dimensional bulk . effective theory in the bulk is described by the @xmath2 chern - simons theory , while on the boundary chiral excitations are localized . conformal field theory describing these excitations was first derived from anomaly consideration @xcite . ] in various field theories with solitonic objects in them ( see e.g. @xcite ) , in string and m - theory @xcite . thus , in the brane - world models a number of questions arises : 1 . [ item:1 ] can one distinguish between a purely 4-dimensional anomaly - free theory , and a theory ( also anomaly - free ! ) in which anomaly is canceled by a small inflow from extra dimensions ? what new observable effects appear in the latter situation ? 2 . [ item:4 ] is it possible to discriminate at low - energies between two different completions of anomalous 4-dimensional effective theory : the one , where anomaly is canceled by inflow from higher dimensions and purely 4-dimensional one , where at higher energies there exist additional particles , which ensure anomaly cancellation ( c.f . [ item:2 ] for generic values of hypercharges the minimal standard model is anomalous . however , for any hypercharges , electrodynamics is still vector - like . can this anomaly nevertheless be observed if this anomalous theory is embedded into a consistent higher - dimensional one ? [ item:3 ] what are experimental constraints for the parameters describing anomaly mismatch ? can the above observations serve as tests in the search for extra dimensions ? in this paper we will attempt to answer the questions [ item:1][item:2 ] . the question [ item:3 ] will be addressed in our next paper @xcite . we would like to note also that in this paper ( as well as in the subsequent paper @xcite ) we study only gauge anomalies . the global chiral anomalies in brane - worlds and corresponding complicated vacuum structure of gauge fields has been discussed in @xcite . we consider a theory on a brane , which is anomalous , while the bulk effective action possesses a chern - simons - like term , responsible for an inflow current . a naive answer to the question ( [ item:1 ] ) then would seem to be the following : as anomaly means gauge current non - conservation , one could expect that particles escape from the brane to extra dimensions . this would look like a loss of unitarity from a purely 4-dimensional point of view , however higher - dimensional unitarity would still be preserved . anomalous current then is simply a flow of such zero - mode particles into the bulk . however , this answer is incorrect . indeed , phenomenologically acceptable brane - world scenario must have a mass gap between zero modes and those in the bulk for both fermions and non - abelian gauge fields @xcite . alternatively , the rate of escape of the matter from a brane must be suppressed in some way ( see e.g. @xcite ) . as a result , the low - energy processes on the brane simply can not create an excitation , propagating in the bulk . if no matter can escape the brane ( or flow onto it ) what is then an inflow current , which should be non - zero even for weak fields causing anomaly in four - dimensional theory ? one seems to be presented with a paradox . we show that in reality the inflow current by its nature is a _ vacuum _ or _ non - dissipative _ current . what we mean by that is the following . usually in electrodynamics one needs to create particles from a vacuum to generate an electric current . such a process is only possible if the available energy is larger than the mass of the particles in the bulk . therefore , weak electric fields on the brane can not induce a current , carried by the bulk modes . however , anomaly inflow current is different . first , it is perpendicular to the field , does not perform a work and thus is not suppressed by the mass of a particle in the bulk . second , such a current is not carried by any real particles excited from the vacuum , being rather a collective effect , resulting from a rearrangement of the dirac sea . this simple observation is in fact very important . essentially , it means that anomaly inflow is a very special type of brane - bulk interaction . namely , we show that in the presence of parallel magnetic and electric fields the latter changes as if photon had acquired mass . as a result the spatial distribution of an electric field changes , in particular it appears outside the plates of the capacitor . similarly , an electric field of an point - like electric charge changes from coulomb to yukawa , when the charge is placed in a magnetic field . the dynamics of the screening of electric charge happens to be quite peculiar and can be shown to lead to the temporary appearance of _ dipole moment _ of elementary particles . at first , we conduct our analysis in @xmath2 gauge theory with the chiral couplings to a matter . one example of such a theory would be an electrodynamics , where left and right - handed particles have different electric charges . the masses of such particles can then be generated only via the higgs mechanism with a charged higgs field , which means that a photon in such a theory becomes massive . a different example is a theory , where axial current comes from a neutrino with a small electric charge . in such theory photon would remain massless . then we analyze the version of the standard model where electric charges of electron and proton differ . we show that although low - energy electrodynamics in this theory is vector - like , there exist effects , similar to those , described above . we would like to stress again that in the theories with extra dimensions there is no immediate reason to expect that four - dimensional anomaly is so small as it follows already from present experimental constraints on the corresponding parameters . the effects which would follow from the existence of an anomaly inflow are observable , but quite peculiar . they do not change drastically physical content of the theory and can be ruled out or constrained only experimentally . thus in those brane - world scenarios , in which _ higher - dimensional _ theory is automatically anomaly free , we are facing yet another fine - tuning puzzle . for distinctness we will often refer to a simple brane - world : the fermions in the background of a kink , realizing a 4-dimensional brane , embedded into a 5-dimensional space as in @xcite ( although main effects are believed to be model - independent ) . we localize gauge fields in a spirit of @xcite , introducing an exponential warp - factor . such warp - factor can arise dynamically because of the fermionic zero modes @xcite . zero modes of both types of fields are separated by mass gaps from the corresponding bulk modes . the paper is organized as follows . we begin section [ sec : qed ] with describing a model of chiral electrodynamics , embedded into a 5-dimensional @xmath2 theory and analyze the microscopic structure of the inflow and details of anomaly cancellation . next , we turn in section [ sec : anom - sm ] to the question of possible extensions of the standard model , such that it would admit an anomaly . we consider one - parameter family of such modifications , with the free parameter being the difference of ( absolute values of ) electric charges of proton and electron . in such model particles also acquire an anomalous dipole moment . we discuss it in the section [ sec : proton ] . we conclude with a discussion of future extensions of this work and some speculations . finally , we would like to stress that in this paper we conducted analysis on the level of classical equations of motion . we discuss some aspects of quantum theory and the validity of our approach in [ sec : disc ] . we start our analysis considering in this section the simplest model which nevertheless catches main effects we are interested in . namely , we will discuss a four - dimensional @xmath2 theory with anomalous chiral couplings to the fermions . such a theory suffers from a gauge anomaly , meaning that on the quantum level its unitarity is lost @xcite . thus it can not describe physics of a 4-dimensional world . however , if one thinks about such an anomalous theory as living on a brane , embedded in a 5-dimensional theory , situation changes . five - dimensional theory is anomaly - free ( all couplings there are vector - like ) and therefore there should exist a special type of interaction ( _ anomaly inflow _ @xcite ) between the theories on the brane and in the bulk , which ensures the unitarity and consistency of the total system . below we demonstrate this mechanism in details . for definiteness we will illustrate our steps by using a concrete model of localization of both fermions and gauge fields . consider the following theory : , greek @xmath3 . we are using _ mostly negative _ metric . our brane is stretched in @xmath4 directions and is located at @xmath5 . we will often use notations @xmath6 instead of @xmath7 and choose polar coordinates @xmath8 in the plane @xmath9.[fn:1 ] ] @xmath10 here @xmath11 is a five - dimensional charge , with the dimensionality of @xmath12 . there are two fermions @xmath13 , interacting with the gauge fields with the different charges : @xmath14 . the fermionic mass terms @xmath15 have a `` kink - like '' structure in the direction @xmath16 , realizing the brane : @xmath17 . as it is well - known ( see e.g. @xcite ) , in this model one has chiral fermion zero modes localized on the brane and a continuous spectrum in the bulk , separated from the zero mode by the mass gap @xmath18 . localization of the zero mode of gauge field is achieved by introduction of a warp - factor @xmath19 . whenever @xmath20 is convergent the system has a zero mode with the constant profile in the 5th direction . warp - factor @xmath19 looks like an @xmath16 dependent coupling constant in the action ( [ eq:1 ] ) , getting infinite as @xmath21 . the lattice study of @xcite has confirmed that the perturbative and non - perturbative analysis of ( [ eq:1 ] ) give the same result for the zero mode up to the energies of the order of the mass gap , at least for @xmath22 reduction . we will assume that this is the case for @xmath23 . depending on the warp - factor , the spectrum of gauge field may be discrete or continuous with or without the mass gap between the zero and bulk modes @xcite . we will often use @xmath24 in which case the zero mode is separated from the bulk continuum by the mass gap @xmath25 @xcite . the nature of the warp - factor @xmath19 is not essential here , it can originate from that of the metric or can arise dynamically , coming from fermionic zero modes @xcite . a low - energy effective action for the gauge fields of the theory ( [ eq:1 ] ) will contain terms , describing interaction between the brane and the bulk : @xmath26 the origin of two last terms is the following . the computation of the fermionic path integral of our theory is separated into the determinant of the 5-dimensional bulk modes of the fermions and that of the 4-dimensional chiral fermionic zero modes . in the former case one can show ( essentially repeating the computations of @xcite ) , that after integrating out the non - zero modes of the fermions , the 5-dimensional effective action of the model ( [ eq:1 ] ) acquires an additional contribution of chern - simons type @xmath27 compared to the usual case @xcite , chern - simons term in models like ( [ eq:1 ] ) acquires additional factor @xmath28 . it is proportional to the difference of the charges of the five - dimensional fermions and depends on the profile of the domain wall . in the limit of an infinitely thin wall and the large fermionic mass gap @xmath29 . below we sketch the derivation of this function . terms like generically appear in the theories with anomaly inflow @xcite . in situations with more than one extra dimension an effective action will have similar structure , with the fifth coordinate being suitably chosen radial coordinate . such terms can also appear for non - abelian fields . evaluating of the determinant of the 4-dimensional dirac operator in the background of the gauge field ( and , possibly , higgs field , see below ) , we get @xmath30 . let us denote the _ 4-dimensional _ charges of the left and right - handed fermionic zero modes as @xmath31 and @xmath32 . they are proportional to their 5-dimensional counterparts @xmath33 and @xmath34 and thus @xmath35 . as a result @xmath30 is _ not gauge invariant_. the manifestation of this fact is the anomalous divergence of the gauge current at one - loop level : @xmath36 the action ( [ eq:3 ] ) is also non - gauge invariant , and the variations of these two terms cancel each other ( as we will demonstrate below ) . divergence of the current ( [ eq:5 ] ) is proportional to the transversal profile of the fermion zero modes . in this paper we will be interested in the characteristic energy scales much less than the mass gap of the gauge fields @xmath25 , which is in turn much less than the fermionic mass gap : @xmath37 . therefore the profile of the fermion zero mode can be approximated by the delta - function @xmath38 , which appears in eq . ( [ eq:5 ] ) . equations of motion coming from the action ( [ eq:2 ] ) are : @xmath39 and @xmath40 in the right hand side of eq . ( [ eq:6 ] ) there is a current of the zero modes @xmath41 ( we will omit brackets @xmath42 from now on ) . the variation of the effective action @xmath30 with respect to the @xmath43 is equal to zero .. zero modes of @xmath44 and @xmath45 have definite 4-dimensional chirality . therefore @xmath46 and analogously for @xmath45 . ] the current @xmath47 in the right hand side of eqs . ( [ eq:6])([eq:7 ] ) is defined via @xmath48 note that in the right hand side of eq . ( [ eq:8 ] ) there is a term , proportional to the derivative of @xmath28 . this term is non - zero only in the vicinity of the brane ( as @xmath49 is proportional to the delta - function of @xmath16 ) . thus , essentially it corresponds to the modification of definition of the brane current @xmath50 ( [ eq:5 ] ) . this additional term is responsible for making anomaly of the zero modes _ gauge invariant _ @xcite , as the inflow current ( [ eq:5 ] ) is always gauge invariant in the @xmath2 case . thus , terms proportional to @xmath49 correspond to the local counterterms , shifting between the _ covariant _ and _ consistent _ anomalies @xcite . in what follows we will always consider covariant anomaly of the brane modes and ignore terms in the bulk , proportional to the @xmath49 . the chern - simons current @xmath51 is a non - dissipative one as it does not perform a work due to its antisymmetric structure . in addition to that it does not depend on the mass of the fermions in the bulk and hence is not suppressed by the mass gap . if a gauge field has four - dimensional components with @xmath52 , there exists a current @xmath53 , flowing onto the brane from the extra dimension @xcite . trivial consequence of eqs . ( [ eq:6])([eq:7 ] ) is @xmath54 it means that if coefficients in front of the @xmath55 and @xmath30 were not correlated , by virtue of ( [ eq:5 ] ) the solutions of eq . would be field configurations with @xmath56 . however , to preserve the gauge - invariance in the action the divergences of currents in the left hand side of should be equal . from the definition ( [ eq:8 ] ) one can see that @xmath57 comparing ( [ eq:10 ] ) with eq . ( [ eq:5 ] ) we can choose the coefficient of @xmath58 of the function @xmath28 so that the current @xmath47 cancels the anomalous divergence of the current @xmath41 and make the theory ( [ eq:2 ] ) consistent and gauge - invariant . namely @xmath59 microscopical calculations of @xmath28 confirm that this is indeed the case . the relation between 5-dimensional fields @xmath60 , entering eqs . ( [ eq:6])([eq:7 ] ) and the 4-dimensional ones is given by : @xmath61 here @xmath62 is a _ four - dimensional @xmath2 coupling constant _ , related to the 5-dimensional one via @xmath63 ( to see that , one should substitute @xmath60 , independent on the 5th coordinate , into the kinetic term in ( [ eq:2 ] ) and integrate over @xmath16 ) . field @xmath64 satisfies the 4-dimensional maxwell equations @xmath65 with the current in the right hand side given by : @xmath66 this current is conserved : @xmath67 as a consequence of eqs . ( [ eq:7 ] ) and ( [ eq:9 ] ) . in the theory ( [ eq:1 ] ) fermionic zero modes on the brane are massless . to make this model more realistic , we should add mass to these zero modes as well . the only way to make the electro - dynamics anomalous is to take left and right moving fermions with different electric charges . thus one can only introduce a mass term via the higgs mechanism with an electrically charged higgs field , i.e. to add the following term to the bulk lagrangian ( [ eq:1 ] ) : @xmath68\:,\ ] ] where @xmath69 and the higgs mass @xmath70 is negative at @xmath71 and tends to the positive constant in the bulk , as @xmath72 ( figure [ fig : higgs ] ) . thus higgs field has a non - trivial profile @xmath73 , with @xmath74 and @xmath75 at @xmath76 . so the symmetry is restored far from the domain wall . the higgs mass far from the brane is taken to be large , so that the scalar is localized on a brane . the higgs field has an expectation value , which on the brane is @xmath77 ( we take the scalar s self - coupling @xmath78 to be large , so that higgs mass @xmath79 is much bigger than @xmath80 yukawa mass of the fermions on the brane ) . as the higgs field is charged , the photon should acquire mass for @xmath81 ) . to determine it exactly one needs to find the spectrum of the theory with action given by the sum of ( [ eq:1 ] ) and ( [ eq:16 ] ) , but for us it is only important that @xmath82 , where parameter @xmath58 was defined in ( [ eq:11 ] ) . in section [ sec : anom - sm ] we will analyze the consequences of anomaly inflow in the standard model , where yukawa mass of the fermions does not necessarily lead to the massive photon . dynamics of the higgs field @xmath83 simplifies in the limit of the large ( compared to the energy scale of the processes on the brane ) vev @xmath84 . in this limit it is convenient to work in the _ unitary gauge _ , in which the phase of the higgs field is fixed ( e.g. via @xmath85 ) . in the regions of space where the gauge field has non - trivial @xmath60 , it is impossible to have @xmath86 @xcite . however , variations of the @xmath87 are suppressed by the higgs s vev , we can neglect them and take @xmath88 on the brane . now the vector field is massive and longitudinal component becomes a physical degree of freedom . from equations of motion one can obtain the following condition on the field @xmath89 : @xmath90 eq . ( [ eq:17 ] ) is just an expression of the gauge invariance of the total action . divergences of @xmath91 and @xmath47 precisely cancel each other ( [ eq:9 ] ) . therefore eq . ( [ eq:17 ] ) implies that @xmath92 once we have added the mass to the zero mode fermions , we can integrate them out and obtain effective action @xmath30 ( [ eq:2 ] ) , written in unitary gauge . the variation of this action with respect to @xmath93 gives gauge invariant current @xcite : @xmath94 we will often call this current _ the dhoker - farhi current _ denote it @xmath95 . in the unitary gauge eq . ( [ eq:19 ] ) becomes @xmath96 this current has a divergence , equal to that of the chern - simons current ( @xmath97 ) : @xmath98 a purpose of this work is to explore in details this anomaly cancellation mechanism . namely , _ what constitutes the inflow _ ? zero modes of fermions are confined to the brane and can not propagate in the bulk , while bulk modes are massive and can not be excited at low energies . another question is _ whether 4-dimensional observer can distinguish _ between an anomaly - free , intrinsically 4-dimensional theory and 5-dimensional ( anomaly - free ! ) brane - world theory , in which a very small anomaly of the 4-dimensional fermions on the brane is canceled by an inflow from the bulk . as we have seen , on the microscopic level anomaly inflow is a collective effect of reorganization of the dirac see of the massive fermions in the bulk in the presence of a field configuration with non - zero @xmath99 . the simplest way to obtain non - zero @xmath100 is to choose both @xmath101 and @xmath102 to be parallel to each other and constant ( in @xmath103 ) in some region on the brane . we realize such fields by two parallel infinite plates in the vacuum , separated in the direction @xmath104 by the distance @xmath105 a capacitor with initial charge densities @xmath106 on plates , placed in the magnetic field . magnetic field is created by an ( infinite in the @xmath107-direction ) solenoid with the radius @xmath108 . before these fields are turned on , the vacuum structure on the classical level is unchanged and there is no difference with non - anomalous theory . also , the effects of anomaly are controlled by the parameter @xmath58 , which is naturally assumed to be very small . that is why the most natural setup to study the anomaly inflow is qualitatively the following . we assume that the vacuum current is zero and an electric field is equal to its value for non - anomalous theory . then we turn on a parallel magnetic field and expect that small ( perturbative in @xmath58 ) inflow currents start to change the charge distribution on the brane trying to compensate the anomalous field configuration . according to this picture , in section [ sec : perturb ] we will choose initial condition describing the absence of anomalous vacuum structure and will consider an initial stage of inflow ( linear in time ) perturbatively , in the first order in @xmath58 . we will see , however , that this naive approach does not work and perturbation theory breaks down . therefore , we will have to study the full non - linear equations non - perturbatively in @xmath58 . then , as we will show in section [ sec : final - state ] , a non - analytic in @xmath58 static solution exists . the properties of this solution will define the main observable effect of the anomaly inflow , studied in the present paper . it is , nevertheless , instructive to study the dynamical problem to understand qualitatively how this state is formed . as we will see in section [ sec : old - part ] , from the 4-dimensional point of view the picture of the inflow given by perturbation theory is qualitatively correct and knowing the non - perturbative static solution one can define it properly . as in general we expect the parameter @xmath58 to be extremely small , we will try to solve the maxwell equations ( [ eq:6])([eq:7 ] ) by perturbation theory in @xmath58 . to find linear in time corrections to the fields at @xmath109 , we should specify the initial values of all fields . our initial conditions should satisfy the gauss constraint , @xmath110 , @xmath111 and choose polar coordinates @xmath8 in the plane of the capacitor @xmath9 ( c.f . footnote [ fn:1 ] ) . ] @xmath112 where @xmath113 and @xmath114 are expressed in terms of components @xmath115 and are both proportional to the @xmath58 ( c.f . ( [ eq:8 ] ) and ( [ eq:20 ] ) correspondingly ) . because we have added a charged higgs field ( [ eq:16 ] ) and the electrodynamics became massive , specifying initial values of @xmath115 is not enough and one should also set values of the gauge potential at @xmath109 . we choose all but @xmath116 components @xmath117 initially equal to zero ( see appendix [ app : linear ] for details ) . the time derivatives of @xmath117 can then be extracted knowing electric fields at @xmath109 and using condition ( [ eq:18 ] ) . in this paper we restrict our analysis only to such initial conditions . in general initial conditions on the vector field @xmath117 will depend on how one turns on electric and magnetic fields . corresponding analysis ( similar to that of @xcite ) will be presented elsewhere . in addition to these initial condition we will choose as a zeroth approximation such configuration of electromagnetic fields @xmath115 that @xmath118 . the reason for that is as follows : from ( [ eq:8 ] ) one has @xmath119 . in the absence of anomaly @xmath120 would be equal to zero for our configuration of charges . therefore , it is natural to assume that initially @xmath120 is equal to zero in the full theory as well . initial conditions @xmath121 and @xmath122 mean that we should prepare an electric field configuration , satisfying 5-dimensional gauss constraint in the theory with @xmath123 ( we will mark these fields by the symbol @xmath124 ) . then we find ( perturbative ) corrections to these fields due to anomalous current and analyze effects arising on the initial ( linear ) stage of inflow . the details of the computations can be found in appendix [ app : naive ] , here we only quote the main results . with our choices for the fields @xmath125 , initially only time component of the dhoker - farhi current changes with time ( see appendix [ app : naive ] ) : @xmath126 if the theory were purely four - dimensional , eq . ( [ eq:35 ] ) would have described the rate of anomalous particle production , essentially the measure of non - unitarity of the theory . in the 5-dimensional brane - world the interpretation of eq . ( [ eq:35 ] ) is quite different . namely , one should not think about the inflow current , as `` bringing particles to the brane '' . this is obvious as all the light ( zero mass ) particles are confined to the brane and do not propagate in the bulk , while the fermions , which live in the bulk are too massive to be excited at low energies . thus , inflow current ( [ eq:8 ] ) is essentially a _ vacuum current _ redistribution of the particle density without actual creation of the charge carriers . this fact can be checked by the direct microscopic computations of the vacuum average of @xmath127 in the full theory ( [ eq:1 ] ) . the divergence @xmath128 is non - zero only in the position of the brane @xmath129 and thus modifies the charge density only there ( which is reflected by the delta - function in eq . ) . one may think of the effect of the vacuum current as of a dielectric susceptibility of the vacuum of four - dimensional theory , embedded in the 5-dimensional space - time . notice , that since there is no @xmath113 , eq . ( [ eq:35 ] ) means that _ the total electric charge density changes with time_. if the charge density changes in the finite volume on the brane , this leads to the change of both @xmath130 and @xmath131 . ( if initial field @xmath130 and @xmath132 uniformly filled all the brane and were localized in @xmath133 direction , then the appearance of the anomalous density would not lead to the change of the @xmath107 component of the electric field . the change in @xmath131 would be unobservable , since this field is antisymmetric in @xmath133 and so @xmath134 ) . for the case when @xmath100 is non - zero only in the finite volume on the brane we explicitly find the change in @xmath130 by solving maxwell equations ( [ eq:6])([eq:7 ] ) : @xmath135 now one can see that _ our perturbation theory approach has a problem_. indeed , although @xmath136 is proportional to @xmath58 , it grows in @xmath133 as @xmath137 while @xmath138 decays as @xmath139 . the correction term ( [ eq:109 ] ) becomes larger than @xmath138 at @xmath140 . ( [ eq:109 ] ) gives correction to the _ 5-dimensional _ electric field . the corresponding 4-dimensional field is obtained via ( [ eq:12 ] ) . as shown in appendix [ app : naive ] , the integral in ( [ eq:12 ] ) is saturated at @xmath141 where characteristic @xmath142 . therefore in the physically relevant region @xmath143 thus , we see that the naive form of perturbation theory does not work . it means that something was wrong with our qualitative picture of the inflow and that we have to study full equations non - perturbatively . we have seen in the previous section that effects , caused by anomaly inflow are non - perturbative . as the full system of non - linear maxwell equations is too complicated to solve exactly , we will compute a static solution . this means that we do not study how anomaly inflow modifies the initial fields , but rather describe a final state of the inflow . we will consider the simplest case : a capacitor with infinite plates , when these equations are reduced to the @xmath1 ones . below we will only sketch the computations , for details reader should refer to appendix [ sec : final - state - compute ] . we will return to the perturbation theory in the section [ sec : old - part ] . for static solution we can describe the electric field in terms of the electro - static potential @xmath144 : @xmath145 , @xmath146 . then the gauss constraint can be re - written as ( see appendix [ sec : final - state - compute ] for details ) : @xmath147 where @xmath148 is a 3-dimensional laplacian in the coordinates @xmath149 . components @xmath150 can all be expressed via single function @xmath116 ( because we choose all the gauge connection to be @xmath151 independent ) . the maxwell equation ( [ eq:6 ] ) for @xmath152 becomes then an equation for @xmath116 : @xmath153 we will analyze this solution as follows . first , we notice that if the right hand side of eq . ( [ eq:46 ] ) were identically zero , then there would be a solution @xmath154 , @xmath155 , @xmath156 . then one can easily find an explicit solution of eq . ( [ eq:45 ] ) ( under these assumptions gauss law becomes identical to that of the @xmath1 dimensional case with the point - particle as the source ) . after that we show that corrections to the found solutions of eqs . ( [ eq:45])([eq:46 ] ) due to the non - zero right hand side of ( [ eq:46 ] ) are of the order @xmath157 . note , that potentially the first terms in the right hand sides of both ( [ eq:45 ] ) and ( [ eq:46 ] ) can be non - perturbative , as they are proportional to the @xmath137 . this is precisely why the naive perturbation theory of the section [ sec : perturb ] did not work . after the fourier transform in @xmath149-directions : ( @xmath158 ) and a substitution @xmath159 equation ( [ eq:45 ] ) becomes in eq . ( [ eq:47 ] ) as compared to ( [ eq:45 ] ) . @xmath114 is proportional to @xmath160 , and therefore it can be treated perturbatively , as any function which is proportional to @xmath58 and does not have a growing profile in @xmath133 direction . ] @xmath161 where operator @xmath162 does not depend on @xmath163 : @xmath164 let us denote the eigen - function of ( [ eq:48 ] ) by @xmath165 and its eigen - values by @xmath166 : @xmath167 . one can show ( see appendix [ sec : final - state - compute ] ) that @xmath168 ( recall that @xmath169 is a _ four - dimensional coupling constant _ ( [ eq:13 ] ) ) . notice that @xmath170 depends only on the four - dimensional quantities ( electro - magnetic coupling constant @xmath169 and the magnetic field @xmath171 measured by a 4-dimensional observer ) ! the eigen - function for the eigen - value @xmath172 is given by @xmath173 here @xmath174 is a modified bessel function of the second kind ` . ] and @xmath175 is a normalization constant . the solution of eq . ( [ eq:45 ] ) can be easily found for any charge distribution @xmath176 . for the point - particle in the uniform magnetic field an electrostatic potential , created by the particle will be of the yukawa form : @xmath177 where the mass @xmath170 depends on the anomaly coefficient and the magnetic field @xmath132 ( [ eq:49 ] ) . the profile of the solution ( [ eq:58 ] ) in the @xmath133 direction @xmath178 is proportional to @xmath179 ( see also appendix [ sec : final - state - compute ] ) : @xmath180 ( where in the last equality we assumed that @xmath181 ) . function @xmath178 is sharply localized in the region @xmath182 indeed , asymptotics of the bessel function @xmath174 is @xmath183 for @xmath184 , therefore outside the specified region potential @xmath144 decays as an exponent @xmath185 . thus , the function @xmath186 exponentially decays in @xmath133 direction on the scale proportional to the @xmath187 and depending on @xmath58 ( potentially very small ) only logarithmically , while the scale @xmath188 of the exponential decay of the potential in the @xmath107 direction does not depend on @xmath25 and is proportional to the @xmath189 . as a result @xmath190 for any physically plausible values of magnetic field @xmath171 . notice , that the function @xmath174 is non - regular for small @xmath191 and the same is true for @xmath178 as a function of @xmath58 . now it is easy to show that equation ( [ eq:46 ] ) for @xmath116 only gives small corrections to the constant magnetic field @xmath171 . indeed , the right hand side of eq . ( [ eq:46 ] ) is proportional to the @xmath157 . in case of eq . ( [ eq:45 ] ) such term was non - perturbative , because it was divided by @xmath192 and thus could get arbitrarily large as @xmath139 . in the case at hand , however , the right hand side is proportional to the @xmath193 . as @xmath144 decays in @xmath133 much faster than @xmath192 , this term is always small and therefore all the corrections to the @xmath194 are of the order @xmath157 . similarly , there are corrections to the @xmath195 , which are of the order @xmath196 due to the presence of dhoker - farhi term @xmath114 . to summarize , we see from eq . ( [ eq:58 ] ) that anomaly inflow causes screening of an electric charge with the screening radius being @xmath188 . this means , that total amount of anomalous charge , which inflows on the brane is equal to an initial electric charge @xmath197 of a particle . similarly to eq . ( [ eq:58 ] ) , in case of capacitor with infinite plates one can see that for @xmath198 the expression @xmath144 is given by an electro - static potential created by two infinite charged plates in 3 spatial dimensions for a _ massive _ electric field with the mass @xmath170 : @xmath199 ( i.e. @xmath200 ) . constructing 4-dimensional electric field from ( [ eq:51 ] ) we get : @xmath201 } \ : . \ ] ] here @xmath202 is a value of 4-dimensional field , which would be created between the plates of the capacitor in the theory without anomaly @xmath203 . in the presence of anomaly distribution of an electric field changes . inside a capacitor it diminishes : @xmath204 but appears outside : @xmath205 as one can see , inside the capacitor the field is smaller , than would be in the absence of anomaly , however , it appears outside the capacitor . if @xmath58 is such that @xmath206 , the field outside the capacitor ( for @xmath207 ) is almost constant , given by @xmath208 the expression ( [ eq:54 ] ) is the main result of this paper . it shows that effect of anomaly inflow leads to the drastic change of the electromagnetic fields on the brane . namely , in the presence of magnetic field , the electric field behave as if photon has become massive , with the mass @xmath170 dependent on the magnetic field . the same is true for a point - like charge , placed in magnetic field ( [ eq:58 ] ) . in particular , the electric field , created by an infinite capacitor , appears outside its plates ( eq . ( [ eq:57 ] ) ) . as we will argue in our next paper @xcite , this effect can be used as a signature of extra dimensions . although we have already obtained a non - perturbative final state , we would still like to see how this solution is formed . therefore let us come back to the dynamical problem . although the naive perturbation theory did not work , the parameter @xmath58 is very small , and it is natural to assume that if we will choose a more appropriate zeroth approximation , we can describe time - dependent effects of initial stage of anomaly inflow perturbatively . part of the wave - function @xmath209 plus perturbative series in @xmath210 in a pre - exponential term . ] the results of the previous section explain why the naive perturbation theory of section [ sec : perturb ] did not work . essentially , this is due to the fact that the solution in the bulk is non - perturbative in @xmath58 . therefore , we will try to find a dynamic solution ( linear in time ) , based on the non - perturbative zeroth approximation , derived from the static solution ( [ eq:51 ] ) . at first sight solution ( [ eq:51 ] ) differs drastically from electric fields of section [ sec : perturb ] . for example , the component @xmath211 and a charge density distribution are symmetric with respect to an inversion @xmath212 , while the correction @xmath136 to an electric field is anti - symmetric . this is due to the fact that the full ( time - dependent ) maxwell equations do not possess such symmetry ( unlike the equations for the static case ) . thus , for example , the point particle can have a ( time - dependent ) dipole moment at initial stage of anomaly inflow ( see below , section [ sec : dipole - moment ] ) , while such a dipole moment will be absent in the static solution ( [ eq:58 ] ) . additionally , static solution behaves as @xmath213 for @xmath214 outside the region ( [ eq:53 ] ) around the brane , which is very different from the @xmath133 dependence of the fields @xmath138 and @xmath215 in section [ sec : perturb ] , which had a power law decay ( see @xcite ) . however , we can still develop a perturbative ( in time ) expansions for electric fields . as a zeroth approximation we will take functions @xmath138 and @xmath215 , whose profiles in @xmath133 direction is @xmath178 and @xmath216 correspondingly ( that is they go to zero exponentially fast for @xmath217 ) , unlike functions @xmath138 and @xmath215 in the section [ sec : perturb ] ) . namely , we express @xmath215 and @xmath138 via a function @xmath218 such that @xmath219 and @xmath220 . the function @xmath218 will be chosen in the form ( [ eq:51 ] ) : @xmath221 where @xmath202 is a value of a _ 4-dimensional electric field _ between the plates of the capacitor . we stress that @xmath218 is just an auxiliary function , which is not related to the component @xmath222 of the gauge field . the initial field @xmath223 is created by an infinite capacitor in @xmath224 dimensions . the reason for such a choice is the following . the relaxation time in the bulk is determined by the mass of the fermions there ( fermionic mass gap ) . this mass is much larger than any other scale in our theory , therefore one can think that the bulk fermions form an incompressible fluid and have a relaxation time @xmath225 much faster than any time scale in the theory . therefore the profile in @xmath133 direction settles over the time of turning on of a magnetic fields . as the magnetic field @xmath132 is already taken at its stationary value in our problem , we also take static profile in @xmath133-direction for the electric fields . on the contrary , the relaxation in @xmath107-direction is of the order @xmath226 , which is incomparably smaller than the afore - mentioned scale . the difference of these two scales justifies an ansatz ( [ eq:59 ] ) . we will find corrections to the electric field ( [ eq:59 ] ) for the times @xmath227 . the resulting expression has @xmath133 dependence of the form @xmath228 ( see appendix [ app : correct ] for details ) . thus , it will decay in the same region , as a full solution and a problems of the perturbation theory of the section [ sec : perturb ] are removed . in the linear stage the result of the anomaly inflow is the following . the anomalous electric density appears between the plates of the capacitor : @xmath229 where @xmath171 is the value of magnetic field @xmath230 in the center of the solenoid . this means that the _ capacitor acquires anomalous electric charge , linearly changing with time_. this implies an existence of the stationary 4-dimensional current @xmath231 , given by @xmath232 ( where @xmath222 is given by ( [ eq:59 ] ) ) . this picture has an apparent contradiction with causality , as a non - zero current @xmath233 instantaneously appears at spatial infinity at @xmath109 . this is due to our choice of initial conditions : the magnetic field @xmath234 was turned on instantaneously in the whole space . in reality there is a transitional period , depending on the speed with which we are turning on magnetic field , and the current will appear at infinity only when magnetic field will reach it . we see that in accord with our qualitative expectations , electric charge is initially accumulating in the region of space where initial @xmath235 . such a charge of course creates an electric field outside the plates of the capacitor . for @xmath236 and @xmath237 the electric field outside of the capacitor behaves as @xmath238 comparing correction @xmath239 with @xmath202 , we see that the linear stage of anomaly inflow is valid for the time @xmath240 where @xmath241 we discussed in section [ sec : observ - inflow ] the simplest possible way to create configuration of electro - magnetic field with parallel @xmath101 and @xmath102 and therefore observe effects of anomaly inflow . actually , the electro - magnetic field created by any charged particles with spin has non - zero @xmath242 . as we will presently see , anomaly inflow in this case leads to an appearance of an anomalous dipole moment of the particle . to estimate this effect , we start from the quasi - classical expressions for electric and magnetic fields , created by a particle with an electric charge @xmath62 and magnetic moment @xmath243 @xmath244 thus , in the region of the space where expression ( [ eq:77 ] ) is non - zero , inflow current ( analogous to that of section [ sec : observ - inflow ] ) creates non - zero anomalous charge density @xmath245 . such density will be positive in one half of the ball , surrounding a particle , and negative in another one . as a result , the total charge of the particle does not change . however due to the inflow any particle acquires an anomalous _ electric dipole moment _ : @xmath246 ( here @xmath247 is a compton radius of the particle , given by @xmath248 in our system of units ) . charge density @xmath245 is given by ( [ eq:65 ] ) with electric and magnetic fields estimated by @xmath249 and @xmath250 . substituting all these values into eq . ( [ eq:78 ] ) we get : @xmath251 i.e. as a consequence of anomaly inflow a particle acquires a dipole moment , which has the absolute value growing with time . notice , however , that this result was obtained in perturbation theory as a linear in time approximation ( section [ sec : old - part ] ) . it is valid for @xmath252 where @xmath170 is given by an analog of formula ( [ eq:49 ] ) for @xmath253 : @xmath254 for times , much bigger that this characteristic time , field configuration around the particle approaches static solution . as it has already been discussed ( see the beginning of the section [ sec : old - part ] ) , static equations are symmetric with respect to inversion @xmath255 , therefore , there can not be any dipole moment for @xmath256 . electric field configuration will be significantly modified from the usual coulomb to yukawa form ( similar to ( [ eq:58 ] ) ) at the distances larger than @xmath188 , with @xmath170 given by ( [ eq:98 ] ) . this means that the electric charge of the particle _ gets completely screened _ and the total amount of an anomalous charge which appeared on the brane is equal to the charge of the particle . in this section we apply the logic of section [ sec : qed ] to the standard model . indeed , if the standard model fields are localized on a brane in a 5-dimensional world , there is no apparent reason to expect a separate anomaly cancellation for them . let us try to see what kind of new effects we should expect for the standard model if 4-dimensional anomaly cancellation condition is not imposed and there is an inflow from extra dimensions . we will see that such effects do exist and discuss experimental restrictions for corresponding parameters . for simplicity we consider the electroweak @xmath257 theory with only one generation of fermions ( the addition of extra generations does not change the analysis ) . the action for the @xmath258 and @xmath2 gauge fields is similar to that in eq . ( [ eq:1 ] ) . the warp - factor @xmath19 is such that there is a zero mode , localized on the brane . we also add a higgs @xmath83 field , which is an @xmath258 doublet . as in section [ sec : qed5d ] its mass @xmath70 is negative at @xmath71 and tends to the positive constant in the bulk as @xmath72 ( figure [ fig : higgs ] ) . therefore it has a non - trivial profile @xmath73 , with @xmath74 and @xmath75 at @xmath259 . so the @xmath257 symmetry is broken down to @xmath260 on the brane and restored far from it . therefore the vector boson acquires mass @xmath261 which becomes zero at large @xmath262 . , as such a constant would have removed the zero mode from the spectrum . indeed , in this paper we consider such warp - factors @xmath19 that there is a mass - gap @xmath25 between the zero mode and the modes in the bulk . therefore no state with the mass @xmath263 can exist away from the brane . ] we also add fermions in the bulk , charged with respect to @xmath257 . their zero modes are localized on the brane with mass gap @xmath18 much larger than that of the gauge fields . we consider the extension of the standard model in which the @xmath264 hyper - current @xmath265 becomes anomalous @xmath266}{16\pi^2 } \epsilon^{\mu\nu\lambda\rho}\cf_{\mu\nu}\cf_{\lambda\rho } + \frac{{\mathop\mathrm{tr}}[y_l]}{16\pi^2 } \epsilon^{\mu\nu\lambda\rho}{\mathop\mathrm{tr}}g_{\mu\nu } g_{\lambda\rho}\ : . \ ] ] here @xmath267 is a @xmath268 field strength of the @xmath2 field @xmath269 ( hyper - photon ) : @xmath270 ; @xmath271 is an @xmath258 non - abelian field strength , @xmath272 and @xmath273 are @xmath268 and @xmath258 coupling constants correspondingly . the first term in comes from the diagram fig . [ fig : anomaly]b and is proportional to the sum of cubes of hypercharges of _ all _ particles . the hypercharges in the standard model are chosen in such a way that @xmath274={\mathop\mathrm{tr}}[y_l ] = { \mathop\mathrm{tr}}[y]=0 $ ] @xcite . we choose a one - parameter extensions of this model , in which @xmath275=-\frac12{\mathop\mathrm{tr}}[y_l]=\kappa_0 $ ] . the second term comes from the diagram fig . [ fig : anomaly]a , with one @xmath268 and two @xmath258 vertices . along with anomaly there is also a non - conservation of @xmath258 current in the background @xmath2 and @xmath258 fields ( its coefficient is proportional to the same diagram as the second term in ): @xmath276}{8\pi^2 } \epsilon^{\mu\nu\lambda\rho}g^\alpha_{\mu\nu } f_{\lambda\rho}\ : . \ ] ] current in the presence of non - trivial @xmath258 background * ( a ) * ; or in a background of two @xmath2 fields * ( b)*. ] recall that @xmath268 and @xmath258 fields , entering and , are the fields _ above _ the electroweak symmetry breaking scale . at lower energies it is convenient to re - express these anomalies in terms of of electro - magnetic field @xmath277 and _ neutral _ field @xmath278 , which can be obtained from the @xmath2 and @xmath258 fields @xmath279 and @xmath280 via the rotation @xmath281 and @xmath282 . , etc . ) because we use different normalization of the action for the gauge fields the one with the coupling constant in front of the kinetic term . ] the electro - magnetic current @xmath283 is given by @xmath284 and _ neutral current _ @xmath285 , where @xmath286 is the 3rd component of the @xmath258 triplet @xmath287 , @xmath288 . using ( [ eq:92 ] ) and ( [ eq:94 ] ) one can easily see that _ ( i ) _ electro - magnetic current is conserved in the arbitrary background of electro - magnetic fields ( as one expected , because the electrodynamics remains vector - like in our model ) ; _ ( ii ) _ there is an anomalous @xmath289 _ coupling _ @xmath290 which implies the non - conservation of the neutral current in the parallel electric and magnetic fields ( @xmath291 is the number of generations ) ; _ ( iii ) _ another important consequence of the presence of @xmath289 coupling is the non - conservation of the electro - magnetic current in the mixed electro - magnetic and @xmath292-backgrounds : @xmath293 as we will see this leads to the effects similar to those , described in sections [ sec : observ - inflow][sec : dipole - moment ] . consider again the setup of section [ sec : observ - inflow ] : place a capacitor in the strong magnetic field @xmath294 , such that electric field @xmath295 , created by the capacitor is in the same direction @xmath104 . our choice of hypercharges implies the non - conservation of @xmath296 current . this also means that there exist chern - simons terms in the bulk action , which ensure an anomaly inflow , canceling anomaly ( [ eq:96 ] ) . indeed , chern - simons term ( [ eq:3 ] ) , originally written in terms of the hypercharge field @xmath269 and @xmath258 field @xmath297 , can also be re - expressed in terms of electro - magnetic and z - field and it creates a inflow of z - current in the parallel electric and magnetic fields . this creates anomalous density of z charge on the brane . this distribution of z charge creates an anomalous z field in the 5th direction and as a result inflow of _ electro - magnetic _ current to cancel an anomaly ( [ eq:97 ] ) . such an inflow creates anomalous distribution of electric charge on the brane and modifies electric field inside and outside the capacitor . similarly to the case of electrodynamics ( section [ sec : qed ] ) we can find this static configuration of an electric field . it will be given by the same expression as in section [ sec : final - state ] ( see appendix [ sec : ew - final - state - compute ] for details ) . as a result , distribution of an electric field inside and outside the capacitor is given by expressions ( [ eq:55])([eq:57 ] ) , with @xmath170 proportional to the magnetic field @xmath171 and anomaly parameter @xmath58 : @xmath298 notice , that this result does not depend on the mass scale of the extra dimension and would be true even for planck scale @xmath25 . it also does not depend on the mass of the z boson @xmath299 . this feature , as well as the result ( [ eq:237 ] ) itself , is valid only assuming that @xmath300 . this can be understood as follows : anomalous density of z charge creates electric z field in the 5th direction . this field is , of course , decaying on the scales larger than @xmath301 . however the inflow comes from the region ( [ eq:53 ] ) which is much smaller than @xmath301 . as a result , in the leading order in @xmath302 mass of the z boson does not modify the effect . for details see appendix [ sec : ew - final - state - compute ] . the consequence of anomaly is the appearance of anomalous dipole moment of the particle ( c.f . section [ sec : dipole - moment ] ) . namely , a background with non - zero @xmath303 can be created around any particle which has a spin and also electric and @xmath292-charges . unlike the electro - magnetic case , @xmath292-field , created by the particle , is short - ranged . in the physically interesting case of a particle , lighter than @xmath292-boson ( e.g. proton or electron ) , anomaly inflow is concentrated in the region smaller then compton wave length of the particle and , therefore , it can not be treated quasi - classically . therefore we will only give the order of magnitude estimate here . compared to the analysis of section [ sec : dipole - moment ] the result is suppressed by @xmath304 : @xmath305 again , this result does not depend on the parameters of the extra dimension , but in this case it depends on the mass of the particle and on @xmath299 . in this paper we analyzed a brane - world scenario , when fields on the brane possess an anomaly , canceled by inflow from the bulk . we stress that in such setup there is no reason to require cancellation of anomaly separately on the brane . we show in a simple model of electrodynamics that inflow results in the concrete observational effects , even in the situation when the real escaping of the matter from the brane is not possible . this logic can also be applied to the standard model ( sm ) , if one considers it as an effective theory of some fundamental theory in higher - dimensional space - time . anomaly cancellation condition which is usually assumed in sm becomes an unjustified fine - tuning and one should study the theory without it . easing this condition leads ( in the simplest case ) to the one free parameter electric charge difference of electron and proton . this is one more legitimate phenomenological parameter imposed onto the sm from the point of view of extra - dimensions . we show that with this parameter ( called @xmath306 in sections [ sec : anom - sm][sec : proton ] ) being non - zero , the phenomenology of the standard model is modified in such a way as to admit an anomaly . it is interesting to note that this modification does not affect electrodynamics sector of the sm : even for different charges of electron and proton it remains non - anomalous . the presence of anomaly inflow exhibits itself on a brane via a subtle mechanism , leading to the dynamical screening of an electric field in the presence of a parallel magnetic field , e.g the change of an electric field in a capacitor , placed in a magnetic field . similarly , anomaly inflow leads to the screening and change with time of the electric charge of any elementary particle with electric charge and spin . the inflow , before the screened static solution is formed , such a particle acquires also anomalous dipole moment , which , however , disappears again in the final state , where the electric field is parity symmetric , but screened . these effects would not be present if the theory were purely 4-dimensional , with anomaly canceled by addition of new ( chiral ) particles at higher energies . thus the described effects can in principle serve as a signature of extra dimensions . of course , there can exist other low - energy ( as well as high - energy ) effects , through which anomaly inflow exhibits itself . for example , one can show that photon , propagating in the magnetic field , would becomes massive , depending on its polarization . we plan to study other signatures of anomaly inflow elsewhere . [ [ radiative - corrections . ] ] radiative corrections . + + + + + + + + + + + + + + + + + + + + + + the analysis in this paper was conducted at the level of classical equations of motion . the natural question arises : to which extent this theory can be considered weakly coupled and radiative corrections can be neglected ? a most sensible quantity seems to be the mass of the photon , which is constrained strongly by a number of different experiments . a simple power counting applied to a diagram , shown on fig . [ fig : twocs ] , containing two triangles in the _ anomalous four - dimensional _ electroweak theory with a non - standard hypercharge choice , gives an order of magnitude of the photon mass @xmath307 . then , for an ultraviolet cutoff of the order of @xmath308 tev one finds a very strong constraint on an anomaly coefficient @xmath58 . if the same consideration were true for our case , when the anomalous electroweak theory is coming from anomaly - free theory in 5 dimensions , the classical effects described in this paper would be subleading and non - observable . however , this is not the case because of the following reason . the same diagram , considered now in full 5-dimensional theory , where both massive and zero modes run in loops , does not lead to generation of the photon mass simply because the theory is free from anomalies and is gauge - invariant . on the language of the effective field theory , the insertion of triangular diagrams will not produce any irregularities , as any contribution coming from the chern - simons vertex is precisely canceled by the same process with an insertion of a dhoker - farhi vertex . [ [ effective - theory . ] ] effective theory . + + + + + + + + + + + + + + + + + our main result , given by eqs . ( [ eq:54 ] ) , ( [ eq:237 ] ) , does not depend on the mass scale of the 5th dimension @xmath25 and on the higgs vev . so , the effect stays even when these parameters are sent to infinity , in other words , when the only low energy particle residing in the spectrum is the photon . this can also be seen from eqs . ( [ eq:3 ] ) and ( [ eq:20 ] ) , which do not contain any information about the heavy particles which generated them and are not suppressed by any cut - off . this may seemingly contradict to the usual logic behind effective field theories which is based on the conjecture ( often known as `` decoupling theorem '' and proven for quite a general class of four - dimensional theories in @xcite ) that effects of massive fields on the _ renormalizable _ low - energy effective action only exhibit themselves in renormalization of charges and fields , while all additional interaction are suppressed by some positive power of @xmath309 ( characteristic energy of processes over cut - off @xmath310 ) . it is known , however , that the `` decoupling theorem '' does not always hold . in particular , it is not true in theories , leading to chern - simons @xcite or dhoker - farhi @xcite interactions . in the latter case the low energy theory is _ non - renormalizable _ , which is true for our case as well . so , in addition to the chern - simons and the dhoker - farhi term in the effective theory one must introduce infinitely many other terms and counterterms to remove divergences . to determine them one needs the knowledge of the fundamental theory . the construction of this type of theory goes beyond the scope of the present work . the low energy description of our theory contains 5-dimensional terms ( see e.g. action ( [ eq:2 ] ) ) . it would be interesting to construct a low - energy effective action entirely in terms of 4-dimensional fields . naively , one could try to integrate over the massive modes of the field @xmath311 in the action ( [ eq:2 ] ) . however , as this work demonstrates , the chern - simons term in ( [ eq:2 ] ) can not be treated perturbatively , therefore it is not clear how one can perform such an integration . [ [ fine - tuning - of - anomaly - mismatch . ] ] fine - tuning of anomaly mismatch . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the idea of realization of the standard model on a brane is very popular and wide - spread nowadays . as we showed in this paper , any such model should either produce a mechanism , which prohibits the anomaly on the brane ( due to symmetry reasons or dynamics ) or incorporate it into the model . at the same time , it is known experimentally that ( if non - zero ) the charge difference between electron and proton is extremely small . as a result , one is presented with a novel type of a fine - tuned parameter and should try to find a reason for its existence . as it is usually the case , this fine - tuning can tell us something new about the physics beyond the standard model . certainly , the answers to the questions of anomalous standard model and fine - tuning of its anomaly coefficient will depend on various ingredients of the brane - world models : the higher - dimensional theory , the details of localization of the fields , etc . therefore it is important to consider semi - realistic brane - worlds , where at least some sector of the standard model ( including gauge fields ) is localized . a simple example of such sort was proposed in @xcite , where the @xmath2 theory was localized on the vortex in six dimensions . in that model the fermionic content on the brane could have anomalous chiral couplings ( while the theory in the bulk was always anomaly free ) . this theory can be considered as a particular realization of the scenario of section [ sec : qed ] and effects , similar to those , described in section [ sec : observ - inflow ] , should be present there . one should notice also that its anomaly coefficient can be of the order of unity , as it should on general grounds . more realistic models should include non - abelian fields localized which is rather non - trivial . in all these models it is important to analyze the question of anomaly ( non)cancellation on the brane and the reason of the fine - tuning of corresponding parameters . in string theory , being a theory with extra - dimensions , this question also arises . a popular approach there is to obtain the standard model fields on an intersection of various branes ( see e.g. recent paper @xcite ) . on the other hand , there are many examples in string / m theory with branes when world volume theory is anomalous and there is an inflow ( see e.g. @xcite ) . again , there should be a special reason to have so exact anomaly cancellation for the standard model realized in this way . anomaly analysis is usually very instructive in string theory and a string theoretical solution of this problem would be very interesting . one should also apply this logic to any extension of the standard model ( e.g. to the minimal supersymmetric standard model ( mssm ) ) appearing in a brane - world setup . in this case there could be more free parameters which may allow for some other phenomenological effects ( e.g. non - zero electric charge of neutralinos ) . we would like to thank i. antoniadis , x. bekaert , v. cheianov , t. damour , s. dubovsky , j. harvey , s. khlebnikov , m. libanov , j .- h . park , s. randjbar - daemi , s. shadchin for useful comments . the work a.b . and m.s . was supported by the swiss science foundation . o.r . would like to acknowledge a partial support of the european research training network contract 005104 `` forcesuniverse '' and also warm hospitality of ecole polytechnique fdrale de lausanne where a part of this work was done . would like to acknowledge the hospitality of institut des hautes tudes scientifiques . to compute initial , linear in time , change of electromagnetic fields on the brane , one should specify values of the fields at @xmath109 . as discussed at the beginning of section [ sec : perturb ] , one should specify not only values of electromagnetic field ( to be discussed below , in sections [ app : naive][app : correct ] ) , but also initial values of gauge potential . in this paper we consider the following case . @xmath312 values of the time derivative at @xmath109 can then be determined , knowing field strengths and using eq . ( [ eq:18 ] ) . there are two non - zero derivatives : @xmath313 our initial conditions mean that at the initial moment the longitudinal component of the field is not excited . indeed , at @xmath109 our field @xmath314 obeys : @xmath315 and @xmath316 ( we denote by @xmath317 ) , which means that it is transversal . notice , that the condition ( [ eq:24 ] ) implies that transversality of @xmath314 will not hold for @xmath318 . for these initial conditions the only non - zero component of the dhoker - farhi current is @xmath319 first , we attempt to solve the maxwell equations ( [ eq:6])([eq:7 ] ) by perturbation theory in @xmath58 . in addition to the initial condition ( [ eq:30 ] ) , we take initial electric fields to satisfy gauss constraints in the theory with @xmath123 ( see the discussion in the beginning of section [ sec : perturb ] ) : @xmath320 we will mark such fields by the symbol @xmath124 . our choice in particular imply that @xmath321 . although the zero mode of the gauge field has constant profile in fifth direction , the solution of ( [ eq:31 ] ) gives @xmath322 and @xmath323 , both decaying at @xmath324 . the rate at which these fields decay depends on the warp - factor @xmath192 . a solenoid creates non - zero field @xmath325 as well as @xmath326 , again , both decaying at infinity in @xmath133 ( see @xcite for details ) . according to the definition of the inflow current ( [ eq:8 ] ) such electric and magnetic fields generate the inflow current from the fifth direction : @xmath327 this current flows onto the brane from both sides in the @xmath133 direction . two other non - zero components of the chern - simons current are @xmath328 for our configuration of electric and magnetic fields the dhoker - farhi current ( [ eq:20 ] ) satisfies the following property : @xmath329 this property suggests that the dynamics of the theory may be separated into two parts : dynamics in the directions @xmath330 and in @xmath331 . we will see below that this is indeed the case at the initial stage of the process . for the initial conditions of section [ sec : perturb ] , the anomaly of the dhoker - farhi current is split equally between @xmath332 and @xmath333 , both being equal to the right hand side of eq . ( [ eq:34 ] ) . in particular @xmath334 i.e. we see that an electric charge is accumulating on the brane between the plates of the capacitor . at initial stage of anomaly inflow , only electric fields appears in the left hand side of the maxwell equations ( [ eq:6]) ( [ eq:7 ] ) . the reason for that is the following . as we have shown , initially the total charge density grows linearly in time ( [ eq:86 ] ) . therefore , it creates electric field , also growing linearly in time . as a consequence of the bianchi identities ( @xmath335 ) all the magnetic components @xmath336 . thus , if we keep only terms at most linear in time in these equations , we obtain : @xmath337 all the gauge fields in the left hand side are of the order @xmath196 , while those in the right hand side are of the zeroth order in @xmath58 and are explicitly multiplied by @xmath58 . ( [ eq:37 ] ) gives @xmath338 and similarly from eq . ( [ eq:39 ] ) @xmath339 we see that _ for any _ @xmath58 the correction @xmath340 will become bigger than @xmath341 for large @xmath214 , as @xmath342 for @xmath139 . notice , that solution is a _ 5-dimensional _ electric field . to discuss the observational consequences , we should switch to the 4-dimensional fields , defined as in ( [ eq:12 ] ) , namely @xmath343 four - dimensional electric field @xmath239 is given by @xmath344 integration in can be done explicitly in the region @xmath237 . as shown in @xcite @xmath345 is a constant as a function of @xmath133 for @xmath346 . notice , that even in case of @xmath347 integral ( [ eq:43 ] ) is convergent . indeed , for any warp - factor @xmath192 , decaying at infinity , @xmath348-component of an electric field decays at infinity as @xmath349 ( see @xcite for details ) . we will be using this approximation through this section . the error from approximation of @xmath350 by its value at the origin @xmath351 can be estimated as @xmath352 ( see @xcite for details ) . in case of the warp @xmath353 the integral ( [ eq:43 ] ) is saturated in the region @xmath354 . the solution of this section has an apparent problem : for any @xmath58 , e.g. @xmath136 becomes bigger than @xmath138 for @xmath214 large enough . see the discussion in the section [ sec : perturb ] . to compute perturbatively in @xmath58 an initial changes of electric field of a capacitor , we should choose a zeroth approximation different from that of the previous appendix . namely , we express @xmath215 and @xmath138 via an auxiliary function @xmath218 such that @xmath219 and @xmath220 with @xmath218 given by ( [ eq:51 ] ) : @xmath355 where @xmath202 is a value of a _ 4-dimensional electric field _ between the plates of the capacitor . the initial field @xmath223 satisfies 4-dimensional poisson equation : @xmath356 . unlike the case of section [ sec : perturb ] , the _ initial _ chern - simons charge density @xmath113 is not equal to zero ( as one can see by substituting @xmath341 and @xmath357 into the gauss constraint ) . this implies that component @xmath358 is nonzero ( it can be read off the right hand side of the gauss constraint ( [ eq:22 ] ) using the definition ( [ eq:8 ] ) ( see also ( [ eq:158 ] ) ) : @xmath359 where @xmath172 is given by ( [ eq:49 ] ) . however , its explicit form will not be needed for the analysis below . the first correction to the field @xmath211 is given by @xmath360 ( recall , that our initial conditions for the dhoker - farhi currents are those of eq . ( [ eq:30 ] ) , and therefore @xmath361 ) . similarly @xmath362 notice that according to the definition ( [ eq:12 ] ) the _ 4-dimensional _ electric field , @xmath239 is determined by only the first term in ( [ eq:61 ] ) : @xmath363 the difference of equation ( [ eq:63 ] ) with that of ( [ eq:43 ] ) in the section [ sec : perturb ] is in the fact that here the integral over @xmath133 in effectively restricted to the region @xmath364 . in this region one can substitute @xmath345 with its value @xmath171 in the center of the solenoid . as a result one gets : @xmath365 one can check that in the linear in @xmath58 order the energy is conserved . indeed , in this order the change of energy ( in the full 5-dimensional theory ) is given by @xmath366 substituting the solutions ( [ eq:61])([eq:62 ] ) we see that the integrand of ( [ eq:74 ] ) is equal to zero . from the point of view of the 4-dimensional observer , the energy is defined as @xmath367 and therefore its change in the linear in @xmath58 order is @xmath368 because the @xmath369 is an even function of @xmath107 , while @xmath370 is an odd one , this integral is equal to zero . thus although there is an inflow of chern - simons currents to the brane , the energy on the brane does not change . it simply gets redistributed in the space , as field appear outside the capacitor and diminishes inside . according to the definition ( [ eq:110 ] ) @xmath371 , where the function @xmath372 equals to 1 for @xmath373 and zero otherwise and @xmath374 ( see appendix [ sec : final - state - compute ] ) . as a result we see that _ the inflow current creates a non - zero charge density on the brane in the region where ( 4-dimensional ) @xmath375 _ ( in our case between the plates of the capacitor ) : one should take into account that there is a contribution , coming from the term @xmath376 . as mentioned after the eq . ( [ eq:34 ] ) , divergence of the dhoker - farhi current is equally split between @xmath377 and @xmath331 components . therefore , the coefficient in front of the integral in ( [ eq:65 ] ) is twice as little as that in eq . ( [ eq:64 ] ) . this can be easily verified by directly computing @xmath378 from eq . ( [ eq:38 ] ) and substituting it together with ( [ eq:64 ] ) into ( [ eq:65 ] ) . ] @xmath379 thus the capacitor accumulates over time an additional electric charge @xmath380 ( where area @xmath381 ) . this means that from a 4-dimensional point of view there is an electric current flowing from infinity . indeed , according to the definition ( [ eq:14 ] ) the 4-dimensional current in the direction @xmath107 is given by @xmath382 ( both @xmath383 and @xmath384 are equal to zero in the linear in time approximation ) . as we will now see , this current @xmath233 is non - zero as @xmath385 . using definition ( [ eq:33 ] ) we get : @xmath386 where @xmath387 . if anomalous electric charge @xmath388 changes with time , this means that @xmath389 or equivalently that @xmath390 ( we have substituted integral over the @xmath8 plane with the area @xmath391 of the plates of the capacitor ) . using eq . ( [ eq:110 ] ) one can see that indeed @xmath392 . this gives @xmath393 the result ( [ eq:68 ] ) can be obtained from the conservation of the 5-dimensional current . namely , one can see that there is a flow of the chern - simons current from infinity . in the approximation @xmath394 we should also take @xmath395 ( @xmath396 has the same @xmath133-dependence as @xmath397 ) and therefore we can neglect radial chern - simons current . as a result the analysis becomes effectively @xmath1 dimensional in coordinates @xmath398 . consider a box @xmath399 , @xmath400 , where both @xmath401 . the total amount of chern - simons current , inflowing through the boundaries of this box is @xmath402 from eqs . ( [ eq:32])([eq:33 ] ) it follows that the first integral equals to zero : @xmath403 as @xmath404 for @xmath405 ( see ( [ eq:110 ] ) ) . at the same time the second integral in ( [ eq:69 ] ) is finite : @xmath406 as @xmath407 for @xmath408 . as a result we again recover ( [ eq:68 ] ) . ) together with the initial conditions @xmath409 and the fact that the total charge density is equal to @xmath114 implies that @xmath410 in ( [ eq:69 ] ) is equal to @xmath388 . ] the anomalous charge @xmath388 creates an additional electric field eq . ( [ eq:43 ] ) . that is we see that due to the anomaly inflow an electric field appears outside of the capacitor . in the region close to the plates , ( i.e. @xmath411 and @xmath237 ) this field is constant in space and grows linearly in time according to the law @xmath412 inside the capacitor @xmath239 changes linearly from its value ( [ eq:72 ] ) on one plate of the capacitor to the opposite of it on another . we see that the electric field , created due to the anomaly inflow depends on the anomaly coefficient @xmath58 and _ does not depend _ on the parameters of the extra dimension . from eq . ( [ eq:72 ] ) we can determine the characteristic time during which the linear approximation is valid . it is given by @xmath413 and our solution ( [ eq:40 ] ) is valid for @xmath414 . consider the system of maxwell equations , describing a _ static _ solution in @xmath415 dimensions : @xmath416 ( we will be mostly interested in the case when @xmath417 or @xmath418 ) . we are searching for the axially - symmetric static solution , therefore all the derivatives with respect to time and angle @xmath151 were put to zero in eqs . ( [ eq:153])([eq:157 ] ) . the components of chern - simons current ( [ eq:8 ] ) in the cylindrical coordinates @xmath149 are equal to @xmath419 for static solution we can describe the electric field in terms of the electro - static potential @xmath144 : @xmath145 , @xmath146 . in this case one can rewrite expression @xmath420 in the following way : @xmath421 as a result we solve the system ( [ eq:153])([eq:157 ] ) by the following ansatz . we express @xmath120 , @xmath422 , @xmath423 via @xmath144 and @xmath424 , using eqs . ( [ eq:154 ] ) , ( [ eq:155 ] ) , ( [ eq:157 ] ) : @xmath425 then the gauss constraint ( [ eq:153 ] ) can be re - written as @xmath426 where @xmath148 is a 3-dimensional laplacian in the coordinates @xmath149 . the equation ( [ eq:156 ] ) becomes ( we introduce gauge potential @xmath116 : ) @xmath427 we will find a solution of the eqs . ( [ eq:165])([eq:166 ] ) as follows . first we will put right hand side of eq . ( [ eq:166 ] ) to zero . then there is a solution @xmath154 , @xmath428 , @xmath429 . in this case one could easily find an explicit solution of eq . ( [ eq:165 ] ) see eqs . ( [ eq:167])([eq:176 ] ) below . after that we will show that corrections to the found solutions , which are due to the non - zero right hand side of are of the order @xmath157 and thus can be neglected . under assumptions @xmath430 , @xmath431 , @xmath156 gauss law becomes identical to that of the @xmath1 dimensional case . after the fourier transform in @xmath149-directions : ( @xmath432 ) and a substitution @xmath159 equation ( [ eq:165 ] ) becomes in eq . ( [ eq:9 ] ) as compared to ( [ eq:165 ] ) . it accounts for the perturbative corrections of the order @xmath196 . ] @xmath433 = \frac{36{\,\mathbf{e}}^4\kappa_0 ^ 2 { \text{\textbf{\slshape h}}}_0 ^ 2}{\delta^2(z ) } \psi_p(z)+{\,\mathbf{e}}^2 \tilde q(\vec p ) \delta(z ) \:,\ ] ] where @xmath434 and @xmath435 is a fourier transform of @xmath176 . for our main example the warp - factor @xmath436 , we have @xmath437 . let us re - write eq . ( [ eq:167 ] ) as @xmath438 where operator @xmath162 does not depend on @xmath163 : @xmath439 [ [ sec : spectrum ] ] spectrum of @xmath162 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let us denote eigen - functions of ( [ eq:169 ] ) and their eigen - values by @xmath165 and @xmath166 : @xmath167 . introduce new _ dimensionless _ variable @xmath440 and denote by @xmath441 . then the eigen - value problem for operator ( [ eq:169 ] ) reduces to @xmath442 this is the standard form of the equation for the bessel functions of the second kind . a solution of this equation , which is regular at @xmath443 ( i.e. @xmath444 ) is given by @xmath445 , which behaves at infinity as @xmath446 . as a result we see that the solution of eq . ( [ eq:170 ] ) is given by @xmath447 eigen - values @xmath166 are determined from the gluing condition at the origin , following from the presence of @xmath448 in the operator ( [ eq:169 ] ) @xmath449\bigr|_{z=0 } = -2{m}\psi_n(0)$ ] : @xmath450 where @xmath451 was defined in ( [ eq:171 ] ) and @xmath452 . using the properties of the bessel function ( see e.g. @xcite ) one can show that there is a single solution of this equation for real @xmath453 ( corresponding to the smallest eigen - value ( @xmath454 ) and infinitely many solutions for imaginary @xmath453 s ( corresponding to all the other eigen - values @xmath455 ) . the easiest way to solve eq . ( [ eq:172 ] ) is by iteration around @xmath456 . as a result one finds that , for which a zero mode exists , that operator @xmath162 has the lowest eigen - value @xmath457 . ] @xmath458 the wave - function of the lowest level is given by @xmath459 where @xmath174 is a modified bessel function of the second kind and @xmath175 is a normalization constant , given by @xmath460 . the green s function of the operator ( [ eq:168 ] ) is given by the standard expression @xmath461 the solution of eq . ( [ eq:165 ] ) can be easily found for any charge distribution @xmath176 . in case at hand ( capacitor with the infinite plates ) one gets : @xmath462 we will also often use the function @xmath463 . it is given by : @xmath464 and has the following important properties : @xmath465 and @xmath466 consider the system of two couple @xmath2 fields : a massless ( electro - magnetic ) one , whose gauge potential and field strength we will denote by @xmath314 and @xmath467 and _ massive _ ( @xmath468 field ) , with gauge potential @xmath469 and field strength @xmath470 . these fields interact via chern - simons terms : @xmath471 where ( compare ( [ eq:96 ] ) ) @xmath472 fermions interact with both electro - magnetic and z fields . according to eqs . ( [ eq:92 ] ) , ( [ eq:94 ] ) both electro - magnetic current @xmath473 and z - current @xmath296 are not conserved : @xmath474 and @xmath475 this means that there are two types of inflow currents : z - current @xmath476 and electro - magnetic chern - simons currents : @xmath477 the expression for the dhoker - farhi currents @xmath478 and @xmath479 are similar to that of eq . ( [ eq:20 ] ) with an obvious modifications following from eqs . ( [ eq:217])([eq:226 ] ) . consider the system of maxwell equations , describing a _ static _ solution in @xmath415 dimension . for static solution we can describe the electric field in terms of the electro - static potential @xmath144 : @xmath145 , @xmath480 . we will denote electro - static potential for the electro - magnetic and z - field as @xmath481 and @xmath482 correspondingly . then we obtain : @xmath483 where @xmath148 is a 3-dimensional laplacian in the coordinates @xmath149 . we will be mostly interested in the case when @xmath417 or @xmath418 . for the z - field we have similar system of equations : @xmath484 we are searching for the axially - symmetric static solution , therefore all the derivatives with respect to time and angle @xmath151 were put to zero in eqs . ( [ eq:179])([eq:224 ] ) . to find a solution of this non - linear system of equations , let us recall the electro - magnetic case first ( appendix [ sec : final - state - compute ] ) . we saw there that of all the magnetic components @xmath485 and @xmath486 only @xmath487 was not suppressed by a power of anomaly parameter @xmath58 . , @xmath488 and @xmath489 are of the same order , we will use notation @xmath58 when speaking about the order of magnitude value of various expressions . ] we will try to find a solution of the system ( [ eq:179])([eq:224 ] ) under a similar assumption : only terms , containing @xmath487 , give dominating ( non - perturbative in @xmath58 ) contribution to these equations , while all other terms account for small ( suppressed by powers of @xmath58 ) corrections . in addition we take all components of the magnetic z - field to be much smaller than those of electro - magnetic field . then the system of equations ( [ eq:179])([eq:224 ] ) reduces to the gauss constraints for potentials @xmath482 and @xmath481 , _ in which we have left only terms , proportional to @xmath487 _ : @xmath490 and @xmath491 other relevant parts of the system ( [ eq:179])([eq:224 ] ) are equations for the component @xmath120 : @xmath492 and @xmath493 : @xmath494 from eqs . ( [ eq:229 ] ) we immediately see that @xmath495 similarly to the electro - magnetic case ( compare eq . ( [ eq:164 ] ) ) . solution of eq . ( [ eq:230 ] ) depends on the @xmath133-dependence of the mass - term @xmath496 . as an example we will consider @xmath497 ( localization scale of the higgs field is of the same order as that of the fermions on ) . in this case , one can easily see that the solution of eqs . ( [ eq:230])([eq:231 ] ) does not differ from the massless case : is an odd function of @xmath133 and therefore @xmath498 and mass term disappears from ( [ eq:231 ] ) . one could in principle add to the solution ( [ eq:225 ] ) the term @xmath499 , however , this term grows as @xmath137 as @xmath324 and therefore the energy of such a solution would be infinite . thus , we conclude that @xmath500 , mass terms in eqs . ( [ eq:230])([eq:231 ] ) drop out and the solution for @xmath493 is given by ( [ eq:225]).]@xmath501 is a limiting case , in which mass dependence drops out of equation ( [ eq:232 ] ) and subsequent does not enter at all into gauss constraint for an electric field . one can show that for arbitrary profiles @xmath496 , spreading into the 5th direction , eq . ( [ eq:232 ] ) receives corrections due to mass . in particular the effect , described in this section disappears as @xmath502 . ] @xmath503 and as a result we obtain the following system of equations : @xmath504 we will solve this system by assuming that the last terms of both equations ( [ eq:233 ] ) and ( [ eq:234 ] ) can be treated perturbatively . the results of our computations then will confirm this assumption . let us start with eq . ( [ eq:233 ] ) . with exception of the last term we have already encountered similar system in section [ sec : final - state ] ( see also appendix [ sec : final - state - compute ] ) . for @xmath505 the solution is given by analogs of eqs . ( [ eq:51])([eq:52 ] ) : ) is easy to obtain , knowing the spectrum of the operator ( [ eq:169 ] ) and writing its green s function ( compare ( [ eq:175 ] ) ) . ] @xmath506 ( where by @xmath507 we have denoted fourier transform of the function @xmath508 with respect to @xmath107 ) . function @xmath509 satisfies the poisson equation with the mass term @xmath510 with the mass @xmath511 the profile of @xmath481 in the @xmath133 direction @xmath178 is given by @xmath512 ( here @xmath174 is a modified bessel function of the second kind ) . this profile is sharply localized in the region @xmath513 the last term in ( [ eq:239 ] ) can be treated perturbatively in @xmath306 if a @xmath133-profile of the field @xmath482 is decaying faster than @xmath192 . to demonstrate that this is indeed the case , let us insert the solution ( [ eq:239 ] ) into the gauss constraint ( [ eq:228 ] ) . we will find that expression for @xmath482 is given by expression , similar to the ( [ eq:239])([eq:241 ] ) : @xmath514 ( by @xmath515 we have denoted fourier transform of the solution ( [ eq:239 ] ) with respect to @xmath107 ) . here @xmath516 is a solution of a differential equation similar to ( [ eq:167 ] ) but with the modification of the factor in front of the @xmath448 : @xmath517 . this modifies gluing condition ( [ eq:172 ] ) , which determined the eigen - values of the equation in @xmath133 direction : @xmath518 . as a result expression for @xmath519 is given by : @xmath520 ( the corrections to this result are of the order @xmath521 and @xmath522 , where @xmath523 ) . as the first term in eq . ( [ eq:106 ] ) is much bigger than the second one gev and arbitrary @xmath524 , recalling that 1 gauss@xmath525 and the strongest magnetic fields , achievable in a laboratory are @xmath526 gauss . ] we neglect the latter and find the mass @xmath527 in contrast with ( [ eq:240 ] ) . notice , however , that the profile of the @xmath482 in @xmath133 direction is very similar to that of eq . ( [ eq:241 ] ) , still decaying as @xmath185 at the distances of the order ( [ eq:242 ] ) . finally , function @xmath528 , entering ( [ eq:243 ] ) satisfies poisson equation with the mass term of the z - boson : @xmath529 now we can analyze corrections due to the last term in eq . ( [ eq:243 ] ) . from solution ( [ eq:239 ] ) we see that both terms ( and thus the whole @xmath530 ) have @xmath133-profile proportional to @xmath178 . similarly , @xmath133-profile of the field @xmath482 is proportional to @xmath531 and again decays as @xmath185 for @xmath214 outside the region ( [ eq:242 ] ) . one can easily find that integrals over @xmath532 in both expressions ( [ eq:239 ] ) and ( [ eq:243 ] ) behave as @xmath533 as an index of a bessel function in expressions for @xmath534 and @xmath516 . ] and therefore these terms give small ( of the order @xmath535 corrections to the functions @xmath509 and @xmath536 . substituting expressions for @xmath481 and @xmath482 into ( [ eq:225 ] ) and ( [ eq:232 ] ) , one can easily see that these terms are indeed proportional to @xmath58 , have the profiles , decaying as @xmath185 i.e. can be treated perturbatively . thus , we have shown that the last terms in eqs . ( [ eq:227 ] ) and ( [ eq:228 ] ) can be treated perturbatively in @xmath58 and therefore can be neglected . after that equation for @xmath481 becomes identical to that of the section [ sec : final - state ] and thus the static configuration of an electric field in a capacitor will be the same ! 99 c. n. yang and r. l. mills , `` conservation of isotopic spin and isotopic gauge invariance , '' phys . rev . * 96 * , 191 ( 1954 ) . d. j. gross and r. jackiw , `` effect of anomalies on quasirenormalizable theories , '' phys . d * 6 * , 477 ( 1972 ) ; s. l. adler , `` axial vector vertex in spinor electrodynamics , '' phys . rev . * 177 * , 2426 ( 1969 ) ; j. s. bell and r. jackiw , `` a pcac puzzle : pi0 @xmath537 gamma gamma in the sigma model , '' nuovo cim . a * 60 * , 47 ( 1969 ) . g. t hooft , `` symmetry breaking through bell - jackiw anomalies , '' phys . lett . * 37 * , 8 ( 1976 ) . v. a. rubakov , `` monopole induced baryon number nonconservation , '' iyai preprint iyai - p-0211 , april 1981 ; `` adler - bell - jackiw anomaly and fermion number breaking in the presence of a magnetic monopole , '' nucl . b * 203 * , 311 ( 1982 ) . c. g. callan , `` disappearing dyons , '' phys . d * 25 * , 2141 ( 1982 ) ; `` dyon - fermion dynamics , '' _ ibid . _ * 26 * , 2058 ( 1982 ) . h. georgi and s. l. glashow , `` gauge theories without anomalies , '' phys . rev . d * 6 * , 429 ( 1972 ) . e. dhoker and e. farhi , `` decoupling a fermion in the standard electroweak theory , '' nucl . b * 248 * , 77 ( 1984 ) ; e. dhoker and e. farhi , `` decoupling a fermion whose mass is generated by a yukawa coupling : the general case , '' nucl . b * 248 * , 59 ( 1984 ) . v. a. rubakov and m. e. shaposhnikov , `` do we live inside a domain wall ? , '' phys . b * 125 * , 136 ( 1983 ) . k. akama , `` an early proposal of brane world , '' lect . notes phys . * 176 * , 267 ( 1982 ) [ arxiv : hep - th/0001113 ] . s. randjbar - daemi and m. shaposhnikov , `` qed from six - dimensional vortex and gauge anomalies , '' jhep * 0304 * ( 2003 ) 016 [ arxiv : hep - th/0303247 ] . l. d. faddeev and s. l. shatashvili , `` algebraic and hamiltonian methods in the theory of nonabelian anomalies , '' theor . math . * 60 * , 770 ( 1984 ) [ teor . mat . fiz . * 60 * , 206 ( 1984 ) ] . c. g. callan and j. a. harvey , `` anomalies and fermion zero modes on strings and domain walls , '' nucl . b * 250 * , 427 ( 1985 ) . x. g. wen , `` chiral luttinger liquid and the edge excitations in the fractional quantum hall states , '' phys . b * 41 * , 12838 ( 1990 ) . j. frohlich and a. zee , `` large scale physics of the quantum hall fluid , '' nucl . b * 364 * , 517 ( 1991 ) . r. e. prange , s. m. girvin , eds . , _ the quantum hall effect _ , ( springer , new york , 1990 ) s. naculich , `` axionic strings : covariant anomalies and bosonization of chiral zero modes , '' nucl . b * 296 * , 837 ( 1988 ) ; j. a. harvey and o. ruchayskiy , `` the local structure of anomaly inflow , '' jhep * 0106 * , 044 ( 2001 ) [ arxiv : hep - th/0007037 ] . a. boyarsky , j. a. harvey and o. ruchayskiy , `` a toy model of the m5-brane : anomalies of monopole strings in five dimensions , '' annals phys . * 301 * , 1 ( 2002 ) [ arxiv : hep - th/0203154 ] . e. witten,``five - brane effective action in m - theory , '' j. geom . * 22 * , 103 ( 1997 ) [ hep - th/9610234 ] . m. b. green , j. a. harvey and g. w. moore , `` i - brane inflow and anomalous couplings on d - branes , '' class . * 14 * , 47 ( 1997 ) [ arxiv : hep - th/9605033 ] . d. freed , j. a. harvey , r. minasian and g. moore , `` gravitational anomaly cancellation for m - theory fivebranes , '' adv . * 2 * , 601 ( 1998 ) [ hep - th/9803205 ] . a. boyarsky and b. kulik , `` a note on the m5 brane anomaly , '' phys . b * 516 * , 171 ( 2001 ) [ arxiv : hep - th/0107041 ] . a. boyarsky , o. ruchayskiy and m. shaposhnikov , `` anomalies as a signature of extra dimensions , '' phys . b * 626 * , 184 ( 2005 ) [ arxiv : hep - ph/0507195 ] . s. khlebnikov and m. shaposhnikov , `` brane - worlds and theta - vacua , '' phys . d * 71 * , 104024 ( 2005 ) [ arxiv : hep - th/0412306 ] . v. a. rubakov , `` large and infinite extra dimensions : an introduction , '' phys . usp . * 44 * , 871 ( 2001 ) [ usp . nauk * 171 * , 913 ( 2001 ) ] [ arxiv : hep - ph/0104152 ] . s. randjbar - daemi and m. shaposhnikov , `` a formalism to analyze the spectrum of brane world scenarios , '' nucl . b * 645 * , 188 ( 2002 ) [ arxiv : hep - th/0206016 ] . s. l. dubovsky , v. a. rubakov and p. g. tinyakov , `` is the electric charge conserved in brane world ? , '' jhep * 0008 * , 041 ( 2000 ) [ arxiv : hep - ph/0007179 ] . m. marinelli and g. morpurgo , `` the electric neutrality of matter : a summary , '' phys . lett . b * 137 * , 439 ( 1984 ) . i. oda , `` localization of matters on a string - like defect , '' phys . b * 496 * , 113 ( 2000 ) [ arxiv : hep - th/0006203 ] . m. e. shaposhnikov and p. tinyakov , `` extra dimensions as an alternative to higgs mechanism ? , '' phys . b * 515 * , 442 ( 2001 ) [ arxiv : hep - th/0102161 ] . m. laine , h. b. meyer , k. rummukainen and m. shaposhnikov , `` localisation and mass generation for non - abelian gauge fields , '' jhep * 0301 * , 068 ( 2003 ) [ arxiv : hep - ph/0211149 ] . g. r. dvali , g. gabadadze and m. a. shifman , `` ( quasi)localized gauge field on a brane : dissipating cosmic radiation to extra dimensions ? , '' phys . b * 497 * , 271 ( 2001 ) [ arxiv : hep - th/0010071 ] . a. n. redlich , `` gauge noninvariance and parity nonconservation of three - dimensional fermions , '' phys . lett . * 52 * , 18 ( 1984 ) ; a. n. redlich , `` parity violation and gauge noninvariance of the effective gauge field action in three - dimensions , '' phys . d * 29 * ( 1984 ) 2366 . l. alvarez - gaume , s. della pietra and g. w. moore , `` anomalies and odd dimensions , '' annals phys . * 163 * , 288 ( 1985 ) . w. a. bardeen and b. zumino , `` consistent and covariant anomalies in gauge and gravitational theories , '' nucl . b * 244 * , 421 ( 1984 ) . j. ambjorn and p. olesen , `` electroweak magnetism : theory and application , '' int . j. mod . phys . a * 5 * , 4525 ( 1990 ) . j. ambjorn , j. greensite and c. peterson , `` the axial anomaly and the lattice dirac sea , '' nucl . b * 221 * , 381 ( 1983 ) . i. s. gradshteyn , i. m. ryzhik , a. jeffrey , d. zwillinger , _ table of integrals , series , and products _ , any edition . t. appelquist and j. carazzone , `` infrared singularities and massive fields , '' phys . d * 11 * , 2856 ( 1975 ) . i. antoniadis and s. dimopoulos , `` splitting supersymmetry in string theory , '' arxiv : hep - th/0411032 .

recent years have witnessed a surge of interest in understanding the microscopic origin of friction as a result of the increased control in surface preparations and the development of local probes like the atomic force microscopes ( afm ) . one of the goals of this research is to understand whether extremely low friction can be obtained by an appropriate choice of the sliding conditions . in particular , commensurability between the sliding lattices is one of the elements that determine friction . for a purely incommensurate infinite contact , theoretical arguments suggest that static friction should vanish @xcite . this effect has been called superlubricity @xcite . commensurability and incommensurability are defined in terms of lattice parameters . however , specific orientations can also lead to ( mis)matches of lattices in contacts . very low friction was found in experiments with finite contacts at very low velocities to depend strongly on the orientation @xcite . coexisting states of very different friction have also been observed in the sliding of antimony nanoparticles @xcite and have been attributed to contamination or ( in)commensurate interfaces . meanwhile , recent theoretical studies @xcite have shown that nanocrystals can slide with constant orientation only for particular orientations . this paper examines systematically theoretically the ( in)commensurability and friction of sliding nanocrystals with a triangular lattice symmetry [ such as au ( 111 ) ] on a triangular or hexagonal substrate ( such as graphite ) . gold nanocrystals sliding on graphite are a prototype system for friction that is being investigated both experimentally @xcite and computationally @xcite . in this work , the friction is investigated analytically through the total potential energy of the contact layer on the substrate . the potential energy corrugation plays an essential role in the survival and appearance of stable ( in)commensurate sliding orientations @xcite , as well as the order of magnitude of the friction . in the regime of low sliding velocity , and low temperature , which is the typical situation in afm experiments @xcite , the friction is of the order of @xmath4 , with @xmath5 the substrate period and @xmath6 the corrugation of the total potential energy of the interface . the interaction between the surface and an atom of the contact layer is modelled with a realistic static potential . the scaling of the friction with contact size is determined , depending on the orientation , and conclusions are drawn from this regarding commensurate and incommensurate orientations . the calculations are repeated also for contacts between crystals with rectangular lattice geometries . an illustration of the systems is shown in fig . [ fig : cartoon ] . an illustration of the systems under study . a nanocrystal lies on a triangular or hexagonal substrate with a contact layer that has triangular lattice symmetry . systems with rectangular lattice symmetry are discussed as an example in the appendix . , width=325 ] in sec . [ sec : geometry ] the geometry of gold on graphite is introduced . an expression for the potential energy is derived in sec . [ sec : maths ] . at first , rigid crystals are considered , followed by correction terms for crystals with relatively stiff , but deformable latices . in sec . [ sec : scaling ] , the scaling of the friction is discussed at different ( in)commensurate orientations and orientations with different scaling are identified and classified . it is found that for perfect crystals there are three types of ( in)commensurability and consequently three possibilities for the scaling of the friction with the size of the contact area @xmath0 , namely @xmath7 , and @xmath0 . conditions for each are derived . these three different scaling behaviors for the friction are a direct consequence of the two - dimensional nature of the contact layer , and are thus possible in all geometries , not just triangular . in the appendix , the calculations are repeated for a rectangular geometry , to exemplify this . the scaling of friction for imperfectly and irregularly shaped nanocrystals is discussed in sec . [ sec : imperfectscaling ] . in sec . [ sec : discussion ] , the implications for experiments are discussed , and a method is proposed for quantifying the commensurability of finite - size crystal interfaces . finally , conclusions are presented in sec . [ sec : conclusions ] . the interaction between the graphite substrate and the gold crystal are dominated by the bottom , contact layer of the gold crystal and the top layer of the graphite . commensurability of these two layers is not only controlled by the ratio of the lattice parameters , but also by their relative orientation . the surface of graphite has a hexagonal lattice of carbon atoms with an inter - atomic distance of @xmath8 . a two - dimensional hexagonal potential for a gold atom at postition @xmath9 on a graphite substrate can be written as @xmath10 where @xmath11 \vec{e}_x + \sin[(l-1 ) \pi/3 ] \vec{e}_y$ ] is the unit vector in the direction @xmath12 with respect to the @xmath13-axis , and @xmath14 . without any load force , the potential corrugation of a single gold atom on graphite is denoted by @xmath15 and is around @xmath16 @xcite . a load force can be accounted for with a higher corrugation . with a negative @xmath15 , this potential can also be used to describe a substrate with a triangular lattice . while in general a substrate potential may not be sinusoidal , it is always periodic and hence can be written as a sum of sinusoidal potentials . the analysis described in this work can also be applied to sums of different sinusoidal potentials . more details of the geometry are shown in fig . [ fig : trapezoid ] . from the triangular shape of the nanocrystal found in experiments @xcite , it can be deduced that the contact layer must have similar triangular symmetry . as bulk gold has an fcc lattice , it is reasonable to conclude that the contact layer must be the ( 111 ) cleavage . in the bulk , the lattice parameter is @xmath17 , and consequently the inter - atomic distance in the ( 111 ) layer is @xmath18 . however , the au(111 ) surface is known to show surface reconstruction under a wide range of conditions @xcite . in vacuum , the reconstructed au(111 ) surface looks similar to the unreconstructed bulk ( 111 ) layers , but has a superstructure with a unit cell of size @xmath19 and an average compression in the ( 110 ) direction . consequently , the rotational symmetry of the contact layer is lost . in addition , the surface gold atoms do not lie in a plane , but are shifted by up to 0.15 . as the interaction between the graphite substrate and the au atoms is weak compared to the interaction between the au atoms , the reconstruction of the contact layer in the au - graphite configuration is likely similar to that of au(111 ) in vacuum . in this work , we simply assume that the lattice is only slightly distorted , with the atoms arranged in scalene triangles . as the interaction between the graphite surface and the gold atoms is weak , the shift of 0.15 orthogonal to the surface does not lead to significant changes in the interactions . as the results derived in this work can be generalized to any regular lattice , they can be applied to any reconstructed surface . in general , the nanocrystal can be flexible , and may be deformed due to forces exerted by the substrate . in this work , the crystal is at first assumed to be rigid , and then corrections are made for the displacements of atoms with respect to their equilibrium position . elastic deformations can be neglected @xcite , as the typical coherence length of gold , about 1@xmath20 m , is much larger than the size of a crystal in friction experiments @xcite . young s modulus , which for gold is @xmath21 gpa , gives the stress - strain response and can be used as an indication of the displacement @xmath22 of a single atom with respect to the equilibrium lattice positions . one may consider the force exerted by the substrate on an atom at position @xmath9 , @xmath23 to work on a single unit cell of the contact layer , of diameter @xmath24 . the displacement of the atom can be estimated as @xmath25 this leads to a typical displacement of a single gold atom in the crystal of about @xmath26 , corresponding to a difference in potential energy two orders of magnitude smaller than the corrugation . though young s modulus is a bulk property , and we are interested in the displacement of surface atoms , it is sufficient for obtaining the order of magnitude . a similar estimate for the order of magnitude of the displacement of a surface atom can be obtained from the total number of neighboring atoms and parameters used in molecular - dynamics simulations with atomistic force fields , such as that of ref . @xcite . in the next section , for the total potential energy of the contact layer , the contribution from the displacement of surface atoms are neglected . in sec . [ sec : systematic_soft ] , corrections that include the displacement of the atoms with respect to their equilibrium positions are also made . it has been shown that sliding crystals rotate to specific orientations , which are stable and remain ( nearly ) constant for all time @xcite . these orientations can be identified with periodic orbits in the dynamics and obtained from the potential energy landscape . they occur when the potential energy averaged over a scan line and its corrugation simultaneously exhibit an extremum as a function of orientation . however , the incommensurate orientations typically have higher average energy than the commensurate ones , and can be destroyed by sufficiently large thermal fluctuations , leading to commensurate sliding and an increase of friction . in experiments of small graphite flakes on graphite , incommensurate orientations were found to decay @xcite . larger crystals were found theoretically to rotate more slowly , so that incommensurate orientations survive . the sliding gold nanocrystals studied in the experiments by dietzel et al . @xcite are sufficiently large for this ( with contact areas between @xmath27 and @xmath28 nm@xmath29 ) , though the smaller crystals studied numerically by guerra et al . @xcite , can rotate easily while sliding . in order to determine the scaling of the friction , and hence the commensurate orientations , as well as apply the theory of ref . @xcite , it is necessary to focus on the potential energy of the contact layer on the substrate as a function of position and relative orientation . diagram of the hexagonal graphite substrate lattice ( left ) and a trapezoidal gold crystal ( right ) with definitions of the size parameters @xmath30 , lattice spacings @xmath31 , shape parameter @xmath32 , substrate lattice vector @xmath33 , orientation @xmath34 , and position @xmath35 . the dark atoms shown complete the trapezoid crystal into a triangular one . we use @xmath36 and @xmath37 . for gold , with the reconstruction that has been observed in vacuum @xcite , the parameters for the lattice are @xmath38 , @xmath39 , @xmath40 , @xmath41 . in a perfectly triangular crystal , @xmath42 . , width=325 ] let us calculate the total potential energy of a trapezoidal crystal . the geometry of such as crystal is displayed in fig . [ fig : trapezoid ] , along with definitions of the diameters @xmath43 and @xmath44 , the inter - atomic distances @xmath45 and @xmath46 , the shape parameter @xmath32 , the orientation @xmath34 , and the position of the center of smallest triangle that contains the trapezoid , @xmath35 . first , we assume that the crystal is rigid , and then make a systematic correction for displacement of atoms from their equilibrium position . the total potential energy of the crystal at position @xmath47 and orientation @xmath34 on the substrate is dominated by the interaction of the contact - layer atoms with the periodic substrate potential . it can therefore be written as a sum over all atoms in the contact layer , @xmath48 where @xmath49 are the positions of the atoms in the contact layer relative to @xmath35 , the center of the contact layer . if the crystal is rigid , the positions of the atoms can be written as @xmath50 with @xmath51 the @xmath52 matrix that performs a rotation over an angle @xmath34 . by substituting the substrate potential , eq . ( [ eq : substratepotential ] ) , and rewriting the @xmath53 in terms of imaginary exponentials , we find that for a rigid crystal , @xmath54~. \label{eq : s}\end{aligned}\ ] ] due to the rotational symmetry of the substrate and of a perfectly triangular crystal , @xmath55 consequently , for an irregular trapezoid , it suffices to determine @xmath56 . finally , the sums in eq . ( [ eq : s ] ) can be evaluated by making use of the relation @xmath57}/{\sin\left({c } \right)}~ , \label{eq : sinratio}\end{aligned}\ ] ] with @xmath58 any real number . one finds that @xmath59\right\}\nonumber\\ & \phantom{\sum_{j=0,1}}\times\frac{\sin\left\ { \frac{\pi b}{\alpha } ( d^\prime+1 ) \left [ \tfrac12 \sqrt{3 } \cos\phi + \left(\gamma + \tfrac12 ( -1)^j \beta \right ) \sin\phi \right]\right\}}{\sin\left\ { \frac{\pi b}{\alpha } \left [ \tfrac12 \sqrt{3 } \cos\phi + \left(\gamma + \tfrac12 ( -1)^j \beta\right ) \sin\phi \right]\right\ } \sin\left(\frac{\pi b}{\alpha } ( -1)^j \beta \sin\phi\right)}~ , \label{eq : pottrap2 } \\ v_{\mathrm{trap . } , d , d^\prime,\alpha,\beta,\gamma}^{(0)}(\vec{r},\phi ) & = \null- \tfrac19 v_0 \sum_{l=1}^3 \sin\left [ \frac{\pi b}{\alpha}\left ( \frac{\vec{r}_\triangle\cdot\vec{e}_l}{b}\right ) \right ] s_{d , d^\prime,\alpha,\beta,\gamma}\left(-\tfrac12 \alpha\vec{e}_l , \phi , l\right ) \nonumber\\ \label{eq : pottrap1 } & \phantom{= \null}{}\null - \tfrac19 v_0 \sum_{l=1}^3 \cos\left [ \frac{\pi b}{\alpha}\left ( \frac{\vec{r}_\triangle\cdot\vec{e}_l}{b}\right ) \right ] s_{d , d^\prime,\alpha,\beta,\gamma}\left(0,\phi , l\right ) ~.\end{aligned}\ ] ] note that the angle - dependent corrugation prefactor contains all the geometric dependence on size and orientation , and does not depend on the position @xmath47 of the center of mass on the substrate . the total corrugation is @xmath60^{\tfrac{1}{2 } } \left ( \tfrac32 \cos \frac{\xi}{3 } + \tfrac12 \sqrt{3 } \sin \frac{|\xi|}{3 } \right)~,\\ \xi & = \arctan\left(\frac{s_{d , d^\prime,\alpha,\beta,\gamma } \left ( -\tfrac12 \alpha\vec{e}_1 , \phi,1 \right)}{s_{d , d^\prime,\alpha,\beta,\gamma } \left(0 , \phi,1\right)^2}\right)~,\end{aligned}\ ] ] where @xmath61 and @xmath62 are used to denote the value of the maxima and minima of the potential respectively . the average potential energy over any scan line vanishes for all scan lines not exactly orthogonal to one of the @xmath63 . in this case , the conditions for the existence of stable orientations of ref . @xcite are met . as is also the case for rectangular lattices in ref . @xcite , the incommensurate stable orientations are distributed evenly over the unit circle and their number grows linearly with the diameter @xmath43 . it should be noted that these results are consistent with the numerical results for the distribution of activation energies described in ref . @xcite for rigid crystals with triangular symmetry on a square lattice . however , because those results were obtained for relatively small clusters , the authors could not distinguish the commensurate orienations with larger corrugation . in realistic nanocrystals , atoms of the contact layer are displaced with respect to their equilibrium position , due to the forces exerted on them by the substrate . from the arguments presented in sec . [ sec : geometry ] , it is clear that this displacement is small for the system studied in this work . however , as the contact layer of typical nanocrystals contain many atoms , the energy effect is not necessesarily negligeable compared to the low total potential energy for incommensurate configurations . to provide some insight into the effects of lattice deformations without making the calculations much more complicated , it is assumed here that each atom couples only to its equilibrium position . within this mean - field approximation , small displacements @xmath64 from the equilibrium position are due to a force equal to @xmath65 where the spring constant @xmath66 can be estimated from eq . ( [ eq : displacementfestimate ] ) to be , @xmath67 because typical sliding velocities in experiments are low , we may limit ourselves to the quasi - static case , where all forces on each atom sum up to zero . the correction to the potential energy consists of the reduction in potential energy due to the displacement on the lattice , and the additional potential energy stored in the spring . to leading order in the displacement , we find a correction to the potential energy of the atom with equilibrium position @xmath9 of @xmath68 with @xmath69 the force exerted by the substrate on an atom at position @xmath9 . by substituting eq . ( [ eq : substratepotential ] ) , using the relation @xmath70 , and using the properties of @xmath63 , one obtains @xmath71 ^ 2\\ & = - \frac{4 \pi^2 v_0 ^ 2}{81\kappa \alpha^2 } \sum_{l=1}^3 \left [ 1 - \cos\left(\frac{4 \pi}{\alpha } \vec{e}_l\cdot\vec{r}\right ) + \cos\left(\frac{2 \pi}{\alpha } \vec{e}_{l } \cdot\vec{r}\right ) -\cos\left(\frac{2 \sqrt{3}\,\pi}{\alpha } \vec{e}_{l+\tfrac12 } \cdot\vec{r}\right ) \right]\end{aligned}\ ] ] this expression is of a form very similar to eq . ( [ eq : substratepotential ] ) . the approach used in sec . [ sec : rigidpotential ] to obtain the total potential energy of a rigid crystal can be applied to this expression as well . hence , one obtains the first - order correction of @xmath72 } \left . \null - v_{\mathrm{trap . } , d , d^\prime,\alpha/2,\beta,\gamma}^{(0)}(\vec{r},\phi ) - v_{\mathrm{trap . } , d , d^\prime,\alpha/\sqrt{3},\beta,\gamma}^{(0)}({\mathcal r}(\pi/6 ) \cdot \vec{r},\phi+\pi/6 ) \right ] ~. \label{eq : firstordercorrection}\end{aligned}\ ] ] though the functional forms of the terms in this correction are very similar to eq . ( [ eq : pottrap1 ] ) , the values of the parameters are different . as is discussed in the next section , this allows the first - order correction to become larger than the leading order in some cases , despite the small prefactor . higher order corrections and corrections for direct coupling to the neighboring atoms in the contact layer could , in principle , be obtained using the same approach . however , these corrections would have to include the nonlinear response of the lattice , as well as second - order derivatives of the potential energy . this would lead to expressions similar in form to eq . ( [ eq : substratepotential ] ) , and therefore to additional terms similar to eq . ( [ eq : firstordercorrection ] ) , but with yet again different parameters and smaller prefacors . from the structure of the function @xmath73 in eq . ( [ eq : pottrap2 ] ) , one can derive important general geometrical and scaling properties . from eqs . ( [ eq : pottrap2 ] ) , ( [ eq : pottrap1 ] ) , and ( [ eq : firstordercorrection ] ) it is clear that there are a number of parameters which control the potential energy corrugation of the surface which the nanocrystal is subjected to . for a perfectly triangular crystal these are the diameters @xmath43 , the ratio of the lattice constants , @xmath74 , and the orientation @xmath34 . in eq . ( [ eq : pottrap2 ] ) , the numerator varies rapidly between -1 and 1 , with @xmath34 , while the denominator provides a size - indepent prefactor that can diverge for particular angles . this can be seen in figs . [ fig : numdenom ] and [ fig : numdenom2 ] , where one of the angle - dependent corrugation prefactors is plotted along with the function @xmath75\right\ } \sin\left(\frac{\pi b\beta}{\alpha } \sin\phi\right)\right|^{-1 } \nonumber\\ & \phantom{= } \null + \tfrac{1}{9 } v_0 \left|\sin\left\ { \frac{\pi b}{\alpha } \left [ \tfrac12 \sqrt{3 } \cos\phi + \left(\gamma- \tfrac12 \beta\right ) \sin\phi \right]\right\ } \sin\left(- \frac{\pi b \beta}{\alpha } \sin\phi\right)\right|^{-1}~.\end{aligned}\ ] ] this function gives an upper bound for the absolute value of the corrugation due to a particular term . it can be used to estimate the typical corrugation of any triangular or trapezoidal crystal . at particular combinations of orientations @xmath76 and lattice parameter ratio @xmath74 , one or both of the denominators vanishes and @xmath77 diverges . the numerator in eq . ( [ eq : pottrap2 ] ) vanishes as well , so that @xmath73 itsself does not diverge . as the arguments of the trigonometric functions in the numerator contain factors of @xmath43 and @xmath44 , the resulting contribution to @xmath73 is an order of @xmath43 or @xmath44 higher than at other orientations . there is some arbitraryness to the choice of numerator and denominator , as both can be multiplied by any function . this can not however lead to additional factors of @xmath43 . for determining the order of magnitude of the corrugation at any orientation , @xmath77 therefore suffices . the angle - dependent corrugation prefactor , eq . ( [ eq : pottrapcor ] ) , for a perfectly triangular nanocrystals on a hexagonal substrate with @xmath78 ( solid line ) , along with @xmath79 ( dotted line ) , which does not depend on the diameter @xmath43 . the ratio @xmath80 . the denominators control the typical size of the corrugation . whenever the denominator nearly diverges near the orientations given by eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) , a peak appears . there , the interface is ( partially ) commensurate . , width=325 ] a logarithmic plot of a peak in the corrugation for a perfectly triangular rigid nanocrystals on a hexagonal substrate with @xmath81 ( thin solid line ) and @xmath82 ( thick solid line ) , along with @xmath79 ( dotted line ) . the denominators control the typical size of the corrugation . in this case , the corrugation diverges with @xmath83 , but , very close to @xmath76 , it remains finite , because the numerator vanishes as well . , width=325 ] the ( in)commensurate configurations can be distinguished and classified by the exponent @xmath84 with which the corrugation scales with the contact diameter . this order of the commensurability can be obtained from the divergence of the size - independent prefactor in the corrugation , which goes as @xmath85 . i.e. , it is equal to the number of derivatives ( starting with the zeroth ) with respect to @xmath34 of the denominators in eq . ( [ eq : pottrap2 ] ) that are equal to zero . the scaling of the friction with the contact area can be directly related to the order of commensurability through @xmath86 for the commensurate orientations generated by the first - order correction in eq . ( [ eq : firstordercorrection ] ) that do not coincide with commensurate orientations of the leading order , the friction is a factor of order @xmath87 smaller . for most parameter combinations and orientations the contact is incommensurate with @xmath88 , the denominator does not vanish at all , and both denominator and numerator are of order @xmath89 . the corrugation of the nanocrystal on the substrate does not scale with the size of the crystal . the sliding friction consequently is of the order @xmath90 . it should be noted , however , that without a load force , @xmath15 for gold on graphite is comparable to @xmath91 at room temperature . temperature effects on the friction can therefore play a role as well in this regime , reducing the friction further . commensurability order @xmath92 can occur when one of the two @xmath93 functions in the denominator in eq . ( [ eq : pottrap2 ] ) vanishes for a particular orientation @xmath76 . for a rigid crystal , this occurs when one of the following conditions is met : @xmath94 & = p~,\\ \label{eq : denominator2 } \frac{b\beta}{\alpha } \sin\phi_{\mathrm{c . } } & = p'~,\end{aligned}\ ] ] with @xmath95 . if @xmath74 is sufficiently large , i. e. large inter - atomic distance in the nanocrystal , solutions exist also for @xmath96 . this implies that there exist nontrivial commensurate - like orientations with high corrugation . at these orientations , there is no obvious lining up of symmetry axes , or other source of commensurability , as there is for the trivial solutions with @xmath97 , but the corrugation and friction nevertheless behave in a similar way . this is illustrated in fig . [ fig : nontrivialangles ] , where the corrugation is plotted for several values of @xmath74 . two commensurate combinations of lattices are also shown , one trivial , and one nontrivial . in the nontrivial one , no lining up is immediately visible , yet the potential energy still scales with the diameter , as it does for the trivial commensurate configuration . for bulk au and graphite , @xmath41 , @xmath98 , @xmath99 , and thus there are commensurate orientations @xmath100 . the corrugation ( a ) as a function of the orientation for crystals with diameter @xmath101 and various @xmath74 and two commensurate orientations with @xmath92 for triangular crystals with @xmath102 and with ( b ) trivial @xmath103 and ( c ) nontrivial @xmath104 . though the nontrivial commensurate configuration shown in ( b ) does not have any obvious lining up of symmetry axes , it nevertheless corresponds to a strong peak in the corrugation in ( c ) . for larger @xmath74 , more nontrivial solutions of eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) exist and hence more orientations with large corrugations . , title="fig:",width=325 ] the corrugation ( a ) as a function of the orientation for crystals with diameter @xmath101 and various @xmath74 and two commensurate orientations with @xmath92 for triangular crystals with @xmath102 and with ( b ) trivial @xmath103 and ( c ) nontrivial @xmath104 . though the nontrivial commensurate configuration shown in ( b ) does not have any obvious lining up of symmetry axes , it nevertheless corresponds to a strong peak in the corrugation in ( c ) . for larger @xmath74 , more nontrivial solutions of eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) exist and hence more orientations with large corrugations . , title="fig:",width=309 ] the first - order correction for lattice deformations [ eq . ( [ eq : firstordercorrection ] ) ] produces additional commensurate orientations , which satisfy one of the conditions @xmath105 & = \tfrac12 p~,\\ \label{eq : denominator4 } \frac{b\beta}{\alpha } \sin\phi_{\mathrm{c . } } & = \tfrac12 p'~,\\ \frac{b}{\alpha } \left [ \tfrac12 \sqrt{3 } \cos(\phi_{\mathrm{c.}}-\tfrac16 \pi ) + \gamma \sin(\phi_{\mathrm{c.}}-\tfrac16 \pi ) \right.&\nonumber\\\left.\null \label{eq : denominator5 } \pm \tfrac12 \beta \sin(\phi_{\mathrm{c.}}-\tfrac16 \pi ) \right ] & = \tfrac13\sqrt{3 } p~,\\ \frac{b\beta}{\alpha } \sin(\phi_{\mathrm{c.}}-\tfrac16\pi ) & = \tfrac13\sqrt{3 } p'~. \label{eq : denominator6}\end{aligned}\ ] ] these conditions contain additional factors of @xmath106 and @xmath107 when compared to eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) , due to the appearance of such factors in eq . ( [ eq : firstordercorrection ] ) . equations ( [ eq : denominator3 ] ) through ( [ eq : denominator6 ] ) yield a set of additional commensurate orientations that are different from the ones produced by a rigid crystal . in fig . [ fig : corrections ] , the corrugation in the total potential energy , consisting of the rigid - crystal term and the first - order correction for lattice deformations , is shown for a very large perfectly triangular crystal . peaks corresponding to both the nontrivial commensurate configurations and the additional commensurate configurations associated with lattice deformations are clearly visible . the the corrugation ( a ) as a function of the orientation for non - rigid triangular crystals with diameter @xmath108 and @xmath109 ( the typical value for au ) , for triangular crystals with @xmath110 , and lattices for commensurate orientations with @xmath92 , for ( b ) a rigid crystal @xmath103 and ( c ) a non - rigid crystal at @xmath111 . the ratio of lattices parameters @xmath110 is incommensurate . the short lines in ( b ) indicate the direction of the displacement of the atoms as a result of the substrate potential , and their length is proportional to its magnitude . the peaks marked `` rigid crystal '' appear already when the crystal is assumed to be rigid , and satisfy eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) . the lower peaks that appear due to lattice deformations are marked with `` first order '' and are the solutions of eqs . ( [ eq : denominator3 ] ) ( [ eq : denominator6 ] ) . , title="fig:",width=325 ] the the corrugation ( a ) as a function of the orientation for non - rigid triangular crystals with diameter @xmath108 and @xmath109 ( the typical value for au ) , for triangular crystals with @xmath110 , and lattices for commensurate orientations with @xmath92 , for ( b ) a rigid crystal @xmath103 and ( c ) a non - rigid crystal at @xmath111 . the ratio of lattices parameters @xmath110 is incommensurate . the short lines in ( b ) indicate the direction of the displacement of the atoms as a result of the substrate potential , and their length is proportional to its magnitude . the peaks marked `` rigid crystal '' appear already when the crystal is assumed to be rigid , and satisfy eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) . the lower peaks that appear due to lattice deformations are marked with `` first order '' and are the solutions of eqs . ( [ eq : denominator3 ] ) ( [ eq : denominator6 ] ) . , title="fig:",width=309 ] finally , @xmath112 can occur only if the two @xmath93 functions in the denominators have coinciding zeros . this happens only for special ratios of the lattice parameters and corresponds to commensurability in two parameters . the conditions of eqs . ( [ eq : denominator1 ] ) and ( [ eq : denominator2 ] ) for the orientation can only be satisfied simultaneously if the lattice parameters are related by @xmath113 ^ 2~ , \label{eq : denominatorx1}\end{aligned}\ ] ] with @xmath95 . from the first - order correction in eq . ( [ eq : firstordercorrection ] ) additional @xmath112 commensurate combinations of parameters are found , @xmath114 ^ 2~,\\ 3 \left(\frac{b}{\alpha}\right)^2 = \left(\frac{p'}{\beta}\right)^2 + \frac43 \left[p - p'\left(\frac\gamma\beta \pm \frac12 \right)\right]^2~. \label{eq : denominatorx3}\end{aligned}\ ] ] equation ( [ eq : denominatorx2 ] ) corresponds to commensurate configurations at angles satisfying eqs . ( [ eq : denominator3 ] ) and ( [ eq : denominator4 ] ) , while eq . ( [ eq : denominatorx3 ] ) corresponds to commensurate configurations at angles satisfying eqs . ( [ eq : denominator5 ] ) and ( [ eq : denominator6 ] ) . a similar structure appears also in the expressions for rectangular nanocrystals on rectangular lattices shown in the appendix and also treated in ref . this is due to the fact that any sum over atoms arranged in a regular way in nanocrystal , subjected to a periodic potential , can be written in terms of double sums of the form of eq . ( [ eq : sinratio ] ) . hence , two @xmath93 functions always appear in the denominator , and similar conditions exist for which the denominator vanishes . the system studied in ref . @xcite , w on naf , however does not have a sufficiently high ratio of lattice parameters to produce nontrivial commensurate orientations . consequently , no nontrivial commensurate orientations were observed there . in experiments of friction of nanocrystals , the crystals are either pulled along a particular direction @xcite , or pushed @xcite . when crystals are pulled , the direction of their motion is enforced externally . when the crystals are pushed , they may move in a direction not precisely along to the force , following some preferred sliding direction . a comparison between eqs . ( [ eq : pottrap1 ] ) and ( [ eq : substratepotential ] ) shows that the potential energy of a nanocrystal depends on the position in a similar way to that of a single atom . consequently , the preferred sliding directions are the same as for a single atom , along vectors othogonal to @xmath63 . the nontrivial commensurate orienations do not lead to any nontrivial preferred sliding directions . in addition , the activation energy is always equal to the same fraction of the corrugation . however , the first - order corrections produce a potential form which is rotated by @xmath115 . for some parameter combinations , this may lead to additional preferred sliding directions parallel to @xmath63 . it is interesting to compare the model used here and the results to what is known about the frenkel - kontorova ( fk ) model @xcite in one and two dimensions . the fk model consists of a chain or sheet of particles , coupled to one - another and a ( quasi-)periodic substrate . the coupling between the particles may be weak or strong compared to the substrate , leading to different types of behavior . when the substrate is weak by comparison , for incommensurate lattice parameters , the interface can slide without friction . in the one - dimensional fk model , commensurate configurations occur for rational ratios of the lattice parameters @xcite . for the nearly rigid case , in the fk model , there is a cascade of commensurate configurations with decreasing friction . for weak coupling between monomers in the chain , there is a breaking of analyticity @xcite , which results in any ratio of lattice parameters yielding nonvanishing friction for infinite contacts . other effects also start playing a role that do not affect stiff crystals , for instance nucleation of incommensurate domains within a large sheet @xcite ) . in the system considered here a similar cascade would appear , if more higher order corrections for lattice deformations were included . more commensurate orientations with increasingly small friction would be found . each subsequent term in the potential energy would however be multiplied by a small prefactor , and so each additional commensurate orientation would correspond to an increasingly small corrugation . the one - dimensional fk model has only two length scales , and therefore only one parameter which controls commensurability . the system studied in this work has two parameters which control commensurability , similar to the frenkel - kontorova ( fk ) model in two dimensions , and the quasi - periodic one - dimensional fk model @xcite . unlike the model described here , the quasiperiodic fk model can not have coexistence of commensurate states with different order of commensurability . the two - dimensional fk model , however , is already too complex to be analytically tractable . hence , most results dealing with extended two - dimensional deformable interfaces are numerical ( see , for instance , refs @xcite ) . recently , though , experiments were performed that model atoms as colloids in suspensions , and reproduce the fk model closely @xcite . in real - world , experimental conditions , crystaline particles are almost never perfectly triangular , or even trapezoidal . at the very least , even if the crystal lattice is in tact , some atoms may be missing and the corners may be rounded . any shape of particle can be written as a sum ( including negative terms , if necessary ) of regularly shaped particles . the number of terms in this sum can be estimated from the number of step edges @xmath116 on the circumference of the contact layer . each term contributes an amount to the total potential energy of the order of @xmath117 . because the total potential energy at a particular orientation varies rapidly with the size of the particle [ see eq . ( [ eq : pottrap2 ] ) ] , one may assume a random phase for every term . the total corrugation for an irregular particle is therefore of the order of @xmath118 , but can never grow faster than @xmath119 . if the shape of the particle is simply scaled , then the number of step edges increases with @xmath120 . such crystals therefore have friction scaling similarly to eq . ( [ eq : scaling ] ) with , @xmath121 i. e. with exponents @xmath122 , and @xmath89 . the effects of lattice distortions at the edge of the contact layer can be included by summing up in a similar way over a small number of contact layers with different sizes to account for the different interactions at the edge of the crystal . therefore , the scaling with the crystal size and the ( in)commensurate configurations is not affected by the edges . it should also be noted that , regardless of their shape , the friction of amorphous crystals always scales @xmath123 @xcite . the three different types of ( in)commensurability , quantified with the values of @xmath124 , can occur at different orientations in a single interface already for rigid crystals . in fig . [ fig : multiplicity ] , such a case is shown . the trivial commensurate orientations have @xmath92 , corrgation scaling with @xmath43 , and are the result of a single vanishing denominator in eq . ( [ eq : pottrap2 ] ) . the nontrivial commensurate orientations are the result of both denominators vanishing at the same orientation , and thus correspond @xmath112 and corrugation scaling with @xmath125 . the corresponding orientations around @xmath126 are particularly interesting , as these are usually assumed to be strongly incommensurate , and are indeed so for most combinations of lattice parameters . the corrugation ( a ) as a function of the orientation for @xmath127 and @xmath102 for a perfectly triangular nanocrystal with @xmath128 , and two commensurate orientations for @xmath127 , at ( b ) @xmath129 and ( c ) @xmath130 . the three different friction branches occur in one interface here . for the case of @xmath131 , the corrugation is of the order @xmath125 , while for @xmath103 it is of order @xmath43 , and at most other orientations the typical corrugation is not dependent on @xmath43 . , title="fig:",width=325 ] the corrugation ( a ) as a function of the orientation for @xmath127 and @xmath102 for a perfectly triangular nanocrystal with @xmath128 , and two commensurate orientations for @xmath127 , at ( b ) @xmath129 and ( c ) @xmath130 . the three different friction branches occur in one interface here . for the case of @xmath131 , the corrugation is of the order @xmath125 , while for @xmath103 it is of order @xmath43 , and at most other orientations the typical corrugation is not dependent on @xmath43 . , title="fig:",width=309 ] for some lattice parameters , the rigid crystal has only two friction branches , but a third branch appears when lattice deformations are taken into account . this is shown for a perfectly triangular non - rigid nanocrystal with @xmath132 in fig . [ fig : correctionsau ] . the first - order peaks appear at the solutions of eqs . ( [ eq : denominator3 ] ) ( [ eq : denominator6 ] ) , and are generated by both rigid - crystal contributions and first - order corrections . however , for this combination of lattice parameters , eqs . ( [ eq : denominator5 ] ) and ( [ eq : denominator6 ] ) are satisfied for @xmath133 . as a result , @xmath112 is possible for the first - order correction terms , but not for the zeroth order . consequently , the small prefactor for the stiffness of the crystal is multiplied by a factor that can be as large as @xmath43 . for sufficiently small crystals with nearly commensurate lattice parameters , the commensurate orientation behaves as if @xmath112 , and its friction grows as @xmath125 . for gold on graphite , with @xmath134 close to @xmath135 , the corrections can easily become as large as the leading order terms . it should however be noted that , as this effect is very sensitive to the lattice parameters , the surface reconstruction , however minimal , can still enhance or decrease the effect strongly . the surface construction of au(111 ) on graphite may indeed itsself be affected by the near match in lattice parameters , possibly leading to an even closer match of the reconstructed surface . the fk model can be used to understand surface recontructions @xcite , and the calculations of this paper can similarly be applied to get a handle on surface reconstructions in an interface . the corrugation ( a ) in the potential energy as a function of the orientation and commensurate lattices ( b ) at @xmath129 , ( c ) at @xmath130 for a perfectly triangular nanocrystal with @xmath108 that is not rigid with @xmath109 and @xmath132 , which is a commensurate configuration for the first - order corrections , this plot is similar to fig . [ fig : corrections ] . the additional peaks that appear because the crystal is not rigid are the solutions of eqs . ( [ eq : denominator3 ] ) ( [ eq : denominator6 ] ) . the orientation shown in ( c ) is usually strongly incommensurate , even though the symmetry axes area clearly lined up , because the potential energy of the interface is not strongly corrugated . due to the matching lattice parameters , the first - order corrections contribute @xmath136 to the corrugation . for rigid crystals this relative orientation is incommensurate for almost all values of @xmath74 . even for @xmath135 , which looks commensurate , a rigid crystal does not have a corrugation that increases with the size . , title="fig:",width=325 ] the corrugation ( a ) in the potential energy as a function of the orientation and commensurate lattices ( b ) at @xmath129 , ( c ) at @xmath130 for a perfectly triangular nanocrystal with @xmath108 that is not rigid with @xmath109 and @xmath132 , which is a commensurate configuration for the first - order corrections , this plot is similar to fig . [ fig : corrections ] . the additional peaks that appear because the crystal is not rigid are the solutions of eqs . ( [ eq : denominator3 ] ) ( [ eq : denominator6 ] ) . the orientation shown in ( c ) is usually strongly incommensurate , even though the symmetry axes area clearly lined up , because the potential energy of the interface is not strongly corrugated . due to the matching lattice parameters , the first - order corrections contribute @xmath136 to the corrugation . for rigid crystals this relative orientation is incommensurate for almost all values of @xmath74 . even for @xmath135 , which looks commensurate , a rigid crystal does not have a corrugation that increases with the size . , title="fig:",width=309 ] the order of commensuratebility , as it is defined in this work , is a general quantity that can be used to characterize commensurability of crystalline interfaces . the @xmath93 functions in the denominator in eq . ( [ eq : pottrap2 ] ) originate from summing over a line of regularly spaced atoms , which must always produce factors of the type shown in eq . ( [ eq : sinratio ] ) . such sums occur in the potential energy of any rigid regular crystalline contact layer . thus , the analysis described here in terms of its derivatives can be used to identify and classify commensurate configurations for any regular nanocrystal on any regular substrate . in the appendix , the expression for the potential energy of a rectangular crystal on a rectangular substrate is worked out as an example . for a two - dimensional contact area , it is not possible to have @xmath84 larger than 2 , because the total potential energy can never grow faster than linearly with the contact area . correspondingly , in one dimension , @xmath84 can not exceed unity . as illustrated by the case of gold on graphite , configurations close to commensurate still show some of the commensurate behavior for finite size crystals . it is therefore worthwhile to consider a quantity similar to @xmath84 , a finite - size order of commensurability , @xmath137}{\ln { n}}~.\end{aligned}\ ] ] for simplicity , or if @xmath138 is not well - known , a simple periodic sinusoidal function can be used in place of @xmath138 . in the limit of @xmath139 , @xmath140 converges to the order of commensurability @xmath84 . as it is impossible to define lattice orientation for a single atom , it diverges for @xmath141 . for finite - size crystals , it can be used as a measure of the commensurability of the contact . in ref . @xcite another measure for commensurability was defined for determining to what extent weakly - coupled atoms line up with the substrate in an incommensurate contact . this measure , however , is not suitable for determining the commensurability of two ( nearly ) rigid lattices . in this work , the ( in)commensurability of finite - size crystalline interfaces was investigated , particularly for crystals with triangular symmetry on triangular or hexagonal substrates such as gold on graphite . the crystals were assumed to be nearly rigid , with the stiffness being used as an expansion parameter . a simple method was developed for determining the commensurate configurations , through quantifying the commensurability by the scaling of the potential corrugation of the crystal with the diameter of the contact area at constant orientation and lattice parameters . this method can be applied not only to the two geometries discussed in this paper ( triangular / hexagonal and rectangular / rectangular ) , but to any regular lattice on a regular substrate . it was found that , due to the two - dimensional nature of the contact layer , and associated two parameters for commensurability ( ratio of lattice parameters and relative orientation ) , two different types of commensurate contacts can exist for the same materials . if the relative orientation is commensurate , but the lattice parameters are not , the potential corrugation scales with the diameter of the contact area , while if both are commensurate , it scales with the contact area . consequently , for some combinations of materials three different friction branches can appear for the same crystal shape and materials , scaling with the contact area as @xmath7 , and @xmath142 for perfect crystals [ see eq . ( [ eq : scaling ] ) ] . when the crystals have irregular shapes , the corrugation and friction scale with @xmath143 [ see eq . ( [ eq : scalingimperfect ] ) ] . nontrivial commensurate configurations were identified in particular for the triangular on triangular on hexagonal geometry . these commensurate orientations exist if the lattice parameter the nanocrystal is sufficiently high compared to the lattice parameter of the substrate . they are also a direct result of the two - dimensional nature of the contact layer . it was found that lattice deformations produce a number of additional commensurate configurations that would be incommensurate for a rigid crystal . interestingly , the prototype geometry of gold on graphite is very close to the parameters which produce such a friction multiplicity . it might therefore be possible to detect all three friction branches with different commensurability in friction experiments of gold nanocrystal on graphite , as well as nontrivial commensurate angles . this work was financially supported by a veni grant of netherlands organisation for scientific research ( nwo ) and by an unga forskare grant from the swedish research council . the author would also like to thank a. fasolino , d. dietzel , and a. schirmeisen for discussions , and o. manyuhina for carefully reading the manuscript . 31 natexlab#1#1bibnamefont # 1#1bibfnamefont # 1#1citenamefont # 1#1url # 1`#1`urlprefix[2]#2 [ 2][]#2 , * * , ( ) . , in _ _ , edited by ( , ) , vol . of _ _ , pp . , * * , ( ) . , , , * * , ( ) . , , , , , , * * , ( ) . , , , , , * * , ( ) . , , , , , , * * , ( ) . , , , , , , , , , * * , ( ) . , , , * * , ( ) . , * * , ( ) . , , , , ( ) , . , , , , * * , ( ) . , * * , ( ) . , , , * * , ( ) . , , , , * * , ( ) . , * * , ( ) . , * * , ( ) . , * * , ( ) . , , , , , * * , ( ) . , * * , ( ) . , , , , , * * , ( ) . , * * , ( ) . , , , , * * , ( ) . , * * , ( ) . , , , , * * , ( ) . , * * , ( ) . , , , , * * , ( ) . , , , * * , ( ) . , , , * * , ( ) . , , , * * , ( ) . , * * , ( ) . , lattice parameters @xmath144 , contact layer size @xmath145 , and the potential energy of a contact layer atom on the substrate . if a contact layer atom lies on top of a substrate atom , its potential energy is @xmath146 . if it lies directly between an atom and its nearest neighbor in the @xmath13 or @xmath147 direction , it has potential energy @xmath148 or @xmath149 respectively . if it lies in the center of a rectangle , at equal distance from four substrate atoms , without loss of generality , we may set the potential energy to 0 . the origin of the coordinate system is chosen to lie on top of a substrate atom . [ fig : rectangular ] , width=294 ] the identification and classification of commensurate(-like ) configurations described in this paper can easily be repeated for the general rectangular lattices , as described for instance in ref . a general potential energy of a single atom , @xmath150 is given by @xmath151 where the geometry and associated parameters are defined in fig . [ fig : rectangular ] . @xmath152 \nonumber\\ & \phantom{= nnnn } \times \frac { \sin\left[\pi ( m+1 ) \left(\frac{b_1}{a_1}\cos{\phi } + ( -1)^j \frac{b_1}{a_2}\sin{\phi}\right)\right ] \sin\left[\pi ( n+1 ) \left(\frac{b_2}{a_1}\sin{\phi } - ( -1)^j \frac{b_2}{a_2}\cos{\phi}\right)\right ] } { \sin\left[\pi \left(\frac{b_1}{a_1}\cos{\phi } + ( -1)^j \frac{b_1}{a_2}\sin{\phi}\right)\right ] \sin\left[\pi \left(\frac{b_2}{a_1}\sin{\phi } - ( -1)^j \frac{b_2}{a_2}\cos{\phi}\right)\right ] } ~,\end{aligned}\ ] ] where @xmath153 originate from the second , third , and fourth term on the right - hand side of eq . ( [ eq : va ] ) . similarly , the higher - order corrections for lattice deformations can also be obtained along similar lines as described in sec . [ sec : firstordercorrections ] . this yields terms with higher perdiodicity , i.e. , a term obtained from the derivative of @xmath154 which looks very similar , except that @xmath155 is replaced by @xmath156 . similarly , there is a term derived from @xmath157 with @xmath158 replaced by @xmath159 , and two terms from @xmath160 with both @xmath155 and @xmath158 replaced . from these expressions it is easy to obtain the conditions for vanishing denominators , and thus for @xmath124 . the three terms @xmath153 each produce a different set of conditions for commensurability at different angles and ratios of lattice parameters . for @xmath154 , one finds that @xmath92 commensurability occurs if one of the following conditions is met , @xmath161 for some @xmath95 . both of these conditions can be met simulataneously , leading to @xmath112 commensurability , if @xmath162 as each of @xmath153 has a different prefactor that depends on the energy barriers of the potential , the different terms do not , in general , cancel each other out . it is interesting to note that often for square lattices it is assumed that @xmath166 . in this case , @xmath160 vanishes and the commensurate configurations associated with it disappear , leading to lower friction .

it has been believed some time ago that gauge symmetries with anomalies cancelled by the green - schwarz mechanism @xcite in @xmath1 dimensions could be spontaneously broken in such a way that , after compactifying to four dimensions , a global symmetry inherited from the gauge symmetry could , in principle , survive , @xmath2 , @xcite . the survival of a global gauge symmetry even after the gauge bosons have acquired a higgs field induced mass is quite mysterious because the vacuum is not gauge invariant under the gauge symmetry with constant gauge function ( @xmath3 ) due to the kalb - ramond field ( which plays a major role in this phenomenon ) . despite this , it seemed as though there remains a phase transformation symmetry for the fields carrying the family - dependent @xmath4-charge when gauge symmetry is spontaneously broken in this remarkable way . we emphasise in this article that , precisely this phase transformation symmetry is _ not _ a true symmetry if one takes into account the world sheet instantons . we shall argue below that if the compactifying space is of order of the fundamental scale , then the local and global gauge symmetry gets totally broken . however , it is unrealistic to compactify the space so close to the fundamental scale . the crucial point is that the effect of the world sheet instantons will be exponentially suppressed . in the case of very strong breaking ( of the order of the fundamental scale ) it would mean that we could not apply the green - schwarz anomaly cancellation mechanism in @xmath0 dimensions , @xmath5 , for the application to the large hierarchical yukawa coupling constant structures . on the other hand , if we let the compactifying scale be much below the fundamental scale , there is another odd feature : there is very light ( abelian ) gauge particle from the fundamental scale point of view and correspondingly we get that the condition @xmath6 for not having anomalies is fulfilled identically . this would lead to a rather strange electrodynamics . if the validity of the usual @xmath7 is not maintained , the dynamics could open the possibility for space - time foam causing the break down of the global charge conservation due to wormholes . this article is organised as follows : in the next section , we review the green - schwarz anomaly mechanism , and in section @xmath8 the world sheet instantons , and we also discuss the string coupling constant . section @xmath9 contains the various discussions including the suspected effects of wormholes . finally , section @xmath10 contains our conclusions . let us review the green - schwarz anomaly cancellation mechanism focusing on the kalb - ramond field in @xmath1 dimensions and then the application for the @xmath0 dimensions . for the purpose of making phenomenological fits of the quark and ( charged ) lepton masses and mixing angles it is very useful to have some approximately conserved charges @xcite ( in addition to the gauge charges of the standard model ) so that most of the masses get suppressed due to the differences in quantum numbers of the right- and left - handed weyl components . it is very attractive , and needed due to the effects of wormholes @xmath11 @xcite , to let such mass suppressing charges be gauge charges . there are many gauge charges in superstring theory so such a picture is not unnatural in this theory . working in @xmath0 dimensions one would at some level expect to obtain a @xmath0 dimensional field theory with gauge fields which could be described as renormalisable . that in turn would imply that the triangle anomalies resulting from the various chiral fermions in the effective @xmath0 dimensional model should cancel , @xmath5 , no violation of gauge symmetry would be caused . otherwise this effective model would not be renormalisable . now , however , it became very popular to use the inspiration from the superstring theory to suggest models in which this `` usual '' gauge anomaly cancellation does _ not _ take place . from the four dimensional point of view this avoidance , which is usually needed for renormalisation , gauge- and mixed anomaly cancellation conditions seems quite extraordinary : a certain coefficient field @xmath12 in an expansion for the kalb - ramond field @xmath13 in the @xmath1 dimensional theory , couples as an axion field . that is to say it couples via the lagrangian density term of the form @xmath14 where @xmath15 . in the superstring theories ( type ii and heterotic strings ) there is a kalb - ramond anti - symmetric tensor field with two indices on the potential @xmath16 and three on its field @xcite @xmath17 here the three forms @xmath18 and @xmath19 are given by @xcite . in these theories there is a very sophisticated way of cancelling the gauge , gravitational and various mixed anomalies , first by having the right number of chiral fermions but in addition some terms , which are gauge non - invariant when alone , in the action for zero mass particles are used , @xmath20 \nn , \\ \label{eq : gsaction3 } & s_3 = & \phantom{- } c \int \left ( \frac{1}{8 } \ , b \ , \mathrm{tr}{}\ r^2 \ , \mathrm{tr}{}\ f^2 + \frac{1}{48}\ , \omega_{3 l}^0 \ , \omega_{3 y}^0 \ , \mathrm{tr}{}\ r^2 - \frac{1}{24 } \ , \omega_{3 y}^0 \ , \omega_{3 l}^0 \ , \mathrm{tr}{}\ f^2 \right . \nonumber \\ & & \phantom{s_3 = } - \frac{2}{3}\ , \omega_{3 l}^0\ , \omega_{7 y}^0 + \frac{1}{12}\ , \omega_{3 y}^0 \ , \omega_{7 l}^0 \left . \phantom{\mathrm{tr } { } r^5 f^2 } \hspace{-1.4 cm } \right ) \nn,\end{aligned}\ ] ] to cancel the remaining part of the anomalies . here @xmath21 is numerical constant @xcite . in order to get chiral fermions as is phenomenologically required to obtain the standard model it is needed to break the parity symmetries that only makes reflections in the compactifying dimensions ( for instance by having non - zero magnetic field in the extra dimensions ) . one may typically make use of calabi - yau spaces as the @xmath22-dimensional compactifying space . for pedagogical reasons , just to illustrate the idea we may in the present article think of a compactifying space being the cross product of three spheres , each of topology @xmath23 and each with a magnetic field on them corresponding to a `` magnetic monopole in the centre of the @xmath23 sphere '' . let us imagine that the equations of motions have led to that the vacuum has @xmath23 rotation invariant fields on the different @xmath23 s . then we may symbolically use these rotational invariant rotation symmetry is , we have in mind , that the topological @xmath23 is represented by a sphere which then has the symmetry under @xmath24 rotations ( about a point outside the sphere ) with respect to the fields assumed to be present . ] field strength @xmath25 @xmath26 . the first term in integrand of @xmath27 ( eq . [ eq : gsaction1 ] ) which by itself gauge breaking , will contain a contribution of the form @xmath28 imagine expending the variation over the @xmath23 sphere ( @xmath5 , @xmath29 and @xmath30 dependence ) on `` spherical harmonics '' or `` eigenfunctions '' . suppose we arranged one of spherical harmonic or eigenfunction to be dominant in the weakly exited state . we describe the effective four dimensional theory by means of the expansion coefficient @xmath12 to this term : @xmath31 really we could define such @xmath12 by integrating the two form @xmath32 over a homotopically non - trivial @xmath33-cycle . this would then require that we imposed other terms in the expansion of @xmath34 to be restricted to zero . taking the magnetic fields in the compactifying dimensions as constants we end up with an effective term in the four dimensional lagrangian density which up to the over all constant is of the form ( [ eq : axionlag ] ) . from the kinetic term for the kalb - ramond field , @xmath35 where @xmath36 is the gravitational coupling constant , @xmath37 the dilaton field , @xmath38 the yang - mills gauge coupling constant in the lagrangian density , we obtain a kinetic term for the coefficient field @xmath12 . due to the @xmath18 term in eq . ( [ eq : hgs ] ) it comes together with an abelian part of the yang - mills potential in an expression of the form @xmath39 this is a gauge invariant combination for the @xmath40-field gauge transform @xmath41 while @xmath42 where @xmath43 is the gauge function for an invariant @xmath44 subgroup of the left over symmetry group , not spontaneously broken . for simplicity we imagine that the presence of the extra dimension fields @xmath25 @xmath45 represents a break down to a subgroup which still contains at least one invariant abelian subgroup called @xmath4 . then we may concentrate on the gauge field associated with this subgroup @xmath4 and denote the gauge function for it as @xmath43 . we shall discuss a rather extraordinary behaviour of the theory from the four dimensional point of view with the axion field @xmath12 ( see also sec . [ sec : foureffec ] ) . suppose that @xmath12 does not quantum fluctuate so widely that it totally looses an expectation value . this means that there is a spontaneous break down of the gauge symmetry for @xmath4 . since even the constant @xmath43 gauge transformation is spontaneously broken due to the additive transformation property of @xmath40 , the spontaneous breaking situation is just like that of the higgs case . however , that means , one would expect that particles such as fermions carrying @xmath4-charge quanta would be able to make transitions into ( sets of ) particles with a different number of such charges ( together ) . at this point , however , one has often found the belief that the global symmetry and the noether conservation of the charged particles is _ not _ violated . in the perturbative approximation this belief is well - granded . we will discuss this question in the following in non - perturbative approximation . although at first it seems as if there is no way to cause the global @xmath4-charges on particles to be created or annihilated , it was shown in @xcite that such a violation of the charge was indeed occurring due to world sheet instantons . these `` world sheet instantons '' refer to the tunnelling of a string so as to have a `` time track '' during tunnelling which encloses in our simple scenario the @xmath23 involved with the @xmath34 . in the real general case we should have the tunnelling go around a @xmath33-cycle homotopical to the @xmath33-cycle(s ) used for extracting @xmath40 from @xmath32 . the important point for the present discussion is as follows : when such a world sheet instanton exists , there are zero modes for the fermions ( as well as bosons ) which have @xmath4-charge . these zero modes cause the @xmath4-charge to change . in this anomalous way similar to the qcd - instanton the global charge gets also violated after all . according to @xcite this is much more natural than not having the global @xmath4-charges broken . in reality the effect of this zero - mode effect is described by an effective lagrangian term @xmath46 here the @xmath47 are various @xmath4-charged fields and the product @xmath48 could , for instance , be @xmath49 where @xmath50 is a field for which @xmath4 has the role of a chiral mass protecting charge . the whole term is to be multiplied by the amplitude of the world sheet instanton . the factor @xmath51 comes from the exponentiated string action eq . ( [ eq : insttunn ] ) , which may cause the damping and the factors @xmath52 are needed to make the whole term gauge invariant . in supersymmetric theories , where these models are usually considered , such an axion field must occur together with a corresponding field in a complex combination . for instance the anomalous fayet - iliopoulos d - term was studied @xcite in the context of the heterotic string theory : they consider a dilaton chiral supermultiplet , @xmath53 , adding to the khler potential a part , @xmath54 , which transforms under the @xmath4 gauge group as @xmath55 the khler potential is with appropriate vector kinetic term @xmath56 the gauge coupling constant thus depends on the dilaton , @xmath57 . the theory also has an axion coupling proportional to @xmath58 . with the shift transformation of the axion field under @xmath4 gauge group ( see eq.([eq : u(1)xtra ] ) ) , this term serves to remove the anomaly proportional to @xmath59 . the d - term potential is using eq . ( [ eq : stringyk ] ) @xmath60 we should emphasise that in the weak coupling limit , where dilaton field goes to zero ( @xmath61 ) the fayet - iliopoulos d - term vanishes , @xmath5 , there is a supersymmetric minimum . if we contrary to the just given argumentation assumed that after all non - zero fayet - iliopoulos d - term were stabilised @xmath5 , there were at least the metastable minimum for non - zero @xmath62 then for the weakly - coupled heterotic string according to ref . @xcite with @xmath63 , there is a fayet - iliopoulos d - term given by @xmath64 where @xmath65 is the planck mass and @xmath66 is the sum of @xmath4-charges . thus the potential energy becomes @xmath67 so that @xmath68 is obviously negligible at very small dilaton field even if one wishes to have a model in which @xmath66 is non - zero , a typical value of @xmath69 to @xmath70 ( see @xmath2 @xcite ) . according to @xcite the d - term ( [ eq : anomalousfi ] ) may induce some ( non - zero ) expectation value of a scalar field called @xmath71 , and thus cause a spontaneous breakdown of the global ( part ) of the @xmath4 symmetry . supposing no other scaler field break this group , then the non - zero vacuum expectation value of @xmath71 may be made a connection with the fayet - iliopoulos d - term in following way : @xmath72 where the @xmath4-charge of the higgs field @xmath71 , @xmath73 , is taken to be @xmath74 . in this way one could obtain an expansion parameter for the fermion mass matrices from @xmath75 : in the case of @xmath75 being not too large ( see eq . ( [ eq : anomalousfi ] ) ) , in other words , @xmath76 is not too small and the sum of @xmath4-charges ( @xmath66 ) is of order @xmath77 to @xmath70 , we could use such a mechanism to produce a good order of magnitude for the breaking of the global charge conservation for @xmath4 . the @xmath4 symmetry would be broken one or two orders of magnitude below the fundamental scale ( planck or string scale , depending on model ) . this situation could be using the fundamental scale as the planck scale , @xmath78 @xmath5 , an expansion parameter for the fermion mass matrices which is identified with the cabibbo angle . if we have identified the string coupling constant , @xmath76 , with a dynamical field the dilaton field @xmath79 the ground state will be found by adjusting this coupling @xmath76 to be zero , which obviously means the disappearance of the global charge breaking effect . in this situation , which is the expected one , we have thus at first no complain from world sheet instantons about the possibility hoped for in literature of having the global charge totally conserved although derived from a higgs gauge symmetry . however , if the string coupling constant @xmath76 is really going to zero , then the models in which this happens have zero string coupling constant , and they become _ totally free _ , since in string theory all interactions are finally derived from the string interaction @xmath76 . unless one can somehow live with a tiny breaking of supersymmetry at very high energy scale allowing a small but non - zero @xmath76 , the models avoiding a fayet - iliopoulos d - term would become totally free as string theories ! the wish that the adjustment lets the fayet - iliopoulos d - term vanish by adjusting the @xmath79 to be zero is further supported by the often favored phenomenological call for supersymmetry not to be broken except at low energy scales , typically close to 1 tev , because it can be of help for the hierarchy problem . from the point of view of the high scale of energy , where we have the higgsing of the @xmath4 group , a phenomenologically useful supersymmetry breaking scale would be tremendously small and essentially count as zero . this argument further disfavours models which would have troubles due to the fayet - iliopoulos d - term . but if the problem is solved by adjusting to the supersymmetry which is only extremely weakly broken by having the string coupling almost zero , then the type of model is even more problematic , because it lacks the interaction altogether . as a resume of the above discussion let us compare the two logical possibilities concerning the size of the string coupling constant @xmath76 of being of order unity @xmath80 or sufficiently small ( @xmath81 ) both being considered in the presence of a non - trivial green - schwarz anomaly cancellation so that the fayet - iliopoulos d - term becomes non - zero unless @xmath82 : 1 . consistence 2 . the possibility @xmath80 is strictly speaking inconsistent because the fayet - iliopoulos d - term ( eq . ( [ eq : anomalousfi ] ) ) drives the @xmath76 , which is effectively dynamical ( related to @xmath62 ) , to zero so that only @xmath81 is consistent the expansion parameter @xmath83 suitable for small hierarchy ? 4 . a reasonable sized @xmath84 could give a good expansion parameter @xmath83 , however , if @xmath85 of course @xmath83 becomes exceedingly small and not useful for fitting the small hierarchy . supersymmetry for hierarchy problem phenomenology ? 6 . @xmath80 leads to extremely high scale supersymmetry breaking by the fayet - iliopoulos d - term such that the supersymmetry is of no use for solving the hierarchy problem . ( explain any weak scale supersymmetry phenomenology for that matter . ) while for a very small @xmath85 one may get supersymmetry accurate enough for hierarchy problem purposes . 7 . freeness of the whole string theory 8 . for @xmath85 the string theory becomes free while for @xmath80 there are interactions . conservation of the global charges of @xmath4 10 . there are two possibilities with @xmath76 in the relative large range which we should mention : 1 . @xmath76 is large : in this case the world sheet instantons and the fayet - iliopoulos d - term may be too large so that the breaking , which they cause , would also be too large ( @xmath5 the expansion parameter @xmath83 ) . in this case @xmath4-charge can not be used to the small hierarchy problem of the fermion masses . @xmath76 is not too large : in this case @xmath76 could give a good expansion parameter @xmath83 ( eq . ( [ eq : epsilon ] ) and see sec . [ sec : order ] ) . then one could hope for the desired breaking of global charge _ without _ invoking further breaking mechanisms . + in the case when @xmath76 is really small the global charge is very well conserved , but one can imagine it broken by other means so that it would be no problem higgsing 12 . the higgsing of the local part of the gauge group uses the kalb - ramond field and that works independent of the size of @xmath76 ( unless of course the whole theory becomes free that one can conclude that nothing happens at all ) . anomaly cancellation for @xmath4 from four dimensional point of view 1 . a strong @xmath76 : in this case the world sheet instanton breaking would be strong and there would essentially be not even a global charge approximately conserved . 2 . a bit weaker @xmath76 : there is an approximately conserved global charge which does not have the cancellation by triangle diagrams which is required for a gauge charge . the four dimensional model is derived from an although non - renormalisable as all theories in ten dimensions then at least gauge invariant theory . it is therefore surprising that it does not satisfy the usual conditions on numbers of fermion species and their charges needed for the anomaly cancellation . it is the reason why there is the wess - zumino term in eq . ( [ eq : hgs ] ) . however , how does that remove the need for anomaly cancellations ? one might wonder how the situation of the anomaly cancellation would be , if we had that the mass scale @xmath86 of the @xmath4-photon ( eq . ( [ eq : ymkin ] ) ) were very low compared to the scale of energy at which we consider the situation . from the point of view of such a high energy scale compared to @xmath86 , it would seem that the kinetic term for the axion field , @xmath40 , were very close to having zero coefficient , @xmath5 , an auxiliary field . the effect of integrating out @xmath40 functionally would be to produce a functional @xmath87-function with the effect of imposing the constraint @xmath88 with such a constraint imposed on the gauge field it would of course be no wonder if one gets no anomalies . in fact it would mean that the anomaly @xmath89 had been constraint to be zero . such a constraint will lead to interactions between photons which are of course in the next approximation in @xmath90 understandable as due to exchange of the @xmath40-particles , @xmath5 , @xmath91 . however , notice that diagrams , like fig . [ fig : photonb ] , have the @xmath40 field propagator which has the contribution of a factor @xmath92 . this propagator contribution is tremendous compared to @xmath93 since @xmath94 is in the range ( orders of magnitude ) above the @xmath4-photon mass scale , @xmath86 , @xmath5 , @xmath95-scatterings have extremely strong interactions . therefore , they may be able to provide the constraint forces which uphold the constraint eq . ( [ eq : ffdual ] ) ( or eq . ( [ eq : ffmatrix ] ) ) . photonb ( 40,25 ) [ fig : photonb ] we have an interacting @xmath4-photon theory with the lagrangian @xmath96 which in addition has the term ( [ eq : ymkin ] ) when we do not consider @xmath86 so small that we can ignore the term in eq . ( [ eq : ymkin ] ) . the equations of motion become , in addition to eq . ( [ eq : ffdual ] ) derived by varying @xmath40 , @xmath97 including charged matter and noticing that the no - matter terms in the euler - lagrangian equation can be written in a form more familiar , we obtain @xmath98 where @xmath99 is the `` matter current '' , and we may define the short hand notation @xmath100 for future discussions it is convenient when we express the equation of motion ( eq . ( [ eq : eomb ] ) ) with the electric field , @xmath101 , and magnetic field , @xmath102 , @xmath103 hereby we have applied @xmath104 . a mathematical point worth noticing in connection with the somewhat unusual electrodynamics , which we discuss here , is that the condition ( [ eq : ffdual ] ) can be shown by trivial algebra to imply that even @xmath105 for all combinations of the indices of @xmath106 and @xmath107 . above , we have reviewed an anomalous horizontal @xmath4 model using fayet - iliopoulos d - term to cause a spontaneously break down of @xmath4-charge . this effect would arise if the dilaton field @xmath62 were non - zero which is , however , not achievable due to supersymmetry being driven to be an exact symmetry . however , if we have a non - zero dilaton field , we also have world sheet instanton effects breaking the global charge conservation using green - schwarz anomaly cancellations . thus , this suggests to use this instanton effect in stead of a fayet - iliopoulos d - term . could this effect be adjusted for a phenomenologically reasonable global charge breaking of order of the cabbibo angle @xmath83 ? baring a totally mysterious cancellation of various contributions from different world sheet instantons the order of magnitude of the @xmath4 breaking by world sheet instantons is estimated as the exponent of the supersymmetry associated term to the field @xmath40 , which makes up a complex field . the possibility of surprising cancellations would have to be ignored @xcite if it were not for the findings that this indeed easily can occur @xcite . let us divide the discussion into the following possibilities : = 0.3 mm fndiagram ( 320,55 ) [ fig : fndag ] _ ( 1 ) _ : : all the quantities , including @xmath63 , are very strictly of order unity and the breaking of the charge conservation is also of order unity in spite of the fact that it is exponentially suppressed as an instanton tunnelling effect . in this philosophy the charge conservation is strongly broken : let us then imagine that the mass or the effective yukawa coupling for a quark or a charged lepton is obtained via a chain diagram ( fig . [ fig : fndag ] ) in which a series of fundamental scale vector coupled fermion propagators are linked by higgs fields or world sheet instanton caused transition symbols . if the strength of the charge violating world sheet instantons is of just the same order of magnitude as the ( typical ) fundamental scale fermion masses , then there will be no suppression , and the @xmath4-charge considered will be of no help in explaining the suppression of some effective yukawa couplings ( at experimental scales ) compared to others . however , taking `` everything '' especially the compactifying space dimensions to be very close to unity in `` fundamental '' units , such that even exponents are accurately of order unity is presumably not likely to be true . _ ( 2 ) _ : : the other possibility is that there are some quantities which can not be considered order unity in the very strong way discussed under point _ ( 1)_. the compactifying dimensions turn out to be the quantities of importance for the strength of the @xmath4-charge violation . let it be clear that there are really two scales of breaking of the @xmath4 symmetry to be discussed : + _ ( a ) _ ; ; there is the @xmath4-photon mass scale , @xmath5 , the mass @xmath86 of eq . ( [ eq : ymkin ] ) . _ ( b ) _ ; ; the mass scale @xmath108 that appear as the mass obtained for the fermions which are mass protected _ only _ by the @xmath4 and get their mass via the world sheet instanton caused term ( [ eq : instop ] ) in the case of @xmath109 . it is this mass scale @xmath108 which divided by the `` fundamental '' fermion masses @xmath110 gives the suppression factor @xmath111 which is used as a factorisation parameter for the fermions mass matrices . we discuss in the following the variation of the scales of breaking _ ( a ) _ and _ ( b ) _ above : the mass square factor in eq . ( [ eq : ymkin ] ) goes back to the term ( [ eq : kinkr ] ) in as far as the @xmath40 field is a coefficient on a term in @xmath32 which in turn has its derivatives go into @xmath112 in eq . ( [ eq : kinkr ] ) . it is remembered that the function multiplying @xmath40 to get the @xmath34 contribution must be normalised so that a shift in @xmath40 by @xmath113 , @xmath114 corresponds to adding to @xmath115 the shifting by a single monopole flux field through the cycle @xmath23 shifted by @xmath116 . if all the couplings are taken to be of order unity , one finds that scaling the dimensions of the @xmath33-cycle as the typical compactifying space dimension , @xmath117 , squared @xmath118 . the area of the @xmath33-cycle is proportional to @xmath118 . this means that that @xmath119 . thus we see that the mass scale of the @xmath4-photon goes as @xmath120 where @xmath117 is really the length scale of the two - cycle . the tunnelling suppression amplitude of the world sheet instanton @xcite is @xmath121 where @xmath122 is the regge slope , @xmath123 is the pfaffian , and @xmath124 and @xmath125 are kinetic operators for the bosonic and fermionic fluctuations , respectively . the `` @xmath126 '' on the pfaffian and determinant denotes that the zero modes are to be omitted . moreover , @xmath127 is the area , which of course @xmath128 again , with @xmath117 being the @xmath117 relevant for the two - cycle . introducing a fundamental mass scale , @xmath110 , we thus have the scale at which the @xmath4-charge conservation violates @xmath129 this crude estimate assumed @xmath76 to be of order one . now , we may go into crude phenomenology , still taking @xmath76 of order unity . in theories with compactified extra dimensions it is quite natural to take this as being the reason for the fine structure constants being weak , compared to the `` self - dual '' strength ( defined as the value of fine structure constant which makes it equal to the corresponding monopole coupling constant , associated by the dirac relation ) . the self - dual strength , @xmath130 , is approximately @xmath131 since @xmath132 for a formal monopole @xmath133 . from this point of view we can claim that a typical , say gut , coupling of order @xmath134 is weaker by a factor @xmath135 than the abelian self - dual . though this is exaggerating and should be corrected at least by the factor @xmath136 . roughly taking anything of this order , we would now expect that @xmath117 measured in `` fundamental units '' would be of the order @xmath137{12}\simeq1.5 $ ] ( see fig . [ fig : energylev ] ) . this would mean @xmath138 and @xmath139 , @xmath5 , @xmath140 , a very useful suppression factor indeed . this is namely a typical order of magnitude for an @xmath83 with which one fits the fermion mass spectra @xcite . energylev ( 60,30 ) [ fig : energylev ] in this way it looks very promising to get very naturally a good scale for violation strength for phenomenological fitting . we can say that , strictly speaking , a scale @xmath117 of the compactifying dimensions just @xmath141 times bigger than the fundamental length scale @xmath142 is so close to being of order unity , that there is hardly any call for any special explanation for this `` deviation '' from it being the fundamental size . that we can notice some small numbers in both the fine structure constants and in the suppressed violation of the @xmath4-charge is due to respectively the @xmath22 dimensional compactifying space and the exponentiation because of the world sheet instanton effect needed . the energy scale gap in which we have the funny electrodynamics with the constraint @xmath143 is in the just sketched scenario ( with @xmath144 ) reduced to a scale factor in the @xmath141 region . that is such a very small range that one would hardly be able to claim that anything strange will happen . independent of whether world sheet instantons do or do not break global charge , there is another mechanism threatening the global charge conservation : the effect of wormholes . we shall in fact show below in subsec . [ sec : wormholeii ] that our estimation of the effects of gravitational wormholes present in the vacuum ( or similar space - time foam effects ) _ will give rise to a significant violation of the global charge in the four dimensional theory with the non - trivial green - schwarz anomaly cancellation , although it has resulted from a gauge charge _ and thus at first might have been suspected of not being violated by space - time foam effects . over most of the energy scale ( logarithmical counted , see fig . [ fig : energylev ] ) between the fundamental scale and the @xmath4 violation scale @xmath108 , we have a globally conserved charge for @xmath4 while the gauge particle `` already '' had its mass at @xmath145 . now , however ( see subsec . [ sec : wormholeii ] ) , according to arguments in ref . @xcite charges which are not gauge protected may disappear into wormholes or baby universes . therefore , one may speculate whether such a charge conservation which is unprotected by gauge fields will not get its conservation spoiled by the wheeler space - time foam . actually such effects of breaking the global charge are expected at energy scales smaller than @xmath86 . but we will see in subsec . [ sec : wormholeii ] that wormholes at even higher energy scales than @xmath86 shall provide such breaking when the monopoles are used . the wormholes of low energy ( smaller than @xmath86 ) , @xmath5 , of large ( length ) are suppressed exponentially with some exponent related to @xmath146 . in fact we would ( naively ) estimate that for a space - time foam ingredient such as a baby universe of size @xmath147 ( in length ) we would have an action of order @xmath148 , and a suppression factor of @xmath149 . it happens under the assumption @xmath150 to be just the same order of magnitude in the exponent suppression factor as the one present in the world sheet instanton suppression factor . for this coincidence to occur it was quite crucial that the estimate just used for the baby universe action was of the form @xmath151 ( @xmath152 is baby universe tube ) as being obtained by use of einstein - hilbert action and not @xmath153 which would have been the case if the cosmological term in the action were significant here . that is to say , we used that the cosmological constant is zero , but at these scales one is closer to the ( running ) cosmological constant which is relevant for short distance and which may have another value than the long distance one which is practically zero . in any case even if the cosmological constant were relevant , it would suppress the baby universes at the scale @xmath86 even more . the space - time foam non - conservation is expected to be at most what the world sheet instanton effect would have been if the @xmath76 were of order unity , contrary to true expectation . now , however , since we truly do not expect the world sheet instantons to provide breaking then the wormhole effects could easily take over as the dominant effect . rather we should say that we do expect an appreciable wormhole breaking ( see the argument below ) . it could be that the world sheet breaking effect could actually dominate even in the case of @xmath154 , but the result would in this case not be so different with respect to the order of magnitude of the ( exponent of the ) effect , and thus the conclusion would be the same . however , in the case that there is no significant world sheet instanton effect , the wormholes can very well work and completely and dominantly break the charge conservation . in this subsection we will prove for that the @xmath4-charge is violated as a global charge due to the wormhole effects . in the foregoing subsection we saw that one could give immediate arguments both in favour of and against the global @xmath4-charge being violated by the wormhole or space - time foam effects . in this subsection we want to deliver an argument which shows that indeed _ the global @xmath4-charge is broken due to the wormholes_. we should , however , stress that this breaking is exponentially suppressed but _ only with the suppression corresponding to the scale of the higgsing @xmath86 _ , which is essentially the compactification scale . in the realistic models this is a rather high energy scale and the breaking from wormholes is thus expected to be quite large . compared to what one gets from world sheet instantons , which do not work after all in the case of @xmath76 being of order one , this could be much bigger than the latter . it is most easy to organise the non - conservation of the coulomb field for the @xmath4-charge by use of virtual wormhole entrances with magnetic fluxes radiating out . these form effective virtual magnetic monopoles in the vacuum . the reason that it is profitable with monopoles violating the charge conservation for the @xmath4-charge is that we indeed can derive some formulas for the variation / development of the @xmath4-electric - charge relating it to the variation of @xmath40 on the sites of the monopoles ( see eq . ( [ eq : diverad ] ) ) , @xmath155 since @xmath156 is the charge density this formula tells us that , for instance , when a monopole is present , the variation rate of the charge @xmath157 contains a term @xmath158 on the site of a monopole whenever @xmath40 varies , @xmath5 , @xmath159 . the monopoles to be used here do not have to be genuine monopoles . they could be entrances to wormholes with magnetic flux going through @xcite . we would not even have to use such genuinely existing monopolic wormhole entrances . it is rather sufficient to consider entrances which are virtually present in the vacuum . we shall imagine that there are many such wormholes virtually present with magnetic flux and that the entrances give rise to interactions with the various fields in the theory . we assume that interactions can be described by effective terms in the lagrangian density . at first they are only at the places where the entrances to the wormholes are . we shall , however , integrate over all the possible positions or movements for the wormholes as part of the functional integration of feynman path integral . this has the implication that one can achieve naturally that such a model of wormholes can become effectively translational invariant . in fact one has to integrate over the positions of the wormhole or baby universe entrances with a translational invariant measure . that can actually be supported by a heisenberg - inequality type argument using that at least baby universes can not transport energy and momentum , because the information about these quantities is safely stored in the gravitational field at long distances from a ( supposed to be ) little baby universe . even for wormholes it is reasonable to integrate over all positions with a translational invariant measure . since it is clearly possible that any sort of particle could be scattered into a wormhole , the effective lagrangian density contribution from the entrance to a wormhole can contain terms annihilating or creating any combination of particles . thus any combination / product of fields is _ a priori _ possible and will come with some coefficient in the effective wormhole and baby universe lagrangian . now , however , there can be processes that can not really take place due to coulomb fields left behind . by this we mean that if we propose terms violating gauge symmetries for charges with light gauge particles associated there remains information outside . in fact there will be a coulomb field left , carrying the information about the charge of the particles which have gone into the wormhole . even if the particle goes deeply into the wormhole and maybe even out somewhere else far away , there will remain electric flux lines exiting from the entrance of the wormhole and even if no appropriate may be different particle is pulled out the entrance the wormhole itself will behave as a charged particle . in this way we can only have effective lagrangian density terms conserving the gauged charges corresponding to gauge particles with compton wave lengths which are long compared to the wormhole sizes . really what matters is , whether the gauge field around the wormhole can keep the information about what went into it . in the case of the conserved @xmath4-charge the @xmath160 , that should have ensured the stability of the coulomb field , does not correspond to a conserved current as in the usual electrodynamics . it is rather the current corresponding to the @xmath161 ( eq . ( [ eq : shandred ] ) ) that is conserved . in fact we have just seen that if the axion field @xmath40 varies on the sites of the virtual monopoles we can / will have that @xmath160 and thus the charge varies . in this way it should now be _ allowed _ to have lagrangian density terms due to the monopolic wormholes violating the @xmath4 . once such terms are allowed they are expected to be there and we will generate masses for particles which are only mass - protected by @xmath4 group . once we have the symmetry strongly broken at the planck scale which is now expected there will no longer be a sign of the conservation and thus also no problem with the anomalies . the symmetry seems to be dynamically broken not only spontaneously because the effective lagrangians representing the wormhole and other space - time foam effects really have to be interpreted as dynamical breaking . we must also expect that it is rather impossible to keep the @xmath4-photon mass @xmath86 to be light compared to @xmath108 under such conditions . we conclude that seriously taking wormholes into account in this way , it results that the green - schwarz anomaly cancellation scheme _ does not work _ in @xmath0 dimensional limit . what happens is that the very strong constraint ensuring forces due to very large @xmath40 propagators lead to the possibility of coulomb fields around a wormhole entrance modified with time . this modification possibility in turn allows the effective lagrangian density corresponding to the absorption of the @xmath4-charges into wormholes . anomaly cancellation by the green - schwarz mechanism in the case of a certain @xmath0 dimension limit of a higher dimensioned string theory is questioned : we consider the gauge symmetry ( needed to the @xmath4 subgroup ) that allegedly results from breaking a larger string theory gauge group using a field @xmath40 derived from kalb - ramond @xmath162 that takes on a non - vanishing vacuum expectation value and thereby higgses the gauge field @xmath163 . this is manifested phenomenologically as an approximately conserved current without having the usual triangle summation anomaly requirement for avoiding gauge and mixed anomalies . this is referred to as green - schwarz anomaly cancellation because the special way of having anomaly cancellations for string theory mass states is inherited to the @xmath0 dimensional limit . if we have supersymmetry and such a charge with green - schwarz anomaly cancellation ( effectively in the @xmath0 dimensions ) , then according to the calculations of ref . @xcite we have a fayet - iliopoulos d - term ( [ eq : anomalousfi ] ) which drives the dilaton field that is essentially in correspondence with the effective string coupling constant @xmath63 to be zero . that means that the theory becomes _ free _ and thus rather unrealistic . this is a severe trouble in itself for the models with a non - trivial anomaly cancellation mechanism . however , contrary to this argument , if we should assume that in some mysterious way it were possible to get a string theory after all with supersymmetry surviving and having a significant d - term and string coupling in spite of such green - schwarz cancellation , then the world sheet instantons would prevent the breaking strength of the surviving global charge conservation from going to zero . it is a major point of the present article that it is actually not achievable for the world sheet instantons do violate the @xmath4-charge conservation . in this way we could avoid the mystery of having a gauge theory breaking from a spontaneous symmetry breaking , which does not break the current conservation . provided that the world sheet instanton effects do not mysteriously cancel ( which is though less safe to assume than naively expected ) , this mystery would disappear . the authors working with the green - schwarz anomaly cancelling charges ( usually ) have in mind supersymmetric models . if the world sheet instanton effects would appear even in supersymmetric theories , the green - schwarz anomaly cancellation would become less suspicious of being pretended to behave strangely from the general point of view . however , due to the fayet - iliopoulos d - term expected when we have such charges , the string coupling constant @xmath76 may be driven to zero and the whole effect of world sheet instantons would disappear . this may though be not realistic because of bringing the whole string theory appears to be free . if this kind of effect could work causing a breaking of the global charge conservation and thus avoiding some of the strangeness one may still seek ( using the world sheet instantons ) to declare a @xmath4-charge needing the green - schwarz anomaly cancellation to have become less suspicious of being pretended to behave strangely from general point of view . however , one may still declare the way to be suspicious , in which the anomaly troubles are avoided in the energy range above the mass scale @xmath86 of the gauge particle : the gauge fields are constrained to never have the configuration leading to anomalies ! if one gets a breaking of charge @xmath4 due to the wormholes ( or world sheet instantons ) , one could imagine as an interesting possibility to _ use this breaking _ instead of some spontaneous breaking induced from @xmath2 the fayet - iliopoulos d - term ( as @xcite proposes ) to provide the soft breaking which is needed to use the charge @xmath4 as a mass protecting charge to implement the large ratios of quark and lepton masses . we further expressed our worry and suspicion whether such an electrodynamics which is constrained by @xmath164 can really be considered as realistic on general physics grounds or whether it represents an unrealisable speculation concerning the relative orders of magnitude . however , we thought there are reasons to believe that the wormholes would give violation of the @xmath4 even above its higgsing scale . that could mean that wormholes break the @xmath4 symmetry completely under the use of @xmath164 . the point is that a very strong coupling of the @xmath40 field causes that coulomb field around the wormholes endings are not stable . 99 m. b. green and j. h. schwarz , phys . b * 149 * ( 1984 ) 117 . e. witten , phys . b * 149 * ( 1984 ) 351 . j. e. kim , phys . b * 207 * ( 1988 ) 434 . m. dine , n. seiberg and e. witten , nucl . b * 289 * ( 1987 ) 589 . j. j. atick , l. j. dixon and a. sen , nucl . b * 292 * ( 1987 ) 109 . m. dine , i. ichinose and n. seiberg , nucl . b * 293 * ( 1987 ) 253 . l. e. ibez , phys . b * 303 * ( 1993 ) 55 [ arxiv : hep - ph/9205234 ] . l. e. ibez and g. g. ross , phys . b * 332 * ( 1994 ) 100 [ arxiv : hep - ph/9403338 ] . p. bintruy and p. ramond , phys . b * 350 * ( 1995 ) 49 [ arxiv : hep - ph/9412385 ] . e. dudas , s. pokorski and c. a. savoy , phys . b * 356 * ( 1995 ) 45 [ arxiv : hep - ph/9504292 ] . y. nir , phys . b * 354 * ( 1995 ) 107 [ arxiv : hep - ph/9504312 ] . p. bintruy , s. lavignac and p. ramond , nucl . b * 477 * ( 1996 ) 353 [ arxiv : hep - ph/9601243 ] . e. j. chun and a. lukas , phys . b * 387 * ( 1996 ) 99 [ arxiv : hep - ph/9605377 ] . p. bintruy , s. lavignac , s. t. petcov and p. ramond , nucl . phys . b * 496 * ( 1997 ) 3 [ arxiv : hep - ph/9610481 ] . k. choi , e. j. chun and h. d. kim , phys . lett . b * 394 * ( 1997 ) 89 [ arxiv : hep - ph/9611293 ] . j. r. ellis , s. lola and g. g. ross , nucl . b * 526 * ( 1998 ) 115 [ arxiv : hep - ph/9803308 ] . j. k. elwood , n. irges and p. ramond , phys . rev . * 81 * ( 1998 ) 5064 [ arxiv : hep - ph/9807228 ] . k. choi , k. hwang and e. j. chun , phys . rev . d * 60 * ( 1999 ) 031301 [ arxiv : hep - ph/9811363 ] . m. bando and t. kugo , prog . * 101 * ( 1999 ) 1313 [ arxiv : hep - ph/9902204 ] . q. shafi and z. tavartkiladze , phys . b * 473 * ( 2000 ) 272 [ arxiv : hep - ph/9911264 ] . a. s. joshipura , r. d. vaidya and s. k. vempati , phys . d * 62 * ( 2000 ) 093020 [ arxiv : hep - ph/0006138 ] . j. m. mira , e. nardi , d. a. restrepo and j. w. valle , phys . b * 492 * ( 2000 ) 81 [ arxiv : hep - ph/0007266 ] . m. kakizaki and m. yamaguchi , jhep * 0206 * ( 2002 ) 032 [ arxiv : hep - ph/0203192 ] . w. buchmller and t. yanagida , phys . b * 445 * ( 1999 ) 399 [ arxiv : hep - ph/9810308 ] . m. tanimoto , phys . b * 501 * ( 2001 ) 231 [ arxiv : hep - ph/0010088 ] . j. sato and k. tobe , phys . rev . d * 63 * ( 2001 ) 116010 [ arxiv : hep - ph/0012333 ] . n. maekawa , prog . phys . * 107 * ( 2002 ) 597 [ arxiv : hep - ph/0111205 ] . h. k. dreiner and m. thormeier , arxiv : hep - ph/0305270 . k. s. babu , t. enkhbat and i. gogoladze , nucl . b * 678 * ( 2004 ) 233 [ arxiv : hep - ph/0308093 ] . c. d. froggatt and h. b. nielsen , nucl . b * 147 * ( 1979 ) 277 . g. gilbert , nucl . b * 328 * ( 1989 ) 159 . b. zumino , y. s. wu and a. zee , nucl . b * 239 * ( 1984 ) 477 . b. zumino , _ lectures given at les houches summer school on theoretical physics , les houches , france , aug 8 - sep 2 , 1983_. l. alvarez - gaum and e. witten , nucl . b * 234 * ( 1984 ) 269 . m. dine , n. seiberg , x. g. wen and e. witten , nucl . b * 278 * ( 1986 ) 769 . m. dine , n. seiberg , x. g. wen and e. witten , nucl . b * 289 * ( 1987 ) 319 . c. beasley and e. witten , jhep * 0310 * ( 2003 ) 065 [ arxiv : hep - th/0304115 ] . e. witten , jhep * 0002 * ( 2000 ) 030 [ arxiv : hep - th/9907041 ] . c. w. misner and j. a. wheeler , annals phys . * 2 * ( 1957 ) 525 .